Place the same for-the-badge invite under the Trendshift badge in each
translated README. The image label stays English; alt text follows the
language of each file. The href is https://discord.com/invite/GkcryMnJDM.
Place a for-the-badge Discord link in the English README hero, under
the Trendshift badge and above the website links. The button uses
https://discord.com/invite/GkcryMnJDM.
Co-authored-by: morluto <[email protected]>
* fix(browser): keep negative script columns Chrome reports
Chrome reported a Debugger.scriptParsed event with startColumn -60 and
endColumn -15 for an inline script in WordPress's helper frame. CDP declares
these columns as plain integers, but REA's admission required min(0), so one
script failed the whole observe-web-execution run with unreadable_output.
CDP resource columns now share one representation, cdpResourceColumnSchema,
for script start/end columns, initiator call frames, parser initiators and
event listeners. Positions on a script's first line include its start column,
so keeping the reported values keeps the documented script-relative
conversion consistent. The domain contract and web runtime docs state that a
negative column does not locate a column in the resource.
Fixes#8674
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(browser): apply the CDP column policy to passive capture
inspect_web_page and discover_webmcp_tools still read CDP columns through
cdpSourceCoordinates, whose own >= 0 check turned a negative column into
null and reported invalid_protocol_value. WebMCP discovery also dropped the
tool's whole registration_source.
cdpResourceColumnSchema now lives with the domain's CDP observation schemas.
Runtime admission, the passive coordinate helper and the runtime, console,
initiator and WebMCP output schemas all use it. Lines keep their
non-negative rule.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(reference): Exclude Git pointer files from source identities
Match .git pointer files in the default import policy after the existing
directory rules, preserving ordinary-checkout exclusion provenance and Git
HEAD discovery. Cover linked-worktree identities, repeated imports, nested
metadata, similarly named source files, and configured-secret precedence.
Independent review: parent independent review APPROVE for Draft CI on patch
61bddf108952c77858fa71c333b4bb8a05ab33e8b03f5c6bccd879cd19c7c379 and source
tree 2d2372a9ec. Review triggers: delegated
shipping output, source inventory/file I/O behavior, and more than 50 lines.
Validation: npm run check passed with Node 24.22.1/npm 11.16.0 on Windows;
tracked Vitest regression 1 passed/3 Windows skips; test discovery passed.
Eight offline Linux production imports passed on cached Node 22.19.0 as
supporting evidence only. Pinned Linux/macOS Vitest cases await hosted CI.
Fixes#7551
* test(reference): Isolate worktree fixtures from host Git settings
Route every fixture Git command through one wrapper with a filtered child
environment, owned empty global config/template/hook paths, and no system
config reads. Keep resource cleanup under the existing temporary owner.
These settings are scoped to synthetic child processes; host Git settings
and this repository's commit/push checks are unchanged.
Targeted independent review: parent review APPROVE on patch
4be9a2af3e40edac3d64f687170e966a87fcf762f1877f2c55016e120bcd90ac and tree
a2ba91ee9c. The reviewer confirmed the native
proof helper hash, complete fixture command coverage, and owned cleanup.
Review triggers: delegated follow-up and fixture file/process boundaries.
Validation: npm run check:fast passed; targeted Vitest 1 passed/3 native
Windows skips. On pinned Node 24.22.1, the actual helper rejects inherited
hooks/templates while commit, worktree creation, HEAD and status checks pass.
The unisolated control copies the injected template and its commit fails.
Refs #7551
Preserve the twenty-entry diagnostic prefix and report the full count,
omitted count and truncation fact at the owning report projection. Document
the recovery limitation and cover the public HAR-to-comparison workflow.
Fixes#7621
Validation: 121 source tests; three selected CLI regressions (five other
cases unselected); nine README conformance cases; full --noEmit typecheck;
documentation facts/metadata/schema/licenses; module boundaries and
architecture guards. VitePress rendering is blocked by Windows EPERM on its
worktree-local Vue symlink. Full aggregate coverage remains for Draft CI.
Independent code review: APPROVE, relayed by parent task
01a17716-6299-75fe-87e9-b0398b8c9019. Trigger: >=50 changed lines / >=3 files.
Reviewed tree: ba964efcd6
Reviewed full-index patch SHA256:
972dcfc219dd7b9606175b1848068f92210fa3f7d0a8a244c1ac7ce296ae32d9
* fix(flutter): read unlabeled Dart Version::String() from engines
Release libflutter.so embeds Dart's Version::String() without the
standalone VM's "Dart VM version:" label, so identify_flutter_build left
dart_version null on real APKs. Accept that channel-bearing form beside
the labeled scanner path.
* fix(flutter): qualify unlabeled version candidates
---------
Co-authored-by: morluto <[email protected]>
Chrome DevTools HAR exports declare creator WebInspector 537.36 and
report IPv6 serverIPAddress values in brackets: its HAR writer strips
the port with lastIndexOf(':') but its bracket removal,
replace(/\[\]/g, ''), only matches the literal pair '[]'. The pinned
schema then rejects the whole archive at the first such entry.
Extend the existing Playwright 1.58.2 adaptation to this declared
producer. Brackets are removed only in the disposable validation copy
and only around a valid IPv6 literal; reported spellings are retained
and the per-entry limitation names the declared producer.
Application analysis preserves the complete literal request endpoint,
including the empty string, and the event seed already accepts a
retained empty name. The endpoint seed still required a non-empty
value, so trace-javascript-semantics refused to select a retained ""
endpoint with invalid_request.
The endpoint seed now accepts any string and keeps exact matching.
Boundary-field seeds keep their minimum because boundary nodes are only
produced from method names.
Found while fixing #6458; the event half of that report landed in
cf5e4f6fa.
Co-authored-by: Claude Fable 5.1 <[email protected]>
* feat(binary-layout): admit ELF64 AArch64 and name the refused host
inspect_binary_layout refused every AArch64 ELF, and on a non-x64 host the
refusal read "supports Linux x64 only", which looked like a target refusal
(#8709 ran on arm64 WSL and suspected an e_machine check).
The pwntools adapter now admits little-endian ELF64 AArch64 EXEC/DYN/REL
objects alongside the existing profiles, pwntools must report aarch64/64/
little, and the Evidence subject names arm64. The host refusal now names the
detected platform/arch and says the target was not read; the host check
itself is unchanged.
verify:binary:layout cross-compiles an AArch64 executable and relocatable
object (never executed) and checks sections, symbols and relocations
against GNU readelf, decoding ELF64 r_info as a 32-bit type under a 32-bit
symbol index. The real-binary-layout workflow installs the toolchain.
Refs #8709
* test(binary-layout): refuse RISC-V as the unsupported machine and assert the host refusal path
verify:binary:layout used e_machine 183 (AArch64) as its unsupported
target, which this branch now admits; use EM_RISCV (243) instead.
The unsupported-host provider test now asserts the refusal names this
host's platform/arch and says the target was not read, so reverting the
decoder to the old message fails it.
* fix(native): stop calling an off-screen window absent
The native UI helper looked up its selected window with
`.optionIncludingWindow`, which returned nothing while that window was off
screen. An existing window owned by the selected process was reported as
`window-mismatch`, before any permission check, so the verifier could not reach
its permission boundary. Select from all windows instead; absent and foreign
windows still fail.
The verifier's changed-bytes check also read `error.message`, but since
cc90771af a selection failure carries its helper code in `error.issues`, so the
check could not pass. Read the issue. Together these let the permissions lane
complete on main.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(native): own UI test processes from the moment they launch
The off-screen test registered cleanup only after parsing the ready record, so a
startup failure left `/usr/bin/open` and its app running. The lane test killed
only the verifier on a timeout, leaving its launcher and detached fixture alive
until they expired.
Register the off-screen owner immediately after the launch; the app inherits the
run token, so the sweep needs no ready record. In the lane test, every stand-in
records itself and leaves on one release the test's cleanup awaits.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(debugger): let REA exit when a debugger could not be stopped
`rea debug-with-gdb` and `rea debug-with-rizin` printed cleanup_incomplete
and set exit status 1, but the retained child and its pipes kept the event
loop alive, so the process never exited. A stdio MCP server whose client
disconnected hung the same way.
closeOwnedDebuggerSessions is the owner both managers finish through. It now
unreferences retained sessions, as supervised providers already do after
incomplete cleanup, restores the references on retry, and names the retained
process IDs. The one-shot commands report those IDs, since no later command
can retry. Debugger ownership, signals and detach behavior are unchanged.
GDB cleanup requests also no longer hold the host until their 120 s command
deadline, because cleanup already bounds its own wait.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(debugger): name retained debuggers when an MCP server shuts down
An MCP server that exits with a retained debugger logged only "REA provider
cleanup failed" and "could not close cleanly", so nothing named the process
left for the caller to check. The PIDs were on OwnedDebuggerCleanupError, but
each shutdown reporter read only the outer aggregate's message, and the debug
failure_cause description dropped the field too.
The failure type now lives in its own dependency-free module with one reader
that walks composed failures. The shutdown report, the server's onclose log,
the debug description and the one-shot CLI all use it, and the shutdown report
writes the same remediation the CLI returns.
The fake debuggers now model each request exactly, with real reply shapes, and
a request they do not model fails the test.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(debugger): compare retained-debugger guidance with its formatter
The new checks froze the wording "Check process ... and its target".
docs/testing.md asks guidance checks to compare with the owning formatter,
alongside independent checks of the facts. They now do, and each still proves
the process ID is named, because a formatter comparison alone would pass if the
formatter dropped it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Static checkpoints were admitted against the same budget as semantic and
application results, half of the parent's free heap. Once those charges used
it, the next checkpoint stopped the analysis and every remaining source lost
its static facts, with most of the parent heap still free.
Semantic and application results still stop at that budget. Static checkpoints
may now continue to three quarters of the free heap while the live heap also
has room for another copy of what it holds. Without that room, admission falls
back to the result budget. The live condition is needed because the limit V8
reports counts young-generation capacity that retained objects cannot use:
without it, the parent aborted on heap exhaustion at small heap limits.
Reserved relationships are given up only when the larger share runs short. A
stop still fails with a typed error, which now reports the limit that applied,
what was available, the live heap in use and what the checkpoint needed.
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(doctor): report provider settings a client registration does not record
Provider checks read the terminal running doctor, so a registration with no
env block looked ready while its server reported not_configured. Compare what
an available provider relied on with the registration's own file and report the
setting names in a non-blocking registration_environment check. The client's
effective environment stays unknown, so health and the exit status are unchanged.
Fixes#8202
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(doctor): write the Cordis fixture at its literal path
The test resolved the path through supportedClients, whose default environment
argument CodeQL traces to the fixture's writes even though the test passes {}.
It only needed the path, so spell it like the other fixtures in this file and
drop a leftover directory it never used.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* feat(evidence): discover missing Evidence links without transferring records
An agent could only learn that a retained record's source was absent by reading
complete records or by having a downstream analysis refuse. The summary said
nothing about links, and complete-record retention was easy to mistake for
source closure.
- Add detail: links to get_evidence_bundle and CLI inspect-evidence-bundle:
paged link rows with retained or missing status and no record payloads, for
selected records against the whole session or file.
- Add retained-closure traversal, the walk composed workflows require, reporting
complete or incomplete. Direct reads make no closure claim.
- Count direct and missing direct links in the summary.
Fixes#8454
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(evidence): count a repeated Evidence link once in every delivery
The Evidence schema does not deduplicate evidence_links, so the summary counted
a repeated link twice while link rows and counts reported it once. Both now read
one helper for a record's distinct direct links.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(evidence): validate link delivery against the advertised schemas
The test never called tools/list, so neither the shared validator nor the MCP
client checked its replies against what each profile advertises. A schema that
rejected the links result could have passed. Fetch the advertised definition and
validate every request and reply against it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* feat(native): select retained batch decompile results through analysis views
A completed batch_decompile could only be read whole. When it did not fit the
MCP response budget, or an agent wanted one function, inspect_analysis_view
refused the parent as unsupported and the only recovery was exporting the
session.
- Own the batch result schema in the domain, replacing the contract's copy and
the workflow's hand-written item types.
- Add summary and batch-decompile views: body-free rows keyed by original index
with exact filters and opt-in bodies, and UTF-16 pages of one body.
- Report the decompiling provider from the upstream analysis profile, since the
Evidence provider is the composing workflow.
- Keep batch limitations visibly aggregated rather than attributing them to one
successful item.
Fixes#8498
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(native): deliver batch decompile as a retained summary
A batch that fits still floods the caller's context with every body, and one
that does not fit only named an export. Selecting a body that cannot fit named
an export too, even though its text pages can.
- Accept detail: summary on batch_decompile: retain the complete batch first and
return its counts, identities and sizes. The default stays complete, and the
selection never reaches the Evidence parameters.
- Offer a ready-to-call body-free first page when a complete batch overflows.
- When a selected batch page overflows, continue from the same retained parent
with fewer rows, then with one body's text pages.
Refs #8498
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(native): count every unavailable entry label in a batch summary
label_unavailable only counted failed lookups, so a batch whose provider
answered null for every label reported zero. The identity contract defines a
null name as an unavailable label in both cases.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(native): continue an oversized batch view from the caller's own source
A batch view read from an Evidence file or inline record advertised a
retained-evidence continuation for a parent the session never retained, so
following it failed. A view's continuation now repeats the caller's source and
only narrows the selection. An inline record offers none, because it cannot be
repeated within the reply it overflowed. Non-batch views skip the selection
parse, since delivery asks about every view it returns.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(native): require advertised batch schemas through the shared validator
A local validator map dropped any tool without an output schema, so removing an
advertised schema would have left these contract tests green. Require both named
schemas and validate through createMcpSchemaValidation, as docs/testing.md asks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(native): offer file view continuations on servers that retain nothing
A fixed-target server reads an Evidence file through inspect_analysis_view but
records nothing, so its overflow skipped the continuation and returned generic
CLI guidance. A view's continuation names the caller's source and needs no
session, so it no longer depends on recording. Retained references still do.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(cli): keep captured source Evidence when one-shot delivery never commits
A completed observation that was cancelled at final close, failed cleanup, or
never reached bundle commit returned its evidence links but dropped the source
records the invocation had already captured. The caller could neither read those
records nor tell whether they were missing, captured or written.
Report the closure at publishCliAnalysisEvidence, the shared boundary every
--evidence_output command uses. details.source_closure carries authentic captured
records within 64 KiB and explicit omissions: not_captured for a link this
invocation holds no record for, and not_inlined for a captured record outside this
response. A committed bundle keeps its receipt instead, nothing is written after
cancellation, and successful delivery is unchanged.
Fixes#8589
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(cli): keep the typed failure when collecting source closure cannot finish
A captured source with a data-sized evidence_links array made
pending.push(...links) exceed V8's call-argument limit. On final-close
cancellation or a refused publication that RangeError replaced the original typed
failure, which is the one outcome this delivery step exists to preserve.
Walk the closure through a Set, which is a deduplicating breadth-first queue, so
no link becomes a call argument and each link costs one lookup however often it
repeats. undeliveredSourceClosure now owns authentication and never throws: if
examining the closure fails, it reports collection_failure with the bounded error
type and an unknown closure state, and the original failure is still returned.
Fault the final-close tests on the scope the command gave to matching, after
matching completed, and assert that close was reached, so a moved close fails
instead of passing on an earlier one.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Preserve the dynamic symbol's visibility attribute (default or protected) when extracting ELF dynamic exports, and expose it across complete Evidence and projected analysis views.
Historical Evidence records lacking visibility normalize safely to 'unknown', and projected views omitting visibility do not inject default values into partial row projections.
Fixes#8517
* fix(javascript): emit no event operation for a literal non-function listener
`EventEmitter` validates a listener before registering or removing anything:
if (typeof listener !== 'function')
throw new ERR_INVALID_ARG_TYPE('listener', 'function', listener);
The throw is synchronous, before any listener exists, but the analysis reported a
`register` or `remove` operation for those calls.
Measured on Node 24.22.1:
emitter.on(1) -> THREW ERR_INVALID_ARG_TYPE
emitter.on("x", 1) -> THREW ERR_INVALID_ARG_TYPE
emitter.on("x", null) -> THREW ERR_INVALID_ARG_TYPE
emitter.on("x", true) -> THREW ERR_INVALID_ARG_TYPE
emitter.on("x", {}) / ([]) -> THREW ERR_INVALID_ARG_TYPE
emitter.once / addListener / prependListener / prependOnceListener /
off / removeListener with a literal -> THREW ERR_INVALID_ARG_TYPE
--- accepted ---
emitter.on("x", () => {}) -> ok
emitter.on("x", function () {}) -> ok
`rejectsEventListener` drops the operation only for an exactly readable literal
that cannot be a function. An identifier, a call result or a spread may still
supply a function and keeps its operation, and a class expression passes Node's
own `typeof` check.
`emit` and `removeAllListeners` are untouched: `emit` passes its remaining
arguments as data to listeners that already exist, and `removeAllListeners` takes
only an optional event name.
* fix(javascript): price a non-finite listener and keep a shifted spread
Two review findings on the event-listener guard.
**A non-finite numeric literal was missed.** Babel stores `1e999` as a
`NumericLiteral` holding `Infinity`, and the exact-value reader refuses to price a
non-finite number, so the guard never saw it:
e.on("x", 1e999) -> THREW ERR_INVALID_ARG_TYPE
A numeric literal is always a number and therefore never callable, so the guard now
decides on the literal node directly.
**A leading spread shifts the listener position.** `e.on(...args, 1)` does not pass
`1` as the listener: Node spreads the array and ignores the trailing argument once
the listener is satisfied. The guard read `arguments[1]` regardless, so it dropped a
real registration. It now leaves the position undecided when argument 0 is a spread.
Five regressions cover both forms; all five fail without the fixes.
* fix(process): re-read unreadable tokens before signaling and name the process
Darwin KERN_PROCARGS2 samples a process's argument length and stack top
before it reads the address space, and exec publishes the new length
before it copies the new strings out (xnu sysctl_procargsx,
exec_copyout_strings). A read during exec can return a zero-filled or
stale strings area, and exec keeps the PID and start time that bracket
the read. The final sweep already re-read such tokens, but the
pre-signal group check read each member's token once, so one transient
read of a live group member failed the whole cleanup.
Move the bounded re-read into one shared fact-reading helper and use it
for both the sweep and the pre-signal group check. The group check binds
re-reads to the generation it first observed, so a reused PID stays
unresolved; the sweep keeps classifying the PID's current process.
An unreadable-token failure now carries the native identity read after
the token and the summary names the candidate by its native name,
parent, group and start identity. The Darwin helper and Linux procfs
reader report the short native process name (p_comm/comm) for display.
Fixes#8337
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(process): refresh other pre-signal members after token re-read waits
The pre-signal group check re-read only members whose token was
unavailable, so a readable member kept the relation it had before the
retry waits. A member that moved to another process group during those
waits could be validated on its old group, and the old group signaled
while the moved member kept running.
After any retry wait, the shared re-read helper now reads every other
live pre-signal fact once more with the same generation binding. A moved
member then takes the existing changed-group path, which retains its
ownership evidence and fails cleanup. An inconclusive refresh replaces
the stale fact so it fails closed. Nothing changes when no re-read ran,
and the sweep keeps classifying current processes because every group
it finds is revalidated before signaling.
Also cover PID reuse that happens before the re-read starts, where both
re-read identities agree and only the first-read generation rejects the
replacement's token.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(process): keep known identity changes final across token re-reads
The pre-signal re-read treated a first read without a same-generation
identity as unbound and replaced it with the re-read. On Linux the first
read can already establish PID reuse (an ESRCH token whose exit check sees
another generation); the re-read then classified the replacement, so a
zombie replacement could let cleanup succeed.
A pre-signal check now re-reads only same-generation facts. Changed,
exited, exiting and foreign-token facts are final. Each attempt also
re-reads the group's owned members instead of one separate refresh after
the loop, so a member that moves groups or turns unreadable while another
waits is seen within the same bound. The caller's run token selects the
owned members, and the sweep keeps classifying current processes.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(setup): refuse Hopper on macOS below the release's minimum
Hopper setup planning accepted any macOS 12+ host as supported, and the
installer never compared the host with the selected release. The current
publisher-signed Hopper 6.6.1-demo requires macOS 12.4 (feed
deployment_target and the app's LSMinimumSystemVersion), so setup on
12.0-12.3 installed an app the host cannot run.
The installer now takes the minimum from the release that owns it: the
feed's deployment_target before the package download, and the signed
app's LSMinimumSystemVersion (decoded in-process after signature
verification) before copying. A host below either fails with the new
incompatible_host_version code and an installer detail naming both
versions; an incomparable app value is a bundle failure. Planning keeps
REA's own macOS 12 floor but reports support as unknown, since the
release minimum is read only after approval, and the plan discloses the
check. Info.plist decoding is shared with app-bundle resolution.
Fixes#8502
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(setup): compare the macOS Hopper plan detail with its owning disclosure
The planning test froze an explanatory sentence from the setup plan. The
macOS install detail now lives in macHopperInstallDisclosure beside the
commands and origins the installer owns, the plan uses it, and the test
compares the plan with that constant while keeping its independent
support-state assertions. An installer test also pins that the signed
app's requirement is read only after both signature checks pass.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(setup): compare macOS Hopper refusal details with their owners
The installer's refusal details were template literals inside
installMacHopper, so tests copied whole sentences. The installer now
formats them through exported macHopperHostVersionDetail and
macHopperIncomparableMinimumDetail. Tests compare each diagnostic and the
setup remediation with those owners and SetupInstallFailure's recovery,
and separately assert the URL, versions, setting name, failure code and
request order. The Info.plist type refusal is checked by error type, key
and path instead of a sentence fragment.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
The handshake check ran only inside startSession, which a query reached
only when the profile committed a startup input, for PE or for an existing
project. Any other session started lazily inside callTool, so analysis seeds,
disabled analyzers, extension results, MIPS soft-float storage or a RISC-V
calling convention that the profile did not commit were never checked and
the query returned its result.
Every query now runs the checked start first. The client shares one startup
attempt, so there is still one launch, and a provider restarted within the
session is checked again. That leaves startSession as the only start in the
provider client, so the PE data-type path reuses its session info.
Recovery guidance is unchanged: the former eager-start condition now selects
only whether a completed handshake counts as a running provider.
Fixes#8612
Co-authored-by: Claude Opus 5.5 <[email protected]>
* feat(native): select retained batch decompile results through analysis views
A completed batch_decompile could only be read whole. When it did not fit the
MCP response budget, or an agent wanted one function, inspect_analysis_view
refused the parent as unsupported and the only recovery was exporting the
session.
- Own the batch result schema in the domain, replacing the contract's copy and
the workflow's hand-written item types.
- Add summary and batch-decompile views: body-free rows keyed by original index
with exact filters and opt-in bodies, and UTF-16 pages of one body.
- Report the decompiling provider from the upstream analysis profile, since the
Evidence provider is the composing workflow.
- Keep batch limitations visibly aggregated rather than attributing them to one
successful item.
Fixes#8498
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(native): count every unavailable entry label in a batch summary
label_unavailable only counted failed lookups, so a batch whose provider
answered null for every label reported zero. The identity contract defines a
null name as an unavailable label in both cases.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(native): require advertised batch schemas through the shared validator
A local validator map dropped any tool without an output schema, so removing an
advertised schema would have left these contract tests green. Require both named
schemas and validate through createMcpSchemaValidation, as docs/testing.md asks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
After a validated seed, trace_native_values turned a cleanup-incomplete
failure from a call-target read or a followed callee dossier into a
plain-text unknown, kept dispatching reads, and returned successful
Evidence without the cleanup_incomplete status, cleanup reason or owned
resources.
The trace now classifies child failures with the shared stopsTraversal
predicate that call graph traversals use. A cancelled child still ends
the trace with its cancellation. A child whose provider cannot confirm
cleanup stops the trace: the provider's own error is the outcome, so its
resources and bound cleanup retry stay intact, and the work admitted
before the stop is returned as the trace's normal shape, with traversal
truncated by cleanup_incomplete and a located unknown at the stopping
read. That stop stays primary over a concurrent caller abort and over
the output-size check. Ordinary child failures remain unknowns.
MCP already retains such partials as workflow Evidence. The one-shot CLI
dropped them, so runDirectAnalysis now retains the partial workflow
Evidence on the failure through the same shared owner.
Fixes#8371
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(cli): describe one-shot timeouts for an invocation whose session has ended
A one-shot CLI analysis that reached a provider deadline told its caller
that the Ghidra session remained usable and reported a runtime-editable
limit applying to the next operation, although the command had already
released its provider session and runtime.
The shared error projection now takes an explicit caller lifetime.
Providers keep guidance that holds only for a continuing caller (session
reuse, same-session runtime updates) apart from the request outcome, and
the presentation adds it only for a continuing caller. One-shot owners
(runDirectAnalysis, runProviderAnalysis, finishAnalysisCommand and the
JavaScript application command) project for the invocation: details.limit
reports scope rea-invocation, applies_at next_invocation and the launch
environment seed, and the remediation leads with that setting for a new
invocation. MCP output keeps its runtime-update and same-session guidance.
Covers the Ghidra decompile timeout, decompiler failure, regex stack
exhaustion and startup timeout, the shared regex deadline, the Kuna
decompile timeout and the Hopper startup timeout.
Fixes#8311
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(cli): project composed child failures for the caller lifetime
Cleanup composition embedded its primary and cleanup failures as
continuing-caller projections, and baked the primary's message and the
cleanup's remediation into the composed error. A one-shot command whose
Ghidra startup timed out while shutdown also failed therefore still
reported a rea-runtime limit and update_analysis_settings in
details.diagnostics.primary_error.
childFailureDiagnostics now snapshots each embedded child at composition
for both lifetimes when its projection depends on the caller, so cleanup
history keeps its composition-time state and the final projection selects
the caller's form. composedCallerGuidance re-attributes message and
remediation with their continuing-only parts kept separate. Error copies
and redaction keep the invocation diagnostics. The Hopper startup timeout
now carries its continuing guidance on the error like the other producers.
Invocation limits keep agent_editable and rea_restart_required, with
values that hold for a new invocation, so only scope and applies_at
change and environment is added.
In existing-project mode a decompile timeout or a cancelled session now
says the next session copies and opens REA_GHIDRA_PROJECT_PATH again
rather than repeating import and analysis.
Tests compare delivered guidance with the owning formatters and constants
instead of freezing prose.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(discovery): carry provider repair instructions into capability discovery
Provider readiness reported only a failure description, so capability
discovery copied that description into tool_availability remediation and
dropped repairs the owners already had: the JEB discovery-deadline setting,
the ADB/Apktool execute-permission repair kept in diagnostics, and the IDA
and Kuna configuration recovery.
Unavailable ProviderAvailability now requires the owner's remediation
next to its reason, and readiness-derived unavailable capability
descriptors carry it too. Every producer supplies it (ADB, Apktool, JADX,
JEB, Ghidra, Hopper, IDA, Kuna, pcap, owned-process prerequisites, profile
rejection); the inventory uses it verbatim, falling back to the reason
only for static capability facts that declare no repair. ADB/Apktool
configuration failures own remediation as a field instead of diagnostics,
and execution refusals project it as remediation.action. JADX doctor reads
the availability's remediation instead of re-deriving it.
binary_session complete candidate availability and unavailable capability
groups expose remediation (CLI capabilities/providers too); summary omits
it and lists it in omitted_fields. The superseded IDA
configuration_remediation and windows_native_failure.remediation
diagnostics are removed.
Fixes#8506
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(discovery): keep owner repairs through provider selection and capability maps
Address review findings on the readiness-remediation contract:
- Provider selection rejections carry the availability owner's
remediation, CLI/MCP error details project it, and an explicit
selection refused by readiness uses it as remediation.action. IDA's
unreadable registration and Ghidra's Windows readiness repair no
longer degrade to generic selection advice now that the diagnostics
copies are gone.
- Ghidra's Windows readiness and existing-project capability maps carry
the same repair as the availability.
- pcap readiness keeps the limiter owner's repair (now a typed
remediationAction) and falls back to Wireshark guidance only when the
failure has none.
- ADB/Apktool repairs are owner constants that are complete sentences.
The execution refusal keeps its previous message-plus-repair recovery,
which the compiled CLI contract asserts.
- Regression tests compare with the owning formatter or constant and
add public MCP and CLI coverage for an explicit IDA selection.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* feat(evidence): read selected complete records without the whole bundle
get_evidence_bundle could discover a record by exact ID, but reading it
required the complete canonical bundle: complete + filters was rejected,
and inspect_analysis_view refuses parents without a view, such as
annotations, pseudocode sources and native traces.
- Add detail: records, a page (offset, limit; default 0, 8) of the
complete matching records in canonical bundle order. Records keep their
original IDs, subjects, providers and readback; missing_evidence_links
names direct links of the returned records that are not retained.
- The selection is labelled as historical and not an importable bundle;
complete keeps returning the canonical bundle, and offset/limit
require detail: records.
- Filters keep one owner shared by summary and records delivery, for the
MCP session and the CLI's inspect-evidence-bundle.
- An oversized record page now advises a smaller limit or one
evidence_id; whole-bundle overflow guidance mentions record reads.
Fixes#8307
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(evidence): compare record-selection guidance with its owners
Assert the record-selection limitations and oversized-page guidance
against their owning constants, and check the reported limit facts
separately, instead of pinning explanatory phrases.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(setup): report the failed Hopper request URL and HTTP status
Hopper download callers collapsed every transport failure into a bare
`ok: false`, so HTTP 404 and 503 from the macOS feed and DMG, the Linux
publisher feed, the public key and the launch-time package download produced
identical generic advice without the failed URL.
The shared downloader now returns a typed failure (url plus http_status,
timeout, network, response_too_large or url_refused) and keeps caller
cancellation as a rejection. Installers attribute it to the request
(release_feed, package or public_key); HopperReleaseError carries it, and
setup projects it once into `download` and a remediation naming the URL,
outcome and next step. Replaces the Linux-only "Package download returned
HTTP" special case.
Fixes#8503
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(setup): refuse credentialed Hopper URLs and keep known HTTP outcomes
Native Fetch rejects a URL with userinfo using a TypeError that echoes the
whole URL, so a credentialed redirect reached `download.detail`, setup
remediation and publisher messages as a network failure. The downloader now
refuses request and redirect URLs carrying credentials as
`url_refused: credentials` before requesting them, and reports the caller URL
through the shared userinfo-removing `sanitizeBrowserUrl`.
Releasing an unadmitted body no longer replaces the outcome it was released
for: an errored (truncated) body after HTTP 503, or a failed cancel at the
size limit, keeps `http_status` or `response_too_large`.
Tests compare delivered remediation with the owning formatters and assert
request, URL, status and recovery class independently.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* feat(javascript): page retained feature traces through analysis views
A bidirectional trace_application_feature result can exceed one MCP
response, and inspect_analysis_view refused trace parents as unsupported,
so the only recovery was exporting the whole session.
- Add a shared feature-trace view owner: a summary with the trace's
identity, producer counts and coverage, and pages of paths, nodes,
edges, seed_matches and native_handoffs in retained order with exact
filters. Node rows are compact unless observations are requested.
Page coverage never upgrades trace coverage.
- Accept an optional MCP view on trace_application_feature that retains
the complete trace before returning the selected facts; the CLI keeps
complete JSON and reads saved traces with inspect-analysis-view.
- Offer a ready-to-call first paths page when a retained complete trace
exceeds the MCP response budget.
- Share the distinct authority/state/confidence basis helper with
module references.
Fixes#8296
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(javascript): name the retained parent when a selected view overflows
Review follow-up for the feature-trace views.
- Share one delivery owner between trace_application_feature and
compare_source_to_bundle: stage the selected view against the
completed parent, keep the established contract that a cancelled
request commits no session facts, then commit sources and the parent.
- A selected view that exceeds the response budget now references the
retained parent and offers the same selection at half its limit,
instead of naming the view record, which cannot be a view parent.
- Cover two native handoffs paged one at a time, the oversized selected
page, and cancellation after the trace completes.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(javascript): bind feature-trace pages to their collection rows
Advertise each feature-trace page with exactly its own row schema, so a
paths page cannot validate with node, edge, seed-match or handoff rows.
The native-trace view in #8533 received the same review finding; the two
sibling views now share one strict page contract.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(javascript): mix valid rows from another collection into a trace page
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(native): find procedures by the secondary names exact lookup accepts
search_procedures matched only each procedure's primary name, while exact
lookup also accepts Kuna inventory aliases (versioned-symbol and generated
sub_ names) and Ghidra symbols at an exact function entry. A search for a
name procedure_address resolves returned nothing, and trace_feature
silently missed the procedure.
Both providers now match the same names their exact lookup accepts. A
procedure still appears once under its primary name; the optional
matched_aliases field lists the secondary names that matched, so aliases
never become procedures of their own and list_procedures is unchanged.
The shared domain helper owns that row rule for Kuna and for the Ghidra
host-side Unicode folding pass. The Ghidra analysis profile records the
new search semantics so older snapshots cannot replay alias-blind results.
Fixes#8330
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ghidra): match leaf spellings of procedure names in search
procedure_address accepts each entry symbol's leaf or namespace-qualified
spelling, but procedure search tested only qualified names, so a regex
such as ^audit_alias$ missed a secondary symbol ns::audit_alias (and ^same$
missed a primary alpha::same) that exact lookup selects.
Lookup and search now share one bridge definition of the names that select
a procedure, so they cannot drift. Matching secondary symbols are still
reported by their qualified names; because a leaf ends its qualified
spelling, the host's Unicode literal pass stays exact over those names.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(session): clear provider-inferred unavailability after recovery
A provider adapter or process failure marked every routed operation of
that provider unavailable, but a later success cleared only the operation
that succeeded. Sibling operations, and composed tools built on them, kept
advertising the stale failure until each was retried individually.
Move runtime availability into one owner that records whether each entry
was observed for the operation or inferred from another operation's
provider-wide failure. A success clears its own failure and that provider's
inferred entries while keeping failures observed on other operations or
providers and declared refusals. Outcomes are ordered by dispatch to the
provider client, so an older in-flight success no longer erases a newer
failure while a request queued behind a failure still clears it.
Fixes#8404
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(session): keep newer successes when older failures complete last
Runtime availability kept one merged entry per operation and forgot
successful dispatches. An older provider-wide failure that completed after
a newer success re-inferred unavailability for siblings the newer success
had already proved working, and a superseded refusal could survive its own
later success.
Record only the newest failure and newest success dispatch per operation
and per provider, and project unavailability from them: an operation is
unavailable while its own newest failure is newer than its own newest
success, or, as inferred, while its provider's newest provider-wide failure
is newer than any success from that provider. The projection no longer
depends on reply completion order.
The recovery tests now bound held-reply acknowledgements, release and join
every held call in finally, and compare remediation with the owning
formatters instead of copying their wording.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(enhanced): stop call graph traversals at unconfirmed provider cleanup
A relationship read whose provider could not confirm cleanup of resources it
owns was projected into a graph node or call-path failure, the traversal kept
dispatching queued reads, and get_call_graph / trace_call_path returned
success with the cleanup_incomplete status buried inside the result.
Both traversals now read relationships through one shared owner,
readCallRelationships, which classifies a cancelled read or a read with
unconfirmed owned cleanup as terminal. On such a cleanup stop the workflow
returns the provider's own error as the outcome, so its cleanup owner, retry
binding and live resolution stay intact, and keeps the work admitted before
the stop as the workflow's normal result shape with a caveat naming the
stopped procedure and the unexpanded count. The composed-workflow layer
turns that admitted result into partial workflow Evidence on the failure,
and the MCP adapter retains it through the shared partial-error delivery.
Ordinary read failures still produce successful partial results.
Fixes#8369
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(enhanced): retain every carried partial Evidence on enhanced failures
Enhanced-tool failures used the shared partial-error delivery only when this
PR's workflow partial was attached. A failing read that already carried its
own partial Evidence, and analyze_function dossier failures, were projected
inline and never recorded with the session writer, so readback and bundle
export could not recover them. Every enhanced execution failure now goes
through one delivery owner that uses toPartialErrorToolResult with the
session writer; it still falls back to inline when no writer or no valid
Evidence exists.
Tests compare the traversal caveat against its owning formatter instead of
copying the sentence.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Scheduled deep static checks on main fail knip because the
DarwinHelperCache type alias added in #8318 has no consumer. Remove
the alias; callers already use createDarwinHelperCache's inferred
return type.
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix: retain address context cleanup failures
* fix!: preserve address context input errors
BREAKING CHANGE: Strict clients that cached the previous full inspect_address_context output schema must refresh tools/list and accept the typed error field on unavailable facets.
* test: use session contract for address context
Reject native view sources whose predicate disagrees with the selected operation. Use the original predicate for existing Evidence; do not alter a record without recomputing its identity.
Resume refused every identity mismatch with one generic message and
generic input-schema recovery. Report the differing target, profile,
layout, or provider facts with the journal and current values so callers
can choose the matching provider/profile or a new output directory.
Fixes#8349
Bind typed service creation, observed availability and retryable cleanup in one composition owner. Reuse selected CLI/MCP factories and canonical contract groups without copying provider operation inventories.
Closes#8429. Refs #740.
Validated paired real application workflows, actual local Codex CLI Terra/high use, complete Evidence and Unknown checkpoint parity, compiler-backed import boundaries, SDK schemas and final real mixed routing. Automated review findings were fixed at their producing owners. Existing exact-base CodeQL alerts are documented on the PR under the owner-authorized unrelated-check exception; no global alerts or security rules were changed.
Assisted-by: Suika (gensokyo-skills:suika-scatter-gather-planning)
Separate engine cleanup, investigation reset and terminal disposal in the shared application owner. Keep stable Evidence IDs and full Unknown revision history through engine close and target changes; serialize reset with all record writers and cleanup owners, optionally acknowledging a complete checkpoint before clearing. Bound retained representation without eviction.
Validation: 43 focused files / 533 distinct tests, source/type/lint/schema/documentation checks, nine genuine source cases across eight families, real Ghidra session identity and cleanup, and nine actual local Codex CLI terra/high runs with independent full-result and checkpoint audits. See docs/verification/investigation-lifecycle.md for measurements, partial attempts and verification limits.
Fixes#8347.
BREAKING CHANGE: close_binary retains investigation records. Use reset_investigation to clear them. Cross-process recovery requires an explicitly acknowledged Evidence bundle checkpoint.
Assisted-by: Suika (gensokyo-skills:suika-scatter-gather-planning)
The DOS verifier compared a fresh CLI session with an MCP session that
its address-boundary helper had renamed and restored, so the entry's
symbol source differed (imported vs user_defined). Compare the CLI with
the pre-edit MCP window, then assert the restoration changes only the
symbol provenance.
Fixes#8189
Co-authored-by: Claude Opus 5.5 <[email protected]>
* feat(native): label binary overview counts with their inventory scope
`binary_overview` described its counts as "exhaustive", and its result was
a pair of bare scalars. An agent could read them as a discovery of every
function and string in the file, or as comparable across providers. They
are exact totals of the rows each provider's inventories return.
The shared `projectBinaryOverview` now adds one provider-neutral
`inventory_coverage` record, so Ghidra's direct summary and the
listing fallback carry the same record:
`{ scope: "provider_inventory", discovery_coverage: "unknown" }`.
The record reuses the `discovery_coverage` vocabulary that
`list_procedures({detail:"summary"})` already reports, now owned by one
domain schema. The tool description states the scope and that counts are
not comparable across providers. Counts, segment omission semantics and
limitations are unchanged.
Fixes#8287
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(native): label overviews in the workflow so older snapshots stay valid
Analysis snapshots replay the projected inventory summary. A snapshot saved
before inventory_coverage existed would then fail the overview output schema.
The record describes what overview counts mean, whoever produced them, so
the binary_overview workflow now adds it on both its provider-summary and
listing paths. Provider values and projectBinaryOverview are unchanged, and
the composed overview's linked atomic record stays as the provider
reported it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(native): refresh composed overviews from before inventory_coverage
Bump the binary_overview workflow revision to 5, so snapshot and CLI replay
recompute a revision-4 composed overview from its atomic record. Replay
would otherwise return the cached overview without its count labels. The
revision-4 case joins the existing refresh scenario. The atomic Evidence
is not rewritten.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(kuna): omit Mach-O data symbols from the procedure inventory
Kuna's `functions --json` registers every Mach-O export as a function
entry, so globals in `__DATA,__data`, constants in `__TEXT,__const` and
`__mh_execute_header` were listed, counted and decompiled as procedures.
The Kuna session now classifies each inventory entry with the target's own
section table, read by the shared `inspect_macho` load-command reader from
the header bytes kept while its private copy is hashed. Entries in sections
that declare no instructions, or in the header and load commands, are data:
listings and searches omit them and report per-region counts, and
procedure operations refuse them by name or address while `address_name`
and `xrefs` still accept them. Instruction sections, symbol stubs and
symbol-pointer import slots are kept; addresses outside every section stay
with an explicit limitation, an unreadable section table keeps the whole
inventory unclassified, and ELF inventories are unchanged. Composed
workflows such as `decompile_all` and `binary_overview` follow the listing.
Fixes#8168
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(kuna): keep an entry's classification caveat on direct answers
Resolution now returns the selected entry's Mach-O placement and the
caveat it leaves open, built by the same constructor as the listings: the
one-entry unplaced wording, or the unreadable section table's reason.
Pseudocode, assembly, analyze_function (including the dossier's own
limitations), procedure_address, call queries by name and address_name
carry it; code and import-slot entries carry none.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(mcp): name the close_binary retry for retained cleanup
A close_binary that retired its target but could not confirm cleanup
returned cleanup_incomplete with retryable: false and only the generic
"Review the reported local resources" remediation, although the session
keeps every failed owner and the next close_binary on the same connection
retries them. Agents could abandon the connection that owns the only retry.
The MCP close tool now passes its retry operation to the session. The
session binds that typed AnalysisCleanupRetry to a failed retirement only
while it retains the cleanup for its next close; the close tool binds it to
process-capture failures its own scope retains and to compositions whose
session part was offered. Projection then reports
recovery.kind=retry_cleanup and appends a conditional instruction: keep the
connection, let the condition settle or resolve it, call close_binary
without snapshot_path and check the result. The hint disappears once that
cleanup resolves. retryable keeps describing the failed operation.
One-shot CLI closes, open_binary, analysis tools and failed snapshot saves
(which keep the target open) offer no retry. Cleanup composition copies the
owner's remediation without a retry offer, so the text appears once.
Fixes#8211
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test: derive close retry guidance from its owning formatters
The new close_binary retry assertions froze ProviderCleanupError's
explanatory sentence, so a harmless wording edit would break them. The MCP
test now compares delivered remediation with analysisErrorRemediationAction
applied to the delivered typed error plus cleanupRetryRemediation. The
domain tests use the same formatters, check that unbound and released errors
carry no retry text, and check that a bound composition adds the offer once.
Recovery fields, cleanup state and retry behavior keep separate assertions.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* perf(process): reuse a verified Darwin ownership helper across processes
Every REA process on macOS compiled the Swift ownership helper into a fresh
temporary root and then paid macOS's first-launch assessment of that new
executable: 408 ms + 325 ms of a 1.4 s `rea inspect-macho /bin/ls`.
With the default xcrun compiler and module cache, the reader now keeps one
content-addressed copy under ~/Library/Caches/rea/process-helper/<key>/.
The key covers the helper sources, toolchain selection, architecture and
Darwin release. A copy is executed only after the cache root, entry and
files are verified as current-user owned, not group- or world-writable,
not links, and digest-matched to the entry manifest. Malformed entries
owned by the user are removed and rebuilt; linked or foreign entries are
left untouched and the per-process compile runs as before. Any cache
failure falls back to that compile, so the cache never fails an ownership
operation.
A miss compiles exactly as before, then publishes a copy through an atomic
rename, keeping the first concurrent publisher, and runs the shared copy.
That way its first-launch assessment is paid once. Per-run roots keep their
existing cleanup; published entries are never removed by close or exit.
An explicit REA_DARWIN_SWIFTC_PATH or REA_DARWIN_MODULE_CACHE_PATH keeps the
per-process compile and writes nothing to the cache. `rea uninstall
--purge-data` removes the namespace.
Fixes#2483
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(process): harden the shared Darwin helper cache
Address review findings on the shared ownership-helper cache:
- Verify every real parent of the cache root: each must belong to the user
or root and not be writable by others (sticky root-owned directories
excepted), so no other user can rename a verified entry. Execute through
the real path.
- Open manifest and helper through the owned non-blocking regular-file
reader, so a FIFO cannot block preparation, and retain handles whose close
fails.
- Key entries on the swiftc and SDK that xcrun itself resolves, including
their real paths, sizes and modification times. DEVELOPER_DIR, TOOLCHAINS,
xcode-select and in-place updates therefore select new entries.
- Never remove or replace an invalid or published entry. Removal could race
another process that has verified or is running that path, so such an
entry falls back to the per-process compile.
- Give the cache a close() that retries retained staging removals and
handle closes and reports ProviderCleanupError while any remain. The
reader closes it with its per-run root.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(process): keep cache lookups owned and fall back from unlaunchable helpers
Address the second round of review on the shared ownership-helper cache:
- Pass the preparation signal to both xcrun toolchain lookups, and wait for
every started lookup before reporting the first failure. Cancellation and
close then stop a stalled lookup instead of waiting out its deadline.
A failed or cancelled key computation is retried by the next lookup.
- When the verified shared copy cannot launch (EACCES, EPERM, ENOEXEC,
ENOENT or an unrequested SIGKILL, for example from a noexec home volume),
the reader stops using the cache. It reuses its own compiled helper or
compiles one, and retries the read once.
- Anchor the cache's returned cleanup errors as cleanupRoot does, so every
error returned while cleanup was pending resolves after a retry succeeds.
- Build the cache path with appendCallerPath, so a home with symlinks and
`..` keeps the caller's meaning instead of being normalized lexically.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(process): keep concurrent fallbacks and dual cleanup failures
Each helper read now records before awaiting whether it launched the shared
copy, so a concurrent read's fallback cannot hide its own. When the per-run
root and the cache's retained resources both fail one close, the returned
error composes both. It stays incomplete until both owners recover.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(process): refuse shared helpers that an ACL lets others write
macOS ACLs can grant writes that the POSIX mode bits do not show. On
Darwin, lookup and publication list the real ancestors, the entry and its
files with `ls -lde`, and refuse sharing when any ACL entry allows a
write-class right. The default `group:everyone deny delete` entries on
macOS home folders are unaffected. The unlaunchable-helper test now removes
the execute bit from a verified copy, which refuses launch on both Linux
and macOS.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* feat(guidance): distinguish focused matches from broad parent sections
Guidance search treated a parent section, including every descendant, as one
match. Terms scattered across unrelated subsections made a 49K-character root
the only result for "switch provider annotations", and every ancestor of a
focused hit was listed alongside it.
The section parser now records each section's own text (heading and body
before its first subsection) and its parent. Search keeps the same admission
rule but classifies each match by match_scope: "section" when every term is
in the section's own text, "subsections" when the terms are only spread
across subsections. Ancestors whose terms all sit inside one subsection are
no longer listed, section matches rank first, and subsections matches carry
own_text_terms plus bounded greedy pointers to the subsections holding the
remaining terms. Every match reports content_length so callers can judge a
read before making it. Read still returns the complete section.
Own text is tokenized once per request instead of each expanded subtree.
Fixes#8290
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(guidance): require the scattered-match checks to run
Count the subsections matches the real-bundle test inspects and fail when
none were checked, and document that a parent whose own text holds every
term stays listed beside its matching subsection.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
swiftc rejects assembly sources as inputs, so verify:ghidra:swift failed
during fixture preparation before any provider check. Assemble both
assembly fixtures to objects with xcrun clang and link only objects.
Fixes#8188
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(decompile): keep provider no-pseudocode facts on batch items
A batch_decompile item whose provider answered without pseudocode was
reported as unreadable output with doctor guidance, and the provider's
reason, such as a Kuna export failure for that entry, survived only in the
batch-wide limitations. The workflow read only the observation's result.
Read each item's complete observation. A null answer is now a
capability_unavailable item naming the provider, and every item answered
without usable text keeps that observation's own limitations, so a
batch-only Evidence export retains the association. Empty or non-text
pseudocode stays unreadable output; typed provider failures and successful
items are unchanged, and the batch Evidence still aggregates every caveat.
The batch workflow revision changes so earlier snapshots are not replayed
with the previous item semantics.
Fixes#8226
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(decompile): match null-pseudocode guidance to reported caveats
A null answer without caveats told the caller to read item limitations
that the batch omits when the list is empty. The guidance now depends on
whether that observation reported caveats, and says the provider gave no
further explanation when it did not.
The batch tests now select their contracts by name from the enhanced and
session families instead of the full catalog.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
trace_native_values set one truncated flag for two different facts: the
returned page omits retained nodes, and a budget stopped or narrowed the
traversal. A last page of a complete trace and a node- or edge-capped trace
both reported truncated: true with next_offset null and no unknown naming the
cap, so a caller could not tell whether to follow pages, raise a budget, or
treat the examined graph as exhausted.
Results now carry page_truncated for page omission and a traversal object
that is complete, or truncated with the observed reasons in schema order:
max_depth, max_functions, max_call_sites, max_nodes, max_edges,
node_payload_bytes, provider_pcode_truncated and ambiguous_call_candidates.
truncated keeps its previous meaning as the union of both.
A new budget owner admits nodes and edges and records every cause together
with a located unknown, only after a budget rejects work; an exactly filled
budget is no longer truncation. Depth and ambiguous-candidate limits are
resolved after traversal, so a destination analyzed or queued through a
shallower path is no longer reported as depth-limited. Once an edge is
rejected, remaining call sites in that procedure are not resolved.
Fixes#8283
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(ghidra): report unbacked file-offset addresses without schema repair
address_to_file_offset projected a well-formed address in uninitialized
or unmapped Ghidra memory as generic invalid input: the headline said the
arguments were invalid and recovery pointed at the input schema, while the
real reason survived only in a nested issue.
The bridge now gives each outcome its own code (address_unmapped,
address_uninitialized, file_mapping_absent, file_mapping_ambiguous,
file_offset_unrepresentable) instead of folding them into invalid_request.
The Ghidra projection keeps the bridge reason as the headline, points to
inspect_native_load_image, records the caller's exact address and the
file_mapping outcome, and scopes diagnostics to the request.
AnalysisInputError gains conformsToInputContract so a conforming request
omits recovery.kind=inspect_input_contract. Malformed addresses and other
bridge refusals keep their existing projection.
Fixes#8244
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* test(ghidra): match fixture block ends to bridge conventions
The fake bridge treated .text's exclusive list_segments end (0x1100) as an
inclusive bound, so address_to_file_offset mapped an address outside the
block. Give the fixture's blocks one owner (start and size) and derive each
reply from the production bridge convention: exclusive list_segments ends,
inclusive byte-search coverage, half-open containment for file offsets.
Add an MCP boundary test that reads the block edges from list_segments and
search_bytes, maps the last file-backed byte and refuses the first address
past each block with its outcome code.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(ghidra): decide batch names independently of request order
The Ghidra bridge applied set_addresses_names in JSON member order, each
address in its own transaction, so a reordered but equivalent map could give
a shared name, or a name released by another selected address, to a
different address.
The bridge now resolves every destination (namespace and leaf name) against
the names held before the batch. Addresses that request the same destination,
or a name that another selected address holds, are rejected without applying;
the remaining edits are independent and apply as before. Both naming
operations report each failure's canonical address, reason and related
addresses. The adapter projects the bridge reply to the unchanged boolean
results and keeps the reply as raw_result only when an address failed.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(ghidra): report each contested batch name once
A batch conflict listed every other requester on each failure, so k
addresses competing for one name produced k(k-1) related addresses, and
the adapter retained the whole reply, booleans included, as raw_result.
The bridge now reports each contested destination once in a conflicts
table with its namespace, name, requesting and holding addresses, and
each conflict failure refers to it by index. Conflicts are ordered by
their first requester, so indexes do not depend on member order. Single
and batch naming share one planner and reply layout. The adapter
validates the table in linear time and keeps only failures and conflicts
in raw_result.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Declare logical Evidence sources once and resolve exact retained inputs through
shared application workflows across 20 composition tools. Preserve domain-owned
authentication, source closure, caller choices, coverage and resource admission.
Validation includes public CLI/MCP regressions, local Codex terra/high trials
and fresh-import real-artifact workflows. Measurements, failed attempts and
coverage limits are recorded in docs/verification/evidence-input-composition-2026-10-31.md.
Fixes#8293.
Fixes#8281.
Assisted-by: Suika (gensokyo-skills:suika-scatter-gather-planning)
* feat(artifacts): inventory and extract UF2 and DfuSe regions
open_binary refused UF2 and DfuSe firmware containers as an unsupported
format, so inspect_artifact and extract_artifact could not show which
bytes a drag-and-drop or DFU update writes at which addresses.
UF2 is recognized by a complete first block with all three magic
numbers. Valid main-flash payloads are joined into address-contiguous
regions per familyID, named family-0x<id>/0x<start>-0x<end>.bin, and
known IDs are named from a small table copied from Microsoft's
uf2families.json at a pinned commit. Numbering is counted per familyID
and block count: absent, repeated, out-of-order and out-of-range block
numbers are reported, and absent blocks leave coverage unknown.
Not-main-flash blocks are excluded and reported; file-container blocks
are reported and leave coverage unknown; invalid blocks and a partial
final block are excluded with their position. A rewrite of one address
with different bytes fails closed, naming both block spans.
DfuSe is recognized only by its prefix together with a DFU 1.1a suffix.
Each element becomes one region at its file range with its target's
alternate setting and name. Structural faults fail closed with a byte
offset; overlapping elements stay separate and are reported. A suffix
CRC mismatch follows integrity_policy: fail refuses with a dfu-crc32
integrity error, and record-and-continue records a root contradiction
that extraction also carries.
HEX, UF2 and DfuSe share one retained-region reader and one run
merger, so the HEX decoder and reader lose their private copies.
* docs(artifacts): document UF2 and DfuSe images and their esptool lane
Describe recognition, region names, metadata, numbering, failures, the
DfuSe suffix CRC under integrity_policy, and limits, and link the page
from the CLI and firmware guides and the agent skill reference.
npm run verify:uf2 runs esptool merge-bin --format uf2 on generated
inputs at known addresses, with and without MD5, with 256-byte
payloads, and on a concatenation of two chips' outputs. Inspection and
extraction through the built CLI must return every input at its address
under the familyID esptool itself reports.
* fix(artifacts): accept both DfuSe image size conventions
UM0391 calls DFUImageSize the file length, and dfu-util's dfuse-pack.py
writes that value, while ST's DfuFileMgr, MicroPython's dfu.py and pydfu
leave the 16-byte suffix out. The decoder accepted only the second form,
so it refused dfuse-pack.py output as malformed.
Accept either value and keep the declared one in image_size; any other
value still fails with the suffix offset and the file length. The tests
pin both producers' output for one image, CRC included, from their
build() functions transcribed and run under Python 3.
* fix(artifacts): keep reused UF2 block numbers from hiding missing blocks
esptool numbers every merged input from 0, so two inputs with the same
block count share one numbering group. A block lost from one input left
its number present in the other, and the file reported complete
coverage. Each block number now records the target addresses it was
seen at: equal address counts are complete copies, a resend to the same
address is one block, and unequal counts make coverage unknown with a
limitation and ambiguous_numbering_group_count. sequence_count becomes
numbering_group_count, and region block positions become
first_file_block and last_file_block.
Review follow-ups in the same change:
- compare_artifacts compares a DfuSe element's alternate setting and
target name, so bytes moved to another device memory are changed;
- the shared reader keeps no region bytes outside its map after close;
- one domain type replaces four identical decoder failure types;
- the UF2 decoder returns a typed limit instead of throwing on bytes
beyond its admitted size, and caps retained regions, not runs;
- DfuSe caps elements with bytes only, names empty elements and
element-less targets, drops the constant bVersion and bcdDFU fields,
and says whether the CRC matches when DFUImageSize fits neither
convention;
- classification evidence carries a suffix only when the probe saw the
end, and root classification reads the tail only for a DfuSe prefix;
- the UF2 domain region owns family_name, and the root records the
family registry revision; the table's MIT notice is preserved;
- docs pin the producer revisions (dfu-util ee6b393d and f9a537c9,
MicroPython cb99ca98), give the MD5 structure's real meaning, and
describe container contradictions in the MCP contract guide;
- tests and the esptool lane add the equal-count inputs and a block
lost from one of them, accepted boundaries beside each refusal, a
DfuSe root larger than the classification prefix, and CLI refusals.
* feat(artifacts): name the UF2 families that LibreTiny builds for
LibreTiny writes UF2 for Realtek, Beken and Bouffalo Lab modules. The
Microsoft registry revision REA already cites (cf3bd7f9) lists eight of
those families: RTL8710A, RTL8710B, RTL8720C, RTL8720D, BK7231U, BK7231N,
BK7251 and BL602. Their regions now carry those names. LN882H, which
that revision omits, keeps its familyID with no name.
* fix(artifacts): count DfuSe declarations without bytes instead of listing them
Empty DfuSe elements have no entry, so DFUSE_MAX_ELEMENTS does not cap
them, and the target walk kept a description string for each one. At 8
bytes per element, a file under the 128 MiB ceiling can declare about
16.7 million of them. The walk now counts every declaration without bytes
and builds only the first NAMED_ELEMENTS descriptions, which is all that
the limitation shows. The limitation text and empty_element_count do not
change. The guide now says that both DfuSe limitations name the first four.
On this host, decoding 4,000,000 empty elements (a 32 MB file) ran out of
memory under a 128 MB V8 heap before this change, and it peaked at 1.59 GB
RSS with the default heap. It now decodes in 114 ms with a 220 MB peak.
---------
Co-authored-by: morluto <[email protected]>
* feat(artifacts): inspect classic iNES container regions
* fix(artifacts): preserve iNES source disappearance as integrity failure
* fix(artifacts): revalidate iNES source before parsing header
* fix(artifacts): validate admitted roots before classification
* fix(artifacts): normalize source loss during root hashing
* fix(artifacts): bind iNES extraction admission to inventory
* test(artifacts): clarify non-executing inspection wording
* fix: preserve binary plist detection with extended magic probe
* test(browser): wait for popup document loads before capture
* fix(artifacts): honor cancellation across source validation
* fix(artifacts): prefer artifact magic over implicit database suffix
* fix(artifacts): select extraction readers from scanned format
* fix(artifacts): complete prefix reads and share ZIP format admission
* fix(browser): budget scenario cleanup for content drain and shutdown
* test(artifacts): admit valid universal inputs before routing
* test(artifacts): verify iNES workflows on macOS
* test(ci): own iNES macOS coverage and portable fixture paths
* fix: preserve cleanup retries and avoid duplicate inventory parsing
* fix(contracts): share ZIP target formats across MCP lifecycle schemas
* fix(target): preserve typed cancellation during descriptor admission
* refactor(artifacts): use JSON-safe inventory producer output directly
* test(artifacts): share file read setup and satisfy CI checks
* fix(artifacts): classify changed admission as integrity failure
* ci: run Mach-O summary delivery coverage on macOS
* test(javascript): align parameter provenance with main contract
* test(browser): await discoverable fixture pages before runtime verification
* fix(artifacts): pass target metadata to format detection
* fix(artifacts): bind extraction to opened target
* fix: repair the static-check failures on the default branch (#2895)
* fix: repair the static-check failures on the default branch
The default branch fails its own required `Static checks` job three ways, so
every pull request reports that job as failed regardless of its own change.
- `npm run format:check` reports `src/domain/functionComparison.test.ts` and
`src/ida/IdaOperationRunner.ts`.
- `npm run lint` reports a duplicate `import { join } from "node:path"` in
`tests/boundary/mcp/ghidraInventoryMcp.test.ts` (lines 4 and 17).
- `npm run typecheck` reports `format: "devin"` missing from the client format
union in `src/application/SupportedClients.ts`, and an unchecked index in
`tests/boundary/cli/clientProfileLifecycle.test.ts`, whose pair array is
inferred as `string[][]`.
Apply the pinned formatter, drop the duplicate import, list the `"devin"` format
that its own definition already uses, and mark the pair array `as const`.
Validation: `npm run check:fast` and `npm run format:check` pass.
* fix(setup): keep Devin on the JSON configuration dialect
---------
Co-authored-by: 吴佳浩 <[email protected]>
Co-authored-by: morluto <[email protected]>
(cherry picked from commit df3d96da80)
* fix(artifacts): bind session inventories to target identity
* test(browser): bound fixture readiness acknowledgements
---------
Co-authored-by: 64579525 <[email protected]>
Co-authored-by: 吴佳浩Alben <[email protected]>
Co-authored-by: 吴佳浩 <[email protected]>
* feat(ghidra): record the RISC-V float-ABI calling-convention decision
Decision D3 (#7477), provisional: admit ELF32 soft-float (ilp32) RISC-V
images through a lane that keeps the pinned RISCV:LE:32:default language and
corrects the calling convention in the bridge.
This commit is the contract only; admission is unchanged.
- The domain owns the e_flags float-ABI vocabulary and parses
.riscv.attributes and Tag_RISCV_arch into a declared, absent or
unavailable ISA declaration.
- The Ghidra profile owns the lane table (class and float ABI), lane ids
built from fields, the ilp32-prototype-v1 policy and its __rea_ilp32
prototype, and the extensions Ghidra 12.1.4 decodes with p-code.
- The session handshake accepts the bridge's calling-convention report.
- docs/ghidra-riscv.md keeps the decision record and per-ABI policies for
ilp32 and ilp32f; a test keeps it in step with the tables.
- tests/conformance/c/riscv-float-abi.c is the source the policies were
measured on.
* feat(ghidra): admit RISC-V ilp32 ELF with a derived calling convention
ESP32-C2, C3, C5, C6, C61, H2 and H21 builds are ELF32 soft-float (ilp32)
images, which the RISC-V lane refused at its double-float check. Ghidra
12.1.4's generic RV32 compiler specification models ilp32d, so its stock
prototype would read float and double arguments from fa0-fa7.
- Target resolution reads the one SHT_RISCV_ATTRIBUTES section through the
same bounded ELF declaration reader as range validation, and records
Tag_RISCV_arch as declared, absent or unavailable.
- Admission selects a lane from the class and float-ABI table. The soft-float
lane also requires an RV32I Tag_RISCV_arch that declares only extensions
Ghidra decodes with p-code; the refusal names any others (Zcb, Zcmp, Zcmt,
Zb*, xesp*).
- Soft-float profiles commit float_abi, arch, riscv_calling_convention
ilp32-prototype-v1 and the lane elf32-exec-le-soft-float-generic-v1.
Double-float profiles are unchanged.
- ReaGhidraRiscvCallingConvention.java runs after analysis. It derives
__rea_ilp32 from the default prototype without floating-point entries and
rules, with fs0-fs11 killed by calls, self-checks its storage and assigns
it to every non-thunk function, external function and function
definition. Its report is written last.
- The provider client requires that report exactly when the profile commits
the policy, before any operation, and keeps it as session limitations and
caller-project history.
ELF32 single-float (ilp32f), quad-float and ELF64 soft- or single-float
images stay refused, now with a reason built from the lane table.
* test(ghidra): verify RISC-V ilp32 against real Ghidra
verify:ghidra:riscv now also builds tests/conformance/c/riscv-float-abi.c
with ESP32-C3's -march=rv32imc_zicsr_zifencei -mabi=ilp32, typed with DWARF
and linked again with --strip-all. Every allocated section of the stripped
image must keep its address and bytes.
- riscv-checks.mjs holds the ilp32 storage of each fixture function together
with the words the compiler's call site writes there. An LLVM objdump
reading of the caller must find every word at its stated location: an
unaligned named double, the a7/stack split, two floats in an integer pair,
a large structure by reference, aligned variadic pairs and an indirect
call.
- Through CLI and stdio MCP, the soft-float builds must report the soft-float
lane and ilp32-prototype-v1, __rea_ilp32 with exactly that storage (one
integer word per location when stripped), caller pseudocode that passes the
source constants and names no floating-point register, the prototype
outcome in retained Evidence, and decoded fence.i and csrr.
- The RV32 double-float image relabelled as soft-float, refused before, is
admitted through the new lane. RV64 soft-float, single-float and unsupported
bits are refused with their reasons.
- riscv-checks.test.ts runs the checks offline against recorded LLVM and
Ghidra output, including Ghidra's stock ilp32d prototypes as a negative
control.
* fix(ghidra): pass RISC-V ilp32 values over 8 bytes by reference
The psABI passes any value wider than 2 x XLEN by reference, scalars
included. __rea_ilp32 converted only structures and unions over 8 bytes to
pointers, so a long double (16 bytes) was joined into a1-a4. Measured on
Ghidra 12.1.4, rea_riscv_abi_quad(long double value, int sel) read value
from a4:4,a3:4,a2:4,a1:4 and sel from a5, where clang passes a pointer in
a1 and sel in a2.
- ReaGhidraRiscvCallingConvention.java now has one derivation driven by a
per-policy row (floating-point entry size, preserved floating-point
registers). Every policy inserts a rule before the input list's final
join that converts any value of 9 bytes or more to a pointer; a return
that size already becomes a caller buffer. The self-check requires a
16-byte value to pass by reference.
- The fixture gains rea_riscv_abi_quad. The call-site oracle now checks the
words the by-reference pointers point at, for the large structure and the
long double, and the recorded data comes from an LLVM 19 build.
- docs/ghidra-riscv.md: the decision record and the ilp32 policy row.
ilp32-prototype-v1 keeps its id: no release carries it yet.
* test(ghidra): check RISC-V lane limitations by their facts
The limitation tests froze a full explanatory sentence, the note count and
array positions, so a harmless rewording failed them: the pattern Greptile
flagged on #7486. Assert the facts the notes must carry instead: the pinned
language, the static-linkage boundary, one note naming the declared ABI,
the ABI the pinned language models and the prototype REA applies, and the
declared ISA.
* test(ghidra): check return storage in stripped RISC-V ilp32 builds
The stripped branch of assertRiscvIlp32Storage compared parameter storage
only, so an inferred return that lost the high word of a register pair,
such as rea_riscv_abi_wide at a0:4 instead of a1:4,a0:4, still passed.
Stripped returns must now use the typed build's storage, or none when the
result goes through caller memory. Recovered type names are still not
compared: a test accepts other type names and refuses the dropped high word.
* docs(ghidra): record decision D3 as decided in #7477
* feat(artifacts): inventory and extract ESP32 firmware images
ESP32-family app and bootloader images (magic 0xE9) had no reader, so
open_binary refused OTA downloads, app partitions and bootloaders.
Add the esp-image container format behind inspect_artifact and
extract_artifact. A pure decoder reads each byte once and keeps only
fixed-size fields. Derived facts cover the chip, from a table cited to
ESP-IDF v6.1; segment classification as ESP-IDF's should_map and
should_load; checksum and appended SHA-256 results; app and bootloader
descriptors; Secure Boot v2 signature blocks, observed and unverified;
and undescribed trailing byte ranges. Truncation and undescribed bytes
make coverage unknown. ESP8266 images are refused with a specific
reason, and ciphertext stays unrecognized.
Add the format to every target, Evidence, snapshot, live-summary and
artifact vocabulary, document it, and add an optional verify:esp-images
lane that compares esptool-built images with esptool image-info and
espsecure signature-info-v2.
Refs #3615
* test(artifacts): run the ESP image lane on esptool v5.6.0
The lane compared descriptor fields only when esptool image-info printed
one, so a descriptor that REA reports and esptool does not print passed
without comparison. Compare the descriptor kind in both directions.
Record that the lane passed for all 13 images with esptool and espsecure
v5.6.0, the version the layout facts cite, as well as with 4.12.0.
* refactor(artifacts): define ESP image layout sizes once
The decoder and the derived facts each defined the segment-header,
appended-digest and flash-sector sizes, the erased-byte scan and the
offset formatter. Define them once with the other fixed-layout structures
and import them in both places. The signature slot count is the one
constant that drives the slot scan.
Correct the observedEnd comment: it is the end of the last chunk read.
* fix(artifacts): report ESP-IDF's segment length limit by chip
ESP-IDF's verify_segment_header rejects a segment of 16 MiB or more
unless the chip maps flash through 32-bit cache addresses
(SOC_SPI_MEM_SUPPORT_CACHE_32BIT_ADDR_MAP: ESP32-S3, P4, C5, C61, H4 and
S31 in v6.1). Record that capability in the chip table and add a
limitation for such segments; unknown chips get none.
`loaded` repeats should_load. It does not mean the bootloader accepts the
segment: verify_load_addresses still rejects one outside the chip's RAM,
which REA does not model. Say so in the docs, the skill reference and the
inventory limitation.
* test(artifacts): check ESP lane tools first and verify extraction
The lane started work before checking its tools, so a missing RISC-V
target, ld.lld, esptool or espsecure surfaced as a raw spawn or command
error. Check each one before building and name the variable to set. Warn
when esptool is not v5.6.0, the release the layout facts cite.
Assert each segment's class for the ESP32-C3 app, its variants and the
bootloader, as the Xtensa image already did. Extract the app and the
signed image through the CLI and compare every file with the bytes at
esptool's segment offsets or in the appended signature sector; the
source's SHA-256 must be unchanged and equal the reported root_sha256.
* docs(artifacts): state what the ESP image lane does not establish
The Verification section listed the lane's checks but not its gaps: only
ESP32-C3 and ESP32 images come from a real producer, no real project
build, OTA download or flash dump is inspected, no signature is verified,
and the lane does not run in CI.
Record the timing of the final lane run on esptool v5.6.0.
* fix(artifacts): compare ESP segment declarations
Segment entries hash only their data, so compare_artifacts reported a
segment that moved to another load address with the same bytes as
unchanged; only the image itself changed. Compare each segment's
load_addr and data_len as its declaration, as Node SEA and PyInstaller
members are compared, and count a changed declaration as a metadata
change, also when content is unavailable.
Header offsets are layout and the classification is derived, so neither
is compared.
* test(artifacts): run ESP lane commands as owned process trees
The lane ran its tools through execFile. A timeout killed only the direct
child, so a linker or assembler that Clang or GCC had started could keep
running while the lane removed its workspace. Run every command through
createVerifierCommands, check the tools inside withVerifierCleanup, and
remove the workspace only after commands.close() has stopped every tree;
a failed close keeps the workspace.
Only a nonzero exit now counts as a refusal or an expected esptool
failure; timeouts and spawn failures propagate. Supervising each command
roughly doubles the lane's run time.
* test(artifacts): cancel ESP decoding after its first decoded chunk
The cancellation test requested cancel or close on setImmediate, which
can land while the file is still being opened and checked, so it never
showed that decoding had started. It also passed against a reader that
ignored both requests while decoding. Acknowledge the decoder's first
chunk, then cancel or close, join the operation, and assert that no
later chunk reaches the decoder and that the reader ends closed.
* docs(artifacts): record which ESP metadata comparison ignores
Segments are compared by their declaration. The image's and the
signature blocks' format_metadata are not: every field comes from their
own bytes, which content comparison already covers, and a block's
image_digest_matches follows the image's content.
* feat(artifacts): inventory and extract ESP flash layouts
A whole-flash dump of an ESP32-family chip was read as its bootloader
image, with the partition table, apps and data left as undescribed bytes,
or refused when it began with erased flash. Read it as an esp-flash-image
container instead.
The partition table is decoded as esp_partition_table_verify accepts it
and is found at the caller's partition_table_offset, at the default
0x8000, or by scanning 0x1000-aligned sectors for a table whose MD5
verifies, which the inventory reports as derived. The bootloader is
located at its chip's ROM bootloader offset. Each partition is an entry
with its declaration in flat format_metadata; the bootloader and app
partitions carry ESP image facts from the existing streaming decoder; the
otadata records are reported with a selection derived without rollback or
anti-rollback, and the running partition stays unknown. One unreadable
partition does not hide the others, and extraction writes every partition,
nvs_keys included, as ordinary data.
Flash layouts are detected before single ESP images. The scan runs only
when an ESP image sits at a ROM bootloader offset, and is refused when a
bootloader at another chip's offset shows that the dump does not start at
flash address 0. open_binary gains flash_offset and partition_table_offset,
which require format esp-flash-image and are part of the snapshot target
identity; the CLI takes them with --target-format esp-flash-image.
Comparison treats a partition's label, type, subtype, range and flags as
its declaration.
* test(artifacts): verify ESP flash layouts with gen_esp32part, merge-bin and otatool
Add the optional verify:esp-flash lane. It builds default and relocated
partition tables with ESP-IDF's gen_esp32part.py, assembles dumps with
esptool merge-bin, including one written from flash 0x1000, and writes
OTA records with otatool through flash_file_tool.py, which replaces
parttool's esptool calls with reads and writes of the dump file.
It compares REA's bootloader and app facts with esptool image-info, every
partition with gen_esp32part.py's decoding of the extracted table, and the
OTA records and derived slot with otatool read_otadata after each switch.
It also checks extracted bytes and source integrity, the scan and caller
table offsets, flash_offset on the 0x1000 dump, a corrupted app beside
valid partitions and a truncated dump. ESP32 linker scripts place the
conformance bootloader and app in ESP32 address ranges.
* fix(artifacts): read ESP flash layouts as the bootloader does
Read the partition table as bootloader_utility_load_partition_table
does: the last factory, otadata and same-subtype OTA declaration wins,
and a limitation names repeats. Plan each otadata record within its
partition and the dump. An otadata partition smaller than two sectors
selects none, the bootloader's INVALID_INDEX; a record the dump lacks
leaves the selection unknown while the other record stays reported. The
fallbacks follow the bootloader's indexes: ota_0-fallback and
factory-fallback name no partition when the table declares none, and the
assumptions add GPIO holds and deep-sleep wake.
Refuse a dump whose bootloader contradicts its origin before the default
table offset is tried, whether flash_offset was omitted or given, and keep
the ESP image reader's ESP8266 refusal. A table never lies inside the
located bootloader image, so single app images are no longer scanned. An
explicit esp-flash-image selection is decided before other container
signatures, and the native commands' --target-format no longer takes it.
Say when no bootloader confirms the default origin.
Exempt the primary bootloader and partition_table entries of ESP-IDF 5.3
from placement limits, report every region inside a longer one, and name
partitions by index when labels repeat ignoring case. Keep an undecodable
label's bytes in label_hex and compare them. Word undescribed ranges as
starting at a byte that is not erased, and treat a short erased region as
erased.
Add an MCP compare_artifacts test for partition declarations, reader
identity and close tests, scan edge cases, the 0x2000 ROM offset and a
literal otatool CRC.
* test(artifacts): verify ESP primary layout entries and the otadata fallback
Build a table with the primary bootloader and partition_table entries
that gen_esp32part.py fills from --primary-bootloader-offset, and check
that REA reads it with complete coverage, no overlap or misplacement
limitation, and a table that agrees with gen_esp32part's decoding. Erase
the otadata with otatool after the slot switches and check the records
against read_otadata and the factory fallback.
* fix(artifacts): keep ESP flash coverage complete for a dump cut short
An esptool merge-bin dump ends after the last part it wrote, usually inside
its last partition. Such a partition made the layout's coverage unknown, so
artifact comparison of two merge-bin dumps reported every unchanged
partition as unknown.
The partition table names every partition whatever the file's length, so a
partition cut short or left out by the end of the file hides no entry. As
the ESP image reader does for a cut-short segment, its entry holds the bytes
the file has and an entry limitation gives the declared size; a partition
with no bytes stays in partitions_outside_dump. Coverage stays complete
unless undescribed bytes remain or no bootloader is located. An MCP
compare_artifacts test of two dumps that end inside ota_0 now classifies
every partition, and fails on the old rule.
* fix(artifacts): report written bytes in ESP partitions, erased slots included
An app slot is erased for ESP-IDF's bootloader when its image header bytes
are erased flash, and REA reported only that status. A slot can still hold
data further in: a dump read from a device held a complete older app image
1 MiB into an OTA slot reported as erased, with no limitation.
Each partition entry now records where the bytes the dump holds are
written, relative to the partition: written_offset, the first byte that is
not 0xFF, written_end, just past the last, and written_bytes, how many are
not 0xFF. They are null for a partition outside the dump. The status stays
erased, which is the bootloader's view, and a limitation names an erased
app slot that holds written bytes and says they are not read as an image.
The reader reads every partition byte once more to find them.
* fix(artifacts): derive the ESP boot selection for every partition table
The boot selection lived inside otadata, so a table without an otadata
partition, the common factory-only layout, had none. ESP-IDF's
bootloader_utility_get_selected_boot_partition returns the factory index
when the table declares no otadata partition.
Move it to a top-level boot_selection that every layout has. Its basis
names the bootloader's branch: otadata, factory-fallback and ota_0-fallback
from the records, no-otadata for a table without otadata, none for the
bootloader's INVALID_INDEX, and unknown when the dump lacks a record the
bootloader reads, which replaces the null selection. running_partition
moves with it, and otadata keeps its partition index and records. The
assumptions name what each branch leaves out: no records assumption
without otadata, and none for none or unknown. The real-tool lane checks
no-otadata on a table gen_esp32part built without otadata.
* fix(artifacts): say when an ESP image is a flash dump read from 0x1000
An ESP32 dump read from flash 0x1000, as read-flash 0x1000 or merge-bin
--target-offset 0x1000 writes it, opened automatically as its bootloader
image with a megabyte of undescribed bytes and no next probe. The flash
reader refuses that origin, but nothing told an agent which flash_offset to
pass.
espFlashDumpOrigin owns the inference: an image of a chip whose ROM loads
the bootloader from flash R, at file offset F, is the bootloader of a dump
read from R - F when a partition table sits where that origin puts 0x8000.
The ESP image reader asks it for a non-app image at file offset 0, so the
result does not depend on how the reader was created, and records
flash_dump {flash_offset, table_file_offset}, null otherwise, with a
limitation that says to reopen the file as esp-flash-image with that
flash_offset. inspect_artifact proposes the same open_binary call as a next
probe, and the flash reader's misplaced-origin refusal names the offset.
* perf(artifacts): summarize nested ESP images in the flash layout summary view
The summary view returned an ESP flash layout's root facts whole, so each
app partition and the bootloader carried every segment and signature
block, about half of a dump's summary.
The summary view now gives each nested image a summary instead of its
facts: status, chip, entry point, segment count, load ranges (each mapped,
loaded or unclassified segment's address and declared length), the
checksum and appended SHA-256 statuses, coverage and limitations. Complete
detail keeps the full facts, every segment included, and the retained
Evidence is unchanged. On a 16 MiB device dump the summary's root falls
from 5,141 to 2,657 bytes, and the summary Evidence from 10,038 to 7,554.
* test(artifacts): expect the misplaced image the ESP bootloader search reports
8a748d3cc added misplacedImage to the bootloader search, which the
locator uses to name the origin a misplaced bootloader implies, but one
test still compared the search result without it. Expect the image's
chip and file offset, and no image for an unknown chip.
* fix(artifacts): give every ESP partition from the dump's start an entry
After ecbc1a772 kept coverage complete for a dump cut short, a partition
wholly past the end of the file still had no entry, only a
partitions_outside_dump row. Comparing a merge-bin dump with a full device
read then reported that partition as added or removed, as if one table
did not declare it.
Every partition from the dump's first byte on is now an entry holding the
bytes the file has: none for one past the end, with a limitation saying
the file holds none of its bytes, no image facts and null written fields.
partitions_outside_dump keeps only partitions that begin before the dump's
first byte, and coverage is unknown when one does, since it has no entry.
Two dumps that lack the same partition now compare it as unchanged, and a
dump against a longer one reports the partition as changed; an MCP test
of a merge-bin dump against a full one checks that no path is added or
removed.
Unlike the ESP image reader, which marks a truncated image's coverage
unknown because later segments and signature blocks may be hidden, the
flash reader can list every partition from the table whatever the file's
length; only the entry limitation follows the image reader.
* fix(artifacts): select no ESP app from a partition table the bootloader rejects
boot_selection followed bootloader_utility_get_selected_boot_partition but
not the table check before it. For a table whose MD5 does not match, or
that has a second MD5 entry, esp_partition_table_verify fails, the
bootloader selects no app and resets. REA still derived a slot from the
otadata records, although the table's own limitation said the bootloader
rejects it.
The selection now has basis none, with no assumptions, for a table the
decoder marks invalid. It also follows two more branches of the
bootloader: an otadata partition at offset 0, which the bootloader takes
for none, selects the factory index as no-otadata, and a table without an
OTA app selects its fallback even when the dump lacks a record, since the
bootloader then ignores the records.
The bootloader also rejects a table whose partitions end past the flash
size in its header. REA does not model that check; the format guide says
so.
* fix(artifacts): require the ESP dump origin's table to follow the image
espFlashDumpOrigin named a later flash origin whenever a partition table
decoded where that origin puts 0x8000. Bytes inside the image's own
segments could be read that way, so an ESP32 bootloader whose segment data
held table-shaped bytes at file offset 0x7000 was reported as a dump read
from 0x1000.
The table must now start at or after the image's end, the rule the locator
already applies to the default table. The ESP image reader passes the
image_end it decoded, or the file size for a truncated image; the flash
reader's refusal passes the end of the misplaced image's segments. The
bootloader search drops its misplaced flag, which only restated
misplacedImage, and the next probe formats offsets with offsetText.
Reader tests cover the 0x1000 and 0x2000 ROM offsets, a standalone
bootloader, bytes that begin like a table, an app with a table at 0x7000
and a table inside a segment; without the boundary check, three of them
fail.
* fix(artifacts): give the ESP flash summary its own kind and each image's identity
The summary view's ESP flash root kept kind esp-flash-image, so a consumer
bound to kind could not tell an image's summary from its full facts, and
the advertised schema held two objects with the same kind const. It is now
esp-flash-image-summary.
The per-image summary also dropped the descriptor, which is what tells two
app slots apart, and named each image only by its partition index, which
the summary view, having no entry rows, could not resolve. Each nested
image now carries its entry path from the layout plan, in both views. The
summary adds the descriptor kind with its identifying fields (an app's
project_name, version, idf_ver, secure_version and app_elf_sha256; a
bootloader's version, idf_ver and secure_version), the signature block
count and each load range's classification. Its enums and fields derive
from the schemas that own them.
An MCP test reads a dump with an app in ota_0 and an erased ota_1 that
holds older bytes: it checks both summaries, that the summary's inventory
Evidence is the one the complete view returns with full facts, and the
erased slot's written range.
* test(artifacts): cover cancel and close during the ESP written-bytes scan
The written-bytes pass reads every byte of each partition the dump holds,
checking that the reader is open before each chunk, but only app decoding
had a cancellation and close test. A data partition, which only that pass
reads, now holds 9 MiB; once the scan has examined its first chunk, a
cancel rejects preparation with the caller's reason and a close with
unavailable, and the scan examines no further chunk.
The erased-slot test now asserts the whole set of erased-slot limitations,
so factory, ota_0 and nvs, which the bootloader reads or which are not app
slots, are shown to have none.
* docs(artifacts): re-time the ESP flash lane and state only observed causes
The lane ran again after its assertions changed; both esptool runs passed,
and the timed run's figures replace the earlier ones. The erased-slot
paragraph no longer names a cause for the older bytes, which REA cannot
observe, and the skill reference states that every table has a
boot_selection in its own sentence, with the none basis for a table the
bootloader rejects.
* fix(artifacts): retire ESP flash members as the ESP image reader does
#7498's maintainer commits made the ESP image reader keep a member owned
until its consumer closes it, retire fully buffered bytes when
cancellation arrives after production, and refuse to finish an inventory
paused across close. The ESP flash reader was copied from the earlier
reader. Main's shared stored-member lifecycle cases failed two of eight
against it: a paused inventory completed after close, and buffered member
bytes stayed readable after cancellation.
Give the flash reader the same lifecycle and run the shared cases against
it.
* docs(artifacts): re-time the ESP flash lane after merging main
The lane passed again with esptool v5.6.0 and 4.12.0 on the merge of upstream main. Record the v5.6.0 run's time and peak resident set.
* fix(artifacts): read ESP image Evidence recorded before flash_dump
The flash layout work made `flash_dump` a required field of the
`esp-image` root metadata. Evidence recorded before it has no such
field, so newer REA rejected records that were valid when they were
made: `compare_artifacts` and `inspect_analysis_view` reported
`evidence_integrity_mismatch` for them.
The field is now optional when read. A record without it parses as it
is, so the graph digest it commits to still matches; nothing fills the
field in. The image reader still writes it for every image, and the
inspection probe reads a missing field as no hint.
A regression test recommits an image inventory without the field and
passes it inline to `compare_artifacts` and `inspect_analysis_view`.
Against a server built with the required field, `compare_artifacts`
rejects it.
* fix(artifacts): stop the ESP flash gap scan when the reader closes
The scan for the first written byte in each undescribed gap checked
only the caller's signal. `close()` marks the reader closed and waits
for preparation, so it waited while the rest of a long erased gap was
read.
The gap scan now takes the reader's open check, as the written-bytes
scan and the image decoder do, and checks it before each chunk. Tests
cancel or close the reader after the scan's first chunk of a 9 MiB
erased gap and confirm that no later chunk is examined.
* fix(artifacts): name ESP partitions with Windows device labels by index
The safe-label rule accepted labels such as `NUL`, `con` or `COM1`, so
extraction wrote `partitions/NUL.bin`. Windows reserves those names
before any extension, and ordinary Windows tools cannot use such a
file.
A label that is a Windows device name (`CON`, `PRN`, `AUX`, `NUL`,
`COM0`-`COM9` and `LPT0`-`LPT9`, any case, alone or before a dot) now
takes the existing index fallback. The partition metadata and the
entry's source path keep the label.
* fix(artifacts): read ESP flash partitions' written bytes in one pass
Each partition got its own scan of the bytes the dump holds, so
partitions declared over the same bytes read them again and again.
Overlaps only add limitations, so a crafted table reached this normally:
94 overlapping partitions over a 32 MiB dump would examine about 3 GiB
before the first entry.
The reader now reads the bytes the partitions cover once. Partition
boundaries cut them into segments; one pass counts each segment's
written bytes and bounds, and each partition's extent is the sum of the
segments inside it. Every partition keeps its own declaration, extent
and limitations. The pass checks the reader's open state before each
chunk, as the per-partition scan did.
A test declares 60 partitions over one MiB, 40 over all of it and 20
over its first half, and checks each partition's extent and the
overlap limitations. The reader reads 1,113,256 bytes of the
1,114,112-byte dump; one scan per partition read 52,493,480.
* test(artifacts): fail the ESP lanes on failed commands and refusals
`otatool` turned a nonzero exit into returned text, which its callers
discarded, and `inspect()` returned `{ refused }`, which the flash lane
never checked. A failed command then surfaced later as a slot
assertion, an output-parser error or a TypeError, hiding its cause.
Commands the lanes need to succeed now reject with the output they
exited with: `otatool` and `inspect()` do so through a shared
`failedCommand`. An expected refusal is opt-in: `refusal()` returns
REA's output and fails the lane when REA opens the file. The ESP image
lane's two ESP8266 cases use it.
Checked by calling `otatool` on a file without a table, `inspect()` on a
file REA refuses and `refusal()` on a file REA opens; each rejects with
its cause.
* fix(artifacts): require a partition entry in an ESP partition table
The decoder counted the slot, not the partitions, so a lone MD5 entry
over the empty prefix, followed by the terminator, decoded as a table
with no partitions. Default and caller-selected detection then accepted
that empty layout, although the guide requires one partition.
A table now needs one partition entry before its terminator. An MD5
entry alone is refused as `entry 1 at 0x20 ends the table before any
partition entry`. The cases cover the decoder and both automatic and
caller-selected location.
* fix(artifacts): decode an image shared by ESP app partitions once
Each app partition got its own decode, so app partitions declared over
the same bytes decoded and hashed the same image again and again, the
same pattern the written-bytes scan had. With 60 app partitions over
one 960 KiB image the reader read 64,029,736 bytes of a 1,114,112-byte
dump.
Regions with the same file range now share one decode; each keeps its
own copy of the facts. Partitions with different ranges still decode
separately, because truncation can change the facts: at most one decode
per partition the table holds, each within its partition.
A test declares 60 app partitions over one image, 40 holding all of it
and 20 cutting it short, and expects three decodes: the bootloader, the
whole image and the cut one. The 60-partition dump now reads
2,161,864 bytes, as a dump with one app partition does.
Since 651afa445 target opening refuses every ELF declaration beyond EOF,
including carved copies, and ac6e0e258 moved the carved MIPS boundary tests
to that contract. The real-Ghidra MIPS lane still expected a carved image to
resolve with section-table limitations and be refused later by Ghidra
admission, so verify:ghidra:mips failed at its first carved fixture.
The lane now requires open_binary and the CLI to refuse the carved copy as
target_unavailable (file_format_rejected) with the exact declared
section-table range and captured size. The carve and that expectation live in
the shared MIPS checks, and a boundary regression applies them to the
production resolver and error projection, so a later contract change fails
without real Ghidra. The MIPS and PSP guides describe the current contract.
* fix(ghidra): report a missing existing-project program as a configuration error
When REA_GHIDRA_PROJECT_PATH selects a project that has no root program
with the selected name, Ghidra's headless analyzer aborts with
"Requested project program file(s) not found: <name>". REA reported this
as a generic execution_failure with rea doctor advice, and the reason was
visible only in the captured diagnostics.
Recognize that refusal in the operation-scoped launcher output, matching
only the exact name REA passed to -process, and return a
ConfigurationError for REA_GHIDRA_PROGRAM_NAME. The message says whether
the name was configured or defaulted from the target file name, and the
remediation explains how to select the exact root program. The selected
project is unchanged and there is no fallback to a fresh import. Other
launch failures keep their existing projection.
Fixes#8216
* test(ghidra): assert missing-program guidance through its formatter
Compare the projected setting constraint and remediation with
ghidraMissingProgramGuidance instead of freezing explanatory phrases,
following docs/testing.md. The program name and configured-or-defaulted
origin are still checked independently.
Expose static Electron protocol call facts through the shared application graph and existing CLI/MCP Evidence views. Preserve canonical case-insensitive scheme identity, original spellings, exact source provenance and explicit unknowns; keep registration success and runtime routing unproven.
Integration fixes also enforce module-only privileged API ownership, reuse the shared finding-key owner, and align the owning type documentation with the Evidence semantics. Validation and bounded real Obsidian CLI/MCP measurements are recorded in the PR.
Co-authored-by: UnknownHacker1 <[email protected]>
Co-authored-by: gensokyo-dev <[email protected]>
In fresh-import mode, the startup-timeout remediation now points to
REA_GHIDRA_PROJECT_PATH for reusing a completed analysis, next to the
existing REA_GHIDRA_STARTUP_TIMEOUT_MS and REA_GHIDRA_DISABLED_ANALYZERS
advice.
In existing-project mode, the message no longer says Ghidra was
importing and analyzing the target, and the remediation no longer
suggests REA_GHIDRA_DISABLED_ANALYZERS, which configuration validation
rejects together with REA_GHIDRA_PROJECT_PATH.
Refs #3751
* fix(javascript): emit no socket for a literal out-of-range port
`net` and `tls` validate a numeric port before creating a socket, so a literal
port Node rejects throws synchronously and no socket exists — but the analysis
reported an `acquire` operation.
Measured on Node 24.22.1:
net.connect(0) / 80 / 65535 -> Socket created
net.connect(65536) -> THREW ERR_SOCKET_BAD_PORT
net.connect(-1) / 1.5 -> THREW ERR_SOCKET_BAD_PORT
net.connect(NaN) / Infinity -> THREW ERR_SOCKET_BAD_PORT
net.connect(1e400) -> THREW ERR_SOCKET_BAD_PORT
net.connect(0n) / true / false -> THREW ERR_INVALID_ARG_TYPE
net.connect(null) -> THREW ERR_INVALID_ARG_TYPE
net.connect("") -> THREW ERR_SOCKET_BAD_PORT
net.connect("/tmp/x.sock") -> Socket created (Unix socket path)
net.connect({ port: 80 }) -> Socket created
net.connect(undefined) / {} / [] -> THREW ERR_MISSING_ARGS (undecided)
--- tls mirrors net ---
tls.connect(80) -> Socket created
tls.connect(65536) / -1 / NaN -> THREW ERR_SOCKET_BAD_PORT
A string is a Unix socket path and an object may carry a path, so only an
exactly readable non-port literal is decided. `undefined` and an empty object
throw `ERR_MISSING_ARGS`, which is a missing option rather than a bad port, so
they stay undecided. `NaN` and `Infinity` are identifiers rather than literals
and are decided only through `isUnmodifiedGlobalReference`.
* chore: retrigger title and readiness checks after the title was corrected
* fix(javascript): place the port check away from the fs flag check
`#8125` adds an fs flag guard at the top of `collectAcquisitions` and this branch
added a socket port guard in the same spot, so the two conflicted on every
rebase. The port check now filters the completed candidate list from
`collectJavaScriptSemanticResources` and its helpers sit at the end of the file,
which cannot collide with the fs flag guard.
* fix(javascript): drop the redundant in-loop port guard
The caller now filters the completed candidate list, so the loop guard became
dead code. Removing it also keeps this branch from editing the same lines as the
fs flag guard in #8125.
Resolve exact external method targets from directly encoded DEX invokes.
Preserve complete caller identity, bounded position evidence and follow-up
selectors. Report aggregate position coverage independently of the caller
inventory, and reuse bounded prototype facts within the exact session.
Absorbs #8096. Fixes#7811.
Co-authored-by: hhang <[email protected]>
* feat(dotnet): expose field constant defaults
* test(dotnet): guard field default comparison warning
Compare actual ordinary and constant-bearing observations independently of the limitation producer. Preserve warning delivery and unchanged matching assertions while allowing harmless wording changes.
Refs #5150
* test(mcp): retain structured bundle path diagnostics
Keep absolute-path refusal and all rejected parameter facts independent of legacy flat or current structured SDK diagnostic rendering. Retain relative CLI, absolute MCP and no-Evidence controls.
Refs #5150
* Revert "fix(ci): integrate main and repair inherited static checks"
This reverts commit afd4d571f6, reversing
changes made to 317bf3fe23.
* fix(ci): replay reviewed field-default integration
* fix(analysis-view): keep inferred confidence on views of inferred Evidence (#8099)
createAnalysisViewEvidence chose view confidence from the view result kind
only, so every summary, page or item view of an inferred parent
(compare_source_to_bundle, inspect_evm_interface, partially resolved
inspect_il2cpp_metadata) was sealed as derived while keeping the parent's
authority. A projection never makes its facts more certain: an inferred
parent now yields inferred views through CLI and MCP alike. Observed and
derived parents keep their existing view confidence and identities.
Fixes#8098
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(apple): read flattened framework Info.plist from Resources (#8101)
A macOS framework without Versions/ keeps its binary, _CodeSignature/ and
Resources/Info.plist directly in the .framework, as Electron and Gecko apps
ship it. Bundle anatomy looked for a shallow bundle's plist only at its root
and reported info_plist_path: null under complete coverage.
Resolve a shallow bundle's root Info.plist first, then Resources/Info.plist,
the same layout a framework version directory uses. Keep the iOS platform
rule tied to a root-level Info.plist so a Resources/-only application root
is not classified as iOS.
Fixes#8100
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(desktop): charge retained rows at their measured representation (#8105)
* fix(desktop): charge retained rows at their measured representation
The desktop scan reserved 1024 + 24 bytes per path character for every
row, omission and the root against its 32 MiB representation budget. Real
rows cost 800-940 heap and 490-640 JSON bytes, so the charge overstated
them 6-13 times and large macOS bundles stopped at 3,700-7,200 entries,
far below max_entries.
Charge each path at the larger of its UTF-16 heap size and its escaped
JSON size, plus the existing fixed row and omission overheads, in one
helper. The bound still covers the retained heap and the serialized
report, including control-character paths that expand six times in JSON.
Fixes#8104
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(desktop): charge every retained copy of a path
The tighter row charge counted each path once, but rows repeat it in
convention-fact values and the next probes those facts offer, and host
failure issues name the absolute path. Control-character paths under a
convention such as Content/Paks or resources/app.asar therefore serialized
47-80 MiB of reports within the 32 MiB budget.
Reserve each copy through the same text helper where it is stored: path
convention facts with their row, the probes any fact offers (at most two,
pinned by a projection test) with that fact, retained read ranges before
their read, and probe failure issues before they are recorded.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(source-maps): resolve empty sourceRoot sources against the map URL (#8107)
webpack 5 and tsc write "sourceRoot": "" for map-relative sources. Since
5f702dd6c the shared join added "/" for any non-null root, so these
sources resolved from the URL root: rxjs's ../../../src/internal/Observable.ts
under /npm/[email protected]/dist/cjs/internal/ became /src/internal/Observable.ts,
and xterm's webpack://xterm/./src/... became https://host/webpack://xterm/....
Treat an empty root like a null root in sourceMapReference, the single join
used by trace-web-source-location, analyze-javascript-application and the
web source-map fetcher. Chrome, Firefox and trace-mapping skip an empty root,
and tc39/ecma426#183 wrote the current steps to match them. Non-empty roots,
including "/" and repeated slashes (#6726), keep their existing behavior.
Fixes#8106
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(javascript): stop reporting self-defaulted receivers as module loads (#8109)
The CommonJS loader search treated any repeated state on its own ancestor
chain as an unknown loader. A self-default such as
`session.messages = session.messages || []` forms that cycle, because a
member read inside the assigned value was attributed to the later call's
read site and selected the same write again. Each `push` on the value was
then reported as an unresolved `require` load, and graph coverage became
partial.
Record a member read inside a flowed value as a depth-1 captured selection
at its own position, as member mutations already do for flowed origins.
Treat a revisited state as adding no source; the cycle's other entries are
still expanded once each. Remove the superseded ancestor tracking.
Fixes#8102
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(captures): admit browser-serialized HAR request URLs (#8112)
* fix(captures): admit browser-serialized HAR request URLs
[email protected] checks request.url with ajv-formats' strict RFC 3986
`uri` format. Browsers record and send URLs as WHATWG serializes them,
which keeps characters such as `|` and `{` unescaped in the query, so one
Google Fonts or GraphQL GET entry made inspect-web-network-capture and
environment-build reject the whole HAR.
HAR 1.2 defines request.url only as the absolute request URL. Own that
correction in the HAR validator: every URL the RFC 3986 check accepts
stays accepted, and a URL that WHATWG URL serialization reproduces
unchanged is accepted too. Spellings WHATWG would rewrite (a backslash
authority, a tab, a space) and relative references still fail.
Fixes#8110
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(captures): state that RFC 3986 request URLs stay admitted
The HAR request URL rule is "passes RFC 3986, or WHATWG serialization
reproduces it unchanged". A spelling WHATWG would rewrite therefore fails
only when it also fails the RFC 3986 check; an uppercase host still
passes. Say so in the docs and the validator comment, and pin the
accepted uppercase-host case in the CLI test.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(apple): charge the XML plist decoder's DOM before decoding (#8115)
* fix(apple): charge the XML plist decoder's DOM before decoding
The XML preflight charged 4 bytes per character and 256 bytes per start
or end tag. The decoder builds an @xmldom/xmldom DOM that measured
1,000-1,160 bytes per element and up to 240 bytes per text, comment or
CDATA node, which were not charged at all. A 1.75 MB plist of 230,000
<true/> lines was admitted under the 64 MiB decode budget and peaked at
249 MiB of heap; shipped Apple plists peaked at 80-104 MiB.
Charge each element and each text, comment, CDATA or DOCTYPE node at a
measured upper bound, keeping the character and number-projection terms.
The binary estimator already bounds its decoder and is unchanged.
Fixes#8113
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(apple): charge every XML element, attribute and reference
The XML preflight recognized only element names starting with an ASCII
letter, so names such as é, _, a CJK or astral character escaped the
element charge: 100,000 <é/> elements (0.5 MB) built an 84 MiB DOM before
the plist decoder rejected the tag. Attributes and entity or character
references were also uncharged; 20 attributes per element measured about
210 bytes each, and each reference about 26 bytes beyond its characters.
Classify markup by the character after "<" instead of a name pattern, so
anything other than an end tag, comment, CDATA, PI or declaration counts as
an element (saxes rejects malformed markup before the DOM is built). Charge
each unquoted "=" in a start tag as an attribute node and every "&" as a
reference.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(apple): charge XML references only where the parser expands them
The previous commit charged every "&" in the document as a reference,
including literal ampersands inside CDATA sections and comments. A valid
plist with one CDATA string of 1,000,000 ampersands was refused although
it is a single node the decoder holds as one string.
Count "&" during the existing token walk, only in text runs and in the
attribute values of start tags, so comments, CDATA, PIs and declarations
add no reference charge and the scan stays a single pass.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(fit): preserve leading U+FEFF in property strings (#8083)
* fix(environment): retain conflicting scenario MIME uncertainty (#8082)
* fix(environment): retain conflicting scenario MIME uncertainty
* test(mcp): assert structured absolute-path diagnostics
* test(mcp): assert typed selector and stdio diagnostics
* fix(electron): keep literal IPC sends ambiguous beside unresolved handlers (#8120)
* fix(electron): keep literal IPC sends ambiguous beside unresolved handlers
A main handler on an unresolved channel can receive any literal renderer
send of a compatible mode, but pairing ignored it unless it had a scoped
receiver. The send was then reported unpaired, or paired with another
handler. Treat an unresolved-channel handler as a competitor that never
pairs.
The wrapper pass seeds only calls with a literal first argument, so it now
records which invocations of a dynamic site it traced. The syntax finding
gains `invocation_channels: "literal"` only when every invocation is a
traced direct call of an unexported local function supplying a literal
channel. Such a site is already represented by its per-invocation findings
and does not compete.
Fixes#8118
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(electron): keep guarded wrapper registrations competing in IPC pairing
A guarded wrapper registration keeps its predicate on a call path and has
no IPC finding. It still counted toward `invocation_channels: "literal"`,
so pairing lost it: a send on that channel was reported unpaired, or
paired uniquely with a direct handler. The enumerated site now carries
these channels as `guarded_invocation_channels`, and each one competes for
its own channel without pairing.
Index each binding's direct call sites by appending in place. Spreading
the earlier calls on every reference was quadratic in a file's call count.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(ci): remove unused exports that fail deep static checks (#8095)
* fix(ci): remove unused exports that fail deep static checks
Knip on main reports three analysis-view hub re-exports, two type
re-exports, an unused artifact-format guard and an alias constant. Drop
the dead re-exports and guard, and give the debugger control-reply budget
its own literal: it is a separate budget from the caller command-output
default, as its comment already states.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(config): read the Rizin control budget from its own constant
The Rizin control-reply admission limit still took its value from the
caller command-output default, while GDB used DEBUGGER_CONTROL_REPLY_BYTES.
Read the dedicated constant in both so the control budget has one owner.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* perf(javascript): keep one member-effect record per stored fact (#8103)
Member-mutation indexing appended a binding write or escape record for
every traversal state that reached it. States that differ only in
coverage fields the record never stores (origin, owner initializer,
getter widening) produced identical records, and each later receiver
query rescanned all of them. Record each stored fact once, alongside the
existing required-module effect deduplication. Propagation is unchanged.
Effect ordering also re-walked the same top-level statement ancestry for
every candidate record. Cache it with the other initializer positions,
and test the selected path before the costlier receiver comparison.
On the #7774 reproducer the semantic IR and the complete
analyze-javascript-application Evidence are byte-identical, while the
CLI run drops from 146 s to 40 s.
Refs #7774
Co-authored-by: Claude Opus 5.5 <[email protected]>
* feat(native): open extension, framework, XPC and plug-in bundle directories (#8117)
* feat(native): open extension, framework, XPC and plug-in bundle directories
open_binary and the CLI inspection commands admitted only .app
directories and pointed every other Apple bundle at the JavaScript
analyzer. Resolve .appex, .framework, .xpc, .systemextension, .bundle
and .plugin directories through their Info.plist CFBundleExecutable, in
the Contents or flat root layout, plus the versioned framework layout
under Versions/Current. Exactly one layout may declare the bundle and
every resolved path must stay inside it.
Unresolvable bundles fail closed with typed constraints, probed paths
and the Info.plist that was read. Other Apple bundle types are refused as
unsupported instead of being offered the JavaScript route. Resource-root
discovery shares the same layout rule, so inspect_plist reads a
versioned framework's Info.plist. The recognized bundle vocabulary now
has one owner in the domain.
Refs #4310
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(native): refuse kernel extensions and state bundle error fields
Name the Apple bundle types binary routes recognize but do not open
(.dext, .kext, .qlgenerator, .mdimporter) explicitly, so a kernel
extension is refused as unsupported instead of receiving the JavaScript
route. Document which bundle refusals carry probed_paths and info_plist,
and make success tests throw the typed admission error before asserting.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(native): open flattened frameworks through Resources/Info.plist
Electron and Gecko apps ship frameworks without Versions/: the program
file sits at the framework root beside Resources/Info.plist, the layout
#8101 taught bundle anatomy. Read that layout for a framework without
Versions/, and keep the versioned layout for one with it, so a versioned
framework's root Resources link is never read as a second declaration.
The resource root reports it as flattened-framework.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* chore(security): configure dependency and ownership reviews
* perf(javascript): index member-effect paths for projected reads (#8108)
* perf(javascript): index member-effect paths for projected reads
Each projected read scanned every member-write record of its binding and
tested the record path against the selection. On the #7774 reproducer
only 3.7% of scanned records pass that test. Index each binding's
append-only records by path in a lazily built trie that extends itself
with newly appended records, and visit only records whose exact keys a
selected choice names or whose keys are wildcards. Candidates come back
in record order and still pass the exact predicate, now owned next to
the index as semanticEffectPathMatches.
The semantic IR and the complete analyze-javascript-application Evidence
are byte-identical. On the full reproducer each worker phase drops from
about 80 s to 29 s, and to 8 s together with #8103.
Refs #7774
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* perf(javascript): stop escape checks at the first indexed match
The escape check only asks whether any matching escape is unsuperseded,
so it does not need ordered candidates. Share one trie traversal between
the ordered write candidates and an early-exit visitor for this check.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(hopper): distinguish an absent demo dialog from publisher rejection (#8074)
* fix(hopper): distinguish an absent demo dialog from publisher rejection
The Linux demo watcher collapsed an admitted launcher with no first-run dialog into unsupported_demo_dialog. Report that post-authentication timeout as demo_dialog_absent, keep the unsupported dialog code for an observed but unrecognized Hopper window, and allow REA_HOPPER_DEMO_DIALOG_TIMEOUT_MS to widen the fixed wait. Closes#8008.
* fix(hopper): tolerate callers without the dialog timeout argument
* fix(hopper): report an invalid demo dialog timeout as its own failure
An invalid REA_HOPPER_DEMO_DIALOG_TIMEOUT_MS exited 73 and was shown as
invalid_launch_command. Exit 82 is invalid_dialog_timeout. The public error
names the setting, shows the rejected value and source, lists the allowed
range, and says to restart REA.
Drive the demo watcher with controlled windows and time. No window exits 81,
a Hopper window without a supported dialog exits 76, and a dialog that arrives
after the old 10 second limit is still accepted.
* fix(hopper): keep startup stderr capture on AdapterFailure
DemoWatchStop was inserted above with_stderr, so launch failures called a
method they no longer had and became unexpected_failure. Restore with_stderr
on AdapterFailure and reject a watcher fixture that cannot capture stderr.
---------
Co-authored-by: Cursor Agent <[email protected]>
* test(ci): require provider checks and reduce test startup (#8145)
* test(ci): require provider coverage and reuse test sessions
* fix(ci): prepare the runtime before Python producer tests
The fast Python job ran the suite on a fresh checkout, so analysis-efficiency
checks failed without compiled process modules. Install locked dependencies
and prepare the runtime before the suite.
* fix(ci): keep provider test reports inside the checkout
The provider-report reader accepted an arbitrary path from argv. Resolve the
path and reject anything outside the checkout before reading it. The packet
and signal workflows write those reports under .cache.
* fix(ci): require a test module in each named Python bridge family
Discovery still runs every bridge test. The mitmproxy, native, and pwndbg
families must each keep a test module, so deleting one cannot pass through
the others.
---------
Co-authored-by: Cursor Agent <[email protected]>
* fix(javascript): emit no http request for a target that is not a URL (#8147)
* fix(javascript): emit no http request for a target that is not a URL
`node:http` parses a string target with `new URL(...)` before a request object
exists, so a literal that is not an absolute URL throws `ERR_INVALID_URL`
synchronously. The analysis reported `endpoint` at `resolution: "complete"` for
those calls, and the graph then projected `constructs-request` as `resolved`.
Measured on Node 24.22.1:
http.get("nope") -> THREW ERR_INVALID_URL
http.get("example.test/api") -> THREW ERR_INVALID_URL
http.get("/api/items") -> THREW ERR_INVALID_URL
http.get("/api/items", { host }) -> THREW ERR_INVALID_URL
http.get("") -> THREW ERR_INVALID_URL
http.get("//example.test/x") -> THREW ERR_INVALID_URL
http.get("http://") -> THREW ERR_INVALID_URL
http.request("nope") -> THREW ERR_INVALID_URL
https.get("/api/items") -> THREW ERR_INVALID_URL
--- accepted ---
http.get("http://example.test/x") -> OK ClientRequest
http.get("HTTP://example.test/x") -> OK ClientRequest
http.get("http:/x") -> OK ClientRequest
http.get("http:///x") -> OK ClientRequest
http.get("http:x") -> OK ClientRequest
`URL.canParse` is the exact predictor of that branch, so a relative spelling such
as `"http:x"` keeps its operation. It already has precedent in this codebase
(`javascriptPackageImportAliases.ts`).
Two boundaries are deliberate. `fetch` and `new WebSocket` are host-supplied
globals: a browser resolves a relative target against the document URL, so their
spellings are not judged by Node's absolute-URL rule. A protocol mismatch such as
`http.get("ftp://...")` is also left alone because an options object can rescue
it, so it is not a provable throw.
The guard is scoped to `method === "get" | "request"`, which only
`requestMethod`'s `http`/`https` builtin branch can produce.
* chore: retrigger title and readiness checks after the title was corrected
* refactor(ci): unify lane ownership and guard Python tests (#8150)
* refactor(ci): unify lane ownership and guard Python tests
* fix(ci): reject entirely skipped bridge subtests
* docs: correct the raw ARM example and document a Cortex-M language override (#7483)
* docs(raw-binary): replace the STM32-style ARM example
The headline raw example opened ecu.bin at 0x8000000, an STM32 flash
address, while the arm lane decodes in ARM state. Map a 32-bit ARM-state
image at 0x8000 instead, and state beside the example that Thumb-only
images, including Cortex-M firmware, are not supported.
* docs(ghidra): document an interim Cortex-M language override
Ghidra imports little-endian ARM ELF files as ARM:LE:32:v8, which
defaults to ARM state and marks Thumb code only from $t mapping symbols
and odd function-symbol addresses. Unpinned arm ELF already admits
REA_GHIDRA_LANGUAGE_ID, so document ARM:LE:32:Cortex (ARMv6-M, ARMv7-M,
ARMv7E-M) and ARM:LE:32:v8-m (ARMv8-M) as an unverified interim.
State the override's full effect while it is set: raw-binary, DOS, DOL,
PSP and language-pinned lanes such as RISC-V refuse it, and REA asks
Ghidra to import every other target in the ARM language, because it does
not check the override's processor against the target.
* fix(ida): admit architecture families by allowlist (#7485)
IDA refused MIPS, RISC-V, PowerPC, Hexagon and arm64_32 by denylist, so
any family nobody excluded was reported supported, and the label ternary
printed "Hexagon" for any other metadata family. Admit exactly the
families IDA admits today (x86, x86_64, arm, arm64) and name every other
family from its own architecture id. Families that Ghidra admits keep
the pointer to a Ghidra profile.
Refusal text now uses architecture ids, for example "no verified mips
lane" instead of "no verified MIPS lane".
* ci(ghidra): probe M-profile ARM and Xtensa decoding (#7482)
* ci(ghidra): probe M-profile ARM and Xtensa decoding
The embedded plan relies on ARM:LE:32:Cortex, ARM:LE:32:v8-m and
Xtensa:LE:32:default, but CI only exercised ARM:LE:32:v7. Import
source-owned bytes for each language at 0x1000 with -noanalysis and
check how they decode from the language's default context: Cortex and
v8-m must start in Thumb state, v8-m must decode the ARMv8-M MSPLIM
register, and Xtensa must decode entry and retw.n.
Add the probe scripts the job runs, including the existing ARM context
probe, to the ghidra_bridge change scope.
* ci(ghidra): match decode-probe operands as complete operands
The operand expectation searched the instruction text for a substring,
so mrs/4/msp also accepted mrs r0,msplim. Compare it with each operand
Ghidra reports for the instruction, so only a complete operand matches.
The Cortex (msp, lr), v8-M (msplim, lr) and Xtensa (a1) expectations are
unchanged.
* test: guard target format vocabularies and DOL snapshot saves (#7481)
* test(mcp): save and reopen a GameCube DOL snapshot
#7290 reported that DOL Evidence could not be bundled or saved; 50822e1ef
fixed the bundle manifest and added bundle export/import coverage, but no
test saved a DOL session through close_binary. Save a DOL snapshot with
recorded Evidence, check its target and artifact manifest, then reopen it
and replay the read without running the provider again.
* test(binary): require every target format in each target vocabulary
A new target format must be added by hand to the Evidence subject,
bundle manifest, snapshot target and live summary vocabularies; missing
one caused #7290. Give every BinaryTarget format a fixture through a
mapped type, so a new format fails compilation until it has one, and
push each fixture through the producers that persist or project targets.
Caller-selected format hints must name executable target formats.
This is a guard until target formats have a single owner.
* test(mcp): assert the replayed DOL result and the owned replay limitation
The DOL snapshot replay test checked the subject and the provider-call
count but not the answer, so a replay returning null or another value
would pass. Compare the replayed normalized result with the original.
Name the replay limitation as SNAPSHOT_REPLAY_LIMITATION in the snapshot
cache that adds it, and assert that value instead of repeating the
sentence.
* test: list Intel HEX and S-record targets in the vocabulary guard
Upstream added the intel-hex and srec formats (#7136). The guard requires
a fixture for every target format, so after merging main it no longer
typechecked; both formats now have archive fixtures and pass the same
persistence and projection checks.
* test: list UnityFS among the target format fixtures
Upstream #7699 added the unityfs target format, so the vocabulary test's
per-format fixtures must name it; without one the test no longer typechecks
against main.
* test: list uImage among the target format fixtures
Upstream #7234 added the uimage target format, so the vocabulary test's
per-format fixtures must name it; without one the test no longer typechecks
against main.
* test: list the ESP image among the target format fixtures
Upstream #7498 added the esp-image target format, so the vocabulary test's
per-format fixtures must name it; without one the test no longer typechecks
against main.
* fix: correct stale provider refusal and diagnostic text (#7484)
* fix(ghidra): name every admitted family in the architecture refusal
The refusal listed x86, x86_64, arm, arm64, MIPS, RISC-V, raw 6502 and
raw m68k, but the admission set also holds powerpc and hexagon. Build
the text from SUPPORTED_ARCHITECTURES, marking raw-only families, so a
newly admitted family cannot be left out again, and name the refused
family.
* fix(hopper): stop calling the Ghidra Hexagon lane verified
Hopper declined Hexagon targets with "Select a verified Ghidra Hexagon
profile instead", but REA has no verified Hexagon lane. Record whether
each declined family's Ghidra lane is verified, and say that Ghidra
admits Hexagon through an unverified lane.
Name declined families by architecture id, as the Ghidra and IDA
refusals do.
* fix(ghidra): name the pinned language in profile readback refusals
The session readback said "MIPS" whenever a provider-pinned profile
conflicted with the target or with REA_GHIDRA_LANGUAGE_ID, although the
RISC-V, PowerPC and DOL lanes are pinned too. The readback helper was
MIPS-named and re-parsed the language slot that
ghidraProfileLanguageOverride already owns. Move it beside that owner as
a family-neutral readback and name the pinned language id instead.
* fix(ghidra): list DOL among independently verified load images
GameCube DOL load images are independently verified, but the declined
format reason, the inspect_native_load_image capability note and the
Windows P0 guide listed only DOS MZ, DOS COM and raw-binary.
* fix(ghidra): count RV32E's sixteen integer registers
The EF_RISCV_RVE refusal described a 15-register ABI. RV32E has 16
integer registers, x0-x15, as docs/ghidra-riscv.md already says.
* fix(elf): name the refused e_machine
"unsupported ELF architecture" did not say which machine was refused.
Include the numeric e_machine, for example EM_XTENSA 94.
* test(hopper): assert declined-lane facts and owned refusal reasons
The MIPS case banned the phrase "not verified" instead of checking
facts, so wrong advice would still pass. Expose each declined family's
Ghidra lane status, assert those statuses, check that the advice depends
on the status, and compare Hopper's refusals with
hopperDeclinedArchitectureReason. The RISC-V refusal table now compares
the provider's reason with ghidraRiscvUnsupportedReason and expects the
EF_RISCV_RVE declaration name rather than fixed wording.
* test(ghidra): assert verified load-image formats as owned facts
The declined-format test required display phrases such as "DOS MZ" and
"GameCube DOL", so harmless rewording failed it while wrong guidance
containing those names passed. The format list was also spelled three
times: in the attestation's membership check, its reason and the
inspect_native_load_image capability note. Give the verified formats one
owner with the formatters for both texts, assert the list as a fact, and
compare the delivered reason and capability note with those formatters.
* fix(hopper): stop restating which Ghidra lanes REA has verified
Hopper's refusal for families it declines carried its own table of
which Ghidra lanes REA has verified (mips, riscv and powerpc verified,
hexagon not). That restated the Ghidra provider's facts in Hopper and
goes stale whenever they change, the drift that made the Hexagon refusal
wrong in the first place. Reading Ghidra's facts from Hopper would
couple the providers instead; no module in src/hopper imports from
src/ghidra.
The table and hopperDeclinedGhidraLane are removed.
hopperDeclinedArchitectureReason now states only what Hopper owns: it
does not admit the family, and the Ghidra provider's lane reports
whether REA has verified it. The arm64_32 sentence is unchanged.
The test checks every declined family against the formatter: each
reason names the family and Ghidra, and apart from the name all reasons
are the same text, so none states a per-family Ghidra status.
* fix(hopper): route declined families to Ghidra without promising its lane facts
The refusal said the Ghidra lane reports whether REA has verified it, but
Ghidra admits Hexagon with no such note, so a caller could not find the
promised status. Hopper now only states that it does not admit the family
and routes the caller to the Ghidra provider.
---------
Co-authored-by: hhang <[email protected]>
Co-authored-by: Kaoru0822 <[email protected]>
Co-authored-by: Claude Opus 5.5 <[email protected]>
Co-authored-by: ShuaixinHuang <[email protected]>
Co-authored-by: morluto <[email protected]>
Co-authored-by: Henglie <[email protected]>
Co-authored-by: Cursor Agent <[email protected]>
Co-authored-by: 吴佳浩Alben <[email protected]>
Co-authored-by: Entrpi <[email protected]>
Co-authored-by: gensokyo-dev <[email protected]>
* fix(ghidra): name every admitted family in the architecture refusal
The refusal listed x86, x86_64, arm, arm64, MIPS, RISC-V, raw 6502 and
raw m68k, but the admission set also holds powerpc and hexagon. Build
the text from SUPPORTED_ARCHITECTURES, marking raw-only families, so a
newly admitted family cannot be left out again, and name the refused
family.
* fix(hopper): stop calling the Ghidra Hexagon lane verified
Hopper declined Hexagon targets with "Select a verified Ghidra Hexagon
profile instead", but REA has no verified Hexagon lane. Record whether
each declined family's Ghidra lane is verified, and say that Ghidra
admits Hexagon through an unverified lane.
Name declined families by architecture id, as the Ghidra and IDA
refusals do.
* fix(ghidra): name the pinned language in profile readback refusals
The session readback said "MIPS" whenever a provider-pinned profile
conflicted with the target or with REA_GHIDRA_LANGUAGE_ID, although the
RISC-V, PowerPC and DOL lanes are pinned too. The readback helper was
MIPS-named and re-parsed the language slot that
ghidraProfileLanguageOverride already owns. Move it beside that owner as
a family-neutral readback and name the pinned language id instead.
* fix(ghidra): list DOL among independently verified load images
GameCube DOL load images are independently verified, but the declined
format reason, the inspect_native_load_image capability note and the
Windows P0 guide listed only DOS MZ, DOS COM and raw-binary.
* fix(ghidra): count RV32E's sixteen integer registers
The EF_RISCV_RVE refusal described a 15-register ABI. RV32E has 16
integer registers, x0-x15, as docs/ghidra-riscv.md already says.
* fix(elf): name the refused e_machine
"unsupported ELF architecture" did not say which machine was refused.
Include the numeric e_machine, for example EM_XTENSA 94.
* test(hopper): assert declined-lane facts and owned refusal reasons
The MIPS case banned the phrase "not verified" instead of checking
facts, so wrong advice would still pass. Expose each declined family's
Ghidra lane status, assert those statuses, check that the advice depends
on the status, and compare Hopper's refusals with
hopperDeclinedArchitectureReason. The RISC-V refusal table now compares
the provider's reason with ghidraRiscvUnsupportedReason and expects the
EF_RISCV_RVE declaration name rather than fixed wording.
* test(ghidra): assert verified load-image formats as owned facts
The declined-format test required display phrases such as "DOS MZ" and
"GameCube DOL", so harmless rewording failed it while wrong guidance
containing those names passed. The format list was also spelled three
times: in the attestation's membership check, its reason and the
inspect_native_load_image capability note. Give the verified formats one
owner with the formatters for both texts, assert the list as a fact, and
compare the delivered reason and capability note with those formatters.
* fix(hopper): stop restating which Ghidra lanes REA has verified
Hopper's refusal for families it declines carried its own table of
which Ghidra lanes REA has verified (mips, riscv and powerpc verified,
hexagon not). That restated the Ghidra provider's facts in Hopper and
goes stale whenever they change, the drift that made the Hexagon refusal
wrong in the first place. Reading Ghidra's facts from Hopper would
couple the providers instead; no module in src/hopper imports from
src/ghidra.
The table and hopperDeclinedGhidraLane are removed.
hopperDeclinedArchitectureReason now states only what Hopper owns: it
does not admit the family, and the Ghidra provider's lane reports
whether REA has verified it. The arm64_32 sentence is unchanged.
The test checks every declined family against the formatter: each
reason names the family and Ghidra, and apart from the name all reasons
are the same text, so none states a per-family Ghidra status.
* fix(hopper): route declined families to Ghidra without promising its lane facts
The refusal said the Ghidra lane reports whether REA has verified it, but
Ghidra admits Hexagon with no such note, so a caller could not find the
promised status. Hopper now only states that it does not admit the family
and routes the caller to the Ghidra provider.
* test(mcp): save and reopen a GameCube DOL snapshot
#7290 reported that DOL Evidence could not be bundled or saved; 50822e1ef
fixed the bundle manifest and added bundle export/import coverage, but no
test saved a DOL session through close_binary. Save a DOL snapshot with
recorded Evidence, check its target and artifact manifest, then reopen it
and replay the read without running the provider again.
* test(binary): require every target format in each target vocabulary
A new target format must be added by hand to the Evidence subject,
bundle manifest, snapshot target and live summary vocabularies; missing
one caused #7290. Give every BinaryTarget format a fixture through a
mapped type, so a new format fails compilation until it has one, and
push each fixture through the producers that persist or project targets.
Caller-selected format hints must name executable target formats.
This is a guard until target formats have a single owner.
* test(mcp): assert the replayed DOL result and the owned replay limitation
The DOL snapshot replay test checked the subject and the provider-call
count but not the answer, so a replay returning null or another value
would pass. Compare the replayed normalized result with the original.
Name the replay limitation as SNAPSHOT_REPLAY_LIMITATION in the snapshot
cache that adds it, and assert that value instead of repeating the
sentence.
* test: list Intel HEX and S-record targets in the vocabulary guard
Upstream added the intel-hex and srec formats (#7136). The guard requires
a fixture for every target format, so after merging main it no longer
typechecked; both formats now have archive fixtures and pass the same
persistence and projection checks.
* test: list UnityFS among the target format fixtures
Upstream #7699 added the unityfs target format, so the vocabulary test's
per-format fixtures must name it; without one the test no longer typechecks
against main.
* test: list uImage among the target format fixtures
Upstream #7234 added the uimage target format, so the vocabulary test's
per-format fixtures must name it; without one the test no longer typechecks
against main.
* test: list the ESP image among the target format fixtures
Upstream #7498 added the esp-image target format, so the vocabulary test's
per-format fixtures must name it; without one the test no longer typechecks
against main.
* ci(ghidra): probe M-profile ARM and Xtensa decoding
The embedded plan relies on ARM:LE:32:Cortex, ARM:LE:32:v8-m and
Xtensa:LE:32:default, but CI only exercised ARM:LE:32:v7. Import
source-owned bytes for each language at 0x1000 with -noanalysis and
check how they decode from the language's default context: Cortex and
v8-m must start in Thumb state, v8-m must decode the ARMv8-M MSPLIM
register, and Xtensa must decode entry and retw.n.
Add the probe scripts the job runs, including the existing ARM context
probe, to the ghidra_bridge change scope.
* ci(ghidra): match decode-probe operands as complete operands
The operand expectation searched the instruction text for a substring,
so mrs/4/msp also accepted mrs r0,msplim. Compare it with each operand
Ghidra reports for the instruction, so only a complete operand matches.
The Cortex (msp, lr), v8-M (msplim, lr) and Xtensa (a1) expectations are
unchanged.
IDA refused MIPS, RISC-V, PowerPC, Hexagon and arm64_32 by denylist, so
any family nobody excluded was reported supported, and the label ternary
printed "Hexagon" for any other metadata family. Admit exactly the
families IDA admits today (x86, x86_64, arm, arm64) and name every other
family from its own architecture id. Families that Ghidra admits keep
the pointer to a Ghidra profile.
Refusal text now uses architecture ids, for example "no verified mips
lane" instead of "no verified MIPS lane".
* docs(raw-binary): replace the STM32-style ARM example
The headline raw example opened ecu.bin at 0x8000000, an STM32 flash
address, while the arm lane decodes in ARM state. Map a 32-bit ARM-state
image at 0x8000 instead, and state beside the example that Thumb-only
images, including Cortex-M firmware, are not supported.
* docs(ghidra): document an interim Cortex-M language override
Ghidra imports little-endian ARM ELF files as ARM:LE:32:v8, which
defaults to ARM state and marks Thumb code only from $t mapping symbols
and odd function-symbol addresses. Unpinned arm ELF already admits
REA_GHIDRA_LANGUAGE_ID, so document ARM:LE:32:Cortex (ARMv6-M, ARMv7-M,
ARMv7E-M) and ARM:LE:32:v8-m (ARMv8-M) as an unverified interim.
State the override's full effect while it is set: raw-binary, DOS, DOL,
PSP and language-pinned lanes such as RISC-V refuse it, and REA asks
Ghidra to import every other target in the ARM language, because it does
not check the override's processor against the target.
* fix(javascript): emit no http request for a target that is not a URL
`node:http` parses a string target with `new URL(...)` before a request object
exists, so a literal that is not an absolute URL throws `ERR_INVALID_URL`
synchronously. The analysis reported `endpoint` at `resolution: "complete"` for
those calls, and the graph then projected `constructs-request` as `resolved`.
Measured on Node 24.22.1:
http.get("nope") -> THREW ERR_INVALID_URL
http.get("example.test/api") -> THREW ERR_INVALID_URL
http.get("/api/items") -> THREW ERR_INVALID_URL
http.get("/api/items", { host }) -> THREW ERR_INVALID_URL
http.get("") -> THREW ERR_INVALID_URL
http.get("//example.test/x") -> THREW ERR_INVALID_URL
http.get("http://") -> THREW ERR_INVALID_URL
http.request("nope") -> THREW ERR_INVALID_URL
https.get("/api/items") -> THREW ERR_INVALID_URL
--- accepted ---
http.get("http://example.test/x") -> OK ClientRequest
http.get("HTTP://example.test/x") -> OK ClientRequest
http.get("http:/x") -> OK ClientRequest
http.get("http:///x") -> OK ClientRequest
http.get("http:x") -> OK ClientRequest
`URL.canParse` is the exact predictor of that branch, so a relative spelling such
as `"http:x"` keeps its operation. It already has precedent in this codebase
(`javascriptPackageImportAliases.ts`).
Two boundaries are deliberate. `fetch` and `new WebSocket` are host-supplied
globals: a browser resolves a relative target against the document URL, so their
spellings are not judged by Node's absolute-URL rule. A protocol mismatch such as
`http.get("ftp://...")` is also left alone because an options object can rescue
it, so it is not a provable throw.
The guard is scoped to `method === "get" | "request"`, which only
`requestMethod`'s `http`/`https` builtin branch can produce.
* chore: retrigger title and readiness checks after the title was corrected
* test(ci): require provider coverage and reuse test sessions
* fix(ci): prepare the runtime before Python producer tests
The fast Python job ran the suite on a fresh checkout, so analysis-efficiency
checks failed without compiled process modules. Install locked dependencies
and prepare the runtime before the suite.
* fix(ci): keep provider test reports inside the checkout
The provider-report reader accepted an arbitrary path from argv. Resolve the
path and reject anything outside the checkout before reading it. The packet
and signal workflows write those reports under .cache.
* fix(ci): require a test module in each named Python bridge family
Discovery still runs every bridge test. The mitmproxy, native, and pwndbg
families must each keep a test module, so deleting one cannot pass through
the others.
---------
Co-authored-by: Cursor Agent <[email protected]>
* fix(hopper): distinguish an absent demo dialog from publisher rejection
The Linux demo watcher collapsed an admitted launcher with no first-run dialog into unsupported_demo_dialog. Report that post-authentication timeout as demo_dialog_absent, keep the unsupported dialog code for an observed but unrecognized Hopper window, and allow REA_HOPPER_DEMO_DIALOG_TIMEOUT_MS to widen the fixed wait. Closes#8008.
* fix(hopper): tolerate callers without the dialog timeout argument
* fix(hopper): report an invalid demo dialog timeout as its own failure
An invalid REA_HOPPER_DEMO_DIALOG_TIMEOUT_MS exited 73 and was shown as
invalid_launch_command. Exit 82 is invalid_dialog_timeout. The public error
names the setting, shows the rejected value and source, lists the allowed
range, and says to restart REA.
Drive the demo watcher with controlled windows and time. No window exits 81,
a Hopper window without a supported dialog exits 76, and a dialog that arrives
after the old 10 second limit is still accepted.
* fix(hopper): keep startup stderr capture on AdapterFailure
DemoWatchStop was inserted above with_stderr, so launch failures called a
method they no longer had and became unexpected_failure. Restore with_stderr
on AdapterFailure and reject a watcher fixture that cannot capture stderr.
---------
Co-authored-by: Cursor Agent <[email protected]>
* perf(javascript): index member-effect paths for projected reads
Each projected read scanned every member-write record of its binding and
tested the record path against the selection. On the #7774 reproducer
only 3.7% of scanned records pass that test. Index each binding's
append-only records by path in a lazily built trie that extends itself
with newly appended records, and visit only records whose exact keys a
selected choice names or whose keys are wildcards. Candidates come back
in record order and still pass the exact predicate, now owned next to
the index as semanticEffectPathMatches.
The semantic IR and the complete analyze-javascript-application Evidence
are byte-identical. On the full reproducer each worker phase drops from
about 80 s to 29 s, and to 8 s together with #8103.
Refs #7774
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* perf(javascript): stop escape checks at the first indexed match
The escape check only asks whether any matching escape is unsuperseded,
so it does not need ordered candidates. Share one trie traversal between
the ordered write candidates and an early-exit visitor for this check.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* feat(native): open extension, framework, XPC and plug-in bundle directories
open_binary and the CLI inspection commands admitted only .app
directories and pointed every other Apple bundle at the JavaScript
analyzer. Resolve .appex, .framework, .xpc, .systemextension, .bundle
and .plugin directories through their Info.plist CFBundleExecutable, in
the Contents or flat root layout, plus the versioned framework layout
under Versions/Current. Exactly one layout may declare the bundle and
every resolved path must stay inside it.
Unresolvable bundles fail closed with typed constraints, probed paths
and the Info.plist that was read. Other Apple bundle types are refused as
unsupported instead of being offered the JavaScript route. Resource-root
discovery shares the same layout rule, so inspect_plist reads a
versioned framework's Info.plist. The recognized bundle vocabulary now
has one owner in the domain.
Refs #4310
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(native): refuse kernel extensions and state bundle error fields
Name the Apple bundle types binary routes recognize but do not open
(.dext, .kext, .qlgenerator, .mdimporter) explicitly, so a kernel
extension is refused as unsupported instead of receiving the JavaScript
route. Document which bundle refusals carry probed_paths and info_plist,
and make success tests throw the typed admission error before asserting.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(native): open flattened frameworks through Resources/Info.plist
Electron and Gecko apps ship frameworks without Versions/: the program
file sits at the framework root beside Resources/Info.plist, the layout
#8101 taught bundle anatomy. Read that layout for a framework without
Versions/, and keep the versioned layout for one with it, so a versioned
framework's root Resources link is never read as a second declaration.
The resource root reports it as flattened-framework.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Member-mutation indexing appended a binding write or escape record for
every traversal state that reached it. States that differ only in
coverage fields the record never stores (origin, owner initializer,
getter widening) produced identical records, and each later receiver
query rescanned all of them. Record each stored fact once, alongside the
existing required-module effect deduplication. Propagation is unchanged.
Effect ordering also re-walked the same top-level statement ancestry for
every candidate record. Cache it with the other initializer positions,
and test the selected path before the costlier receiver comparison.
On the #7774 reproducer the semantic IR and the complete
analyze-javascript-application Evidence are byte-identical, while the
CLI run drops from 146 s to 40 s.
Refs #7774
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(ci): remove unused exports that fail deep static checks
Knip on main reports three analysis-view hub re-exports, two type
re-exports, an unused artifact-format guard and an alias constant. Drop
the dead re-exports and guard, and give the debugger control-reply budget
its own literal: it is a separate budget from the caller command-output
default, as its comment already states.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(config): read the Rizin control budget from its own constant
The Rizin control-reply admission limit still took its value from the
caller command-output default, while GDB used DEBUGGER_CONTROL_REPLY_BYTES.
Read the dedicated constant in both so the control budget has one owner.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(electron): keep literal IPC sends ambiguous beside unresolved handlers
A main handler on an unresolved channel can receive any literal renderer
send of a compatible mode, but pairing ignored it unless it had a scoped
receiver. The send was then reported unpaired, or paired with another
handler. Treat an unresolved-channel handler as a competitor that never
pairs.
The wrapper pass seeds only calls with a literal first argument, so it now
records which invocations of a dynamic site it traced. The syntax finding
gains `invocation_channels: "literal"` only when every invocation is a
traced direct call of an unexported local function supplying a literal
channel. Such a site is already represented by its per-invocation findings
and does not compete.
Fixes#8118
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(electron): keep guarded wrapper registrations competing in IPC pairing
A guarded wrapper registration keeps its predicate on a call path and has
no IPC finding. It still counted toward `invocation_channels: "literal"`,
so pairing lost it: a send on that channel was reported unpaired, or
paired uniquely with a direct handler. The enumerated site now carries
these channels as `guarded_invocation_channels`, and each one competes for
its own channel without pairing.
Index each binding's direct call sites by appending in place. Spreading
the earlier calls on every reference was quadratic in a file's call count.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(apple): charge the XML plist decoder's DOM before decoding
The XML preflight charged 4 bytes per character and 256 bytes per start
or end tag. The decoder builds an @xmldom/xmldom DOM that measured
1,000-1,160 bytes per element and up to 240 bytes per text, comment or
CDATA node, which were not charged at all. A 1.75 MB plist of 230,000
<true/> lines was admitted under the 64 MiB decode budget and peaked at
249 MiB of heap; shipped Apple plists peaked at 80-104 MiB.
Charge each element and each text, comment, CDATA or DOCTYPE node at a
measured upper bound, keeping the character and number-projection terms.
The binary estimator already bounds its decoder and is unchanged.
Fixes#8113
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(apple): charge every XML element, attribute and reference
The XML preflight recognized only element names starting with an ASCII
letter, so names such as é, _, a CJK or astral character escaped the
element charge: 100,000 <é/> elements (0.5 MB) built an 84 MiB DOM before
the plist decoder rejected the tag. Attributes and entity or character
references were also uncharged; 20 attributes per element measured about
210 bytes each, and each reference about 26 bytes beyond its characters.
Classify markup by the character after "<" instead of a name pattern, so
anything other than an end tag, comment, CDATA, PI or declaration counts as
an element (saxes rejects malformed markup before the DOM is built). Charge
each unquoted "=" in a start tag as an attribute node and every "&" as a
reference.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(apple): charge XML references only where the parser expands them
The previous commit charged every "&" in the document as a reference,
including literal ampersands inside CDATA sections and comments. A valid
plist with one CDATA string of 1,000,000 ampersands was refused although
it is a single node the decoder holds as one string.
Count "&" during the existing token walk, only in text runs and in the
attribute values of start tags, so comments, CDATA, PIs and declarations
add no reference charge and the scan stays a single pass.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(captures): admit browser-serialized HAR request URLs
[email protected] checks request.url with ajv-formats' strict RFC 3986
`uri` format. Browsers record and send URLs as WHATWG serializes them,
which keeps characters such as `|` and `{` unescaped in the query, so one
Google Fonts or GraphQL GET entry made inspect-web-network-capture and
environment-build reject the whole HAR.
HAR 1.2 defines request.url only as the absolute request URL. Own that
correction in the HAR validator: every URL the RFC 3986 check accepts
stays accepted, and a URL that WHATWG URL serialization reproduces
unchanged is accepted too. Spellings WHATWG would rewrite (a backslash
authority, a tab, a space) and relative references still fail.
Fixes#8110
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(captures): state that RFC 3986 request URLs stay admitted
The HAR request URL rule is "passes RFC 3986, or WHATWG serialization
reproduces it unchanged". A spelling WHATWG would rewrite therefore fails
only when it also fails the RFC 3986 check; an uppercase host still
passes. Say so in the docs and the validator comment, and pin the
accepted uppercase-host case in the CLI test.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
The CommonJS loader search treated any repeated state on its own ancestor
chain as an unknown loader. A self-default such as
`session.messages = session.messages || []` forms that cycle, because a
member read inside the assigned value was attributed to the later call's
read site and selected the same write again. Each `push` on the value was
then reported as an unresolved `require` load, and graph coverage became
partial.
Record a member read inside a flowed value as a depth-1 captured selection
at its own position, as member mutations already do for flowed origins.
Treat a revisited state as adding no source; the cycle's other entries are
still expanded once each. Remove the superseded ancestor tracking.
Fixes#8102
Co-authored-by: Claude Opus 5.5 <[email protected]>
webpack 5 and tsc write "sourceRoot": "" for map-relative sources. Since
5f702dd6c the shared join added "/" for any non-null root, so these
sources resolved from the URL root: rxjs's ../../../src/internal/Observable.ts
under /npm/[email protected]/dist/cjs/internal/ became /src/internal/Observable.ts,
and xterm's webpack://xterm/./src/... became https://host/webpack://xterm/....
Treat an empty root like a null root in sourceMapReference, the single join
used by trace-web-source-location, analyze-javascript-application and the
web source-map fetcher. Chrome, Firefox and trace-mapping skip an empty root,
and tc39/ecma426#183 wrote the current steps to match them. Non-empty roots,
including "/" and repeated slashes (#6726), keep their existing behavior.
Fixes#8106
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(desktop): charge retained rows at their measured representation
The desktop scan reserved 1024 + 24 bytes per path character for every
row, omission and the root against its 32 MiB representation budget. Real
rows cost 800-940 heap and 490-640 JSON bytes, so the charge overstated
them 6-13 times and large macOS bundles stopped at 3,700-7,200 entries,
far below max_entries.
Charge each path at the larger of its UTF-16 heap size and its escaped
JSON size, plus the existing fixed row and omission overheads, in one
helper. The bound still covers the retained heap and the serialized
report, including control-character paths that expand six times in JSON.
Fixes#8104
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(desktop): charge every retained copy of a path
The tighter row charge counted each path once, but rows repeat it in
convention-fact values and the next probes those facts offer, and host
failure issues name the absolute path. Control-character paths under a
convention such as Content/Paks or resources/app.asar therefore serialized
47-80 MiB of reports within the 32 MiB budget.
Reserve each copy through the same text helper where it is stored: path
convention facts with their row, the probes any fact offers (at most two,
pinned by a projection test) with that fact, retained read ranges before
their read, and probe failure issues before they are recorded.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
A macOS framework without Versions/ keeps its binary, _CodeSignature/ and
Resources/Info.plist directly in the .framework, as Electron and Gecko apps
ship it. Bundle anatomy looked for a shallow bundle's plist only at its root
and reported info_plist_path: null under complete coverage.
Resolve a shallow bundle's root Info.plist first, then Resources/Info.plist,
the same layout a framework version directory uses. Keep the iOS platform
rule tied to a root-level Info.plist so a Resources/-only application root
is not classified as iOS.
Fixes#8100
Co-authored-by: Claude Opus 5.5 <[email protected]>
createAnalysisViewEvidence chose view confidence from the view result kind
only, so every summary, page or item view of an inferred parent
(compare_source_to_bundle, inspect_evm_interface, partially resolved
inspect_il2cpp_metadata) was sealed as derived while keeping the parent's
authority. A projection never makes its facts more certain: an inferred
parent now yields inferred views through CLI and MCP alike. Observed and
derived parents keep their existing view confidence and identities.
Fixes#8098
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(kuna): report document failures as typed codes
Kuna JSON, graphs, xrefs, spans, and command checks threw Error and
published the sentence. Those refusals are now KunaDocumentError codes
with the field or path that failed. Output errors keep the code as
reason and the same facts in diagnostics.
* fix(kuna): keep incoming-reference address fields
---------
Co-authored-by: Cursor Agent <[email protected]>
* refactor(javascript): break the semantic analysis import cycle
Split shared semantic facts out of the analysis hub so projectors of
effects, evaluation, and identity no longer import one another in a cycle.
* refactor(javascript): share the import specifier reader
Read static imports in one module used by the boundary checker and both
cycle guards, so those rules cannot drift apart.
---------
Co-authored-by: Cursor Agent <[email protected]>
* refactor(analysis-view): break the analysis view hub cycle
Move request fragments, the request union, and the result union out of
analysisView.ts so projectors no longer import the module that imports them.
The dispatcher re-exports the public schemas and seals projector results.
Co-authored-by: morluto <[email protected]>
* fix(analysis-view): check inferred projector results
Require each projector result to match the unsigned result union before
sealing it, and parse the cycle guard's imports with Babel so comments
and side-effect imports are not misread.
* fix(analysis-view): close IL2CPP and Android symbol view inputs
Admit those projectors through AdmittedView and check test selections
against their request schemas, so misspelled fields fail at the call site.
---------
Co-authored-by: Cursor Agent <[email protected]>
Co-authored-by: morluto <[email protected]>
* feat(debugger): share GDB and Rizin session caps as REA_ settings
Move the duplicated deadline guidance into the owned debugger helper and
replace the shared session, queue, output, and 120s command caps with
launch settings that name their value, source, and range.
* test(gdb): assert the command-output setting on truncated exit evidence
The retained-output limitation now names REA_DEBUGGER_MAX_COMMAND_OUTPUT_BYTES
instead of the old fixed-cap sentence.
* test(gdb): assert command-output limit fields on truncated exit
A truncated GDB exit now carries the command-output setting, value, and
source on the cleanup error. The coverage test reads those fields.
* fix(debugger): keep control replies and evidence hashes inside their budgets
Point the session-cap mutation target at DEFAULT_DEBUGGER_MAX_SESSIONS.
Give startup and cleanup replies a fixed 16 MiB budget so a small caller
output cap cannot truncate them. Lower the command-output maximum so
escaped Evidence JSON stays under Node's string ceiling, and keep GDB
startup timeout diagnostics when the error is rebuilt.
* test(debugger): throw typed Result errors before success checks
Success cases no longer assert ok and return. They throw the typed
error first, then check session, deadline, and close fields.
---------
Co-authored-by: Cursor Agent <[email protected]>
* refactor(admission): own inspector, web, and CLI state per host
Replace the process-global inspector, web-bundle, guidance, and JavaScript
counters with objects a host passes in, so independent instances do not
share admission or cleanup.
* fix(admission): keep worker and file cleanup on the host
Web-bundle and JavaScript workers retry only their host registry, and
each Inspector admission authorizes and closes its own file scope.
Browser providers can share one web-bundle admission.
* test(inspector): gate one admission's file cleanup
The isolation test delayed the first ArtifactResourceScope.retryCleanup
call, so a second call on a shared ElectronFileScope still finished.
Hold first.files.retryCleanup until the other admission's close
finishes, then release and join before restoring the method.
---------
Co-authored-by: Cursor Agent <[email protected]>
* fix(domain): replace bare parser errors with typed codes
OpenStep syntax, canonical JSON digests, managed-graph source checks,
frame fingerprints, and reference-flow invariants threw Error and
published the sentence. Each refusal now carries a code. Tests assert
that code, including OpenStep position, instead of the previous prose.
* test(domain): pin OpenStep refusal codes and positions
The syntax cases accepted any code and any offset. Each container
refusal now expects its code and source position, including the deep
suffix that changes (x from unterminated_array to expected_separator.
Scalar failures stay identical to the pinned decoder.
* test(filesystem): assert OpenStep keyed-archive codes
The deep-suffix keyed archive cases compared ArtifactReaderFailure
message text with the pinned decoder sentence. They now require
reason format and the OpenStep code and source position on the cause.
---------
Co-authored-by: Cursor Agent <[email protected]>
* refactor(adb): move device and client operations out of the provider switch
Keep AdbProvider as the session owner. Device listing, device inspection,
and client identity now run in AdbDeviceOperations, and shared listing
execution construction lives next to the client identity result.
Co-authored-by: morluto <[email protected]>
* refactor(adb): use the shared client and provider identity types
Drop the ClientIdentity and ProviderIdentity aliases so method signatures
name AdbClientResult and AdbProviderIdentity directly.
Co-authored-by: morluto <[email protected]>
---------
Co-authored-by: Cursor Agent <[email protected]>
Co-authored-by: morluto <[email protected]>
* fix(errors): store resource-constraint recovery as one shape
AnalysisResourceConstraintError kept remediationAction, fixedAllowance,
and retryable as independent optionals, so a refusal could claim both a
fixed allowance and a separate remediation. The constructor now stores
the existing ResourceConstraintRecovery union, and the three accessors
read that object.
* test(errors): cover competing resource-constraint recovery fields
A runtime options object can still carry both a fixed allowance and a
remediation action. The constructor keeps the allowance and drops the
other field; the regression now supplies both.
---------
Co-authored-by: Cursor Agent <[email protected]>
* fix(native)!: report NativeAOT failures as reason codes
PeImage admission failures and parser recoveries used to publish
Error.message as NativeAotReport.reason. The report is now a union on
status, and reason is a typed code with the fields that code needs.
Complete reports keep reason null and a located header. Partial and
not_applicable reports carry a code. Caller budget mistakes throw
AnalysisInputError or AnalysisResourceConstraintError with a diagnostics
code instead of a bare Error.
* fix(native): keep out-of-domain hydration pointers partial
An inline pointer outside the unsigned 64-bit domain made
writeBigUInt64LE throw RangeError, and the hydration catch rethrew it
instead of recording a partial report. Those targets are now an
admission detail. The load-image summary schema still accepts the
optional Ghidra extension's string reason.
---------
Co-authored-by: Cursor Agent <[email protected]>
* refactor(artifacts): describe formats in one registry
Reader construction, classification retention, container role, and
extraction now read one descriptor per domain format id. unknown has
no reader, and zip package families share one capability row.
* fix(artifacts): keep reader presence in format descriptor types
Descriptor helpers no longer widen every format to an optional reader,
so ArtifactReaderFormat stays the union of formats that construct one.
---------
Co-authored-by: Cursor Agent <[email protected]>
* refactor(contracts): register CLI and MCP tools by family id
Walk the published tool-family list once for each adapter so a new
family id must supply both a CLI registrar and an MCP registrar.
* fix(cli): keep family registration off the contract tables
Walk family ids from the startup-safe list, and register the web script
commands with the application family that owns their MCP contracts.
---------
Co-authored-by: Cursor Agent <[email protected]>
* fix(adb): verify pushes to long device paths
sha256sum echoes the device path, so a correct digest line for a path
near 1 KiB exceeded the fixed 1 KiB capture budget. The resulting
resource failure was then reported as device_digest_source "unavailable",
the same outcome as a device without sha256sum. Size the budget to the
path, and return a capture-budget failure as a resource constraint
instead of an unverified success.
Fixes#3369
* Keep adbDeviceAnalysis.ts as on main
* test(adb): keep only the long-path digest test
Main already covers an oversized checksum answer as a resource
constraint with the completed push retained.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(native): preserve Mach-O target selection in macOS tools
* fix(native): retain exact Mach-O slice signature coverage
* fix(native): admit x86_64 LIB64 signature selectors
* test(native): match signature checks to actual slices
* fix(native): recognize vtool refusals on private Mach-O aliases
Match vtool's container refusal to the operand the command received so a
parenthesized universal path keeps completed facts and the refusal
limitation. Expect codesign -a only when the target format is universal.
---------
Co-authored-by: Cursor Agent <[email protected]>
* fix(session): align resource-only status and snapshot recovery
* fix(session): report independent workflows during resource sessions
* test(mcp): honor SQLite runtime availability
* test: compare resource snapshot guidance with its formatter
Assert projected error codes and the owning remediation action for
resource-session snapshot refusals, and type MCP error text as a closed
content item.
---------
Co-authored-by: Cursor Agent <[email protected]>
* fix(hopper): reconcile observed caller relationships
* fix(hopper): retain completed caller scans on failure
Attach the native reply and collected forward facts when reconciliation
fails or is cancelled, and check scan stop codes separately from their
wording. Live observations keep REA evidence ids rather than provider
supplied ones, and the real Hopper verifier uses advertised MCP schemas.
* fix(hopper): answer caller-scan methods in the package fixture
The packaged Hopper stand-in echoed unknown bridge methods, so
analyze_function rejected rea_caller_scan_target as an invalid scan
target and the native package baseline returned unreadable_output.
* fix(hopper): share caller identities in failed scan observations
Failure partials repeated each caller identity under every destination.
The page budget counted compact replies, then schema parsing copied the
expanded form. Identities are stored once and destinations cite addresses.
* refactor(config): own the Hopper caller-scan budget default
The environment schema hard-coded the 2000 ms scan default. The domain
limit owner now names that millisecond budget and the schema reuses it.
* test(mcp): assert bundle path issues by field
compare_bundles reports each relative path as an invalid_format issue
instead of one joined diagnostic sentence.
* test(flutter): assert ABI selector refusals as structured issues
inspect_dart_aot reports the ABI label as an issue on path abi. The
message is the selector contract, without an abi: prefix.
* fix(session): retain caller observations that cannot be replayed
A missing document selector blocks snapshot replay. Backward call-graph
and trace reads still keep the native result and forward observations,
and the workflow Evidence links those records.
* fix(session): keep decompile bodies out of the in-memory ledger
Caller and callee walks still stay in Evidence without a document.
procedure_pseudo_code does not, so decompile_all leaves each body in the
file-backed export instead of retaining the whole export in memory.
* fix(hopper): own caller-scan recovery and the page byte allowance
Scan facts stay in the domain. Environment recovery, including the time
budget bounds and the fixed allowances with no runtime override, is
appended from configuration. Setup forwards the budget, and the bridge
uses one byte allowance for instructions, native callers, and pages.
* fix(hopper): retain compact caller references within the scan budget
Recovered call sites stay in the address form the scan already counted.
Expanded identity copies are measured before allocation and refused when
they exceed the retention budget. Instruction lists are not part of the
retained reply.
* fix(session): omit instruction payloads from retained caller evidence
Documentless caller and callee reads stay in the ledger for the session.
Their evidence keeps call sites and drops instruction lists the scan
already consumed.
* fix(hopper): keep the owning caller address on each compact call site
A page reference marked source_procedure "caller" belongs to that page
item's caller. Saved sites record that address, and an owner missing from
the callee inventory is stored once instead of copied onto every site.
* fix(session): leave Hopper reply shaping to the Hopper adapter
Documentless caller evidence stores the adapter's raw result unchanged.
The session no longer reads native_result.instructions.
* test(hopper): throw typed scan failures before reading results
Successful caller-scan results now throw the typed error instead of
returning after an ok check. Expected failures assert structured tags,
stop codes, and reported limits.
---------
Co-authored-by: Cursor Agent <[email protected]>
* feat(artifacts): inventory and extract Bun standalone executable module graphs
Adds a Bun standalone artifact reader for Mach-O, PE, ELF and appended
trailers, supporting the 36-byte and 52-byte module record layouts. It
preserves module names, entry-point details and raw metadata, and
exposes contents, sourcemaps, bytecode and module-info as separate
entries for safe extraction without executing the binary. Parsing
rejects malformed or ambiguous graphs and enforces module-count,
retained-byte and expansion limits. Adds synthetic parser tests, CLI and
MCP coverage, agent skill guidance, and a verification script for
Bun-produced executables.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016y53PdpyMAp9RSW65fatHe
* fix(artifacts): bound Bun section probes and reject truncated sections
Split Bun recognition into a bounded probe that reads only section
headers, length prefixes, and the Offsets/trailer suffix, and a single
payload load used by the reader. Inventory classification and extraction
admission no longer allocate the whole graph.
Match ELF `.bun` section names by comparing only the name bytes and
terminator, so many headers sharing one long name cost constant work each.
Reject sections whose declared extent ends past EOF instead of reading a
silently shortened window.
Launch Bun, Docker, and REA in the standalone verifier through the shared
owned verifier commands, and release the runtime root only after those
command families have settled.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016y53PdpyMAp9RSW65fatHe
* style(artifacts): format Bun reader integration
* fix(contracts): describe static Bun inventory facts
* fix(artifacts): bound Bun metadata reads and combined path growth
---------
Co-authored-by: Matt Van Horn <[email protected]>
Co-authored-by: Claude Opus 5.5 <[email protected]>
Co-authored-by: gensokyo-dev <[email protected]>
* fix(cli): report incomplete debugger cleanup as typed outcomes
debug-with-gdb and debug-with-rizin hand-rolled final cleanup in
try/finally blocks: a failed or interrupted final close only logged
and assigned the process exit status, so callers still received
complete successful Evidence with no typed error or partial
observation.
Both commands now route through the shared close owner used by the
other one-shot CLI commands (deliverCliAnalysisEvidence /
finishAnalysisCommand) via a small closeDebuggerCommandSession
helper. A rejected closeAll projects as ProviderCleanupError, which
the owner turns into cleanup_incomplete retaining the completed
Evidence as partial observation; a signal observed after finalization
projects as cancelled with the same retained observation. The
per-session close failure still only logs, preserving the rule that
the retained owner's closeAll decides completion.
Fixes#7378. Fixes#7429.
* test(cli): reject unmodeled GDB requests in cleanup regressions
* test(cli): match modeled GDB requests exactly
* Update tests/boundary/cli/gdbStartupCleanupCli.test.ts
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
---------
Co-authored-by: morluto <[email protected]>
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
* fix(firmware): preserve symlink target bytes
* fix(apple): preserve undecodable dylib link targets
* style(tests): fix CI formatting
* test(firmware): select the named contract directly
* test(firmware): record raw bytes for a decodable symlink target
* test(mcp): assert bundle path issues by field
compare_bundles reports each relative path as an invalid_format issue
instead of one joined diagnostic sentence.
* test(firmware): stat raw links without following them
access follows a symlink, so a published dangling raw-link still looked
absent. lstat and readlink inspect the link itself.
---------
Co-authored-by: Cursor Agent <[email protected]>
* fix(cli): preserve literal Markdown rendering
* fix(cli): retain literal Markdown attributes
* fix(cli): preserve literal punctuation in markdown
* test(cli): update escaped empty object expectations
* style(cli): format the literal empty-object Markdown expectation
* test(mcp): assert bundle path issues by field
compare_bundles reports each relative path as an invalid_format issue
instead of one joined diagnostic sentence.
* test(flutter): assert ABI selector refusals as structured issues
inspect_dart_aot reports the ABI label as an issue on path abi. The
message is the selector contract, without an abi: prefix.
---------
Co-authored-by: Cursor Agent <[email protected]>
* fix(javascript)!: preserve operator influence in semantic traces
* test(javascript): align semantic influence regressions
* style(javascript): format semantic IR
* test(javascript): follow literal argument seed influence
* fix(javascript): keep one literal node per operator operand
Reuse an existing literal for operator influence, and treat TypeScript
assertion wrappers as transparent when deciding whether a call result is used.
* test(mcp): assert bundle path issues by field
compare_bundles reports each relative path as an invalid_format issue
instead of one joined diagnostic sentence.
* fix(javascript): index operator literals by range and initializer
Repeated string operands rescanned every retained literal and every
initializer flow. Lookup now uses those indexes, and a 1200-declaration
file still projects one literal per operand.
* fix(javascript): keep literal and call facts on their own occurrences
Operator literals match their own source range, so a nested function does
not inherit the enclosing declaration's literal. A direct argument still
reaches the call, while the assigned result follows modeled returns.
for-loop initializers and updates are discarded call results.
* test(javascript): expect conditional operand literals beside binding values
A conditional binding keeps one candidate literal per union value and one
operator literal per operand token. Those are different facts, so the
aggregate payload check counts four nodes and compares literal seeds in
query order.
* perf(javascript): index call return flows by call site
projectCalls filtered every call return for each assigned result. Group
those flows once while projecting returns-to-call edges.
---------
Co-authored-by: Cursor Agent <[email protected]>
* docs(apk): clarify rotated signer agreement
* test(mcp): assert bundle path issues by field
compare_bundles reports each relative path as an invalid_format issue
instead of one joined diagnostic sentence.
* fix(javascript): ignore worker frames for a different source job
A failure frame for another job was dequeued as an output error, so a
later caller abort lost the race and was not reported as cancellation.
* test(flutter): assert ABI selector refusals as structured issues
inspect_dart_aot reports the ABI label as an issue on path abi. The
message is the selector contract, without an abi: prefix.
* fix(javascript): check each worker frame condition once
A ready frame is rejected on its own. A non-ready frame for another job
keeps waiting. The following id comparison repeated that skip, so it
never ran.
---------
Co-authored-by: Cursor Agent <[email protected]>
* fix(browser): preserve source-map declaration uncertainty
* style(tests): fix CI formatting
* fix(browser): retain known source map differences
* fix(web): bind map comparisons to script metadata
* fix(browser): match source maps with partial script metadata
Keep each script's observed hash and declaration constraints instead of
dropping a metadata field for the whole URL group.
* test(flutter): assert ABI selector refusals as structured issues
MCP input validation reports the ABI path and the schema message separately.
The old concatenated "abi: ..." sentence is no longer the delivered error.
* fix(web): prove large source-map changes without factorial search
Groups larger than eight scripts were reported as unexplained, so nine
x.map declarations plus one omission compared as unknown against ten
y.map declarations. Declaration counts reject impossible placements, and
augmenting paths assign the rest without retrying identical scripts.
* test(mcp): assert bundle path issues by field
compare_bundles reports each relative path as an invalid_format issue
instead of one joined diagnostic sentence.
* fix(web): match repeated source-map scripts by count
Identical declarations share one group, and a shared omission pool is
counted once. A 2000-script self-compare no longer builds a dense
assignment graph.
---------
Co-authored-by: Cursor Agent <[email protected]>
* feat(environment): serve the writes callers make without recorded form fields
- A create member a recording sent empty ("", null, false, [] or {}) is no
longer required; a call that leaves it out is applied as sending that value
(body.blank).
- List views render envelope members that list the listed entities' ids
(Paperless's `all`), so creates in such collections are confirmed.
- Updates move their entity in collections kept sorted by a field they write,
and a failed whole-collection read after an update can teach that order.
- A later create that stores an unruled field otherwise is withheld at the
create, and an item read may fill values a create's answer had not loaded.
- An unsent field respelled between null, [] and {} is unchanged.
- The skill asks for one create with every field and one with only what a
person must give.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(skill): show how to read signed-in pages before recording
Agents guessed capture keys (artifacts, dom, final) to see pages behind
sign-in, which inspect_web_page cannot reach.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): accept every empty spelling recorded creates sent
A create that fills only what a person must give sends `null` where the
full create sent a string, and `null` or `false` for the same flag. The
merged plan now accepts each recorded empty value, and a call that leaves
the member out is applied as the earliest call sent it.
Also adds the spec folder for this change.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(skill): keep the vary-fields step neutral on UI or API writes
Agents read "the form" as a cue to write through the UI, where some
applications' request bodies are not exposed; a body-unusable UI write
now re-records once with request steps.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record verification for blank members and sorted writes
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): address review of blank members and sorted writes
- Bulk updates move their entities in turn against the others' earlier
values and install the final order once, as replay does.
- A member's recorded empty values stay accepted once a later call teaches
its default; a left-out member with a default keeps the default.
- Write answers spell empty lists and maps as their recorded answers did.
- A rename can teach a sort when the creates chose one by a field the
update did not write, such as ids that ascend with titles.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(environment): serve the writes that later reads explain
- An item read whose object is marked deleted (deleted, isDeleted or
is_deleted true) agrees with an absent entity, as EspoCRM answers a
deleted record; such an object never seeds or adopts an entity.
- maxSize, max_results and maxResults are page sizes, for whole-collection
list lines and the list query's limit role.
- An item view also shows the fields only reads after its type's first write
show, when no initial entity holds them (Vikunja's task subscription),
without requiring them; write answers leave such a field out for entities
without it, and it no longer withholds the write as incomplete-entities.
- The skill asks updates to change the fields a person would change, and
expects a deleted object to read back as 404 or marked deleted.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer sent empty members as sent, and keep a placed create's order
- A write answer spells an empty list or map the call did not send as its
recorded answer did; one the call sent is answered as sent, so a create
sending `[]` is not answered `null` because the first recorded create was.
- A read listing the whole collection that placed a create leaves the create
only the sorted inserts whose order that read follows, so a later rename
cannot teach a sort an earlier read ruled out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer bulk updates without fields only later reads show
A bulk update's answer renders each entity with the same allowance a single
update's answer has: a field only reads after the type's first write showed,
which no initial entity holds, is left out for an entity without it, so a
bulk update of a seeded task is no longer refused for lacking it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): correct how the EspoCRM contact update task passes
The agent trajectories show gpt-oss passes it by creating a replacement contact
and deleting the original, since the served update refuses emailAddress; the
update itself is refused in both builds.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer a read of a served-deleted entity marked deleted
An application that soft-deletes still answers a deleted record by id, marked
deleted. Serving now keeps an entity a served delete removed from a type whose
recorded item reads show that, until a reset or load and within the size
data/add loads, and answers a read of it with its last fields and the flag
true, instead of refusing the read.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep unloaded create answers, check bulk answers with sent members, narrow inserts only on collection-order reads
- A create whose item read loaded values its answer had not records them
(`observation.read.unloaded`), and a served create's answer keeps them as
the recorded answer showed them, a zero-time owner included, while the state
holds the read's.
- Replay checks a bulk update's recorded answers with each target's sent
members, as serving answers them, so echoed `null` and `[]` no longer make
the plan body-unusable.
- A whole-collection read whose query sorts it shows the query's order, so it
no longer narrows a create's sorted insert candidates.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record the census with main's unknown-clock rule merged
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(environment): serve query numbers, counted values and made objects
- A query parameter accepts a number or boolean, which serving sends as its
JSON text, as agents send page sizes and flags.
- A sequence may start at the value its first create took above every value
its type held (Vikunja's per-project task index), and a later create's
composed, template or started-sequence rule is adopted when it also spells
the first create's value.
- A made-object rule explains an object the application makes with each
created entity: members the request stamped, members holding the entity's
own id, and counts one past the held objects' values.
- A create's answer names the id of the one object a one-element array member
holds, so bulk creates link to their type.
- The skill reads each listed object by id before writing.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(environment): let a linked create take path values it stores
A create linked to its type's own collection, whose path has no parameters,
accepts any path values when its plan stores each one in a field, as a Vikunja
task stores its project; replay and serving apply it alike.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn a sort from a rename only when moving its entities reaches it
Serving moves only the entities a write changes in a collection kept sorted,
so replay now tries a sort after a rename only when the entities no write the
read exercised changed already follow it. Seeded folders listed out of title
order no longer teach a title sort that serving would answer with another
order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep unloaded create answers a sent value did not echo, and learn no sort from query-sorted reads
- A served create keeps a value its recorded answer had not loaded unless the
call sent that member and the plan stores it as sent, so a create sending
`active: false` is answered `false` as recorded while the state holds the
read's `true`.
- A rename no longer teaches a sort from a whole-collection read its query
sorts, whose ties need not follow the collection's own order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): verify query values, counted values and made objects
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn the soft-delete flag in the state view and serve by it
Extraction now records the member later item reads set true on a record the
application soft-deleted, and the item view carries it as `deleted_flag`.
Serving reads that fact instead of scanning recorded bodies itself, so one
owner decides what a soft delete looks like.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record the call-values checks at #727493821024
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn the soft-delete flag from item reads with a query too
A client may read a deleted record back only with a query, such as a
cache-busting parameter. Every successful read of the item tool now shows the
flag; only plain later reads still add fields.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record the call-values checks at #727445832406
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn sorts only from orders a list read confirmed
- A sort retry after a failed whole-collection read requires the entities no
write changed since a list last showed them to follow it already, rather than
excusing every entity the update's tool ever wrote: an earlier rename a list
confirmed in place stays where it was.
- An update takes its creates' sort only from a create a list read showed
placed; replay still tries a sort a create guessed from the initial order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): drop entities read back as deleted and answer soft deletes as deleted
- A held entity a later read shows marked deleted is dropped, not adopted with
the marked object's fields.
- Bulk-update replay renders answers with the optional and later-only fields
serving uses, so a bulk update of created and seeded entities replays.
- A served delete of a soft-deleting type answers with the entity marked
deleted, single and bulk.
- A deleted flag only query reads showed is a field of the item view, so a
served tombstone keeps it where live records omit it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): check made objects as recorded and keep their ids and clocks honest
- A made object's clock counts as a write clock, so a capture with no clock
anchor withholds the create instead of stamping 1970.
- Members holding the entity's identity follow the id written, generated or
sent, through the type's id field.
- A later create's made object is checked as recorded, before its own members
are rewritten, so one naming another entity withholds the plan.
- A later create's template is adopted only when it spells that create's value
too, not just when the first create's text holds its source value.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): move updated entities only by placed sorts and keep query-sorted reads from confirming order
Replay moved an updated entity by a sort a create only guessed, while the
update plan exported only placed sorts, so a read could confirm an update
by a move serving would not make. Both now use the placed sort. A passing
list read clears its entities from the moved set only when the state
decides the order of every entity it lists, so a read whose query sorts
it no longer hides an earlier move from a later sort check.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): count six withheld plans that explain more, not seven
The bulk task create counted among them is served on this branch.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): try a learned sort by moving each pending write in turn, as serving does
#retrySort sorted the whole collection, which a list could match while
serving, moving each renamed entity in turn against the others' values
before their own writes, gave another order. The state now records the
values each write replaced, and the retry replays the writes since a
list in collection order last showed their entities.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer a served delete with the deleted flag its recorded answer showed
An application may answer a delete with the record as it was and mark it
deleted only on later reads. Delete answers, single and bulk, now take
the flag's value from the recorded answer; reads of a served-deleted
entity still show it marked deleted.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): set aside two hunks next to lines main's saved-values change rewrites
Main's #7267 changes the lines beside the blank-member docs and the
empty-spelling check in explainUpdate. Setting them aside lets that merge
apply cleanly; the commit after it puts them back on top of main's text.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): restore the set-aside hunks on main and bump the skill to 51
Puts the blank-member docs and the empty-spelling check back beside main's
slug and composed text and heldSlugRule. Main is at skill 50 with
different text, so this branch's skill changes take 51.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): bump the skill to 52 above #7260's 51
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): put back the call-values hunks main's merge took its side for, and bump the skill to 53
The merge of main kept main's side where it overlapped this branch, so it
could commit without a conflict. This restores this branch's side next to
main's: made-object agreement beside the & slug spellings, sequence starts
in the earlier evidence (now built by earlierOf), the combined value
imports, and recording step 2 beside main's UI-first step 3. Main is at
skill 50 and #7274 at 52, so this branch takes 53.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep sorts a list showed in place ruled out, and created answers' nested members as recorded
A whole list in collection order that showed a renamed entity where a
sort by the renamed field would have moved it now rules that sort out, so
a later rename of the same entity cannot teach it. A create's answer
leaves out members its item read added inside an object the answer
showed, as the recorded answer did; the state still stores them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(environment): serve writes whose calls send different members
Recordings now vary what a type's two creates and updates send, and REA
withheld writes where a later call left out what the first sent. An update
plan's required members narrow to those every call sent; a member one create
sent and another left for the application to number is a sequence that
takes the value a call sends; a PUT's left-out member the application
cleared becomes that member's default; and an update's clock is held as
the read after it spells it. The skill spaces creates (skill 54).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep a later create's made object only with clocks its request stamped
followsMadeObjects only checked that a made object's clocks parsed, so a
later create whose object showed an older clock still passed, while serving
stamps the write's clock. Its clocks must now be in the rule's format and
within the stamp window of that create's request.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): accept sent values either way, check defaults against every call, and store update clocks as reads spell them
A sequence that takes sent values now accepts one a later create sent
after the first left the member to be numbered. A default an update learned
must agree with every update that left its member out, else the plan is
withheld. An update plan whose read-back spells a stamped clock at a coarser
precision keeps that read as observation.read, and serving stores the
update's clocks in its spelling while answers keep the response's.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): check recorded member and sequence values
* fix(environment): compare soft-deleted read fields
* fix(environment): reject contradictory learned sequences
* fix(environment): preserve initially deleted item observations
* fix(environment): index stored deleted entities separately
* fix(environment): replay lists that include deleted records
* feat(environment): serve sorted creates, listed references and toggles
A create a whole listing placed last is placed again by the sort a later
listing shows. A field holding the entities a sent list of ids names is a
reference with `many`, and a create that sent the member empty agrees with
one that named entities. A scenario request whose body Playwright did not
expose and that names no Content-Type sent no body. An item action calls
show flipping a boolean both ways learns a `toggle` rule. Skill 55 asks
recordings to run an action twice on one object.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): count a sequence number a create sends, as replay does
Serving now records a number a call sends for a `sequence` with `sent` in the
same count history it generates from, so after that entity is deleted the next
generated number still follows it, as replay's takeSequences does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record the final gate run in the verification
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): make no count of a made-object member earlier creates held alike
A count only a later create shows starts at the exemplar's value, which is
right only while the exemplar is the one create accepted: creates accepted
after it showed the member unchanged, which a count would have moved. The
merge now refuses such a count, as the top-level sequence check does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): restate the verification after merging #7539
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): check PUT defaults against reads, and respell update clocks by serving's rule
A default a PUT learns must also agree with what the read after an earlier or
the current call showed for the member its answer left out, so a call whose
read showed the member kept withholds a later clearing default.
Replay and serving now choose an update clock's stored spelling by one rule,
readSpellsCoarser: the read after it at a coarser precision in the same zone
and separator. A read in another zone or separator is no longer adopted by
replay, which serving could not store, and a create's plain clocks stay as
answered, as serving makes them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep sorts earlier listings ruled out, and empty added references
A sort a whole listing that placed a create did not follow stays ruled out for
that create, though the entity that showed it is later deleted. A create that
sends a reference's member empty after a later create added the reference
agrees with it, as with a reference the first create showed.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): restate the verification after the review round
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): leave out a reference the observed create left out
A create the plan observes that left out a reference's member stored no such
field, so a served call leaving the member out stores none either, though a
later recorded create sent it empty; a call sending it empty still stores that.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): restate the verification after #7539 merged main
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep the largest sequence value in replay, and store update clocks by the kept read
Replay now keeps the largest value a sequence took, as serving's commitWrite
does, so a lower sent value no longer lowers the next number.
Serving stores every update's clocks in the spelling of the read the plan
keeps as observation.read; replay now does the same through
readSpelledClocks, instead of adopting each update's own read, so a later
update whose read spells its clock otherwise contradicts the plan. The plain
clock branch of restampedFields, which no caller used any more, is gone.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): restate the verification after merging #75393b800a10
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep learned sorts, references, and toggles faithful
Do not adopt a sort an earlier update already ruled out. Learn a list
reference only when every element shows the same nonempty fields. Promote
a toggle only when every call started from a boolean.
* test(environment): compare list-reference fields with matching values
The mixed-field case uses the held tag's color, so rejection comes from
the field set rather than a value the recording never stored.
* test(mcp): assert bundle path issues by field
The compare_bundles rejection reports each relative path as an
invalid_format issue instead of one joined diagnostic sentence.
* test(mcp): assert field issues for bundle paths and Dart ABI
Relative bundle paths and rejected ABI selectors are reported as
input issues on those fields, not as one joined diagnostic sentence.
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Co-authored-by: morluto <[email protected]>
Co-authored-by: Cursor Agent <[email protected]>
* fix(process): corroborate Linux exits after procfs environ read races
A token read failing with ENOENT/ESRCH may race with process exit while
native identity observation still sees a live PID. Recheck native lifetime
and accept absence only when corroborated; keep unverifiable or reused
PIDs fail closed.
Refs #7790
* fix(process): recheck Linux exit after final token retry
A process can exit after the final retry of a failed procfs environment
read. Re-observe native identity before producing final cleanup failures,
and keep PID replacement and uncertain identity fail-closed.
Add five integration-facing sweep regression cases for absence, live,
unavailable, replaced and zombie-replaced candidates.
Refs #7790
* fix(process): bound native exit settlement for token-open races
After a terminal ENOENT/ESRCH run-token read, allow a tightly bounded
native-only settlement window to corroborate pending Linux process exit.
Retain an explicit failure for live/unavailable or changed PID identities,
without inferring exit from the token error itself.
Cover delayed absence, delayed replacement and bounded live retries.
Refs #7790
* fix(process): share native token exit settlement
* fix(process): signal only groups with validated members
* fix(process): retain lineage through exit settlement
* fix(process): type mixed identity observations in exit checks
Map inference collapsed a readable/absent union onto the absent
overload after main tightened ProcessIdentityObservation.
* fix(process): refresh groups whose members exit during revalidation
A same-group child forked after the revalidation snapshot is not an
empty group, so cleanup reads the table again before skipping the signal.
* test(process): arm group-refresh exit from the revalidation snapshot
The regression follows the table read after planning observed the
launcher token, and requires a later refresh to see the forked child.
* test(mcp): assert bundle path issues by field
The compare_bundles rejection reports each relative path as an
invalid_format issue instead of one joined diagnostic sentence.
* test(mcp): assert field issues for bundle paths and Dart ABI
Relative bundle paths and rejected ABI selectors are reported as
input issues on those fields, not as one joined diagnostic sentence.
---------
Co-authored-by: Harshit Sharma <[email protected]>
Co-authored-by: Cursor Agent <[email protected]>
* refactor(macho): extract non-throwing container and slice load-command readers
Split readMachoImage into readMachoContainer, which classifies a file as
thin, FAT or not Mach-O and keeps FAT range issues per slice, and
readSliceLoadCommands, which reads one slice header and frames its load
commands, returning every command framed before a failure.
readMachoImage keeps its strict policy on top of them: commands framed
before a framing defect still decode first, an unsupported FAT slice
still yields to a later malformed one, and the first FAT range issue is
still reported. A characterization test pins that precedence for inputs
with several defects; it passes against both the previous and the new
implementation.
Part of #4306.
* feat(macho): report per-slice encryption status in inspect_macho
inspect_macho now reports the LC_ENCRYPTION_INFO(_64) state of every
slice in a new `encryption` collection, read in process from load
commands without reading code bytes.
- Each slice is encrypted, not-encrypted, absent, malformed or unknown.
Absent requires a completed load-command walk; malformed and unknown
carry a reason and keep commands read before the failure.
- The named range is end-exclusive in slice and input-file coordinates
and maps to file-backed segment fragments, VM addresses and the
overlapping nonempty sections.
- A ciphertext range adds a limitation naming its input-file bytes.
- The in-process read and the native tools are bracketed by one regular
file descriptor and stat identity checks; a change reports
artifact_changed. A missing or inaccessible target keeps the tool
facts and reports encryption as non-exhaustive, never absent.
- Retained Evidence recorded before this field is parsed by the owning
schema and reports encryption as unknown in the macho view and in
managed/native verification.
Fixes#4306
* fix(macho): keep slice facts and failures under large tables and close errors
- Frame load commands through a visitor instead of retaining every
command, restoring the pre-refactor memory profile (2M eight-byte
commands: peak heap 370 MiB -> 0.9 MiB).
- Confirm a large FAT container from a valid slice's Mach-O header so a
sibling's range fault stays per slice instead of hiding healthy slices.
- Name only sections whose bytes the range and their segment both cover.
- Throw inspection failures inside the owned descriptor callback so a
close failure keeps them as its primary error.
* fix(macho): map encrypted ranges only within segment VM extents
mapRange gave every file-backed segment fragment VM addresses, even for a
segment with no VM size, whose file bytes dyld never maps, and past vmsize
when a segment declared more file bytes than memory.
Read vmsize with the segment. A segment with no VM size keeps its fragment
with null VM addresses. A file size beyond the VM size, or a VM range past
the address space, makes the mapping unknown with a reason while the slice
keeps its encryption status.
The docs also claimed the encryption read never reads code bytes. Like
every Mach-O container read, a 0xcafebabe input of at most 16 MiB is read
whole to rule out a JVM class file; say so.
* fix(macho): read encryption through the admitted descriptor facts
Main's withArtifactRegularFile now hands callers the admitted handle and
its initial stats. Use those facts instead of re-stating the handle, and
pass the legacy Evidence payload straight to createEvidence.
* fix(macho): retain a finished report when the target close fails
The file owner keeps the completed callback value when the descriptor close
fails or cancellation arrives during close. inspect_macho now validates that
value and retains the report as the error's partial observation, as
inspect_signature does, instead of returning only the cleanup or
cancellation error.
* test(native): supply readEncryption in merged dyld refusal cases
main's new dyld refusal cases build NativeMachoInspectionContext without
the encryption reader this branch requires.
* test(views): bind the Mach-O encryption page to its facet
main's facet-ownership case builds inspect_macho Evidence without the
encryption coverage this branch requires; the fixture now carries one
slice row and the loop checks the encryption page like the other facets.
* fix(macho): bound encryption commands and range mapping across slices
A slice retains at most two encryption load commands, since a second one
already makes it malformed; the reason keeps the full count. Range mapping
in every slice shares one budget of segment and section records, charged
from each segment command's capacity before it is decoded. A slice past
the budget keeps its encryption status and reports an unknown mapping.
* style(macho): format encryption reader and binding test
* fix(macho): stop retaining segments past the mapping budget
Segment commands are collected during framing only while their mapping
records still fit in the budget shared by all slices, and the budget is
charged only when the slice maps its range.
* docs(macho): clarify retained mapping allowance
* fix(ci): own the Mach-O encryption test in Apple scope
---------
Co-authored-by: t <t@t>
Co-authored-by: gensokyo-dev <[email protected]>
* feat(environment): serve the writes callers make without recorded form fields
- A create member a recording sent empty ("", null, false, [] or {}) is no
longer required; a call that leaves it out is applied as sending that value
(body.blank).
- List views render envelope members that list the listed entities' ids
(Paperless's `all`), so creates in such collections are confirmed.
- Updates move their entity in collections kept sorted by a field they write,
and a failed whole-collection read after an update can teach that order.
- A later create that stores an unruled field otherwise is withheld at the
create, and an item read may fill values a create's answer had not loaded.
- An unsent field respelled between null, [] and {} is unchanged.
- The skill asks for one create with every field and one with only what a
person must give.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(skill): show how to read signed-in pages before recording
Agents guessed capture keys (artifacts, dom, final) to see pages behind
sign-in, which inspect_web_page cannot reach.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): accept every empty spelling recorded creates sent
A create that fills only what a person must give sends `null` where the
full create sent a string, and `null` or `false` for the same flag. The
merged plan now accepts each recorded empty value, and a call that leaves
the member out is applied as the earliest call sent it.
Also adds the spec folder for this change.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(skill): keep the vary-fields step neutral on UI or API writes
Agents read "the form" as a cue to write through the UI, where some
applications' request bodies are not exposed; a body-unusable UI write
now re-records once with request steps.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record verification for blank members and sorted writes
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): address review of blank members and sorted writes
- Bulk updates move their entities in turn against the others' earlier
values and install the final order once, as replay does.
- A member's recorded empty values stay accepted once a later call teaches
its default; a left-out member with a default keeps the default.
- Write answers spell empty lists and maps as their recorded answers did.
- A rename can teach a sort when the creates chose one by a field the
update did not write, such as ids that ascend with titles.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(environment): serve the writes that later reads explain
- An item read whose object is marked deleted (deleted, isDeleted or
is_deleted true) agrees with an absent entity, as EspoCRM answers a
deleted record; such an object never seeds or adopts an entity.
- maxSize, max_results and maxResults are page sizes, for whole-collection
list lines and the list query's limit role.
- An item view also shows the fields only reads after its type's first write
show, when no initial entity holds them (Vikunja's task subscription),
without requiring them; write answers leave such a field out for entities
without it, and it no longer withholds the write as incomplete-entities.
- The skill asks updates to change the fields a person would change, and
expects a deleted object to read back as 404 or marked deleted.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer sent empty members as sent, and keep a placed create's order
- A write answer spells an empty list or map the call did not send as its
recorded answer did; one the call sent is answered as sent, so a create
sending `[]` is not answered `null` because the first recorded create was.
- A read listing the whole collection that placed a create leaves the create
only the sorted inserts whose order that read follows, so a later rename
cannot teach a sort an earlier read ruled out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer bulk updates without fields only later reads show
A bulk update's answer renders each entity with the same allowance a single
update's answer has: a field only reads after the type's first write showed,
which no initial entity holds, is left out for an entity without it, so a
bulk update of a seeded task is no longer refused for lacking it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): correct how the EspoCRM contact update task passes
The agent trajectories show gpt-oss passes it by creating a replacement contact
and deleting the original, since the served update refuses emailAddress; the
update itself is refused in both builds.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer a read of a served-deleted entity marked deleted
An application that soft-deletes still answers a deleted record by id, marked
deleted. Serving now keeps an entity a served delete removed from a type whose
recorded item reads show that, until a reset or load and within the size
data/add loads, and answers a read of it with its last fields and the flag
true, instead of refusing the read.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep unloaded create answers, check bulk answers with sent members, narrow inserts only on collection-order reads
- A create whose item read loaded values its answer had not records them
(`observation.read.unloaded`), and a served create's answer keeps them as
the recorded answer showed them, a zero-time owner included, while the state
holds the read's.
- Replay checks a bulk update's recorded answers with each target's sent
members, as serving answers them, so echoed `null` and `[]` no longer make
the plan body-unusable.
- A whole-collection read whose query sorts it shows the query's order, so it
no longer narrows a create's sorted insert candidates.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record the census with main's unknown-clock rule merged
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(environment): serve query numbers, counted values and made objects
- A query parameter accepts a number or boolean, which serving sends as its
JSON text, as agents send page sizes and flags.
- A sequence may start at the value its first create took above every value
its type held (Vikunja's per-project task index), and a later create's
composed, template or started-sequence rule is adopted when it also spells
the first create's value.
- A made-object rule explains an object the application makes with each
created entity: members the request stamped, members holding the entity's
own id, and counts one past the held objects' values.
- A create's answer names the id of the one object a one-element array member
holds, so bulk creates link to their type.
- The skill reads each listed object by id before writing.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(environment): let a linked create take path values it stores
A create linked to its type's own collection, whose path has no parameters,
accepts any path values when its plan stores each one in a field, as a Vikunja
task stores its project; replay and serving apply it alike.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn a sort from a rename only when moving its entities reaches it
Serving moves only the entities a write changes in a collection kept sorted,
so replay now tries a sort after a rename only when the entities no write the
read exercised changed already follow it. Seeded folders listed out of title
order no longer teach a title sort that serving would answer with another
order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep unloaded create answers a sent value did not echo, and learn no sort from query-sorted reads
- A served create keeps a value its recorded answer had not loaded unless the
call sent that member and the plan stores it as sent, so a create sending
`active: false` is answered `false` as recorded while the state holds the
read's `true`.
- A rename no longer teaches a sort from a whole-collection read its query
sorts, whose ties need not follow the collection's own order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): verify query values, counted values and made objects
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn the soft-delete flag in the state view and serve by it
Extraction now records the member later item reads set true on a record the
application soft-deleted, and the item view carries it as `deleted_flag`.
Serving reads that fact instead of scanning recorded bodies itself, so one
owner decides what a soft delete looks like.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record the call-values checks at #727493821024
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn the soft-delete flag from item reads with a query too
A client may read a deleted record back only with a query, such as a
cache-busting parameter. Every successful read of the item tool now shows the
flag; only plain later reads still add fields.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record the call-values checks at #727445832406
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn sorts only from orders a list read confirmed
- A sort retry after a failed whole-collection read requires the entities no
write changed since a list last showed them to follow it already, rather than
excusing every entity the update's tool ever wrote: an earlier rename a list
confirmed in place stays where it was.
- An update takes its creates' sort only from a create a list read showed
placed; replay still tries a sort a create guessed from the initial order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): drop entities read back as deleted and answer soft deletes as deleted
- A held entity a later read shows marked deleted is dropped, not adopted with
the marked object's fields.
- Bulk-update replay renders answers with the optional and later-only fields
serving uses, so a bulk update of created and seeded entities replays.
- A served delete of a soft-deleting type answers with the entity marked
deleted, single and bulk.
- A deleted flag only query reads showed is a field of the item view, so a
served tombstone keeps it where live records omit it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): check made objects as recorded and keep their ids and clocks honest
- A made object's clock counts as a write clock, so a capture with no clock
anchor withholds the create instead of stamping 1970.
- Members holding the entity's identity follow the id written, generated or
sent, through the type's id field.
- A later create's made object is checked as recorded, before its own members
are rewritten, so one naming another entity withholds the plan.
- A later create's template is adopted only when it spells that create's value
too, not just when the first create's text holds its source value.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): move updated entities only by placed sorts and keep query-sorted reads from confirming order
Replay moved an updated entity by a sort a create only guessed, while the
update plan exported only placed sorts, so a read could confirm an update
by a move serving would not make. Both now use the placed sort. A passing
list read clears its entities from the moved set only when the state
decides the order of every entity it lists, so a read whose query sorts
it no longer hides an earlier move from a later sort check.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): count six withheld plans that explain more, not seven
The bulk task create counted among them is served on this branch.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): try a learned sort by moving each pending write in turn, as serving does
#retrySort sorted the whole collection, which a list could match while
serving, moving each renamed entity in turn against the others' values
before their own writes, gave another order. The state now records the
values each write replaced, and the retry replays the writes since a
list in collection order last showed their entities.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer a served delete with the deleted flag its recorded answer showed
An application may answer a delete with the record as it was and mark it
deleted only on later reads. Delete answers, single and bulk, now take
the flag's value from the recorded answer; reads of a served-deleted
entity still show it marked deleted.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): set aside two hunks next to lines main's saved-values change rewrites
Main's #7267 changes the lines beside the blank-member docs and the
empty-spelling check in explainUpdate. Setting them aside lets that merge
apply cleanly; the commit after it puts them back on top of main's text.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): restore the set-aside hunks on main and bump the skill to 51
Puts the blank-member docs and the empty-spelling check back beside main's
slug and composed text and heldSlugRule. Main is at skill 50 with
different text, so this branch's skill changes take 51.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): bump the skill to 52 above #7260's 51
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): put back the call-values hunks main's merge took its side for, and bump the skill to 53
The merge of main kept main's side where it overlapped this branch, so it
could commit without a conflict. This restores this branch's side next to
main's: made-object agreement beside the & slug spellings, sequence starts
in the earlier evidence (now built by earlierOf), the combined value
imports, and recording step 2 beside main's UI-first step 3. Main is at
skill 50 and #7274 at 52, so this branch takes 53.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep sorts a list showed in place ruled out, and created answers' nested members as recorded
A whole list in collection order that showed a renamed entity where a
sort by the renamed field would have moved it now rules that sort out, so
a later rename of the same entity cannot teach it. A create's answer
leaves out members its item read added inside an object the answer
showed, as the recorded answer did; the state still stores them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(environment): serve writes whose calls send different members
Recordings now vary what a type's two creates and updates send, and REA
withheld writes where a later call left out what the first sent. An update
plan's required members narrow to those every call sent; a member one create
sent and another left for the application to number is a sequence that
takes the value a call sends; a PUT's left-out member the application
cleared becomes that member's default; and an update's clock is held as
the read after it spells it. The skill spaces creates (skill 54).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep a later create's made object only with clocks its request stamped
followsMadeObjects only checked that a made object's clocks parsed, so a
later create whose object showed an older clock still passed, while serving
stamps the write's clock. Its clocks must now be in the rule's format and
within the stamp window of that create's request.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): accept sent values either way, check defaults against every call, and store update clocks as reads spell them
A sequence that takes sent values now accepts one a later create sent
after the first left the member to be numbered. A default an update learned
must agree with every update that left its member out, else the plan is
withheld. An update plan whose read-back spells a stamped clock at a coarser
precision keeps that read as observation.read, and serving stores the
update's clocks in its spelling while answers keep the response's.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): check recorded member and sequence values
* fix(environment): compare soft-deleted read fields
* fix(environment): reject contradictory learned sequences
* fix(environment): preserve initially deleted item observations
* fix(environment): index stored deleted entities separately
* fix(environment): replay lists that include deleted records
* fix(environment): count a sequence number a create sends, as replay does
Serving now records a number a call sends for a `sequence` with `sent` in the
same count history it generates from, so after that entity is deleted the next
generated number still follows it, as replay's takeSequences does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): make no count of a made-object member earlier creates held alike
A count only a later create shows starts at the exemplar's value, which is
right only while the exemplar is the one create accepted: creates accepted
after it showed the member unchanged, which a count would have moved. The
merge now refuses such a count, as the top-level sequence check does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): check PUT defaults against reads, and respell update clocks by serving's rule
A default a PUT learns must also agree with what the read after an earlier or
the current call showed for the member its answer left out, so a call whose
read showed the member kept withholds a later clearing default.
Replay and serving now choose an update clock's stored spelling by one rule,
readSpellsCoarser: the read after it at a coarser precision in the same zone
and separator. A read in another zone or separator is no longer adopted by
replay, which serving could not store, and a create's plain clocks stay as
answered, as serving makes them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep the largest sequence value in replay, and store update clocks by the kept read
Replay now keeps the largest value a sequence took, as serving's commitWrite
does, so a lower sent value no longer lowers the next number.
Serving stores every update's clocks in the spelling of the read the plan
keeps as observation.read; replay now does the same through
readSpelledClocks, instead of adopting each update's own read, so a later
update whose read spells its clock otherwise contradicts the plan. The plain
clock branch of restampedFields, which no caller used any more, is gone.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep a left-out default from numbering, a read's sub-millisecond digits, and references named by member
A member a call left out and the application stored a fixed value for is no
longer turned into a numbered sequence by a later call that left it out.
Serving shortens an update's clock to the read's precision from the answer's
own digits, keeping microseconds. A PUT that names a reference field's entity
by its member supplies that field, for the default check and in serving.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): count a reference a later PUT names toward the default an earlier one cleared
The default check now reads the reference rules the current call shows as
well as the plan's, so a PUT naming the owner by ownerId after one that
cleared it is served, as in the other order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): number a left-out member past every sent value, and never fill it with a sent blank
A later create the application numbered at or below a value an earlier create
sent no longer turns the echo into a numbered sequence, since serving counts
past every sent value. A member a numbered sequence covers is not filled with
an empty value another call sent when a call leaves it out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): count sent values in the sequence's scope before numbering a left-out member
The largest value earlier creates sent is kept by type and by collection, and
a collection-scoped sequence adopted later checks only its own collection, so
a number sent in another collection no longer withholds a create numbered
within its own.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Co-authored-by: morluto <[email protected]>
An Objective-C category whose class field is null, or is bound to a
symbol other than `_OBJC_CLASS_$_`, was recorded as partial with
`category_class_unresolved`, but no issue was raised, so the
`objc_properties_categories` coverage facet reported `partial` with a
null reason. swiftc emits such categories for @objc members of an
extension on another module's Swift class and binds the class field to
the class metadata symbol (`_$s…CN`); Keynote and Notability carry them.
Report the reference as a `category_class_unresolved` issue naming the
bound symbol, or stating that the reference is null, matching the issue
already raised for an unreadable local class reference.
TYPE_NAMES stopped at flat_dt (8), while the U-Boot include/image.h
commit that the tables are pinned to defines image types 0-47. A header
declaring a later type, such as kernel_noload (14) or tfa-bl31 (45),
was reported with type_name "unknown".
Add types 9-47 with their mkimage -T short names from boot/image.c at
the same commit. IH_TYPE_LOADABLE (22) has no short name there and is
named "loadable", as other tables already name identifiers U-Boot does
not list. Identifiers past the enum stay "unknown".
Reuse the pinned SDK's combined array-element and object-member admission
before canonical validation, deferred runtime loading and tool execution.
Share a configurable finite allowance across eager/deferred HTTP and stdio,
and project refusals to bounded, actionable transport resource errors.
Preserve canonical CLI/application contracts, caller-selected input values,
ordinary validation semantics and complete Evidence access. Exercise actual
transport boundaries, real artifact/TShark workflows, packaged installation,
continued service and downstream identity/capacity regressions.
Fixes#7882
BREAKING CHANGE: MCP tool arguments now default to 1,024 combined array
elements and object members per call. Select a larger positive finite
REA_MCP_MAX_INPUT_ELEMENTS allowance and restart for larger inline inputs,
or use retained/file selectors where supported. Inputs are never truncated.
Assisted-by: Reimu (gensokyo-skills:reimu-incident-triage)
Assisted-by: Kogasa (gensokyo-skills:kogasa-surprise-testing)
Assisted-by: Yukari (gensokyo-skills:yukari-boundary-analysis)
Node's CommonJS loader matches registered extensions case-sensitively, so
require("./data.JSON") and require("./addon.NODE") run as JavaScript.
Derive target_json_status and the native loads relation from the exact
loader suffix instead of the inventory's case-folded file kind.
Fixes#7623
inspect-apple-ips and inspect-pe-imports were the two commands in the
binary diagnostics group whose positional target was still a plain
string. An empty argument passed CLI validation, was resolved to the
working directory, and failed later as a non-file input, unlike their
siblings (inspect-elf-notes, inspect-pe-certificates and others) that
already reject it at validation.
Declare both targets with the shared nonempty, NUL-free local path
schema the rest of the file uses. Explicit "." is unchanged.
Fixes#7421
* fix(javascript): emit no fork when a literal stdio array omits the IPC channel
Node's fork() requires an explicit stdio array to carry the IPC channel:
if (!ArrayPrototypeIncludes(options.stdio, 'ipc'))
throw new ERR_CHILD_PROCESS_IPC_REQUIRED('options.stdio')
The throw is synchronous, before any child process exists, but the analysis
reported a resolved fork with stdioMode "array" — byte-identical to the valid
`[..., "ipc"]` form — which projects to a child-process node, a resolved spawns
relation and resolved connects-stdio.
Measured on Node 24.22.1:
fork({stdio: ["inherit","inherit","inherit"]}) -> THROWS ERR_CHILD_PROCESS_IPC_REQUIRED
fork({stdio: ["pipe","pipe","pipe"]}) -> THROWS ERR_CHILD_PROCESS_IPC_REQUIRED
fork({stdio: ["ipc"]}) -> SPAWNED connected=true
fork({stdio: ["inherit","inherit","inherit","ipc"]}) -> SPAWNED connected=true
fork({stdio: "pipe"}) -> SPAWNED connected=true
spawn({stdio: ["inherit","inherit","inherit"]})-> SPAWNED connected=false
A string stdio is normalised with "ipc" appended, so only the array form can omit
it. The same array is valid for spawn, which has no IPC requirement.
Drop a fork candidate only when its stdio array is fully literal and provably
lacks "ipc"; a spread or a non-literal element may still supply the channel and
stays a candidate.
* chore: retrigger title and readiness checks after the title was corrected
* feat(artifacts): inventory and extract U-Boot FIT images
open_binary refused U-Boot Flattened Image Tree files as an unsupported
format, so inspect_artifact and extract_artifact could not list or
recover a FIT's kernel, devicetree, ramdisk and loadable images.
Recognize an FDT v16/v17 with a top-level /images node and parse it with
the existing parseDevicetree. Each /images child becomes an inventory
entry named images/<unit-name>.bin that extracts its stored bytes from
embedded data, data-offset (past the 4-aligned devicetree end) or
data-position (from the file start). Image format_metadata records the
declared type, arch, os, compression, load and entry; root metadata
records the description, timestamp and configurations.
crc32, md5, sha1, sha256, sha384 and sha512 hash nodes are verified in
one streamed pass per image; other algorithms are reported unverified.
Signatures are reported, not verified, and nothing is decompressed,
decrypted or executed. Data outside the file, overlapping or duplicate
image data, a missing or repeated data source and hash mismatches fail
closed. The devicetree is bounded by the fixed 16 MiB allowance;
external image data is streamed from disk.
Refs #7129
* fix(artifacts): bound FIT configuration references and suggest analysis only for plain images
* fix(artifacts): retain buffered FIT stream cancellation
* test(artifacts): gate FIT faults on the observed read phase
* fix(artifacts): verify FIT links and summary identities
* fix(artifacts): bound FIT diagnostics and retain unknown signatures
* fix(artifacts): compare FIT declarations and resolved data links
* fix(artifacts): admit FIT raw probes through the shared loader contract
* fix(artifacts): preserve FIT field availability independently
* fix(artifacts): isolate nested entry metadata from reader facts
---------
Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: gensokyo-dev <342580529+users.noreply.github.com>
Local file commands in the core binary, native, managed and native-tool
CLI groups declared their positional target as a plain string. An empty
argument therefore reached path admission, where it resolved to the
working directory, as docs/cli.md says it never should.
Declare those targets with the shared nonempty, NUL-free local path
schema already used by analyze, so the CLI rejects them as a
VALIDATION_ERROR naming the field before any analysis runner is called.
Explicit "." is unchanged.
Fixes#7436Fixes#7404
* test(firmware): exercise the staging ceiling through the emitted refusal
The ceiling recovery added in #6971 was covered through the wording helper only, so a wiring change that passed the schema ceiling where the selected value belongs, or that dropped the remediation, would have stayed green.
Adds boundary coverage that calls inventoryFirmwareOutput itself: a sparse file reaches the real byte ceiling without storing the bytes and asserts the projected refusal carries the ceiling recovery, and a one-byte tree below the ceiling asserts the raise advice. Verified red on both regressions — passing the schema ceiling for the selected value fails the below-ceiling case, and restoring the single-form remediation fails both.
* test(firmware): compare the emitted recovery with its owner
The wiring test matched three phrases of the emitted remediation, so an ad-hoc or truncated action that kept those phrases would still pass while the emitted recovery had diverged from firmwareStagingBudgetRemediation.
Assert exact equality with the owner's output for the selected value in both cases instead. Verified red: replacing the emitted action with a divergent sentence that keeps the ceiling phrases now fails both cases, where the phrase matching passed.
* chore: retrigger title and readiness checks after the title was corrected
* test(firmware): drive the staging ceiling through the request boundary
greptile and codex both flagged that the ceiling tests called
inventoryFirmwareOutput directly, so they would still pass if request parsing or
the provider wiring dropped or replaced max_output_bytes.
Add a request-level test that records the parsed request a provider receives:
the caller's max_output_bytes arrives unchanged at the schema ceiling and below
it, the documented default applies when omitted, and a budget above the schema
maximum is rejected before the provider runs. That is the wiring the previous
tests assumed rather than asserted.
* fix(javascript): match resource release methods to the acquired handle
Every acquisition family admitted every release method, so `h.close()`,
`h.destroy()` and `h.end()` each resolved as a complete release of whatever the
binding held. Node exposes a different subset per acquisition, and the rejected
pairs throw at the call site instead.
Measured on Node 24.22.1:
| acquisition | value | release members |
| --- | --- | --- |
| `fs.openSync` | number | none |
| `fs.open` (callback) | undefined | none |
| `fs.createReadStream` | ReadStream | `close`, `destroy` (`end` is the byte-offset field) |
| `fs.createWriteStream` | WriteStream | `close`, `destroy`, `end` |
| `fs/promises.open` | FileHandle | `close` |
| `net.connect` | Socket | `destroy`, `end` |
| `tls.connect` | TLSSocket | `destroy`, `end` |
So `fs.openSync(...)` + `fd.close()`, `createReadStream(...)` + `.end()`,
`net.connect(...)` + `.close()` and the FileHandle cases emitted a `releases` /
`resolved` fact for a call that throws.
Carry the specifier with each acquisition candidate, list the release methods
each pair actually has, and emit no release when the pair is not one of them. A
release stays `complete` only when the binding held exactly one acquisition and
that acquisition admitted the method.
`tls.createConnection` is not a Node export at all: `'createConnection' in
require('node:tls')` is false and the ESM import does not link. Drop it from the
table so a `tls.createConnection` call yields no acquisition.
* chore: retrigger title and readiness checks after the title was corrected
* fix(javascript): cover a mixed-family binding before completing a release
greptile asked for the mixed-family case: a binding that can hold either a
FileHandle or a socket, then calls close(). The FileHandle is the only linked
acquisition and the release stays partial.
That case also showed the `candidates.length === 1` guard was dead: an
acquisition of another family is not in `linked`, so `holdsOnlyAcquisitions`
already returns false for it and the release already stayed partial without the
guard. Removing the guard and keeping the test means the behaviour is now pinned
by a test rather than by two overlapping conditions.
* fix(process): normalize descendants of slash-spelled directory roots
A working directory or observation path ending in / or /. kept its
trailing separator as the alias, so descendant paths never met the
delimiter lookahead and stayed literal. Strip those spellings before
building aliases and version the path normalization contract as v3.
Fixes#7584
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(process): normalize filesystem root descendants
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Co-authored-by: morluto <[email protected]>
Field, MethodDef and MemberRef signature issues reported the invalid operand's offset inside its blob, while local signature issues in the same coverage array used artifact file offsets. Translate both through one helper that adds the blob's artifact base.
Fixes#7385
REA_GHIDRA_DISABLED_ANALYZERS parses a blank or comma-only value to no override, but the existing-project conflict check tested the raw value. MCP startup exited and doctor reported config_drift for an input that selects nothing. The check now uses the parsed value.
Fixes#7422
A selected .gpr marker that does not exist and a target that existing-project mode cannot serve both surfaced as execution_failure with no setting and a doctor action that cannot help. Both now return configuration_invalid naming REA_GHIDRA_PROJECT_PATH. The ENOENT diagnostic of the missing marker is kept.
Fixes#7417Fixes#7416
* fix(call-path): reject incompatible coordinate mappings
Use the existing provider-owned coordinate context comparison before composing dossiers. Preserve same-mapping profiles and missing-dossier uncertainty. Fixes#7345.
* test(call-path): assert the structured rejection instead of its wording
docs/testing.md asks for structured error checks. Assert the issue path and reason, and keep the MCP category and code.
* fix(javascript): read spawn env as Node reads it
Node's normalizeSpawnArguments starts with `const env = options.env ||
{ ...process.env }`, so a falsy env is replaced by a copy of the parent
environment. A child started with env: null, undefined, 0, "" or false
therefore inherits everything, exactly as if no env were passed.
REA tested the key's presence alone, so every one of those reported
environment_supplied: true and the graph emitted a resolved supplies-env
relation. The presence test also missed the opposite direction: an env
supplied through an inline object spread reported false although Node uses
it.
Read the declared value the way stdioMode already reads stdio: an explicit
property whose statically known value is falsy does not supply an
environment, and an inline spread that carries env does.
Verified on Node v24.22.1 with a child that prints a parent marker variable:
no options, env: null, undefined, void 0, 0, "" and false all print the
marker, while env: {} prints nothing and env: { A: "1" } prints only that.
* fix(javascript): read falsy numeric literals as Node does
greptile flagged that environmentSupplied still reported a supplied environment for
`env: -0` and `env: 0n`: staticJavaScriptTruthiness recognized neither form, so it
returned undefined and `!== false` turned that into true. Node's
normalizeSpawnArguments uses `options.env || { ...process.env }`, so both inherit.
Extend the shared truthiness reader to cover a BigInt literal and a negated numeric
literal, and add regression cases for `-0`, `0n` and `0.0`.
The reader is shared with the route scan and the state tracker; both treat
`undefined` as "unknown", so a more precise `false` only strengthens them.
The inventory registry gained read_binary_inventory_summary, so the
read-only inventory is fifteen inventory and twelve function operations.
Update the adapter summary and the Windows lane descriptions to match the
registries the Windows verifier derives its coverage from. The historical
Windows verification record keeps its observed 26-operation count.
Fixes#7611
* fix(pe): report exhausted certificate decoder allowances as resource limits
A PKCS#7 payload that exhausted the fixed 64 KiB decoded-text or 16,384-value ASN.1 allowance was refused as malformed input with advice to correct request fields. Keep the allowance and limit on the decoder failure and report it as a nonretryable resource_constraint with fixed-allowance recovery.
Fixes#7531
* test(pe): cover certificate allowances through MCP
---------
Co-authored-by: morluto <[email protected]>
* fix(environment)!: report recorded reads and target uncertainty
BREAKING CHANGE: Environment state and coverage can report method-replay for non-GET safe reads; exhaustive reason consumers must handle it.
* docs(environment): include safe-method replay reason
* fix(ghidra)!: distinguish unlisted instruction addresses
BREAKING CHANGE: Ghidra now returns not-in-listing for in-memory addresses without listed instructions or defined data. Consumers matching the former undecodable value must handle the new status.
* fix(ghidra): invalidate snapshots with old instruction status
* test(ghidra): update raw profile commitment goldens
* feat(artifacts): compare artifacts by Evidence reference and from the CLI
compare_artifacts made agents send back the complete inventory Evidence
that REA already retains, and the CLI had no artifact comparison.
Each side of compare_artifacts now takes exactly one of: complete inline
Evidence (unchanged), {kind: "retained-evidence", evidence_id}, or
{kind: "evidence-file", path, evidence_id?}. The two references are the
sources inspect_analysis_view accepts; both tools now compose one
definition in evidenceReference.ts. A summary still resolves to its exact
retained parent, and every source form gives the same comparison Evidence.
An unresolvable retained reference is refused with a typed missing
reference error.
The application layer owns source resolution and the comparison Evidence,
so MCP and the new `rea compare-artifacts` command share one workflow.
readEvidenceFileSource reports bundle-selection refusals at the caller's
own field path.
* refactor(artifacts): resolve compare_artifacts sides like compare_functions
compare_functions, compare_wasm_interfaces and the JavaScript application
comparisons name each side with evidenceInputSchema and resolve it with
resolvePairedEvidenceRequest. compare_artifacts reached the same result
for inline and retained sides through its own dispatch.
evidenceSourceInputSchema now extends evidenceInputSchema with the
Evidence-file reference, and resolvePairedEvidenceSources sends inline and
retained sides to the same resolveEvidenceInput and file sides to the
shared Evidence file reader. compare_artifacts uses both, so the forms it
shares with compare_functions are parsed and resolved by one owner. The
advertised input schema is unchanged.
* docs(cli): point rea compare to compare-artifacts for inventory contents
rea compare (compare_bundles) compares records by Evidence identity and
explicit pairs and never reads artifact contents, so two build
inventories stay unresolved there. Its help and the CLI guide now name
compare-artifacts for that comparison.
* fix(artifacts): resolve summary parents within Evidence source bundles
* perf(evidence): reuse file-owned immutable authentication proofs
* test(evidence): inject failures at both authentication boundaries
---------
Co-authored-by: Entrpi <[email protected]>
Normalize raw URL separators after admitting an original relative ESM
request, before artifact path validation and root confinement. Preserve
the authored declaration and the existing encoded-separator checks.
Retain Node-backed static and dynamic loader controls and compare the
application graph with the historical importer already present on main.
Narrow PR #4323 to this remaining case after the upstream #4269 fix.
Signed-off-by: Jiakai Xu <[email protected]>
U-Boot's mkimage writes the multi-file size table for script images
(`mkimage -T script`, the boot.scr format) exactly as it does for multi
images, `mkimage -l` lists the script as "Image 0", and the `source`
command skips the table before running the script ("scripts are just
multi-image files with one component", boot/image-board.c).
The uImage reader parsed the table only for IH_TYPE_MULTI, so a script
image reported sub_image_count 0 and exposed only payload.bin, whose
first bytes are the binary size table rather than the script text.
Treat IH_TYPE_SCRIPT like IH_TYPE_MULTI: parse the same bounded table,
expose image-<n>.bin components and report the size-table limitation.
payload.bin is unchanged. A script payload without a valid table now
fails closed, as a multi image already does.
* fix(cli): align MCP doctor completion with its options
* fix(cli): explain zsh completion initialization
* fix(cli): include launcher timing in doctor completion
* style(cli): format doctor completion schema
* feat(artifacts): inventory and extract ESP32 firmware images
ESP32-family app and bootloader images (magic 0xE9) had no reader, so
open_binary refused OTA downloads, app partitions and bootloaders.
Add the esp-image container format behind inspect_artifact and
extract_artifact. A pure decoder reads each byte once and keeps only
fixed-size fields. Derived facts cover the chip, from a table cited to
ESP-IDF v6.1; segment classification as ESP-IDF's should_map and
should_load; checksum and appended SHA-256 results; app and bootloader
descriptors; Secure Boot v2 signature blocks, observed and unverified;
and undescribed trailing byte ranges. Truncation and undescribed bytes
make coverage unknown. ESP8266 images are refused with a specific
reason, and ciphertext stays unrecognized.
Add the format to every target, Evidence, snapshot, live-summary and
artifact vocabulary, document it, and add an optional verify:esp-images
lane that compares esptool-built images with esptool image-info and
espsecure signature-info-v2.
Refs #3615
* test(artifacts): run the ESP image lane on esptool v5.6.0
The lane compared descriptor fields only when esptool image-info printed
one, so a descriptor that REA reports and esptool does not print passed
without comparison. Compare the descriptor kind in both directions.
Record that the lane passed for all 13 images with esptool and espsecure
v5.6.0, the version the layout facts cite, as well as with 4.12.0.
* refactor(artifacts): define ESP image layout sizes once
The decoder and the derived facts each defined the segment-header,
appended-digest and flash-sector sizes, the erased-byte scan and the
offset formatter. Define them once with the other fixed-layout structures
and import them in both places. The signature slot count is the one
constant that drives the slot scan.
Correct the observedEnd comment: it is the end of the last chunk read.
* fix(artifacts): report ESP-IDF's segment length limit by chip
ESP-IDF's verify_segment_header rejects a segment of 16 MiB or more
unless the chip maps flash through 32-bit cache addresses
(SOC_SPI_MEM_SUPPORT_CACHE_32BIT_ADDR_MAP: ESP32-S3, P4, C5, C61, H4 and
S31 in v6.1). Record that capability in the chip table and add a
limitation for such segments; unknown chips get none.
`loaded` repeats should_load. It does not mean the bootloader accepts the
segment: verify_load_addresses still rejects one outside the chip's RAM,
which REA does not model. Say so in the docs, the skill reference and the
inventory limitation.
* test(artifacts): check ESP lane tools first and verify extraction
The lane started work before checking its tools, so a missing RISC-V
target, ld.lld, esptool or espsecure surfaced as a raw spawn or command
error. Check each one before building and name the variable to set. Warn
when esptool is not v5.6.0, the release the layout facts cite.
Assert each segment's class for the ESP32-C3 app, its variants and the
bootloader, as the Xtensa image already did. Extract the app and the
signed image through the CLI and compare every file with the bytes at
esptool's segment offsets or in the appended signature sector; the
source's SHA-256 must be unchanged and equal the reported root_sha256.
* docs(artifacts): state what the ESP image lane does not establish
The Verification section listed the lane's checks but not its gaps: only
ESP32-C3 and ESP32 images come from a real producer, no real project
build, OTA download or flash dump is inspected, no signature is verified,
and the lane does not run in CI.
Record the timing of the final lane run on esptool v5.6.0.
* fix(artifacts): compare ESP segment declarations
Segment entries hash only their data, so compare_artifacts reported a
segment that moved to another load address with the same bytes as
unchanged; only the image itself changed. Compare each segment's
load_addr and data_len as its declaration, as Node SEA and PyInstaller
members are compared, and count a changed declaration as a metadata
change, also when content is unavailable.
Header offsets are layout and the classification is derived, so neither
is compared.
* test(artifacts): run ESP lane commands as owned process trees
The lane ran its tools through execFile. A timeout killed only the direct
child, so a linker or assembler that Clang or GCC had started could keep
running while the lane removed its workspace. Run every command through
createVerifierCommands, check the tools inside withVerifierCleanup, and
remove the workspace only after commands.close() has stopped every tree;
a failed close keeps the workspace.
Only a nonzero exit now counts as a refusal or an expected esptool
failure; timeouts and spawn failures propagate. Supervising each command
roughly doubles the lane's run time.
* test(artifacts): cancel ESP decoding after its first decoded chunk
The cancellation test requested cancel or close on setImmediate, which
can land while the file is still being opened and checked, so it never
showed that decoding had started. It also passed against a reader that
ignored both requests while decoding. Acknowledge the decoder's first
chunk, then cancel or close, join the operation, and assert that no
later chunk reaches the decoder and that the reader ends closed.
* docs(artifacts): record which ESP metadata comparison ignores
Segments are compared by their declaration. The image's and the
signature blocks' format_metadata are not: every field comes from their
own bytes, which content comparison already covers, and a block's
image_digest_matches follows the image's content.
* docs(artifacts): state that ESP mmu_page_size is a base-2 logarithm
The app descriptor's mmu_page_size keeps ESP-IDF's name and value, the
MMU page size in log base 2 as esp_app_desc_t declares it, so 16 means
64 KiB. Say so in docs/esp-image.md and the schema comment; a reader
could otherwise take 16 as 16 KiB.
* fix(esp-image): retain member consumption and share lifecycle regressions
* test(artifacts): scope MCP catalogs and budget full-schema validation
* test(artifacts): verify compact output profiles and document bounded catalogs
* test(artifacts): close readers when cleanup-test admission fails
* test(artifacts): compare ESP image limitations with their owners
The decoder and reader tests froze complete limitation sentences, so a
harmless rewording failed them. The wording now has two owners:
ESP_IMAGE_LIMITATIONS in espImageFacts.ts, which the decoder, the inventory
and the reader use, and ESP_DESCRIPTOR_LIMITATIONS beside the descriptor
decoder. Tests compare delivered limitations with the owner, called with the
facts each case establishes (segment index, declared and available bytes,
offsets, chip, page range), next to the structured assertions they already
make. Delivered text is unchanged.
* test(esp-image): assert structured facts and owned guidance
* fix(esp-image): keep spi_flash_mode and descriptor fields before a truncated tail
ESP-IDF v6.1's esp_app_desc_t declares spi_flash_mode at byte 181, after
mmu_page_size, as the committed reference layout shows; the descriptor now
keeps it as stored. The decoder also required all 256 descriptor bytes,
though it reads none past byte 181, so a file ending in the reserved tail
lost the version, project, digest and revision facts it does hold. The app
descriptor now needs only its fields; the segment's truncation is reported
as before.
* test(esp-image): pin the P-192 public key digest to espsecure's layout
espsecure packs ECDSA signature blocks as "<BBBx32sB64s64s1031x" and
zero-pads a 48-byte P-192 point to 64 bytes, so the signature starts at
byte 101 for every SHA-256 curve. signature_info_v2 hashes sig_block[36:101]
and _digest_sbv2_public_key the curve ID with the padded point. No test built
a P-192 block; this one checks public_key_sha256 against that definition.
* fix(esp-image): keep each app descriptor field the file holds in full
A file can end anywhere inside the 256-byte esp_app_desc_t. The decoder
needed every field byte, so a cut at byte 176 dropped the version, project
and app_elf_sha256 it does hold. Once the 4-byte magic identifies an app
descriptor, each field is now kept when the file holds all of its bytes and
is null otherwise; observed_length records where the descriptor ends, so a
cut field is told apart from text that is not UTF-8. The fields after the
magic become nullable in the metadata schema.
* fix(esp-image): report a signature block whose image digest does not match
A CRC-valid Secure Boot v2 block can carry another image's digest. The
block recorded image_digest_matches false, but no limitation said so, unlike
a checksum or appended-SHA mismatch, so CLI and MCP summaries showed only
the generic unverified note. The mismatch is now a limitation; coverage
stays complete and signature_status stays unverified. The three other
signature-sector limitations move into ESP_IMAGE_LIMITATIONS with their
text unchanged, so all ESP image wording has one owner.
* refactor(esp-image): remove redundant descriptor byte guards
---------
Co-authored-by: gensokyo-dev <[email protected]>
Seal sensitive terminal chunks before deferred rendering, apply one bounded
capture-local literal matcher, and retain exact redacted byte accounting.
Keep private input facts in explicit fields and separate comparison policies.
Retain distinct terminal failures once and verify structured refusal facts.
Fixes#7782
BREAKING CHANGE: sensitive interaction data is empty with sensitive and input_bytes fields instead of a byte-count placeholder. Sensitive raw_data is also redacted. Read input_bytes and recapture both comparison sides when migrating older private-input Evidence.
Assisted-by: Reimu (gensokyo-skills:reimu-incident-triage)
Assisted-by: Kogasa (gensokyo-skills:kogasa-surprise-testing)
Assisted-by: Yukari (gensokyo-skills:yukari-boundary-analysis)
CSP3 parsing keeps the first directive of each name in a policy and ignores
later repeats; Chrome reports "Ignoring duplicate Content-Security-Policy
directive". parseCsp flattened every policy's directives into one stream, so
a header such as `script-src 'self'; script-src 'unsafe-inline'` reported the
ignored `'unsafe-inline'` directive as declared and counted nonces and hashes
from it. Track directive names per policy so only the effective directive is
retained, while the same name in separate comma-separated or repeated-field
policies is still listed once per policy.
* fix(javascript): keep both-direction feature traces from crossing shared containers
trace_application_feature with direction "both" walked the application
graph as if it were undirected. From a seed, one outgoing or incoming step
into a shared container (an asset or the artifact root) followed by steps
in the other direction reached every sibling of that container, so a trace
seeded on one storage key returned the whole connected graph.
"both" now returns the union of the outgoing and incoming closures. Each
walk keeps its own predecessor chains, so a terminal path never changes
direction midway. "outgoing" and "incoming" are unchanged.
* test(javascript): check both-direction trace paths end to end
The existing tests checked predecessor maps but no returned path. Add a
trace whose seed reaches one storage terminal outgoing and another
incoming, and assert each path's complete node_ids and edge_ids. A
pathTo that reads only the first walk's map returns a one-node path for
the incoming terminal and fails this test.
* fix(javascript): confine only ascended containers in both-direction traces
Making "both" the union of separate outgoing and incoming walks dropped
modules that a feature reaches by crossing directions. In the Notes
known-answer evaluation, the notes:export channel is pointed at by both
the preload asset (invokes) and the handler (handles); its preload,
renderer and csv modules are reached only through relationships that
change direction, so the trace kept main.js alone and the feature_modules
claim failed in tests/boundary/mcp/agentKnownAnswerMcp.test.ts.
Walk both directions in one traversal again, and confine only the
containment case the change was meant to stop: a node reached by stepping
up a contains edge from one of its children follows only further contains
steps up. A shared asset or artifact root then no longer leads into its
other children, while relationships across the feature are kept. The
Notes trace returns csv.js, main.js, preload.js and renderer.js again
(30 nodes; the previous main returned the whole 32-node graph).
outgoing and incoming traces are unchanged, and paths come from one
predecessor map again, as before this branch.
* fix(javascript): stop double-expanding application projection estimates
estimateJavaScriptStaticApplicationProjection multiplied the per-record
metadata allowance (2 KB plus the path) by the 8x expansion that it also
applies to checkpoint bytes. A projected record serializes to about 0.46 KB,
so the allowance alone already covers retention; expanded again, the
charge was about 25x the serialized application relationships.
The charge accumulates in the parent's shared result budget. On a
synthetic 2.8 MB Vite-style build (88 files, one 456 KB chunk), the first
10 files used 2.0 GB of a 2.08 GB budget while the parent heap held about
300 MiB. The other 78 files lost their application relationships and were
offered a 0-byte semantic budget. On a 201-file build the charges exceeded
the budget at file 93 and the analysis failed with resource_constraint,
leaving the remaining sources without static facts.
Keep the 8x expansion for checkpoint and repeated bytes and count the
per-record allowance once. The same builds now refuse 14 and 102
projections instead of 78 and a failed run; peak RSS stays at 1.2 GB and
2.0 GB under a 4.2 GB parent heap limit.
workerFailureComposition simulated 64 KiB of parent result budget, which
the corrected charge for its one-finding checkpoint fits. Lower that
headroom to 32 KiB so the case still exercises a refused projection.
* test(javascript): cover application projections near the shared parent budget
The estimate test compares one file's charge with its serialized graph,
but no test put several sources against one parent result budget. Replay
one recorded static checkpoint for five sources against an exact budget:
sources 1-3 fit, source 4 misses by half a source and keeps its static
facts with the reported required and budget bytes, and the last source
fits again once no checkpoint reserve is held back. A worker that charges
the projection twice, or not at all, passes the existing tests and fails
this one.
* test(javascript): cover a malformed source before a valid one under the shared budget
The testing guide asks representation-expansion and capacity changes for
a large malformed input followed by a valid input, and for escaped or
expanded representations.
Under an exact parent budget, a 3 MiB checkpoint cut off inside a string
comes first and is rejected as unreadable output with no static facts.
The valid source after it keeps its application relationships; it fits
only if the rejected bytes took no charge. A worker that charges the
static checkpoint on receipt, before parsing, passes the existing tests
and fails this one.
The estimate ratio test also runs with a nested source path and storage
keys holding quotes, backslashes, a tab and multibyte text, which expand
in serialized JSON; the charge stays within 3x to 8x of the serialized
application graph (5.3x).
* fix(javascript): resolve Vite build-output preload dependencies
With an absolute base, Vite writes __vite__mapDeps entries relative to the
build output directory (for example assets/chunk.js), and its preload helper
prefixes the configured base at runtime. REA emitted these entries as
dynamic-import module specifiers, so a scheme-less path without a leading
dot resolved as a bare package and every entry was reported as an
unresolved "external" module, even when the build output directory was
selected.
The static analyzer now marks such entries with a vite-output: specifier,
as Webpack chunk loads are marked with chunk:. Entries starting with ./,
../ or /, and scheme-qualified entries, keep their previous handling. The
graph resolves a marked entry as a URL path under each ancestor directory
of the importing file; exactly one ancestor holding the file establishes
the build output directory. Otherwise the entry is not-found (or
unavailable when several ancestors match) with a limitation that names
the build output directory.
On a synthetic 7-file Vite build, the 12 preload entries that were
"external" now resolve when the build output directory is selected, and
are not-found with that guidance when only assets/ is selected.
* fix(javascript): keep unanalyzed Vite output candidates
resolveViteOutputDependency considered only resolved ancestors. A file
that exists under an ancestor but was not analyzed (for example
dist/assets/lazy.css for the entry assets/lazy.css) resolves there as
unavailable, and that result was replaced by the root's not-found with a
limitation saying that no ancestor holds the file.
Keep resolved and unavailable ancestor candidates together. One candidate
is returned with its own path and limitations; several are reported as
unavailable with every candidate path and limitation. Only when no
ancestor can hold the file does the root's not-found result stand.
Apktool 2.9.0 replaced SnakeYAML with the YamlWriter that 3.x still uses:
apktool.yml starts with an untagged `version: 2.x.y` line and strings are
written unquoted with Java-style escapes. Only a `version: 3.x.y` header
selected the producer dialect, so 2.9.0-2.12.x output went through YAML
core parsing: a versionName of `1.0 #beta` became `1.0`, escaped quotes kept
their backslashes, an empty name became null, and `1.0: beta` made the whole
decode fail as invalid metadata.
Select the line-oriented reader for 2.9.0 and later as well. Older tagged
SnakeYAML documents keep YAML scalar semantics.
Every source string literal without an established Next root was tested
against /(?:^|\/)app\/(?:[\s\S]*\/)?page(?:\.[cm]?[jt]sx?|-[^/?#]+\.js|)(?:[?#].*)?$/u.
The unanchored search restarted at every app/ segment and rescanned to the
end of the literal, and each page? or page# candidate scanned to the next
line terminator before $ failed. A long literal therefore took quadratic
time on the analysis thread: about 6 s for a 256 KB literal of repeated
app/ segments, and 3.5 s for 224 KB of x/page? followed by a newline.
The first app/ segment admits every later page segment, and . cannot cross
a line terminator, so a query or fragment after the page filename starts on
the last line. Test the tail after the first app/ segment one page segment
at a time instead; results are unchanged.
Use one command owner for setup, client writers, Gemini policy and CLI defaults.
Migrate recognized runner registrations with backups and remove runner prewarm.
Retain unknown identities and exercise real offline npm, CLI and MCP workflows.
Fixes#7757
BREAKING CHANGE: managed MCP registrations use the selected absolute runtime
and REA entry script, including package-runner cache paths. Keep that installation
available or restore it before setup; unknown registrations require confirmed
manual repair. Setup no longer returns prewarm or emits its progress events.
Assisted-by: Reimu (gensokyo-skills:reimu-incident-triage)
Assisted-by: Kogasa (gensokyo-skills:kogasa-surprise-testing)
Assisted-by: Yukari (gensokyo-skills:yukari-boundary-analysis)
* feat(artifacts): add U-Boot legacy uImage reader (part of #7129)
Recognize legacy uImages by their 0x27051956 magic and inventory them as containers: 64-byte header facts (arch/os/type/compression raw and named, load/entry addresses, name, timestamp) with header and data CRC32 verified, stored payload extractable as payload.bin, and multi-file images split from the stored size table into image-<n>.bin. Compressed payloads stay as stored; nothing is executed or decompressed. Wires the format through classification, the artifact reader dispatch, extraction, and the evidence/snapshot format vocabularies. Related to #7129 (uImage part only); HEX/SREC and FIT are out of scope.
* fix(uimage): preserve aligned component ranges and exact header facts
* fix(uimage): retain buffered members through consumption and cancellation
* fix(uimage): preserve active read and validation failure precedence
---------
Co-authored-by: gensokyo-dev <[email protected]>
Admit the final macOS Hopper launcher arguments before application handoff.
Keep prepared-image source identity, return precise typed refusal locations,
and preserve bootstrap/lease cleanup. Update native CLI/MCP recovery
coverage and the installation guide.
Fixes#7727
BREAKING CHANGE: macOS Hopper launches refuse double quotes, backslashes
and line breaks in final helper paths or arguments before application
handoff. Select literal-safe launcher/runtime/database paths and loader
arguments, or a compatible provider. Original executable names can remain
unchanged when their prepared image path is literal-safe.
Assisted-by: Reimu (gensokyo-skills:reimu-incident-triage)
Assisted-by: Kogasa (gensokyo-skills:kogasa-surprise-testing)
Assisted-by: Yukari (gensokyo-skills:yukari-boundary-analysis)
* feat(artifacts): read UnityFS AssetBundle containers
inspect-artifact and extract-artifact now recognize the UnityFS signature
and treat format versions 6 through 9 as containers. Header facts (format
version, player/engine strings, declared size, dataflags, blocks-info
bounds and the declared 16-byte data hash) are retained alongside a
directory of CAB-* and .resS members. Member bytes stream on demand:
only the blocks covering the requested member are read and decompressed
with bounded expansion and cancellation.
Bundles declaring asset-bundle encryption keep header facts and report
unknown directory coverage. LZMA blocks-info reports unknown coverage;
LZMA member reads fail as unavailable while the directory stays
readable. Malformed counts, out-of-stream members and declared sizes
beyond the file fail as format errors. UnityFS classifies by bytes in
root detection, inventory graph construction and desktop inspection,
which emits a unityfs-header fact.
Validation: tsc --noEmit, oxlint (no new findings on touched files),
17 reader tests covering LZ4HC/uncompressed tables, blocks-info at end,
format 6 without padding alignment, multi-block members, malformed
input, unsupported compression, cancellation and abandoned streams;
full artifacts/domain/conformance suites pass. Verified against one
real YooAsset bundle (Unity 2021.3.32f1, format 8, LZ4HC): the
streamed SerializedFile contained its managed assembly byte-identical
to a UnityPy extraction. Other producer variants unverified.
* fix(unityfs): own bounded member reads and preserve partial evidence
* refactor(unityfs): clarify producer revision flag decisions
* fix(artifacts): preserve coverage authority and bound UnityFS block work
---------
Co-authored-by: gensokyo-dev <[email protected]>
Resolve Windows ADB, Apktool and update commands against the selected
PATH/PATHEXT before launch. Keep explicit paths and relative PATH choices,
report lookup failures, and reuse absolute batch invocation and ownership.
Add native launch/argv regressions, real npm metadata probes and focused
CI ownership so these boundaries retain Windows verification on the fast tier.
Fixes#7724
Assisted-by: Reimu (gensokyo-skills:reimu-incident-triage)
Assisted-by: Kogasa (gensokyo-skills:kogasa-surprise-testing)
Assisted-by: Yukari (gensokyo-skills:yukari-boundary-analysis)
Reuse shared padded-base64 facts and derive encoded admission from the existing decoded-byte budget. Refuse overflow before decoded allocation and retain padded syntax. Cover capacity, compact HTTP replies and live MCP state readback.
Fixes#7719
Assisted-by: Reimu (gensokyo-skills:reimu-incident-triage)
Assisted-by: Kogasa (gensokyo-skills:kogasa-surprise-testing)
* feat(artifacts): inventory and extract Intel HEX and Motorola S-record regions
open_binary refused Intel HEX and S-record programming files as an
unsupported format, so inspect_artifact and extract_artifact could not
answer which bytes load at which addresses.
Recognize both encodings by a fully valid first record and decode them
statically into address-contiguous regions, each an inventory entry named
0x<start>-0x<end>.bin and extractable as raw bytes. Root format_metadata
records the encoding and variant, record counts, unique and duplicate
bytes, address bounds, termination, declared start address, S0 header and
validated S5/S6 count. Gaps are never filled; start addresses are declared
facts, not proven entry points.
Checksum, record-type, byte-count, line-length, overlap-conflict, address
wrap and post-termination errors fail closed with the line number; a
missing termination record leaves coverage unknown. Decoding is bounded by
a 128 MiB file ceiling (decoded bytes are at most half the file and are
reserved after the size check) and 65,536 address-discontiguous runs.
Refs #7129
* fix(artifacts): refuse Intel HEX extended address records with a non-zero address
* fix(artifacts): bound HEX region copies and retain stream cancellation
---------
Co-authored-by: gensokyo-dev <[email protected]>
Refresh compatible security fixes and verified dependency notices. Keep
historical HTTP/3 guest verification separate from the refreshed trust store.
Fixes#7700
Assisted-by: Aya (gensokyo-skills:aya-source-investigation)
* feat(environment): serve the writes callers make without recorded form fields
- A create member a recording sent empty ("", null, false, [] or {}) is no
longer required; a call that leaves it out is applied as sending that value
(body.blank).
- List views render envelope members that list the listed entities' ids
(Paperless's `all`), so creates in such collections are confirmed.
- Updates move their entity in collections kept sorted by a field they write,
and a failed whole-collection read after an update can teach that order.
- A later create that stores an unruled field otherwise is withheld at the
create, and an item read may fill values a create's answer had not loaded.
- An unsent field respelled between null, [] and {} is unchanged.
- The skill asks for one create with every field and one with only what a
person must give.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(skill): show how to read signed-in pages before recording
Agents guessed capture keys (artifacts, dom, final) to see pages behind
sign-in, which inspect_web_page cannot reach.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): accept every empty spelling recorded creates sent
A create that fills only what a person must give sends `null` where the
full create sent a string, and `null` or `false` for the same flag. The
merged plan now accepts each recorded empty value, and a call that leaves
the member out is applied as the earliest call sent it.
Also adds the spec folder for this change.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(skill): keep the vary-fields step neutral on UI or API writes
Agents read "the form" as a cue to write through the UI, where some
applications' request bodies are not exposed; a body-unusable UI write
now re-records once with request steps.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record verification for blank members and sorted writes
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): address review of blank members and sorted writes
- Bulk updates move their entities in turn against the others' earlier
values and install the final order once, as replay does.
- A member's recorded empty values stay accepted once a later call teaches
its default; a left-out member with a default keeps the default.
- Write answers spell empty lists and maps as their recorded answers did.
- A rename can teach a sort when the creates chose one by a field the
update did not write, such as ids that ascend with titles.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(environment): serve the writes that later reads explain
- An item read whose object is marked deleted (deleted, isDeleted or
is_deleted true) agrees with an absent entity, as EspoCRM answers a
deleted record; such an object never seeds or adopts an entity.
- maxSize, max_results and maxResults are page sizes, for whole-collection
list lines and the list query's limit role.
- An item view also shows the fields only reads after its type's first write
show, when no initial entity holds them (Vikunja's task subscription),
without requiring them; write answers leave such a field out for entities
without it, and it no longer withholds the write as incomplete-entities.
- The skill asks updates to change the fields a person would change, and
expects a deleted object to read back as 404 or marked deleted.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer sent empty members as sent, and keep a placed create's order
- A write answer spells an empty list or map the call did not send as its
recorded answer did; one the call sent is answered as sent, so a create
sending `[]` is not answered `null` because the first recorded create was.
- A read listing the whole collection that placed a create leaves the create
only the sorted inserts whose order that read follows, so a later rename
cannot teach a sort an earlier read ruled out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer bulk updates without fields only later reads show
A bulk update's answer renders each entity with the same allowance a single
update's answer has: a field only reads after the type's first write showed,
which no initial entity holds, is left out for an entity without it, so a
bulk update of a seeded task is no longer refused for lacking it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): correct how the EspoCRM contact update task passes
The agent trajectories show gpt-oss passes it by creating a replacement contact
and deleting the original, since the served update refuses emailAddress; the
update itself is refused in both builds.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer a read of a served-deleted entity marked deleted
An application that soft-deletes still answers a deleted record by id, marked
deleted. Serving now keeps an entity a served delete removed from a type whose
recorded item reads show that, until a reset or load and within the size
data/add loads, and answers a read of it with its last fields and the flag
true, instead of refusing the read.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep unloaded create answers, check bulk answers with sent members, narrow inserts only on collection-order reads
- A create whose item read loaded values its answer had not records them
(`observation.read.unloaded`), and a served create's answer keeps them as
the recorded answer showed them, a zero-time owner included, while the state
holds the read's.
- Replay checks a bulk update's recorded answers with each target's sent
members, as serving answers them, so echoed `null` and `[]` no longer make
the plan body-unusable.
- A whole-collection read whose query sorts it shows the query's order, so it
no longer narrows a create's sorted insert candidates.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record the census with main's unknown-clock rule merged
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(environment): serve query numbers, counted values and made objects
- A query parameter accepts a number or boolean, which serving sends as its
JSON text, as agents send page sizes and flags.
- A sequence may start at the value its first create took above every value
its type held (Vikunja's per-project task index), and a later create's
composed, template or started-sequence rule is adopted when it also spells
the first create's value.
- A made-object rule explains an object the application makes with each
created entity: members the request stamped, members holding the entity's
own id, and counts one past the held objects' values.
- A create's answer names the id of the one object a one-element array member
holds, so bulk creates link to their type.
- The skill reads each listed object by id before writing.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(environment): let a linked create take path values it stores
A create linked to its type's own collection, whose path has no parameters,
accepts any path values when its plan stores each one in a field, as a Vikunja
task stores its project; replay and serving apply it alike.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn a sort from a rename only when moving its entities reaches it
Serving moves only the entities a write changes in a collection kept sorted,
so replay now tries a sort after a rename only when the entities no write the
read exercised changed already follow it. Seeded folders listed out of title
order no longer teach a title sort that serving would answer with another
order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep unloaded create answers a sent value did not echo, and learn no sort from query-sorted reads
- A served create keeps a value its recorded answer had not loaded unless the
call sent that member and the plan stores it as sent, so a create sending
`active: false` is answered `false` as recorded while the state holds the
read's `true`.
- A rename no longer teaches a sort from a whole-collection read its query
sorts, whose ties need not follow the collection's own order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): verify query values, counted values and made objects
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn the soft-delete flag in the state view and serve by it
Extraction now records the member later item reads set true on a record the
application soft-deleted, and the item view carries it as `deleted_flag`.
Serving reads that fact instead of scanning recorded bodies itself, so one
owner decides what a soft delete looks like.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record the call-values checks at #727493821024
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn the soft-delete flag from item reads with a query too
A client may read a deleted record back only with a query, such as a
cache-busting parameter. Every successful read of the item tool now shows the
flag; only plain later reads still add fields.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record the call-values checks at #727445832406
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn sorts only from orders a list read confirmed
- A sort retry after a failed whole-collection read requires the entities no
write changed since a list last showed them to follow it already, rather than
excusing every entity the update's tool ever wrote: an earlier rename a list
confirmed in place stays where it was.
- An update takes its creates' sort only from a create a list read showed
placed; replay still tries a sort a create guessed from the initial order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): drop entities read back as deleted and answer soft deletes as deleted
- A held entity a later read shows marked deleted is dropped, not adopted with
the marked object's fields.
- Bulk-update replay renders answers with the optional and later-only fields
serving uses, so a bulk update of created and seeded entities replays.
- A served delete of a soft-deleting type answers with the entity marked
deleted, single and bulk.
- A deleted flag only query reads showed is a field of the item view, so a
served tombstone keeps it where live records omit it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): check made objects as recorded and keep their ids and clocks honest
- A made object's clock counts as a write clock, so a capture with no clock
anchor withholds the create instead of stamping 1970.
- Members holding the entity's identity follow the id written, generated or
sent, through the type's id field.
- A later create's made object is checked as recorded, before its own members
are rewritten, so one naming another entity withholds the plan.
- A later create's template is adopted only when it spells that create's value
too, not just when the first create's text holds its source value.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): move updated entities only by placed sorts and keep query-sorted reads from confirming order
Replay moved an updated entity by a sort a create only guessed, while the
update plan exported only placed sorts, so a read could confirm an update
by a move serving would not make. Both now use the placed sort. A passing
list read clears its entities from the moved set only when the state
decides the order of every entity it lists, so a read whose query sorts
it no longer hides an earlier move from a later sort check.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): count six withheld plans that explain more, not seven
The bulk task create counted among them is served on this branch.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): try a learned sort by moving each pending write in turn, as serving does
#retrySort sorted the whole collection, which a list could match while
serving, moving each renamed entity in turn against the others' values
before their own writes, gave another order. The state now records the
values each write replaced, and the retry replays the writes since a
list in collection order last showed their entities.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer a served delete with the deleted flag its recorded answer showed
An application may answer a delete with the record as it was and mark it
deleted only on later reads. Delete answers, single and bulk, now take
the flag's value from the recorded answer; reads of a served-deleted
entity still show it marked deleted.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): set aside two hunks next to lines main's saved-values change rewrites
Main's #7267 changes the lines beside the blank-member docs and the
empty-spelling check in explainUpdate. Setting them aside lets that merge
apply cleanly; the commit after it puts them back on top of main's text.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): restore the set-aside hunks on main and bump the skill to 51
Puts the blank-member docs and the empty-spelling check back beside main's
slug and composed text and heldSlugRule. Main is at skill 50 with
different text, so this branch's skill changes take 51.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): bump the skill to 52 above #7260's 51
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): put back the call-values hunks main's merge took its side for, and bump the skill to 53
The merge of main kept main's side where it overlapped this branch, so it
could commit without a conflict. This restores this branch's side next to
main's: made-object agreement beside the & slug spellings, sequence starts
in the earlier evidence (now built by earlierOf), the combined value
imports, and recording step 2 beside main's UI-first step 3. Main is at
skill 50 and #7274 at 52, so this branch takes 53.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep sorts a list showed in place ruled out, and created answers' nested members as recorded
A whole list in collection order that showed a renamed entity where a
sort by the renamed field would have moved it now rules that sort out, so
a later rename of the same entity cannot teach it. A create's answer
leaves out members its item read added inside an object the answer
showed, as the recorded answer did; the state still stores them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep a later create's made object only with clocks its request stamped
followsMadeObjects only checked that a made object's clocks parsed, so a
later create whose object showed an older clock still passed, while serving
stamps the write's clock. Its clocks must now be in the rule's format and
within the stamp window of that create's request.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): compare soft-deleted read fields
* fix(environment): reject contradictory learned sequences
* fix(environment): preserve initially deleted item observations
* fix(environment): index stored deleted entities separately
* fix(environment): replay lists that include deleted records
* fix(environment): make no count of a made-object member earlier creates held alike
A count only a later create shows starts at the exemplar's value, which is
right only while the exemplar is the one create accepted: creates accepted
after it showed the member unchanged, which a count would have moved. The
merge now refuses such a count, as the top-level sequence check does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Co-authored-by: morluto <[email protected]>
A saved response's decoded size is bounded by its received size, or the
declared length of one served from cache, times 1032 for each gzip or deflate
coding, the most DEFLATE expands. Grafana answers folder creates gzipped, so a
UI step saving a new folder's uid was refused. Other codings and unknown or
zero sizes stay refused before the body is read.
Co-authored-by: Claude Opus 5.5 <[email protected]>
Co-authored-by: morluto <[email protected]>
* feat(environment): serve the writes callers make without recorded form fields
- A create member a recording sent empty ("", null, false, [] or {}) is no
longer required; a call that leaves it out is applied as sending that value
(body.blank).
- List views render envelope members that list the listed entities' ids
(Paperless's `all`), so creates in such collections are confirmed.
- Updates move their entity in collections kept sorted by a field they write,
and a failed whole-collection read after an update can teach that order.
- A later create that stores an unruled field otherwise is withheld at the
create, and an item read may fill values a create's answer had not loaded.
- An unsent field respelled between null, [] and {} is unchanged.
- The skill asks for one create with every field and one with only what a
person must give.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(skill): show how to read signed-in pages before recording
Agents guessed capture keys (artifacts, dom, final) to see pages behind
sign-in, which inspect_web_page cannot reach.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): accept every empty spelling recorded creates sent
A create that fills only what a person must give sends `null` where the
full create sent a string, and `null` or `false` for the same flag. The
merged plan now accepts each recorded empty value, and a call that leaves
the member out is applied as the earliest call sent it.
Also adds the spec folder for this change.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(skill): keep the vary-fields step neutral on UI or API writes
Agents read "the form" as a cue to write through the UI, where some
applications' request bodies are not exposed; a body-unusable UI write
now re-records once with request steps.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record verification for blank members and sorted writes
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): address review of blank members and sorted writes
- Bulk updates move their entities in turn against the others' earlier
values and install the final order once, as replay does.
- A member's recorded empty values stay accepted once a later call teaches
its default; a left-out member with a default keeps the default.
- Write answers spell empty lists and maps as their recorded answers did.
- A rename can teach a sort when the creates chose one by a field the
update did not write, such as ids that ascend with titles.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer sent empty members as sent, and keep a placed create's order
- A write answer spells an empty list or map the call did not send as its
recorded answer did; one the call sent is answered as sent, so a create
sending `[]` is not answered `null` because the first recorded create was.
- A read listing the whole collection that placed a create leaves the create
only the sorted inserts whose order that read follows, so a later rename
cannot teach a sort an earlier read ruled out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep unloaded create answers, check bulk answers with sent members, narrow inserts only on collection-order reads
- A create whose item read loaded values its answer had not records them
(`observation.read.unloaded`), and a served create's answer keeps them as
the recorded answer showed them, a zero-time owner included, while the state
holds the read's.
- Replay checks a bulk update's recorded answers with each target's sent
members, as serving answers them, so echoed `null` and `[]` no longer make
the plan body-unusable.
- A whole-collection read whose query sorts it shows the query's order, so it
no longer narrows a create's sorted insert candidates.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn a sort from a rename only when moving its entities reaches it
Serving moves only the entities a write changes in a collection kept sorted,
so replay now tries a sort after a rename only when the entities no write the
read exercised changed already follow it. Seeded folders listed out of title
order no longer teach a title sort that serving would answer with another
order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep unloaded create answers a sent value did not echo, and learn no sort from query-sorted reads
- A served create keeps a value its recorded answer had not loaded unless the
call sent that member and the plan stores it as sent, so a create sending
`active: false` is answered `false` as recorded while the state holds the
read's `true`.
- A rename no longer teaches a sort from a whole-collection read its query
sorts, whose ties need not follow the collection's own order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn sorts only from orders a list read confirmed
- A sort retry after a failed whole-collection read requires the entities no
write changed since a list last showed them to follow it already, rather than
excusing every entity the update's tool ever wrote: an earlier rename a list
confirmed in place stays where it was.
- An update takes its creates' sort only from a create a list read showed
placed; replay still tries a sort a create guessed from the initial order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): move updated entities only by placed sorts and keep query-sorted reads from confirming order
Replay moved an updated entity by a sort a create only guessed, while the
update plan exported only placed sorts, so a read could confirm an update
by a move serving would not make. Both now use the placed sort. A passing
list read clears its entities from the moved set only when the state
decides the order of every entity it lists, so a read whose query sorts
it no longer hides an earlier move from a later sort check.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): try a learned sort by moving each pending write in turn, as serving does
#retrySort sorted the whole collection, which a list could match while
serving, moving each renamed entity in turn against the others' values
before their own writes, gave another order. The state now records the
values each write replaced, and the retry replays the writes since a
list in collection order last showed their entities.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): set aside two hunks next to lines main's saved-values change rewrites
Main's #7267 changes the lines beside the blank-member docs and the
empty-spelling check in explainUpdate. Setting them aside lets that merge
apply cleanly; the commit after it puts them back on top of main's text.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): restore the set-aside hunks on main and bump the skill to 51
Puts the blank-member docs and the empty-spelling check back beside main's
slug and composed text and heldSlugRule. Main is at skill 50 with
different text, so this branch's skill changes take 51.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep sorts a list showed in place ruled out, and created answers' nested members as recorded
A whole list in collection order that showed a renamed entity where a
sort by the renamed field would have moved it now rules that sort out, so
a later rename of the same entity cannot teach it. A create's answer
leaves out members its item read added inside an object the answer
showed, as the recorded answer did; the state still stores them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Co-authored-by: morluto <[email protected]>
* feat(environment): serve the writes callers make without recorded form fields
- A create member a recording sent empty ("", null, false, [] or {}) is no
longer required; a call that leaves it out is applied as sending that value
(body.blank).
- List views render envelope members that list the listed entities' ids
(Paperless's `all`), so creates in such collections are confirmed.
- Updates move their entity in collections kept sorted by a field they write,
and a failed whole-collection read after an update can teach that order.
- A later create that stores an unruled field otherwise is withheld at the
create, and an item read may fill values a create's answer had not loaded.
- An unsent field respelled between null, [] and {} is unchanged.
- The skill asks for one create with every field and one with only what a
person must give.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(skill): show how to read signed-in pages before recording
Agents guessed capture keys (artifacts, dom, final) to see pages behind
sign-in, which inspect_web_page cannot reach.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): accept every empty spelling recorded creates sent
A create that fills only what a person must give sends `null` where the
full create sent a string, and `null` or `false` for the same flag. The
merged plan now accepts each recorded empty value, and a call that leaves
the member out is applied as the earliest call sent it.
Also adds the spec folder for this change.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(skill): keep the vary-fields step neutral on UI or API writes
Agents read "the form" as a cue to write through the UI, where some
applications' request bodies are not exposed; a body-unusable UI write
now re-records once with request steps.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record verification for blank members and sorted writes
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): address review of blank members and sorted writes
- Bulk updates move their entities in turn against the others' earlier
values and install the final order once, as replay does.
- A member's recorded empty values stay accepted once a later call teaches
its default; a left-out member with a default keeps the default.
- Write answers spell empty lists and maps as their recorded answers did.
- A rename can teach a sort when the creates chose one by a field the
update did not write, such as ids that ascend with titles.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* feat(environment): serve the writes that later reads explain
- An item read whose object is marked deleted (deleted, isDeleted or
is_deleted true) agrees with an absent entity, as EspoCRM answers a
deleted record; such an object never seeds or adopts an entity.
- maxSize, max_results and maxResults are page sizes, for whole-collection
list lines and the list query's limit role.
- An item view also shows the fields only reads after its type's first write
show, when no initial entity holds them (Vikunja's task subscription),
without requiring them; write answers leave such a field out for entities
without it, and it no longer withholds the write as incomplete-entities.
- The skill asks updates to change the fields a person would change, and
expects a deleted object to read back as 404 or marked deleted.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer sent empty members as sent, and keep a placed create's order
- A write answer spells an empty list or map the call did not send as its
recorded answer did; one the call sent is answered as sent, so a create
sending `[]` is not answered `null` because the first recorded create was.
- A read listing the whole collection that placed a create leaves the create
only the sorted inserts whose order that read follows, so a later rename
cannot teach a sort an earlier read ruled out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer bulk updates without fields only later reads show
A bulk update's answer renders each entity with the same allowance a single
update's answer has: a field only reads after the type's first write showed,
which no initial entity holds, is left out for an entity without it, so a
bulk update of a seeded task is no longer refused for lacking it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): correct how the EspoCRM contact update task passes
The agent trajectories show gpt-oss passes it by creating a replacement contact
and deleting the original, since the served update refuses emailAddress; the
update itself is refused in both builds.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer a read of a served-deleted entity marked deleted
An application that soft-deletes still answers a deleted record by id, marked
deleted. Serving now keeps an entity a served delete removed from a type whose
recorded item reads show that, until a reset or load and within the size
data/add loads, and answers a read of it with its last fields and the flag
true, instead of refusing the read.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep unloaded create answers, check bulk answers with sent members, narrow inserts only on collection-order reads
- A create whose item read loaded values its answer had not records them
(`observation.read.unloaded`), and a served create's answer keeps them as
the recorded answer showed them, a zero-time owner included, while the state
holds the read's.
- Replay checks a bulk update's recorded answers with each target's sent
members, as serving answers them, so echoed `null` and `[]` no longer make
the plan body-unusable.
- A whole-collection read whose query sorts it shows the query's order, so it
no longer narrows a create's sorted insert candidates.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* docs(environment): record the census with main's unknown-clock rule merged
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn a sort from a rename only when moving its entities reaches it
Serving moves only the entities a write changes in a collection kept sorted,
so replay now tries a sort after a rename only when the entities no write the
read exercised changed already follow it. Seeded folders listed out of title
order no longer teach a title sort that serving would answer with another
order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep unloaded create answers a sent value did not echo, and learn no sort from query-sorted reads
- A served create keeps a value its recorded answer had not loaded unless the
call sent that member and the plan stores it as sent, so a create sending
`active: false` is answered `false` as recorded while the state holds the
read's `true`.
- A rename no longer teaches a sort from a whole-collection read its query
sorts, whose ties need not follow the collection's own order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn the soft-delete flag in the state view and serve by it
Extraction now records the member later item reads set true on a record the
application soft-deleted, and the item view carries it as `deleted_flag`.
Serving reads that fact instead of scanning recorded bodies itself, so one
owner decides what a soft delete looks like.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn the soft-delete flag from item reads with a query too
A client may read a deleted record back only with a query, such as a
cache-busting parameter. Every successful read of the item tool now shows the
flag; only plain later reads still add fields.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): learn sorts only from orders a list read confirmed
- A sort retry after a failed whole-collection read requires the entities no
write changed since a list last showed them to follow it already, rather than
excusing every entity the update's tool ever wrote: an earlier rename a list
confirmed in place stays where it was.
- An update takes its creates' sort only from a create a list read showed
placed; replay still tries a sort a create guessed from the initial order.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): drop entities read back as deleted and answer soft deletes as deleted
- A held entity a later read shows marked deleted is dropped, not adopted with
the marked object's fields.
- Bulk-update replay renders answers with the optional and later-only fields
serving uses, so a bulk update of created and seeded entities replays.
- A served delete of a soft-deleting type answers with the entity marked
deleted, single and bulk.
- A deleted flag only query reads showed is a field of the item view, so a
served tombstone keeps it where live records omit it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): move updated entities only by placed sorts and keep query-sorted reads from confirming order
Replay moved an updated entity by a sort a create only guessed, while the
update plan exported only placed sorts, so a read could confirm an update
by a move serving would not make. Both now use the placed sort. A passing
list read clears its entities from the moved set only when the state
decides the order of every entity it lists, so a read whose query sorts
it no longer hides an earlier move from a later sort check.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): try a learned sort by moving each pending write in turn, as serving does
#retrySort sorted the whole collection, which a list could match while
serving, moving each renamed entity in turn against the others' values
before their own writes, gave another order. The state now records the
values each write replaced, and the retry replays the writes since a
list in collection order last showed their entities.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): answer a served delete with the deleted flag its recorded answer showed
An application may answer a delete with the record as it was and mark it
deleted only on later reads. Delete answers, single and bulk, now take
the flag's value from the recorded answer; reads of a served-deleted
entity still show it marked deleted.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): set aside two hunks next to lines main's saved-values change rewrites
Main's #7267 changes the lines beside the blank-member docs and the
empty-spelling check in explainUpdate. Setting them aside lets that merge
apply cleanly; the commit after it puts them back on top of main's text.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): restore the set-aside hunks on main and bump the skill to 51
Puts the blank-member docs and the empty-spelling check back beside main's
slug and composed text and heldSlugRule. Main is at skill 50 with
different text, so this branch's skill changes take 51.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* chore(environment): bump the skill to 52 above #7260's 51
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): keep sorts a list showed in place ruled out, and created answers' nested members as recorded
A whole list in collection order that showed a renamed entity where a
sort by the renamed field would have moved it now rules that sort out, so
a later rename of the same entity cannot teach it. A create's answer
leaves out members its item read added inside an object the answer
showed, as the recorded answer did; the state still stores them.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(environment): compare soft-deleted read fields
* fix(environment): preserve initially deleted item observations
* fix(environment): index stored deleted entities separately
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Co-authored-by: morluto <[email protected]>
Check destination kind before overwrite policy and cover directory recovery,
regular-file controls, symlink modes and preservation.
Fixes#7444.
Independent review: APPROVE for patch SHA-256
3d65eb23d01da9b6df786ad0db337dcff79d85923f23678a5168babb88c16080,
relayed by the coordinating thread (reviewer identity not supplied in the
delegated packet). Triggers: file I/O, error handling, three shipping files.
Validation: baseline/candidate runtime builds and actual synthetic CLI probes;
initial focused run 70 passed, 2 Windows EPERM fixture failures, 3 skipped.
Final changed filesystem file 42 passed, 5 skipped; symlink semantics and
full aggregate checks remain unverified. Targeted format/lint/discovery passed.
Ship Claude Code and Codex from one release-pinned npm artifact with bundled runtime dependencies, extraction-only MCP qualification, plugin-aware recovery, and synchronized release metadata.
Integrates #2815 and the original proposal #1977. The original commits remain available on the PR branch; retain their author and assistant credit in this squash.
Validation: required CI passed, real Codex/Claude loading verified, and reviewed packaging/recovery findings addressed.
Co-authored-by: GamerDubz <[email protected]>
Co-authored-by: morluto <[email protected]>
Co-authored-by: OpenAI Codex <[email protected]>
Co-authored-by: Claude Opus 5.5 <[email protected]>
Retire redundant historical JSON copies from the active documentation tree after verifying identical published release snapshots and an immutable route-report archive. Preserve current release data, original evidence, release publication logic and historical access.
Validation: 48 release tests; documentation conformance and rendered archive link/digest; Git blob/SHA-256 checks; selected current-head CI passed. Greptile: 5/5 with no actionable findings. Codex GitHub review unavailable due quota.
Closes#7517.
Migrate MCP session tests to the named lifecycle output schema and preserve raw reply envelopes. Move R8 type/alias checks to the production field resolver and recovery guidance checks to their scoped factory; remove superseded test-only exports and an unused import.
Validation: 99 focused tests, an exact-old-schema negative control, full selected CI including all 8 shards and source tests, and Greptile source review passed. Fresh-head real stdio MCP analyzed [email protected] and retained a complete authenticated Evidence bundle after close; compact/full catalogs are unchanged. Full local typed lint hit the 3 GiB memory guard; scoped typed lint passed. Codex review quota exhausted.
Fixes#7513.
Use one typed-checker worker by default without changing semantic scope or caller overrides. Reuse the bounded CLI worker for pure argument validation while retaining fresh-process coverage and all upstream cases.
Keep registration tests independent of real package-runner downloads while asserting the actual prewarm command and startup context. Pass the advertised binary_session definition by name to package verification, retaining default SDK validation and complete catalog/contract checks while avoiding compilation of unrelated tool outputs. Preserve all four installed-tarball scenarios and existing deadlines.
Strengthen Hermes preview operation/backup assertions and clarify physical-path routing documentation.
Assisted-by: Suika (gensokyo-skills:suika-scatter-gather-planning)
Android overload selection declared a plain number, so the CLI argument
parser coerced the raw token with Number before the schema ran. An empty
token therefore arrived as 0, and hexadecimal, exponent, signed, and
float spellings arrived as their numeric value, leaving both commands to
proceed to artifact admission on a selector the caller never wrote as a
decimal integer.
Reuse the existing decimal token parser that the other integer CLI
options already use. Omitted and explicit decimal zero keep their current
meanings, and the published schema for overloadIndex is unchanged.
Validation: npm run compile, the cli-boundary vitest project for
tests/boundary/cli/cliIntegerArguments.test.ts (56 passed), npm run
typecheck, npm run lint, npm run lint:typed, npm run format:check, npm
run metadata:check, npm run verify:architecture-guards, and npm run
knip:production. The new cases fail against the base source and pass with
the change.
Fixes#7430.
Co-authored-by: S1Mon <[email protected]>
The phone header gave each link an equal minmax(0, 1fr) column and
allowed overflow-wrap: anywhere, so single-word labels such as
"Showcases", "Anleitungen", "Руководства" and "Посібники" broke between
letters at 320-430px. Links now share the row as flex items that never
shrink below their longest word. Labels wrap only at spaces, and a word
wider than the whole row can still break, preserving the enlarged-text
fit from #3598.
Refs #3597
Cache direct return/yield source expressions by immutable AST identity for all existing static and semantic consumers. Weak keys bound retention to AST lifetime; mutable analysis facts and Evidence remain unchanged. Refs #6311; full CastleIO analysis still times out.
Validation: 225 focused tests; baseline cache-regression controls; build, serial check:fast and test discovery passed. CastleIO direct-syntax facts equal with 78,272 warm lookups reduced from 2.72-2.81s to 13.0-13.4ms. Complete sourcemap-codec CLI analyses retained byte-identical Evidence; no end-to-end speedup or token reduction claimed.
CI: source 20,888 passed / 15 skipped; seven built-dependent shards, static, docs, package builds, Linux native baseline and real JS recovery CLI/MCP passed. Shard 6 Hermes path-preview failures both reproduce on exact main b9d1a61dd; Portable tests fails downstream. Greptile 5/5, no findings; Codex quota exhausted. Authorized admin bypass for unrelated baseline CI and unchecked human readiness; no human review attestation made.
Clarify the Objective-C analyzer message, coverage limitation, structured failure diagnostics, and existing opt-out/startup resource controls.
Verification: exact documentation change applied on main 79d2fc054; docs:check passed. Installed Ghidra 12.2 analyzer/scheduler and REA error/launch owners inspected. No real macOS target reproduction; original failure remains unknown. PR CI and Greptile passed; Codex review unavailable.
rea doctor reported a skill copy with an unreadable or unresolved path as identity state unknown but classified its skill:identity check as config_drift, which claims the copy has drifted. docs/installation.md keeps an unreadable skill identity unknown, and the sibling registration checks already classify unknown state as unknown. The skill check and the DoctorScope fallback now do the same.
* fix(hermes): preserve consumers of anchored registrations
Keep original alias values and comments before setup or uninstall replaces an anchored REA node. Preserve original nested bindings and retained sharing through the common YAML writer.
Fixes#3325
* test(hermes): assert retained alias object identity
* fix(hermes): preserve removed keys and surviving alias owners
* test(cli): restore retained-view cancellation phase coverage
* fix(setup): retain copied Hermes anchor comments
* chore(hermes): reverse incidental merge hook changes
Reverse the merge relative to its included-main parent to restore exact upstream bytes. Reapply only the independently reviewed shared-owner and regression patch in a normal follow-up commit. Keep all ordinary history and normal hooks; no remote push of the rejected hook drift.
Refs #3325
* fix(hermes): retain reviewed shared-owner integration
* chore(hermes): reverse incidental merge hook changes
Reverse the merge relative to its included-main parent to restore exact upstream bytes. Reapply only the independently reviewed shared-owner and regression patch in a normal follow-up commit. Keep all ordinary history and normal hooks; no remote push of the rejected hook drift.
Refs #3325
* fix(hermes): retain exact reviewed shared-owner regression
* test(native): align plist command fixtures with main
Adopt the two maintained-main fixture files as of
dd198d037f. Distinguish exact XML lint
validation from conversion commands and retain truthful provenance,
source-value, format, limitations and unknown-scalar assertions.
The original seven provenance failures reproduce before this repair.
The affected three-file group passes all 86 checks afterward. Product
code and installed-package bytes remain unchanged.
Co-authored-by: morluto <[email protected]>
* fix(native): retain bound-target dispatch failure contracts
Adopt morluto's canonical Apple dispatch metadata owner and bounded MCP
and cleanup controls from 41e10b240c.
Keep fresh admission separate from already-bound target changes and
retain access-denied classification, exact physical causes and cleanup
ownership. Update the two stale unit expectations exposed by Source CI.
The original unit pair reproduces before the correction. Current focused
regressions, package verification and installed feature controls pass.
Only the canonical Apple owner and related controls are ported; this does
not integrate unrelated later-main changes or create another contribution.
Co-authored-by: morluto <[email protected]>
* test: align plist provenance and package listing fixtures
Keep the exact XML lint invocation in CLI provenance and include the
required CLI environment owner in the synthetic package member list.
Preserve binary provenance and all LF, CRLF, literal-CR and bytecode checks.
* revert(hermes): restore canonical main before reviewed replay
This reverts commit 575a10d194, reversing
changes made to bb17e90f27.
* test(cli): retain full cancellation oracle with prepared projection
---------
Co-authored-by: hhang <[email protected]>
Co-authored-by: gensokyo-dev <[email protected]>
Co-authored-by: morluto <[email protected]>
* fix(native): classify confirmed dSYM dyld refusals
* style(native): format target refusal handling
* fix(native): confirm precise Mach-O layout refusals before export fallback
* test(cli): observe cancellation at stable source resolution boundary
* fix(native): reject contradictory dyld layout refusals
* fix(native): preserve mixed dyld diagnostic failures
* fix(native): retain the evidence behind contradictory layout refusals
When the imports and exports refusals disagree, or a refusal disagrees with
the parsed Mach-O header, inspect_macho returned a generic output error that
kept only one dyld_info capture. It dropped the header file type and the other
refusal, so the failure could not show why the admitted diagnostic was
rejected.
The contradiction now has its own error. Its diagnostics keep the parsed file
type and every admitted refusal (mode, admitted layout, exit code and output),
and the captured output is still the conflicting command. The test checks the
reason and the retained diagnostics for each contradiction case.
* test(native): allow time for real Mach-O CLI and MCP controls
Link original DEX native declarations to per-ABI function exports and relocated static JNI table candidates through shared CLI/MCP workflows. Preserve global attribution ambiguity, exact provenance, partial coverage and retained Evidence; default MCP delivery to bounded summaries with filtered/batch pages.
Use one immutable ELF relocation decoder with streaming IL2CPP materialization, cooperative bounded export/version parsing and bounded JNI state. Reuse validated session inventories without loading target libraries or executing JNI code.
Validation: relevant type/lint/build and parser/domain/public CLI/MCP regressions; genuine four-ABI NDK fixtures and independent library/slot oracles; actual JDK/JADX Conscrypt workflows with retained-source removal. Full measurements and limitations are recorded in docs/verification/android-jni-bindings-2026-10-27.md.
Closes#6403.
Restore independent PE fixtures and dependency-complete standalone artifact verifier tests. Execute existing real Windows native behavior and inspection recovery proofs before unrelated curated tests.
Validation: 101 focused tests; full typecheck; scoped lint, module/discovery/format/docs checks; valid/corrupt real-image verifier controls; real Windows artifact, behavior, and inspection recovery proof steps passed. Greptile found no actionable issues. Codex review quota exhausted.
CI remains red from exact-base failures: Source 9, Windows 7 (base 15; loader 8 repaired), and built shard 4's 7 matching failures; 7 other shards cancelled by fail-fast. Later Windows package/Node 22/24 checks skipped. User explicitly authorized bypass of established main failures. No claim of complete CI or unavailable coverage.
Fixes#7474.
Declare copied complete browser observations as abbreviated object operands, keeping canonical validation and normalization controls unchanged. Preserve all other tool declarations and Evidence identities.
Refs #4973 and #4081.
Apply caller-selected R8/ProGuard mappings to shared Android inspections, preserving raw identities, exact verification, ambiguity and provenance. Add retained summaries and bounded/cached parsing, with measured real R8/Conscrypt workflows and review fixes.
Fixes#6329.
Co-authored-by: 555aaditya <[email protected]>
Move the retained allocation report and unchanged receipt JSON into verification documentation. Keep one report producer, repair current and archived links, document test-support ownership, and remove retired lint inputs.
Fixes#7458.
Validated focused/Python checks, report consistency, docs rendering, and real Bash CLI/MCP facts against an independent oracle. Greptile reviewed all seven files without comments. CI source/Windows/shard-8 failures match clean-main or existing-main controls; the requester authorized merging with those established failures. Codex review quota was unavailable.
768b58c6d (fix(setup): honor Droid home override) made factoryDirectory
resolve from FACTORY_HOME_OVERRIDE, so setup now writes the registration
and the personal skill copy under that home instead of ~/.factory. The
Factory Droid section still described only ~/.factory, and the page names
every other client's home override (KIMI_CODE_HOME, QODER_CONFIG_DIR,
QWEN_HOME, HERMES_HOME, CODEX_HOME, GEMINI_CLI_HOME, PI_CONFIG_DIR, DSH_HOME),
so Droid was the only one missing.
Match v1 APK certificate and signer serials by signed integer value. Normalize each comparison identity once and preserve raw certificate bytes and Evidence.
37 source/integration tests, 23 real Android SDK verifier cases, SDK-valid padded APK CLI/MCP oracle and serial static checks passed. Hosted real-apk-signature passed. CI27pass7fail2skip7cancel; exact-main controls reproduce17source and14targeted existing-main assertions. NativeWindows/Mac/Node22 baseline gaps and canceled shards are documented. Greptile sole finding is unchanged upstream browser privacy code, scoped and resolved without a fix claim. Requester authorized readiness and unrelated-main failure bypass; no human review is attested.
Co-authored-by: gensokyo-dev <[email protected]>
Extend static manifest declarations and retain precise source locations and historical unknowns. Share bounded selected component views across CLI and MCP with complete parent Evidence.
Verified real ApiDemos, an AAPT2-built fixture, legacy Evidence, default-package metadata and relevant CLI/MCP contracts. Recorded whole-task token estimates, timing/resource limits and independent main-failure controls in the PR.
Co-authored-by: Harsh <[email protected]>
Normalize transport-specific discovery facts at their shared owner; preserve upstream duplicate-ID rejection and correct discovery-time attachment uncertainty in all consumers.
53 focused regressions, serial static checks and real Chrome CLI/MCP fault injection/inspection pass. Greptile final head 5/5. Exact-main controls reproduce all 17 source and 21 shard-8 failures; seven shards cancelled and remain unverified. Requester authorized readiness and unrelated-main CI bypass. Human review is not attested.
Co-authored-by: gensokyo-dev <[email protected]>
Keep cookies with opaque partitions distinct from unpartitioned identities while retaining value hashes and incomplete comparison coverage. Preserve known partition hashes and use the existing shared CLI/MCP fingerprint owner.
Validated with 63 source and 57 storage boundary tests, builds/static checks, and real Chrome156 CLI/MCP controls. Live opaque-cookie emission was not reproduced; its protocol regression is explicit. Hosted CI's unrelated failures were reproduced against exact main a905a71e3; seven fail-fast-cancelled shards remain unverified; Greptile's sole existing-main finding was scoped and documented. Human readiness remains unasserted; bypass authorized by the requester.
Co-authored-by: gensokyo-dev <[email protected]>
Remove the derived package renderer property from new application graphs. Use renderer_declarations for source fields and package loads edges for the selected candidate; previously retained Evidence remains readable.
Add validate-browser-scenario for offline schema and semantic checks of inline JSON, local files and stdin. Return compact validity/action/secret counts without reading secret values or discovering, launching, attaching to or navigating a browser.
Execution and validation share CLI input admission and the canonical domain schema. Preserve existing capture lifecycle, diagnostics, cancellation and caller-relative executable paths.
Integrates #7149 with attribution. Fixes#5488. Validation includes 24 focused tests, static/build/docs checks, and the preserved real PocketBase scenario with its backend stopped and secret variables unset. Verification and resource/token measurements are recorded in docs/verification/browser-scenario-validation-2026-10-27.md.
Co-authored-by: Ace <[email protected]>
Save typed response fields and reuse generated IDs or authentication values across browser scenario requests, UI observations and navigation through the shared CLI/MCP workflow.
Preserve exact numbers, retained Evidence, explicit source/coverage limits and prospective page-value semantics. Bound response decoding, retained spellings and request serialization before allocation, and protect marked response values across source records, follow-up text and planned screenshots.
Integrates Edgar Arakelyan's #7162, including follow-ups through 9f5191ab0. Final local validation: 560 feature-branch tests and 561 current-main integration tests, static/build/docs checks and real PocketBase CLI/MCP privacy workflows. Whole-task comparison and verification limits are recorded in docs/verification/browser-saved-values-2026-10-26.md.
Co-authored-by: Edgar Arakelyan <[email protected]>
Replace overlapping traversals with one bounded relationship owner. Preserve superview direction, treat conflicting or unsupported registry placement as unknown, and retain source facts and coverage.
Keep historical inventory summary variants with their projection owner and share embedded PCK trailer observations while preserving each reader’s admission policy.
Share exact live target variants across open and session responses, preserving historical Evidence and snapshot admission. Consolidate equivalent ELF metadata aliases at their owner.
Pass parsed MIPS program declarations to shared-load analysis and validate IDA string address selectors before enumeration, including empty inventories.
Move shared function, dossier, inventory, reference and output semantics from the legacy Hopper module to native domain owners. Migrate provider, adapter, test and verifier consumers directly.
Reuse the read-side capture validation before publication so exported bundles can be imported. Preserve historical and foreign record identity and existing destination files on failure.
Resolve IDA compatibility facts in one place and migrate capability, session, and verifier consumers. Move native search fields to the domain and reuse residual unknown fields for filters.
Migrate schema, installer, ASN.1, and tooling consumers to their owning definitions. Remove the redundant CI task entry and the deprecated Cutter command-local JSON flag.
BREAKING CHANGE: Replace Cutter --json=true or --json=false with --json-output=true or --json-output=false. Bare --json still controls CLI result formatting.
Use canonical owners for literal matches, tool families, MCP schema presentation and IDA protocol selection; remove unused web-bundle reexports. Remove the duplicate CLI command name and keep the canonical compare workflow.
BREAKING CHANGE: rea compare-bundles has been removed. Use rea compare instead. The MCP compare_bundles tool is unchanged.
Parse JEB failure envelopes at the provider boundary and retain the original response, recovery facts, and selected inputs. Do not infer an unsupported target from undocumented error codes; preserve completed primary observations when metadata requests fail.
BREAKING CHANGE: JEB engine refusals now report execution_failure rather than unsupported_target. Consumers should inspect the retained upstream diagnostics for the engine's reason and recovery facts.
Authenticate exact retained parents against current and documented historical summary projections, including caller-selected display paths. Keep source facts, Evidence commitments and unsupported-projection rejection intact.
Attach to external browsers without implicit default-context overrides, and own temporary focus only on the selected page. Preserve unrelated page media and exact page inventory through CLI/MCP. Remove bootstrap-page retries; retain late/failed cleanup ownership and verified SDK notices.
Assisted-by: Kogasa (gensokyo-skills:kogasa-surprise-testing)
Co-authored-by: gensokyo-dev <[email protected]>
Compare each retained member's exact runtime key and role without treating archive order as semantic identity. Keep one-sided legacy declaration gaps unknown, preserve proven changes with unavailable content, and retain original Evidence.
Record packed and unpacked ASAR file declarations at the reader boundary and compare known storage changes without treating historical omissions as packed. Preserve known differences when bytes are unavailable and retain existing integrity precedence.
BREAKING CHANGE: ASAR file occurrences may now contain unpacked: false. Consumers that validate only true must accept an optional boolean; absent historical fields remain unknown.
Normalize function address and name without requiring an unused upstream size. Preserve complete upstream observations and the explicit unknown body coverage across IDA protocol profiles.
Reuse completed embedded-format observations only after validating the admitted file identity and state, and centralize reader construction across inventory and extraction. Preserve early refusal, selected-path checks, ZIP ownership, and each operation's traversal policy. A source changed between preflight and inventory now fails integrity validation.
Preserve status independently of malformed content and use one declaration reader for route grouping, state extraction and manifest eligibility. Verify the partial facts through CLI and MCP views.
Derive each facet's row schema from the Mach-O producer and validate it against the requested page. Preserve the existing wire shape and historical defaults.
Distinguish omitted slice fields from declarations that cannot be interpreted, and preserve their source in native results and artifact limitations. Keep structural fallback and capacity failures explicit.
Use one owner for exact recorded timestamps and documented native-Date fallback in state and probe replay. Retain the source and disclose native precision and host-timezone dependence in state limitations.
Validated historical chronology, timezone and unknown-duration regressions against prior behavior.
Keep missing or ambiguous scenario request durations unknown, retain learned clock rules, and withhold plans that cannot generate an observed clock. Refuse such writes in older imported states before changing state or generators.
BREAKING CHANGE: The write-plan reason can now be unknown-clock; strict clients must accept the new output value.
Validated state, CLI and advertised MCP workflows, documentation, and focused clock and cleanup regressions.
Retain valid request starts independently of duration and require observed completion for interval-based state and write inferences. Bound indexed interference checks across captured writes while preserving ordinal fallback and unmodified Evidence.
Validated state, comparison, CLI and MCP workflows with known, unknown, overlapping and redacted timing controls; check:fast passed.
Preserve producer-supported recovery guidance after an existing-image DMG attachment refusal, without borrowing or detaching existing mounts. Keep diagnostics, root provenance and unknown coverage; centralize guidance ownership and run the real mounted-image regression in macOS CI.
Co-authored-by: gensokyo-dev <[email protected]>
The collection-paging paragraph says complete Evidence stays accessible through
`get_evidence`, but no tool by that name exists. The session family registers
`get_evidence_bundle`, which is the only name the catalog, the contracts and the
CLI equivalents use.
Every other page already uses the full name, so this was the only place a reader
or agent would look for a tool that cannot be called.
Admit same-major Ghidra releases at or above the verified minor through the existing compatible/unverified policy. Preserve the observed version, verified12.1.4 pin, installation-declared Java floor and separate NativeAOT verification gate.
367 targeted tests/static/docs, all10 real Java bridge-source compilation, and actual Linux12.2/JDK25 CLI/MCP queries and cleanup pass. JDK21 remains refused for12.2. Exact-main controls reproduce all3 unrelated CI assertions; canceled shards and an apt-blocked binary-layout lane remain unverified.
Co-authored-by: gensokyo-dev <[email protected]>
Distinguish invalid supported timestamp spellings from unrecognized legacy formats at the recorded-date owner. Probe replay retains ordinal order for invalid dates without changing the existing legacy fallback.
Validated date parsing, state ordering and comparison workflows; the prior implementation reorders the invalid-date regression.
Retain a limitation when ASAR block integrity fields are present. Whole-file hash verification does not establish block verification or Electron runtime acceptance.
Validated inventory and extraction workflows, including partial and malformed block declarations.
Use the shared recorded timestamp interpretation for replay ordering, including exact fractions and UTC interpretation of supported zone-less forms. Preserve native-Date fallback for unsupported historical spellings.
Validated environment comparison, state ordering and CLI tests; five new cases reproduce prior drift.
Carry reported algorithms and hashes into listing limitations when the declaration cannot be compared, retaining the existing verification and extraction policies. Bound escaped text before allocation.
Validated ASAR inventory, extraction, cleanup and MCP workflows; prior code omits the declaration details.
Read bounded plist format probes and Linux resource-limit reports through positive short reads before classifying their contents. Preserve explicit EOF, report capacity, physical failures and descriptor cleanup ownership.
Validated 66 focused tests and check:fast; nine new fault-injected cases fail against the previous implementation. Linux procfs was not verified on a real Linux host.
Share provider availability, admission, logging and Evidence delivery across browser, Electron and runtime tools. Reuse retained summary delivery for Python and BEAM selected views while preserving preflight and complete parent Evidence.
Validated 47 focused MCP and retained-view acceptance tests and check:fast.
Reuse the positional range owner for target metadata, Hopper Mach-O selection and Kuna spans. Preserve physical failures, completed span evidence and cleanup ownership across cancellation and short reads.
Validated 147 focused file and lifecycle tests; ten new cases reproduced failures against the prior implementation.
Resolve ordered descriptor values at the shared JavaScript fact owner and preserve uncertain intrinsic writes instead of stale exports. Normalize ordinary method values and reuse canonical primitive, prototype, ambient identity and ordering facts across consumers.
384 selected tests, format/static/docs checks, all final CI jobs and reviews pass. Actual Node controls cover 35 bundles; one possibly repeated factory explicitly stays unknown. Shipped Obsidian selected-source complete Evidence matches base/fixed CLI and MCP, with coverage and resource measurements in the PR.
Co-authored-by: gensokyo-dev <[email protected]>
Continue reading Mach-O and Electron ranges after positive short descriptor reads, while preserving explicit EOF, cancellation, file identity and cleanup semantics. Advance fuse scanning and hashing by the bytes actually read.
Validated focused filesystem, Mach-O, dyld, Electron and MCP workflows: 117 tests passed, two Windows-only tests skipped; check:fast passed.
Use the common analysis-view reader for Rust and native inventory validation, preserving historical defaults and original Evidence while retaining schema-specific validation.
Validated 123 focused view and service tests and check:fast.
Centralize completed file observations and cleanup error projection across stable-file providers while preserving each provider’s failure classification and exact cleanup leases. Cover BEAM finalization cancellation and devicetree cleanup retries.
Validated 223 focused lifecycle and MCP tests and check:fast.
Parse shared request and response fields at the CDP boundary while preserving capture, session and runtime admission policies. Treat empty request IDs as unknown so unrelated events cannot be joined, and retain redirect occurrences and partial response evidence.
Validate 221 focused tests, fast checks, and real Chrome network and runtime verifiers.
Derive member roles at the table-of-contents boundary and share declaration and Python-module interpretations across validation, projection, metadata, header reconstruction and nested PYZ traversal. Retain source type codes and generic-file behavior for other printable types.
Validation: 217 focused archive, CLI and MCP tests passed, including mixed-role byte and metadata checks. Typecheck and lint passed.
Share Field scalar facts and use one coded-index tag registry for widths and decoding. Preserve reserved constructor tags, generic-instantiation restrictions, heap failure order and selected-field signature admission.
Derive native import member kind from the validated MemberForwarded coordinate independently of name lookup, retaining known field or method identity when its display name is unreadable.
Validation: 368 focused source, CLI and MCP tests passed; five lost-kind regressions reproduced on the previous commit. Heap-width, coded-width, reserved-tag and selection boundaries are covered. Typecheck and lint passed.
Have the MethodDef row owner derive code type, managed kind and the P/Invoke flag for both body decoding and native-boundary inventory. Remove duplicate masks and unreachable fallbacks while preserving raw flags, zero-RVA precedence and ImplMap-specific declaration policy.
Validation: 280 focused source, CLI and MCP tests passed, including cross-workflow flag combinations and declaration/body policy boundaries. Typecheck and lint passed.
Centralize CDP script/context declarations and frame-claim interpretation across web capture, Electron, V8, active runtime and storage seeding. Preserve each workflow's admission, provenance and retention policies while removing superseded helpers and schemas.
Malformed or conflicting frame claims now leave passive scripts unattributed. Reused context IDs replace future frame authority without rebinding earlier scripts, and source readers reject empty script IDs.
Validation: 153 focused tests and check:fast passed; 13 regression failures reproduced on the prior commit. Real Chrome script export and runtime CLI/MCP verifiers passed, including correctly reported uncertain cleanup after intentional target closure. Electron regression coverage used simulated CDP.
Pass validated blob location and decoded metadata with their admitted descriptor from classification to the reader. Keep root cleanup ownership, freshness checks, format validation order and direct-reader decoding intact, then release borrowed facts after preparation.
Validation: 147 focused source, CLI and MCP tests passed, including real Node blob production; check:fast passed. Optional injected-ELF Docker verification was unavailable because the Docker daemon is not running.
Derive dependency and generic-error facts once per owning task, chunk or multifile stage. Reuse those facts for failure classification and dependency aggregation, preserving unnamed identities, severity precedence and raw diagnostics.
Validation: 70 focused tests passed; 41 platform-guarded tests skipped on macOS. Typecheck and lint passed. Real Unblob/Linux provider behavior was not run.
Share CPython magic framing, version and header facts while keeping standalone marshal coverage and PyInstaller reconstruction eligibility distinct. Preserve legacy headers, opaque unknowns and source bytes.
Validated 272 focused tests across source, CLI and MCP workflows, a 38-test reader recheck, and check:fast.
Parse each frame event using its declared protocol fields and retain loader and ancestry only across same-document updates. This removes alias-based attribution, invalidates malformed committed frame facts, and prevents stale WebMCP tools surviving a same-URL document replacement.
Migrate Electron, screenshot, capture, timeline, and execution consumers to the shared facts owner and retire the old helper exports. Validated 428 focused tests, test discovery, typecheck/lint, and real Chrome CLI/MCP network capture; Electron and WebMCP regressions use simulated CDP.
Page retained web-bundle endpoint observations and inferences through shared CLI/MCP projections, preserving source identities, scoped unknowns, coverage and complete Evidence access.
Based on HarsH’s original implementation in #7213 (6b1d5d42). Real Chromium/CLI/SDK verification and whole-task cost limits are recorded in docs/verification/web-bundle-paging-2026-10-26.md.
Explicit user authorization permits leaving human-readiness declarations unchecked and bypassing reproduced main failures (ADB listing fixture, package catalog expectation, untouched APK verifier formatting). No human-review attestation is made.
Co-authored-by: HarsH <[email protected]>
Retain file identity after Rust, web artifact, shared selected-input and historical capture reads complete before close or cancellation fails. Preserve descriptor-set snapshots in diagnostics alongside sanitized capture observations.
Classify canonical physical capture failures before cleanup composition; source permission errors and bare syscall failures keep their meaning through late cancellation and retry. Keep explicit sensitive-value redaction on retained facts.
Validation: 247 focused cases passed, two Linux-only cases skipped; build, typecheck, lint and formatting passed. Real descriptors with controlled finalization faults establish reader ownership and facts. Worker reply and multi-file database completion semantics remain separate audit work.
Normalize completed stable-file path, digest and length in one shared observation owner. Migrate thirteen identical provider projections while preserving their selected paths, schema checks and cleanup composition.
Keep specialized metadata and worker reply observations at their existing owners. Validation: shared real-descriptor coverage and provider lifecycle tests passed; build, typecheck, lint and formatting passed.
Retain exact completed file path, digest and length in Python pyc and Erlang BEAM failure observations. Translate Python close failures at the provider boundary and bind them to the retained descriptor cleanup leases so later successful retries resolve prior errors.
Exercise actual services and file handles with completed reads, cancellation during finalization and physical read failure followed by cancellation. Assert no fabricated snapshot on early failure and resolution of every prior cleanup report.
Validation: six focused descriptor regressions plus existing filesystem, parser and MCP coverage passed; build and static checks passed. Authored format fixtures establish portable reader behavior, not CPython/Erlang runtime execution.
Count retained UTF-8 text rather than UTF-16 code units, including exact-byte base64 spellings. Bound the decoded version-name map independently from output expansion, releasing replaced entries from its live storage charge.
Add real service/provider boundary controls for accepted and over-budget symbol names, version names and unused version metadata. Existing per-string bounds remain the allocation owner.
Validation: native parser, filesystem and CLI/MCP views passed; build, typecheck, lint and formatting passed. No serialized-report size cap, OOM reproduction or performance improvement is claimed.
Charge PE alias names when storing them, before target interpretation can suppress output rows. Validate unused-slot names as bytes without retaining or decoding strings. Keep row and forwarder accounting separate so names are charged once.
Preserve alias ordering, malformed unused-name issues and ordinal-only suppression. Add production service/provider retention regressions using compact shared-pointer fixtures.
Validation: 43 focused parser, filesystem, CLI and stdio MCP tests; fresh build, typecheck, lint and formatting passed. Three bounded Node 24.22.1 measurement pairs showed median process RSS high-water 154000 to 93168 KiB; timing and sampled family memory have cache/startup/sampling limits. No Windows loader acceptance or OOM reproduction claim.
Capture one environment snapshot per admitted WABT, JADX, firmware and Wakaru operation. Use it for tool selection, session identity and child launches while preserving configuration updates between operations.
Regressions reproduce changed JVM options under an old session key, mixed engine/limiter selection and environment drift between subprocesses. Keep selected-wrapper and cleanup coverage in a shared fixture with separate scenarios.
Validation: 115 distinct focused tests passed across the final five-file selection, including seven selected-command cases after refactoring. Fresh build, typecheck, lint and formatting passed. Tests use real subprocess fixtures; no real JADX/WABT/Binwalk/Wakaru engine or native Windows verification is claimed.
Retain authenticated provider source records independently of snapshot replay eligibility and return them explicitly to plain application callers, including late cancellation. CLI/MCP commit source records before delivering links; function dossiers share the canonical Evidence constructor. Keep directory replies compact and preserve successful retry verification.
Real Ghidra CLI/MCP investigations verify 652-function ls, historical report provenance, cross-session resume, unchanged source/project content and complete offline Evidence. Final full 14,338-function libcrypto comparison matches all metadata/pseudocode facts and independent fresh-import observations: 14,194 written, 143 external, one recorded deadline; 494.03 s, 445.63 CPU s, 1.584 GiB sampled family peak under 2-CPU/8-GiB guards, with no OOM or cleanup failure.
144 focused tests and static/docs/site checks pass; latest-main integration passes 289 cases plus typecheck/lint and actual CLI snapshot source retrieval. Final CI source/static/build/docs/quality/package checks pass. Four ADB shard failures also occur in main CI 40349559391; cancelled coverage remains unverified and author declarations stay unchecked under the user's explicit bypass authorization. No human-review attestation is made.
Parse literal import framing and retain ambiguous boundaries. Consult undefined symbol-table names only when needed to distinguish producer spacing, with import and export captures owned separately.
Validation: 59 native tests passed and real Apple tools on inert arm64 fixtures reproduced the old loss and verified the fix. Fresh build, typecheck and lint passed. Small warm-task median elapsed was unchanged; cold costs and agent tokens were not measured.
Consume Toybox’s single NAME separator without stripping source whitespace. Keep MCP subprocess fixtures independent of ambient PATH and validate calls through the existing named-schema helper.
Validation: focused ADB provider/parser and MCP tests passed; no real Android device was used.
Distinguish local provider readiness failures from target-support refusals in shared CLI/MCP diagnostics, preserving original codes and evidence. Avoid claiming configured providers are unconfigured.
120 focused tests, static/docs checks and real ELF CLI/MCP verification pass. Hosted CI has the same four ADB failures as exact main d9908fd7a; three canceled shards remain unverified. Merge uses the user's explicit authorization to bypass proven main failures. Fresh Codex and Greptile reviews are clear.
Fixes#7214
Co-authored-by: gensokyo-dev <[email protected]>
Respect Node asynchronous exec/execFile method-default stdio instead of interpreting an ignored option. Remove the superseded callback-shape query while preserving caller input, Evidence, uncertainty, argv and fork controls. Independently validated on current main through actual Node pipe capture, matching real-app CLI/MCP facts, 81 targeted tests and static checking. Closes#7217.
Require operator-declared hostnames before serving MCP HTTP outside loopback, preserving Host/Origin gates. Parse hostname identities once, share typed startup recovery across CLI/MCP, and announce an admitted client URL. Generated AgentEnv bundles declare their client/service names and validate DNS/HTTP name identity before export.
Validated through real packaged HTTP requests, direct SDK connection, compiled export refusals, 184 focused tests, static and documentation checks on the refreshed main integration. Final CI and review evidence is recorded in the PR body.
BREAKING CHANGE: non-loopback deployments require MCP_ALLOWED_HOSTS or rea mcp --http --allowed-hosts; bundle names must be unchanged HTTP hostnames within the 63-octet DNS label bound.
Co-authored-by: gensokyo-dev <[email protected]>
Route six artifact verifiers through the canonical complete MCP output and error parsers. Remove duplicated JSON/schema interpretation while retaining Evidence authentication, provider provenance, independent CLI parsing and CLI/MCP parity checks. Validate malformed success/error envelopes at the shared boundary.
Validation: 11 boundary tests, Go compiler verifier (30 cases), and synthetic IL2CPP/Flutter verifier workflows passed; typecheck/lint passed. PCAP readiness passed with stub tools, but capture inspection is unsupported on macOS. Linux-only binary-layout and EVM verifier toolchains were unavailable.
Share release JSON decoding, vendor fields and package integrity checks across macOS setup and Linux publishing/cache reads. Reject malformed UTF-8 instead of repairing macOS metadata, retain established initial BOM tolerance, and preserve complete vendor rows through admission and cache retention. Keep platform channel, signature, installation and cleanup policies unchanged.
Validation: 134 focused tests passed with 21 platform skips; typecheck and lint passed. Linux-only signed-cache cases were updated but skipped on the macOS host. No live Hopper installation performed.
Reuse selected Ghidra projects through verified private read-only copies, with shared CLI/MCP configuration, committed project identity, import-SHA verification and explicit historical/unknown analysis semantics. Preserve exact cancellation and platform capability boundaries.
Actual Ghidra 12.1.4/JDK21 Linux CLI/MCP investigations matched fresh import on 652-function ls and 14,338-function libcrypto inventories and pseudocode; authenticated complete Evidence, mismatch/staleness refusal, saved historical reports and shared /tmp projects verified. 958 focused tests passed with four platform skips; build/static/docs/format passed. Real macOS/Windows reuse remains unverified; Windows reuse is unavailable. See PR verification for measured whole-task costs and final CI attribution.
Co-authored-by: akushonkamen <[email protected]>
Co-authored-by: morluto <[email protected]>
Give the object-structured provider envelope one owner that retains original dictionary members. Remove duplicate schemas, text extraction and JSON-or-text decoding plus the downstream casts that repaired lossy record clones. Keep provider-specific error, truncation, wrapper and payload-selection policies in each adapter.
Validation: IDA/JEB compatibility tests, pinned SDK transport tests and provider regressions passed; check:fast passed. Real IDA/JEB applications were not exercised.
Preserve XML, OpenStep and binary dictionary keys at their format owners. Share validated binary reference facts with materialization while retaining scalar precision, UID identity and bounded expansion. Migrate archive, Interface Builder, native and app-bundle consumers, then remove obsolete key-omission accounting and schema copies.
Validated source regressions and compiled CLI/MCP workflows, including native plist fidelity and archive identity. Typecheck, lint, docs and test discovery pass. No performance benchmarks were required.
Use portable hex escapes for Windows absolute-path separators in the shared local path schema. Preserve host path semantics and caller spelling; validate Annex B/u/v compilation and Node win32 agreement. Closes#7151.
Publish portable browser request-path and HTTP-token patterns at their canonical owner, preserving input spelling and existing policy. Independently reproduced and verified through actual full-profile stdio MCP catalog and domain/MCP regressions on latest main. Closes#7168.
Preserve observed failed-response HTTP status through Linux Hopper setup diagnostics without changing download rejection, retry, or installation policy. Independently validated on latest main through red/green regressions, actual TLS sockets, focused checks and CI. Closes#7182.
Keep binary_session available when Linux owned-child probes lack procps ps. Preserve typed ownership-prerequisite failures in direct execution and report per-provider readiness in status. Independently validated on latest main through real stdio MCP plus focused regression, static and documentation checks. Closes#7164.
Upgrade Zod to 4.6.6 and refresh locked license inventories. Preserve REA's established timestamp spellings through one shared UTC/offset grammar, including completion-verifier consumers. Size the stdin resource regression from the child's actual heap capacity.
Fixes#7173
Includes the dependency and inventory changes proposed in #7184.
Co-authored-by: tijo95 <[email protected]>
Scope MCP discovery, setup, doctor, status and recovery guidance by selected tool families using one contract inventory and shared catalog identity. Preserve default discovery, canonical validation and complete Evidence.
Fixes#3246. Builds on #3026. Real libcrypto workflow verifies all 5,367 exports and identical complete Evidence. Static/source/build/docs/Windows/package/Greptile checks pass; ten assertions reproduced identically on clean main and feature across Node 24/22, with cancelled coverage explicitly unverified. Merge uses the user-authorized bypass for baseline failures and unchecked human-readiness declarations, without attesting human review.
Co-authored-by: 0xLLVEL <[email protected]>
Co-authored-by: Codebuff <[email protected]>
Co-authored-by: Claude Opus 4.6 <[email protected]>
Add opt-in bounded output-schema advertisements while preserving the full default, canonical server validation and complete structured content. Verify real stdio discovery under a 1 MiB receive cap, all 5,367 libcrypto export coordinates, complete Evidence parity, whole-task serialization estimates and resource bounds.
Fixes#2512. Integrates #7054 by Akshad Jogi with original authorship retained.
Co-authored-by: Akshad Jogi <[email protected]>
Move shared request parsing into its application owner, migrate all consumers, and remove six duplicated admission blocks while preserving cancellation and validation policy.
Keep operator-selected JavaScript paths in verifier parent Evidence while retaining canonical comparison guards. Validated with 204 passing tests, 3 Linux-x64-only skips, typecheck, lint, and test discovery.
Share ordered schema positions between Gemini literal checks, DeepSeek regex
checks, and catalog array-shape checks. Reuse the canonical keyword and local
reference owners; keep unreferenced instance literals opaque and unresolved
external or dynamic targets unknown.
Gemini and catalog checks now catch constraints in schemas reached through
references into examples. Preserve source paths, resource scopes, cycles and
regex occurrence counts. Align fixture first-error selection with the native
Gemini check and use consistent bracketed array locations in diagnostics.
Validation: 55 focused tests passed, including full/compact MCP catalogs;
typecheck, lint, discovery, and Node source-loader checks passed. The old
Gemini scanner reproduced the local-reference false negative. Installed
Gemini and DeepSeek clients were not run.
This reverts commit 7336a566ac (#7119).
After #7120 and the other merges of 2026-10-25 grew inspect_analysis_view's
complete input schema (71,762 bytes plus 3,003 bytes of examples), the 64 KiB
fitting from #7119 could no longer keep instance level 3 and fell back to
level 2. That summarized the inline Evidence source and the view selector
collections, so main CI failed in toolInputReferences, toolSchemaValidity,
analysisViewMcp, absoluteLocalPathSchema, javascriptFactsViewMcp and
godotPckMcp. #7119's own CI ran against bc143125e, before that growth.
Reverting restores the complete full-profile advertisement while the size
limit is redesigned against these contracts (#4081 stays open).
Co-authored-by: Claude Opus 5.5 <[email protected]>
Migrate twelve application service files to the existing provider input owner,
preserving schema causes, cancellation precedence, caller selection, and
retained Evidence. Keep flows with distinct error policies separate.
Replace the DeepSeek verifier's recursive object scan with schema-position
traversal using the canonical keyword and local-reference owners. Literal
pattern fields no longer cause false regex failures; patternProperties keys
are checked alongside actual pattern constraints in native regex modes.
Validation: 271 focused workflow/service tests and six verifier regressions
passed; static checks, discovery, and actual Node source-loader smoke passed.
The baseline scanner reproduced the new regressions. Installed DeepSeek client
execution was not run.
Retain validated DOS relocation records in their domain owner and consume
those facts in Ghidra load-image preparation. Preserve ordered duplicate
aliases, overlapping writes, and original/final byte evidence.
Share FAT header facts across generic, native, and Hopper readers without
merging their distinct admission policies. Replace managed method bodies'
two identical mapping lookups with one owned offset-and-extent result.
Validation: 660 focused tests passed across 19 suites; typecheck, lint, and
test discovery passed. Hopper/Ghidra checks used fixtures and simulated
provider observations; no real external provider execution was performed.
Share ELF note headers across note inspection, Flutter build IDs, Node SEA,
and recorded-crash source binding. Decode REL/RELA records once for MIPS
shared-load analysis and IL2CPP linking, removing superseded scalar helpers.
Share the CLR header prefix between desktop detection and managed PE parsing
while preserving their distinct read bounds and admission policies. Cover
those differences and ignored R_NONE fields with production-boundary tests.
Validation: 493 focused tests passed, two existing skips; static checks and
test discovery passed. Public managed/desktop CLI and ELF notes CLI/MCP
workflows exercised; no real external analysis providers were run.
Reuse the existing provider input admission helper across twelve services,
preserving cancellation, schema errors, path policies, and Evidence handling.
Share ELF dynamic records across Dart, MIPS, and IL2CPP readers, and reuse
Mach-O segment and section records for Node SEA location.
Require canonical DOS ownership before GPU PE section discovery so relocation
bytes cannot masquerade as a PE pointer. Cover valid short DOS headers and
the malformed case through both inventory and the public CLI.
Centralize ELF section and Godot trailer declarations while preserving caller admission policies. Route six PE entrypoints through canonical DOS ownership so relocation data cannot establish a PE header, and update NativeAOT interpretation identity and verifier dependency coverage.
Share exact ELF program-header facts across nine consumers while preserving their admission and precision policies. Centralize conformance path and digest binding, retain verifier-specific checks, and route fixture-owner changes to CI qualification.
Allow native matching representation budgets to be selected at runtime or launch while retaining the 96 MiB default and main shared-provenance accounting. Preserve actionable configuration recovery through the setting owner and verify complete CLI/MCP matches for 20k/50k synthetic inventories and real stripped LLVM 14/15 inventories.
Fixes#6406.
Based on #7112 by Kaoru0822; original authorship and co-author credit retained.
Co-authored-by: Kaoru0822 <[email protected]>
Co-authored-by: Claude Opus 5.5 <[email protected]>
This reverts commit 7336a566ac (#7119).
After #7120 and the other merges of 2026-10-25 grew inspect_analysis_view's
complete input schema (71,762 bytes plus 3,003 bytes of examples), the 64 KiB
fitting from #7119 could no longer keep instance level 3 and fell back to
level 2. That summarized the inline Evidence source and the view selector
collections, so main CI failed in toolInputReferences, toolSchemaValidity,
analysisViewMcp, absoluteLocalPathSchema, javascriptFactsViewMcp and
godotPckMcp. #7119's own CI ran against bc143125e, before that growth.
Reverting restores the complete full-profile advertisement while the size
limit is redesigned against these contracts (#4081 stays open).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Refs: #7135
Scope expected Ghidra input failures to the active request after the authenticated startup handshake. Preserve operation output, identity and log coordinates while retaining cumulative output with its session owner; lifecycle and unexpected failures keep full diagnostics.
Share remote input-code classification with domain error projection. Verify startup rejection, healthy continuation, cleanup, unknown errors, OOM diagnostics and real CLI/MCP workflows. The measured unknown-symbol reply falls from 13,170 to 2,209 bytes with unchanged successful facts and Evidence IDs.
Add optional native load-image summary delivery and retained selected-row pages through shared CLI/MCP analysis views. Preserve source verification, provenance, unavailable collections, exact row identity and opt-in byte/check payloads.
Preserve direct caller paths during shared CLI summary publication and cover raw/DOL projection contracts through the actual SDK. Real Ghidra ELF CLI/MCP captures, retained and file-backed pages, focused regressions and static checks passed; the PR records existing main schema failures and the JavaScript resource-recovery CI uncertainty.
A write below a slot that stores a value read from the same slot, such
as `s.c = s.c.p; s.c.i = 1` on a parameter, revisits the binding with
one more segment after the stored key on every turn. The cycle closure
only recognized segments added at the front of the path (`node =
node[key]`), so the work-list never terminated and exhausted the owned
worker heap (CodeMirror 5 `mode/stylus/stylus.js`).
Recognize a contiguous segment inserted after a stable base, check that
repeating it after that base can never be pruned, and escape through it
once, as the self-alias rule already does. Front insertions keep their
existing order and result.
Fixes#7130
Co-authored-by: Claude Opus 5.5 <[email protected]>
Reference-store identity queries resolved the written target as a value,
so once a step consumed the written key the walk continued into the value
read from that slot and recorded it, and everything reachable from it, as
an object the store mutates. Every value stored into one local object at
unknown keys then received an entry for every such store, and each new
store re-queued every earlier one. Ambient-write indexing grew cubically
in the number of such stores.
Receiver queries now drop items whose projection no longer ends at the
written key. Stored-value selection has one owner, which the resolver and
the store fixpoint share. The fixpoint re-queues a dependent only when a
changed entry can reach it with a selection that its receiver query keeps.
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(mcp)!: return a compact binary_session status by default
binary_session {} with no target open returned about 66 KB, most of it
provider rejection diagnostics and candidate capability sets that the
skill's recovery flow re-reads on every retry. 6.8.0 grew it by 21%
because the shared regex capability limitation began repeating setting
bounds and step-by-step recovery for every search operation of every
provider (#6099, #6267, #6301, #7049) and IDA gained that limitation for
its two search operations (#4862).
The regex limitation now states the effective deadline and heap, their
sources and where to change them; the deadline and heap errors keep the
full recovery. binary_session gains detail: summary (default) or
complete. The summary keeps open state, target identity, the binding,
route capabilities, run, activity and health status, each candidate's
availability and target-support status, code and reason, and every
tool's availability, reason, remediation and modes, and lists the
fields it leaves out in omitted_fields. Complete detail returns the
previous shape. The CLI capabilities and providers commands keep
complete provider facts through the same application projection.
BREAKING CHANGE: binary_session {} no longer returns
analysis_provider_candidates[].availability.diagnostics,
target_support.diagnostics or capabilities, or tool_availability group
surface, effects and annotations. Pass {"detail": "complete"} to get
them.
Fixes#7109
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(mcp): key binary_session output fields by detail
The output schema accepted detail: "complete" with summary candidates or
tool groups, and a summary without omitted_fields, so detail did not
tell callers which fields existed. Each session kind is now crossed
with a detail-specific shape: a summary requires omitted_fields and
summary-only candidates and tool groups; complete requires diagnostics,
candidate capabilities and static tool flags and forbids omitted_fields.
Both the canonical schema and the SDK-validated advertised schema reject
mismatched shapes.
The verifier helper accepts summary candidates only for a validated MCP
summary; CLI provider output still requires candidate capabilities.
Regex limitation tests check values, sources and setting names instead
of whole sentences.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(cli): report JSON Lines selection refusals as caller errors
A --format jsonl refusal because the selected value is not a record
collection is the caller's output selection, not unreadable analysis
output. Report it as invalid_request/invalid_input that names the
selection and its observed shape, keep the --format json /
--filter-output remediation, and drop the rea doctor advice.
AnalysisInputError gains an optional producer remediation, mirroring
the other refusal classes. The JSON Lines owner examines its documented
boundaries once for both admission and refusal facts.
docs/cli.md now states that columnar results such as
inspect-sqlite-database --table rows are not record collections.
Fixes#7110
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(cli): carry JSON Lines refusal facts as structured observations
JSON Lines refusals now put the location and shape of each refused value
in details.issues[].observed, and the message is built from those facts.
Tests assert the classification, issue path, observed facts, and the
shared remediation constant instead of fixed sentences.
The filtered analysis-view export check now uses a non-empty request
and compares the emitted rows with the unfiltered export.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Logical branch selection evaluated the left operand to decide which
branches stay reachable, and value evaluation then evaluated the selected
left branch again. In a left-nested `a || b || c ...` chain every level
repeated that, so semantic analysis doubled per operand and a 24-operand
chain exhausted the per-source deadline.
The selection now carries the left operand's value with the branch, and
value evaluation reuses it. Provenance still walks the branch nodes.
Results are unchanged.
Fixes#7115
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(mcp): fit every advertised input schema within the 64 KiB client limit
Claude Desktop drops a local connector tool whose compact-JSON input
schema exceeds 64 KiB. The full input profile advertised
compare_web_captures at 171,867 bytes and inspect_analysis_view at
71,632 bytes, because the self-contained projection inlines the
complete producer capture schemas once per side.
The full advertisement now fits the limit at its producing boundary.
A property whose schema repeats an earlier sibling's points at it, and
schemas below the deepest instance level that fits are summarized as
their types, size constraints and descriptions. Every rewrite widens the
advertisement through the existing relaxation rules for oneOf, not,
if/then/else, contains and unevaluated keywords, so canonical inputs
stay accepted and the canonical contract still validates every call.
Schemas within the limit are unchanged, and the compact profile is
still presented from the complete projection.
Numeric const and enum literals are also relaxed in the advertisement,
like boolean literals, since Gemini accepts only string enum members.
Fixes#4081
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(mcp): keep fitted input summaries widening and at the deepest fitting level
Array summaries took item types only from `items`, so a tuple such as
`{prefixItems: [{type: string}], items: {type: number}}` was advertised
as a number array that rejected `["a", 1]`. Item types now include tuple
positions from `prefixItems`, a draft-07 `items` array, and
`additionalItems`, and stay unconstrained when later elements are open.
A summary can also be longer than the schema it replaces (a nullable
array becomes an anyOf), so candidate size is not monotone in the
retained depth and stopping at the first overflow could skip a deeper
candidate that fits. A schema is now kept whole when its summary is not
smaller, and fitting searches from the deepest level down.
Property tests check that every summary accepts what its schema accepts
over every catalog input subschema and synthetic tuples, and that the
fitted schema is the deepest fitting candidate. The browser comparison
boundary test now expects a malformed `after` capture to be rejected
canonically, since `after` is advertised as the same schema as `before`;
the same malformation on `before` stays advertised-invalid.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(javascript): bind defaulted and destructured Webpack factory roles
A Webpack factory whose module parameter was defaulted (`e = {}`) or
destructured (`{exports: x}`) reported an exact empty export list with
complete coverage, because roles were taken only from plain identifiers.
Bind factory roles once, by position, in javascriptBundlerFactoryRoles.
Defaulted module/exports parameters keep their role (the convention always
supplies the argument), `{exports: alias}` binds the alias to
module.exports, and a repeated sloppy parameter name binds its last
argument. Other role patterns (rest, array, nested or computed keys, a
non-identifier loader) receive no role and count as an unknown finding,
so the file's analysis is partial instead of claiming no exports. The
export collector, the factory require name and the standalone bootstrap
externals all use these parsed roles.
Fixes#7113
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(javascript): keep factory receivers through unordered reads
A defaulted role parameter kept the unused default in the binding facts
passed to write ordering, so a receiver read inside a loop, nested block
or closure was unresolved and its export silently dropped while the
module stayed complete.
The role owner now removes defaults the calling convention never runs
before ordering. Separately, an export-shaped write through a role
binding whose read stays unordered now counts as an unknown finding
instead of being dropped, so an incomplete list is never reported as
complete.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(javascript): treat never-run factory parameter defaults as dead code
Removing unused defaults from binding facts let writes inside a default
expression reach the export collector: `(e = {}, t = (() => {
e.exports.ghost = 1; })())` reported `ghost` although Webpack supplies
both arguments. Loader calls in such defaults also produced require
references and async chunk keys.
The role owner now records the default of every role parameter, any
binding pattern included, as never run. The export collector, the
async-chunk walk and the static source traversal skip those subtrees,
so they yield no exports, references or chunk keys. Defaults of later,
non-role parameters still run and are still reported.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
When cancellation or the deadline interrupted LLDB after it had published
its bridge result, the tracer never read that result, so an explicit
`target_process_created: false` (or `target_process_exit_observed: true`)
was lost. The run kept a `native-target:unknown` marker that no cleanup
retry could release, wedging the tracer for every later trace.
Read the published result once LLDB's process group is stopped, even for
an interrupted run, and apply the same target-gone proof in cleanup retry.
Cancellation or timeout stays the primary outcome; a created target, a
truncated result, or a missing PID alone stays unresolved.
Fixes#5385
Co-authored-by: Claude Opus 5.5 <[email protected]>
* perf(native): charge matching for what it copies, not borrowed retained inventories
Native function matching deep-walked every retained list_procedures record
at about 2.8 KB per Ghidra-shaped row, although the session ledger lends
those immutable records without a copy, and charged each inventory's one
Evidence ID once per function. The report and Evidence copies of every
emitted member were not charged at all.
Charge each inventory row before parsing for the member it becomes,
including its report and Evidence copies, charge the shared inventory ID
once, and skip the walk only for borrowed immutable inventories and closure
records that matching never parses. Inline Evidence, bundle records and
selected observations keep their existing charges. Retained references now
fit about 22,000 rows per side instead of 10,000, and an oversized side is
refused before its rows are parsed.
Refs #6406
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(native): charge the envelope of borrowed matching inputs
Borrowed inventories skipped admission entirely, but matching copies their
limitations into the report and Evidence and digests their profile, so a
retained inventory with few rows and large limitations escaped the
allowance. Charge every envelope field and leave only the result payloads,
which row charging covers or matching never parses, uncharged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
Reduced compact input schemas gave every array property an any-value item
schema, a six-member type list beside nested `items`. Gemini function
declarations accept `items` only beside the single `array` type, so every
request through a Gemini-backed client failed with
`properties[functions].items.items: field predicate failed: $type == Type.ARRAY`.
Reduced array properties now advertise the element types their canonical item
schemas accept. Unknown, tuple, and mixed array elements use the empty item
schema, and properties that may be arrays or another type list typed anyOf
branches. An inline `not: {}` keeps the empty schema. The compact catalog test
now rejects `items` beside a non-array type as well as untyped item unions.
Fixes#5201
Co-authored-by: Claude Opus 5.5 <[email protected]>
* fix(artifacts): admit raw UDRW disk images
Raw (UDRW, UDTO) disk images have no UDIF koly trailer, so target
admission and root inventory both refused them as unsupported binaries.
Recognize disk images in one domain owner (koly trailer, GPT, Apple
partition map, MBR, or an unpartitioned HFS+/HFSX volume header) that
both boundaries consume.
hdiutil reports "has no checksum" for these images; the native reader
now treats that status line as an explicit absent checksum, records the
nonzero verify command, attaches read-only, and reports that integrity
was not verified. Checksum mismatches and unrecognized images are still
refused.
Positional range reads used by these probes now fill until end of file
so short host reads cannot hide a signature.
Fixes#5687
Co-Authored-By: Claude Opus 5.5 <[email protected]>
* fix(artifacts): read disk image signature windows at the recognizer
Target admission passed its first unfilled 4 KiB read to the raw-image
checks, so a short host read could hide a GPT, APM, MBR or HFS+
signature that inventory (which fills its prefix) still observed. The
recognizer now reads its own leading window through the caller's
filling positional read, and target admission fills its format-probe
prefix as well. The controlled short-read regression now covers target
admission alongside inventory.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
---------
Co-authored-by: Claude Opus 5.5 <[email protected]>
inspect_macho reported only the symbol-table spelling (`_main`), while
LLDB, dlsym and observe_native_calls use the runtime spelling (`main`).
Passing a static name as a breakpoint silently selected a different
function, so results from the two tools could not be joined by name.
Keep `name` as the exact symbol-table fact and add a derived
`runtime_name` to every export and import row, produced by one shared
Mach-O symbol-name owner. Managed P/Invoke verification now uses the
same owner instead of its own underscore stripping.
Fixes#5298
Co-authored-by: Claude Opus 5.5 <[email protected]>
Retain complete browser scenario Evidence while delivering compact summaries and selected step, event, transaction and text views through shared CLI/MCP workflows. Preserve initial CLI failures across cleanup retries and validate with real Chromium/TodoMVC.
Co-authored-by: Kaoru0822 <[email protected]>
Co-authored-by: Claude Opus 5.5 <[email protected]>
description:Reverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.
metadata:
version:"49"
version:"59"
---
# REA
@@ -17,6 +17,9 @@ readiness or provider commands.
Installing this skill supplies instructions; it does not register the REA MCP
server, install analysis engines, or add tools to an already-running agent.
The exception is a Claude Code or Codex plugin install: the plugin bundles
this skill and registers its own `rea` server, outside the state that
`rea setup` and `rea doctor` manage.
If REA tools are available and their registration is not known to be stale,
proceed directly to the target. Do not run diagnostics or setup before every
@@ -37,42 +40,23 @@ repeat or known not to have started. If it may already have changed the
target, or it fails again, tell the user and ask them to restart the MCP
client. If REA could not start its connection, retry once, then tell the user
and ask them to restart the MCP client. Do not retry a startup failure that
returns the same unchanged result. Run `rea doctor` when REA could not start
returns the same unchanged result. For plugin installs, follow
| No REA tools in the active session | Inspect registration with `rea doctor --client codex --json`. A skill installation alone does not register MCP. |
| No REA tools in the active session | If REA was installed as a client plugin, follow the plugin section below. Otherwise inspect registration with `rea doctor --client codex --json`. A skill installation alone does not register MCP. |
| Missing, malformed or stale registration | Prepare a scoped setup plan; apply only authorized changes with backups. |
| Aligned registration, but no connected tools | Restart/reconnect the client. Doctor checks registration; it cannot establish the current agent's connection. Inspect the client's launch error if reconnection fails. |
| MCP connection dropped or REA could not start | If the connection drops, retry that call once only when it is safe to repeat or known not to have started. If it may already have changed the target, or the same failure returns, tell the user REA's connection failed and ask them to restart the MCP client. If REA could not start its connection, retry once, then tell the user and ask them to restart the MCP client. Do not retry a startup failure that returns the same unchanged result. Run `rea doctor` when REA could not start or tools stay missing after restart. Doctor cannot restore a live connection by itself. |
| `MCP tool call requires approval, but approval policy is never` | Codex rejected the call before REA executed it. Do not install an engine or repeat setup to solve this client-policy decision. See the approval section below. |
| REA returns `provider_unavailable` or an unavailable tool entry | Read that operation's provider diagnostics or `binary_session.result.tool_availability`. Expand each group's tool names and inherit the shared client requirements unless overridden. Use the operation's availability, reason, and remediation; a complete tool catalog does not imply every operation is available. |
| REA returns `provider_unavailable` or an unavailable tool entry | Read that operation's provider diagnostics or `binary_session.result.tool_availability`. Expand each group's tool names and inherit the shared client requirements unless overridden. Use the operation's availability, reason, and remediation; a complete tool catalog does not imply every operation is available. Request `{"detail": "complete"}` only for provider diagnostics. |
| `rea` is missing in a child shell, or Node is incompatible | Inspect the actual shell PATH and launcher/runtime. A prefix installation may be absent from a login shell's PATH; this does not establish a broken MCP registration. |
If the client cannot launch its shell/sandbox helper or read the installed skill,
@@ -116,6 +116,32 @@ is not a recommendation to change it. Do not invent a setting. Host memory and
-`invalid_request`: correct the reported fields from the current schema. Do
not repeat the unchanged call.
## REA installed as a client plugin
A Claude Code or Codex plugin owns its `rea` server and this skill. Doctor's
`--client` and `--skill` checks and `rea setup` read and write only
setup-owned state, so for a plugin-only install they report `missing`, and
applying setup would add a second `rea` server. If this skill was loaded from
a plugin, or the client lists a `rea` plugin, do not prepare a setup plan for
missing tools. Instead:
- Check that the plugin is installed and enabled with the client's own
listing: `claude plugin list` and `/mcp` in Claude Code, `codex plugin list`
in Codex.
- Restart or reconnect the client, then read its launch error for the `rea`
server if tools are still absent. The plugin starts its bundled runtime; a missing build or version mismatch
requires a reinstall, or a fresh build when developing from source.
- Check analysis engines through the connected server's
`binary_session.result.tool_availability`. `rea doctor --provider <id> --json`
from a shell reports that shell's environment, which can differ from the
environment the client gave the plugin server. If you still run it through
the package runner, use the version pinned in the plugin manifest's server
entry (`rea-agents@<version>`), not `@latest`, so it matches the server the
plugin starts.
Use the setup repair below only when the user wants a setup-owned registration
instead of the plugin, and remove or disable the plugin first.
## Repair registration when needed
For Codex:
@@ -130,7 +156,10 @@ Use the current supported client's ID: `claude_code`, `claude_desktop`, `codex`,
Unpaired observations do not prove additions/removals. Existing matched dossier
pairs include inputs for `compare_functions`; otherwise analyze just chosen pairs.
Aggregate completed comparisons with `find_changed_behavior`, selecting each
comparison's exact `retained-evidence` ID instead of resending full records.
Inline comparison Evidence remains accepted. References belong to the current
server: after handoff, import the original bundle including its source closure.
Do not substitute dossier IDs or summary views. Static differences remain
behavior candidates, with runtime behavior unknown unless separately observed.
For a matching representation refusal, inspect `details.limit` and raise
`native.matching_representation_bytes` with `update_analysis_settings` before
retrying the unchanged MCP request. CLI launch seed:
`REA_NATIVE_MATCHING_REPRESENTATION_BYTES`. The default is 96 MiB; the policy
range is 1 byte to 1 GiB and does not bound process RSS. Keep the complete
inventories: splitting them changes global name/body uniqueness.
## Apple app resources
Use `open_bundle_resources` with the selected app path, then `inspect_plist`,
@@ -316,26 +333,44 @@ run, use `observe_native_calls` with explicit breakpoints and a bounded
`duration_ms`/`max_events`. It launches a new process under LLDB, so the target
runs and may change files or use the network. Hardened-runtime targets need the
`get-task-allow` entitlement.
Function breakpoints use exact LLDB runtime names: a Darwin C `_main` static symbol
may be `main` at runtime. REA does not translate names. Set `module` to the image
Function breakpoints use exact LLDB runtime names: pass an `inspect_macho` export
or import `runtime_name` (`main`), not its symbol-table `name` (`_main`). REA does
not translate breakpoint names. Set `module` to the image
file name when selecting that image only; otherwise loaded system images also
match. Inspect retained locations and module paths before attributing a call.
Use `binary_session` with no arguments to check the open target, selected
provider, and alignment. Its default `result.tool_availability` carries a
complete inventory grouped by shared facts. Expand each group's`names` and
inherit the inventory-level client requirements unless a group overrides them;
availability, reasons, remediation, modes, and flags remain inline. The
`effects` and `annotations` arrays name true flags from fixed vocabularies, so an
omitted known flag is false. `tools/list` remains the canonical source for full
tool schemas when target or provider state changes.
provider, and alignment. Its default summary `result.tool_availability` carries
the complete tool inventory grouped by current status. Expand each group's
`names` and inherit the inventory-level client requirements unless a group
overrides them; availability, reasons, remediation, and modes remain inline.
Provider candidates keep availability and target-support status, code, and
reason. `omitted_fields` names what the summary leaves out; `{"detail":
"complete"}` adds each unavailable candidate's `remediation`, provider
diagnostics, candidate capability sets, and tool surface, `effects`, and
`annotations`. Those arrays name true flags from fixed
vocabularies, so an omitted known flag is false. `tools/list` remains the
canonical source for full tool schemas when target or provider state changes.
Provider capability groups carry provider identity and ordered limitations.
Prepend each set's `shared_limitations` to a group's `limitations` to recover the
operation's full ordered list. Candidate capabilities describe possible
providers. Top-level session capabilities describe the current route, and
operation's full ordered list. Candidate capabilities, returned with complete
detail, describe possible providers. Top-level session capabilities describe the current route, and
`analysis_provider_binding` identifies the selected provider.
## Reuse caller-selected Ghidra analysis
When the caller supplies an expanded project, configure `REA_GHIDRA_PROJECT_PATH`
(the `.gpr` marker) and optionally `REA_GHIDRA_PROGRAM_NAME` (one root program),
then restart/reconnect before `open_binary`. CLI uses the same configuration.
REA opens only a fingerprint-bound owned copy with `-process -noanalysis -readOnly`;
original project files are not opened by the engine or saved. Database edits,
import overrides and repair scripts are unavailable in this mode. Treat original
analysis completeness/settings as unknown and prior edits as caller-project state.
The program's recorded import digest must match the target; changed project content
requires reopening for a fresh profile. See the
[setup and preparation guide](https://github.com/morluto/rea/blob/main/docs/installation.md#reuse-an-existing-ghidra-project).
## Ghidra startup deadline recovery
`provider_timeout` during the first Ghidra query can cover import, auto-analysis,
@@ -399,6 +434,21 @@ encoded in bounded batches.
The 25000 ms Cutter request deadline and the 256-candidate discovery budget
still have no runtime override.
## Selected native facts and call graphs
For a large `get_call_graph`, or repeated `procedure_address`,
`procedure_callees`, `procedure_callers` and `resolve_containing_procedure`
queries, select `detail: "summary"` when the first 64 recorded observations
suffice. Read further `native-observation` pages from the returned parent Evidence
ID. Graph message IDs index that page's exact-text dictionary; keep level,
source index, failed nodes, and partial/unknown relationship coverage.
Read referenced source context through `native-context` facets `overview`,
`analysis-profile`, `limitations` and `result-limitations`. Reuse inspected
context/profile values only for the same digest; until caveats are read,
interpretation beyond the returned facts remains unknown. Complete Evidence
remains retained and exportable. Small complete replies can cost fewer tokens
than additional context/page calls; count the completed task and discovery.
## Static memory and original file bytes
For Ghidra, `read_bytes` reads initialized analysis memory, including loader
@@ -426,6 +476,14 @@ loaded bytes, file mappings, relocations and the entry against the immutable
snapshot. Keep `mismatch` and `unsupported` explicit. Verified import covers the
reported static image, not runtime DOS or PC-98 hardware behavior.
For a large load-image report, select `detail: "summary"` and query its
`parent_evidence_id` with `inspect_analysis_view` kind `native-load-image`.
Choose a collection such as `relocations` or `mappings`, select only needed
fields, and page with offset/limit. Original indices preserve same-address
alternatives. Null collections remain unrecorded or inapplicable; paging does
not upgrade unsupported verification. CLI requires `--evidence_output` for
summary delivery so complete source Evidence remains accessible.
For Ghidra decompilation, `procedure_pseudo_code` also accepts `procedure: "file-offset:0x..."`; CLI `rea decompile <path> --file-offset=0x... --provider ghidra` resolves the source byte offset internally. An unresolved 0x-prefixed numeric procedure selector also falls back to a unique file mapping. Existing symbols and analysis addresses win; multiple mappings fail as ambiguous. Preserve the returned file-offset and address locations and raw selection observations. ELF virtual addresses are distinct from file offsets; never assume an arbitrary offset plus image base is an analysis address.
Complete function body ranges use inclusive ends and can contain gaps. Do not
@@ -469,6 +527,11 @@ names), coverage and per-collection counts. Page rows with
Native DMG traversal automatically uses a read-only mount on supported macOS
hosts, with an owned temporary mount directory and cleanup. It needs no approval
flag. Unsupported hosts or mount failures remain explicit limitations; do not
@@ -554,9 +627,27 @@ semantics are unsupported. `m68k` requires an even entry and decodes the 68040
superset of the 68000 family; Amiga/Atari containers and vector-table entry
discovery are unsupported. Windows Ghidra P0 admits raw images only for `6502`. See [raw binary guide](https://github.com/morluto/rea/blob/main/docs/raw-binary.md).
Intel HEX, S-record, UF2 and DfuSe files are containers, not raw images:
`inspect_artifact` lists address regions (UF2 per familyID, DfuSe per element
with its alternate setting), and `extract_artifact` writes each region. Open an
extracted region as `raw-binary` with its start address as `entry_point`. UF2
`integrity_policy`. See [UF2 and DfuSe](https://github.com/morluto/rea/blob/main/docs/uf2-dfuse.md).
### Function annotations in Ghidra
Use `annotate_native_function` for one function name and/or entry comments, with at least one explicit change. Review its annotation readback and refreshed dossier inline. Changes are atomic and session-scoped; empty comments clear them, omitted fields preserve them. Later MCP calls observe edits until close. CLI `annotate-native-function` returns the updated analysis before discarding the session. Original executable bytes are unchanged; edits invalidate immutable snapshots. Windows P0 does not admit database mutations. For naming only, `set_address_name` / `set_addresses_names` also work with Ghidra: they name function entries or data addresses (primary labels) and return compact per-address success booleans instead of a dossier.
Fresh-import sessions admit annotations; existing-project reuse is read-only.
Use `annotate_native_function` for one function name and/or entry comments, with at least one explicit change. Review its annotation readback and refreshed dossier inline. Changes are atomic and session-scoped; empty comments clear them, omitted fields preserve them. Later MCP calls observe edits until close. CLI `annotate-native-function` returns the updated analysis before discarding the session. Original executable bytes are unchanged; edits invalidate immutable snapshots. Windows P0 does not admit database mutations. For naming only, `set_address_name` / `set_addresses_names` also work with Ghidra: they name function entries or data addresses (primary labels) and return compact per-address success booleans instead of a dossier. Ghidra rejects batch addresses that request the same name or a name another selected address holds; `raw_result.failures` gives each failure's reason.
For `compare_functions`, obtain `analyze_function` Evidence before and after the
edit while the MCP session is open; annotation Evidence is not a comparison
input. Export the needed Evidence before close if comparison will happen later.
Authenticated same-artifact coordinates preserve renamed entries as
`matched / same-entry`, with the canonical `entry_address` inline; name and
exact-text changes remain separate dimensions. This requires compatible
architecture, provider-owned mapping and selected-image commitments. Equal
@@ -28,6 +28,10 @@ Assess brittleness in touched code during implementation, debugging, review, and
When fixing a bug, inspect the producing boundary and affected consumers for duplicated interpretation or unclear ownership. Complete any directly related refactor in the same change and remove superseded code. Passing regression tests alone does not establish that the brittleness is resolved. If related cleanup remains, explain what remains and why.
After confirming a bug, search sibling adapters, consumers and similar workflows
for the same failure. Verify suspected matches, record working comparisons, and
group findings by root cause instead of filing duplicate reports.
When addressing a review thread, inspect the producing boundary and affected consumers for brittleness. Complete directly related refactoring in the same change. Before resolving the thread, explain the root cause, structural fix, validation, and any remaining cleanup.
Parse unknown boundary values and model expected failures with the tagged error algebra and `Result`. Preserve partial facts and meaningful failure reasons. Keep identity and lookup values separate from display formatting; preserve source values and explicit unknowns when normalization loses information. Absolute paths, file URLs, and HTTP paths have different semantics.
@@ -53,9 +57,12 @@ Optimize in this order:
3. CPU and memory. Avoid unnecessary work and copies; bound concurrency,
retained state and serialization growth.
For changes affecting these costs, measure representative completed tasks and
compare equivalent workflows before and after where available. Report token
usage, elapsed time, CPU time and peak process-family memory. Separate cold
Reserve performance measurements for changes that target performance or have a
to helper calls, imports, or representations alone do not require synthetic
benchmarks. When measurement is warranted, measure representative completed
tasks and compare equivalent workflows before and after where available. Report
token usage, elapsed time, CPU time and peak process-family memory. Separate cold
and warm runs, label estimates, and state unmeasured costs. Fewer bytes alone
do not establish fewer tokens or faster analysis.
@@ -72,6 +79,9 @@ Use [.nvmrc](.nvmrc) and `package.json#packageManager` for the pinned toolchain.
Use relevant checks for the change. The complete local gate, `npm run check:pr`, is optional for broad changes or CI diagnosis. See [CONTRIBUTING.md](CONTRIBUTING.md) for development, generated-file ownership, and release conventions.
When changing installers or setup recovery, follow
[Checking installers and publisher downloads](docs/testing.md#checking-installers-and-publisher-downloads).
Coordinate expensive checks across agents. If a test unexpectedly consumes heavy CPU, stop its owned process family and profile a bounded representative case before retrying. Inspect production work, fixture generation, subprocesses, and repeated setup; preserve regression coverage when optimizing. Follow the [test resource guidance](docs/testing.md#developer-commands).
`docs/public/product-catalog.json`, `docs/verification/managed-conformance-*.json`, and `skills/` are ignored outputs. Edit source contracts and `.agents/skills/reverse-engineer-anything/`, then regenerate as needed. `.cache/mcp-tool-catalog.json` is test metadata; runtime builds and source checking do not consume it. Never commit binaries, provider project documents, credentials, `dist/`, `node_modules/`, or local planning artifacts.
@@ -80,6 +90,12 @@ Coordinate expensive checks across agents. If a test unexpectedly consumes heavy
For tool design, follow [docs/tool-design.md](docs/tool-design.md): start from the analyst question, reuse existing contracts, preserve caller choices, and return useful evidence inline. Limits must follow real format, protocol, authority, or resource constraints. Account for representation expansion before allocation; derive completeness from examined coverage.
For user- or agent-facing tool problems, prioritize fixing the owning behavior,
contract, and actionable result or recovery at the point of use. Document the
result in product docs when explanation is needed. Add skill guidance only for
workflow judgment or context that the tool cannot reasonably convey; do not use
a skill as the primary repair for a discoverability or contract defect.
For test changes or provider claims, use [docs/testing.md](docs/testing.md). Select applicable [boundary regressions](docs/testing.md#boundary-regressions) for parser, inventory, snapshot, and projection fixes. Prefer real public CLI/MCP workflows, then production-boundary integration, then producer goldens. Keep module tests for distinct semantics stronger workflows cannot reliably reproduce. Preserve malformed-input, cancellation, cleanup, permission, capacity, and target-identity coverage when pruning. Report unverified coverage explicitly; simulated providers do not establish real-provider behavior.
Use Conventional Commit subjects and PR titles. Mark breaking changes with `!` or a `BREAKING CHANGE:` footer. Describe behavior changes, relevant validation, and real-provider verification scope.
@@ -112,8 +128,8 @@ owner to complete. Never attest to human review on their behalf. Draft status
is optional; preserve the PR's current draft status unless the user asks to change
it. Later pushes do not invalidate the readiness declaration or require another
human sign-off. Keep the PR body and affected validation current as work changes.
The readiness check validates
the filled declaration; CI and automated review results do not establish that
The readiness check validates the filled declaration or the author exemption
listed in CONTRIBUTING.md; CI and automated review results do not establish that
human review occurred, maintainer scope approval, or merge permission.
<a href="https://discord.com/invite/GkcryMnJDM"><img alt="Join the REA community on Discord" src="https://img.shields.io/badge/Join%20us%20on%20Discord-5865F2?style=for-the-badge&logo=discord&logoColor=white" height="40"></a>
[Quick start](#quick-start) · [How REA works](#how-rea-works) · [What you can analyze](#what-you-can-analyze) · [Showcases](#showcases) · [FAQ](#faq) · [Documentation](#documentation)
[Quick start](#quick-start) · [How REA works](#how-rea-works) · [What you can analyze](#what-you-can-analyze) · [Showcases](#showcases) · [Build RL environments](#build-rl-environments) · [FAQ](#faq) · [Documentation](#documentation)
<code>npx rea-agents setup</code>
@@ -62,7 +64,7 @@ See a feature in an app that you want in your own product? Ask your agent to inv
REA connects your agent to tools for inspecting native binaries, JavaScript and Electron apps, .NET assemblies, and websites. You can also use the same tools from your terminal. Analysis runs locally, and results include the evidence and limitations behind each conclusion.
Setup registers REA with your agent and installs matching workflow instructions. Native analysis can use an existing Hopper, Ghidra, or IDA installation; setup can optionally install Hopper with approval. Static JavaScript analysis needs no native analysis engine.
Setup registers REA with your agent and installs matching workflow instructions. Native analysis can use an existing Hopper, Ghidra, IDA or Kuna installation; setup can optionally install Hopper with approval. Static JavaScript analysis needs no native analysis engine.
> **[Visit the REA website](https://rea.tools/)** for setup instructions, illustrated guides, and real case studies.
@@ -83,12 +85,15 @@ npx rea-agents setup
Choose your agents, review the proposed changes, and approve them. Setup adds
REA's MCP server and matching workflow instructions, with backups of existing
configuration. Restart your agent afterward.
configuration. Restart your agent afterward. Setup records the selected runtime and REA
entry script directly. If acquired through a package runner, keep its cache entry
available. Run package acquisition from a trusted directory outside the target
project. Installation and recovery details are in the setup guide below.
Setup supports Claude Code, Codex, Cursor, Gemini CLI, Grok Build and
[other agents](docs/installation.md#supported-agents). See
[installation and setup](docs/installation.md) for provider configuration and
manual MCP registration.
manual MCP registration and Claude Code/Codex plugin installation.
MCP uses compact input schemas by default to reduce context overhead. Set
`REA_MCP_INPUT_SCHEMA_PROFILE=full` in your MCP registration when you need the
@@ -96,6 +101,12 @@ complete advertised field guidance and nested validation details. Both profiles
keep the full tool set and canonical server validation. See
"Compact input schema profile" in the MCP contracts guide.
Complete output schemas are advertised by default. If your client caps the
complete `tools/list` frame, set `REA_MCP_OUTPUT_SCHEMA_PROFILE=compact` to
bound each advertised output schema, or `off` to omit output schemas entirely.
Both settings keep canonical server-side validation and complete structured
results. See "Output schemas" in the MCP contracts guide.
### Ask your agent
```text
@@ -183,7 +194,7 @@ Additional tools and host support depend on the target:
| Target | What REA returns | Requirements and guide |
Record a web app session, turn its API into tools, and export an environment an AI agent can use. Run it with AgentEnv, then add tasks and reward checks for RL training or evaluation.
[Read the guide](https://rea.tools/guides/rl-environments/) ·
[Watch the video](https://x.com/reatools/status/2112574822584918075?s=20)
If you've used REA on something interesting, we'd love to see it. Share your
case in an [issue](https://github.com/morluto/rea/issues) or a
[pull request](https://github.com/morluto/rea/pulls), including the target,
your question, how REA helped, and what you found.
## Build RL environments
Give an AI agent a local environment to practice tasks such as finding an issue or creating a folder, without running the original app. Use it for reinforcement learning (RL) or to test how well an agent completes tasks.
REA turns recorded API requests and responses into tools the agent can call. The environment uses the recorded data and supports changes only when the recording confirms their effects.
With REA set up in your agent, replace the example URL and folder tasks below with your own:
```text
Use REA to build an RL environment from my test web app at
http://127.0.0.1:3000.
Cover listing, creating, renaming, and deleting folders. You may create and
remove test folders. Leave existing data unchanged.
Save it in a folder named my-env. Check which operations work and tell me
how to run it.
```
<details>
<summary>Use the command line instead</summary>
Use these commands if you already have a file containing the app’s recorded browser requests and responses. Running the result also requires AgentEnv and Docker.
```sh
rea environment-build ./session.har ./my-env --name my-env --json
agent-env run ./my-env
```
</details>
REA exports a runnable environment with a basic test to check that it starts and responds. Use it with your RL framework to define task goals, add reward checks, and run training.
[Build an environment](https://rea.tools/guides/rl-environments/) ·
@@ -10,7 +10,7 @@ Do not disclose vulnerabilities, proof-of-concept exploits, credentials,
private binaries, Hopper documents, or private Ghidra projects in a public
issue.
Use **Report a vulnerability** in the repository's **Security** tab to submit the report privately. Do not open a public issue to request a private contact channel.
Use **Report a vulnerability** in the repository's **Security** tab as the preferred way to submit the report privately. If GitHub private reporting is unavailable to you, email [[email protected]](mailto:[email protected]). Do not open a public issue to request a private contact channel.
Include the affected REA version, operating system/distribution, selected
provider and version, impact, reproduction conditions, and any suggested
| npm dependencies | [Complete inventory](licenses/npm-dependencies.txt) of every entry in `package-lock.json`, including transitive, optional platform, development, and website dependencies. [Runtime notices](licenses/npm-runtime-notices.txt) preserve the copyright, license, and NOTICE documents from integrity-verified release archives, plus the MIT notice for Rolldown's generated runtime helpers. Development dependencies are inventoried but are not all distributed in REA's npm package. |
| pefile ordinal tables | Import-hash ordinal names are derived from [pefile 2024.8.26](https://github.com/erocarrera/pefile/tree/4b3b1e2e568a88d4f1897d694d684f23d9e270c4), copyright 2004–2024 Ero Carrera. Preserve the [MIT license](licenses/pefile-2024.8.26-LICENSE.txt). The Python package is not a runtime dependency. |
| UF2 family names | UF2 family IDs and short names are copied from [microsoft/uf2 `utils/uf2families.json`](https://github.com/microsoft/uf2/blob/cf3bd7f9c274b949528f8cd6a19eef2502a687d2/utils/uf2families.json), copyright Microsoft Corporation. Preserve the [MIT license](licenses/microsoft-uf2-MIT.txt). |
| NativeAOT parser | REA's derived parser uses the MIT-licensed [ghidra-nativeaot](https://github.com/Washi1337/ghidra-nativeaot/tree/effeb734fc570c32650f88b159608979dc7b423e) work of Washi and the NativeAOT layout definitions in [dotnet/runtime](https://github.com/dotnet/runtime/tree/a3b2d40328be0be3f8dd43f0e5cc925b8827b044). Preserve [Washi's MIT notice](licenses/nativeaot-parser-MIT.txt) and the [.NET Foundation MIT notice](licenses/dotnet-runtime-MIT.txt). |
| EVM analysis | `[email protected]`, MIT, copyright 2023 Maxim Andreev. The published npm archive omits its license document; REA retains the [original upstream license](third_party/evmole/LICENSE) and [provenance](third_party/evmole/README.md). It is also included in the runtime notices. |
| Windows native addon | The release addon is compiled with MinGW-w64 and statically links GCC runtime libraries. Preserve the [MinGW-w64 runtime notices](licenses/MinGW-w64-runtime.txt), [GPLv3 text](licenses/GPL-3.0.txt), and [GCC Runtime Library Exception 3.1](licenses/GCC-Runtime-Library-Exception-3.1.txt). The exception governs eligible compiled combinations; it does not relicense the runtime itself under MIT. The Node-API build headers are from Node 22.19.0; its [original license and component notices](licenses/Node-22.19.0-LICENSE.txt) are retained. Node itself is supplied by the user. Snapshot provenance and release-toolchain review requirements are in [licenses/README.md](licenses/README.md). |
@@ -59,7 +60,7 @@ pin inspection sources; they are not included in the npm package.
| JADX | [Apache-2.0](https://github.com/skylot/jadx/blob/v1.5.6/LICENSE) | The audited jadx-headless-mcp source uses JADX 1.5.6 for Android decompilation; the engine JAR is separately supplied. |
| Kuna | [Apache-2.0](https://github.com/Noelo-Lab/kuna/blob/v1.833/LICENSE), with its [NOTICE](https://github.com/Noelo-Lab/kuna/blob/v1.833/NOTICE) | Bring your own `kuna`; REA verified 1.833 and never downloads, installs or bundles it. REA invokes it as a separate process. |
| Kuna | [Apache-2.0](https://github.com/Noelo-Lab/kuna/blob/v1.989/LICENSE), with its [NOTICE](https://github.com/Noelo-Lab/kuna/blob/v1.989/NOTICE) | Bring your own `kuna`; REA verified 1.989 and never downloads, installs or bundles it. REA invokes it as a separate process. |
case"list_classes"->"Pages of case-sensitive matching top-level class names; offset and next_offset are inventory positions, limit bounds matching rows. Omitted or empty query inventories all names.";
case"get_class_summary"->"Parsed member metadata for one exact class name; no source generation. Set dex_static_values to include encoded static final field values.";
List<String>limitations=newArrayList<>(List.of("Operand components are Ghidra Listing decoder tokens; effective memory base/index/displacement roles and per-instruction context mode are unavailable. The mode field is the program language variant."));
if(result.get("status").getAsString().equals("not-in-listing"))limitations.add("No listed instruction or defined data covers this in-memory address. REA did not attempt a decode; the memory may be uninitialized or contain valid instructions.");
if(flow.get("kind").getAsString().equals("terminal"))limitations.add("No RETURN operation was observed in the provider's instruction p-code; terminal classification remains generic.");
@@ -103,6 +103,10 @@ existing analysis families directly, for example
| `start_adb_activity` | **mutates device** | Starts an intent (`am start -a` with optional URI, component, and up to 16 typed extras); preserves the activity manager's verdict |
| `stop_adb_app` | **mutates device** | Force-stops every process of a package, ending a dynamic-analysis run |
Screen capture validates the PNG signature, IHDR framing, dimensions, and IHDR
checksum before publishing. It does not validate encoding fields or decode and
validate image contents; that limitation is retained in Evidence.
When `push_adb_file` receives an existing directory, adb places the local file
inside it under the local basename; Evidence reports that effective file path
while retaining the caller's directory path in the input parameters. The
@@ -253,12 +257,12 @@ beyond already-connected devices (`adb connect`) are not exposed yet.
## Real verification
`npm run verify:adb` exercises every operation against a live device. It
`npm run verify:adb` exercises observation and transfer operations against a live device. It
requires `adb` (or `REA_ADB_PATH`) and any attached emulator or phone;
`--serial` selects one when several are attached, `--pull PACKAGE` pulls a
real package and verifies the returned digests against the files on disk,
and`--install-apk PATH` drives a full install → resolve → uninstall
lifecycle with a lane-owned probe.
real package and verifies the returned digests against the files on disk.
The optional`--install-apk` lifecycle is unavailable until the selected APK
identity and device cleanup ownership can be established safely.
Record: verified against adb 34.0.5 on Linux against an Android 14
(API 34) x86_64 emulator — all operations above, a 224-package inventory,
| `no-state-target` | The write's path names no collection or item the capture reads. Read the collection or item it changes. |
| `no-state-target` | The write could not be linked to a modelled target. Check the path template and recorded views; record missing reads. Some route forms may not be supported. |
| `no-call` | No recorded call of this tool was modelled, for example because every call failed. Record a successful call. |
| `body-unusable` | The retained request or response cannot be modelled. Record a successful call with a complete supported request and reproducible response. |
| `no-observation` | Nothing showed the entity or document this write changed. Read a created entity back by id right after the write; for an update or delete the state must already hold the entity, so read it before the write or record the create that made it. |
@@ -175,6 +183,7 @@ what to record next; hints never change what is served:
| `side-effect` | The write changed something besides its own entity, which the environment does not model. |
| `unconfirmed` | No later read showed the effect. Read the entity and its list after the write. |
| `incomplete-entities` | The write is confirmed, but the views that show its entity cannot render it in full. Read the written entity through every view that lists it. |
| `unknown-clock` | The learned rules generate timestamps, but the capture establishes no global served clock start. Record usable start times and completed durations for every request. |
The bundle ships `evidence.json`, the whole capture Evidence, response bodies
included. Inspection excludes the values of the standard authentication headers
@@ -196,7 +205,10 @@ attached and a transaction no record represents, and each transaction URL's
query once. Bodies larger than 4 MiB of UTF-8 text are not
scanned; `credential-scan-incomplete` counts them and lists up to 20 record or
event pointers, including base64 bodies whose decoded scan was cut. An empty name
warning does not establish that these bodies were checked. Names do not establish sensitivity,
warning does not establish that these bodies were checked. A browser scenario's
saved values (`save` and `save_response`) named like credentials but not marked
`secret: true` are recorded unredacted; `credential-named-saved-values` lists up
to 20 of them by parameter pointer. Names do not establish sensitivity,
so the build never refuses on them. Capture
Evidence also records the capture file's absolute path (`subject.local_path`,
@@ -394,7 +406,8 @@ returns that record's recorded response. Selection compares the supplied path
parameters against the record's path segments at the parameter indexes, then
the supplied query values against the record's query string, and keeps the
record with the most matches; ties keep manifest order. A query argument accepts
one string or an array of strings for repeated values; the array is matched
one string, a number or boolean sent as its JSON text, or an array of strings for
repeated values; the array is matched
against every occurrence in captured order. The raw recorded query remains the
replay evidence. The result reports
where the answer came from:
@@ -601,8 +614,9 @@ shape and carries no annotations.
Path parameters are required strings whose
description names the segment index, the templating rule and the observed
values (sampled within the list budget above; the manifest keeps them all);
query parameter names accept an optional string or array of strings (for
repeated values in captured order); write methods accept an
query parameter names accept an optional string, number or boolean (a number or
boolean is sent as its JSON text) or array of strings (for repeated values in
captured order); write methods accept an
optional `body` object described by the recorded top-level key types. Each
input keeps its own argument: a query name equal to a path parameter's name is
advertised as `query_<name>`, a query name longer than 64 characters as
@@ -709,6 +723,8 @@ and `POST` tools the capture shows only reading (a lookup answering as a `GET`
did, an RPC tunnel naming read operations, a session renewal), stay served and
answer from their recordings with the reason `query-body`; they are not write
tools.
`HEAD`, `OPTIONS` and `TRACE` tools also replay their captured responses; their
state views report `method-replay` and coverage counts them as recorded reads.
A line establishes whole-collection membership only from the beginning:
`offset`, `start`, `_start` and `page` must be zero when present. Positive or
@@ -723,14 +739,14 @@ tools are not served, so no call claims a change it does not make; each one's
plan states why in its `reason`. A served write changes the live state, which
| `create` | Adds an entity to the collection the path names, built from the observed entity with each ruled field taken from the call or generated, at the plan's `insert` position. A plan tied to its exemplar's collection writes only there. |
| `update` | Sets the sent members on the entity the path names, constants for members accepted only with their recorded value, and the clock, nested clock, increment, slug, stamp, derived, composed, reference, unique and renewed changes the plan models. |
| `update` (action) | A `POST`, `PUT` or `PATCH` to an item path plus an action segment (`/{type}/{id}/archive`) sets, on the entity before the segment, each scalar the recorded calls showed the action set (`set`), with any sent members; it is served only when a recorded call found those fields already set, so a toggle never is. |
| `delete` | Removes the entity the path names. |
| bulk `update` or `delete` | A `DELETE`, `PUT` or `PATCH` at a collection, or a `POST` to it plus `delete`, `remove` or `destroy`, applies its effect to every entity the body names, as recorded: an array of ids or of objects carrying the id field, at the root or as the body's one array member (`{"IDs": […], "Read": true}`). |
| `document` | Merges the body into the plan's document object, or replaces it for `PUT`. |
| `create` | Adds an entity to the collection the path names, built from the observed entity with each ruled field taken from the call or generated, at the plan's `insert` position. A plan tied to its exemplar's collection writes only there. |
| `update` | Sets the sent members on the entity the path names, constants for members accepted only with their recorded value, and the clock, nested clock, increment, slug, stamp, derived, composed, reference, unique and renewed changes the plan models. |
| `update` (action) | A `POST`, `PUT` or `PATCH` to an item path plus an action segment (`/{type}/{id}/archive`) sets, on the entity before the segment, each scalar the recorded calls showed the action set (`set`), or flips each boolean they showed it flipping both ways (`toggle`), with any sent members; it is served only when a recorded call found what it sets already set, or flipped back what it toggles. |
| `delete` | Removes the entity the path names. |
| bulk `update` or `delete` | A `DELETE`, `PUT` or `PATCH` at a collection, or a `POST` to it plus `delete`, `remove` or `destroy`, applies its effect to every entity the body names, as recorded: an array of ids or of objects carrying the id field, at the root or as the body's one array member (`{"IDs": […], "Read": true}`). |
| `document` | Merges the body into the plan's document object, or replaces it for `PUT`. |
An update's unsent changes follow the plan's rules. A `held-clock` is a clock
the client sends back as it read it and the application stamps anew (Ghost's
@@ -741,7 +757,9 @@ the entity already names under its `created…by` counterpart into a
`modified…by` or `updated…by` field. A `derived` field takes a value showing
the one sent member's text that the entity stores nowhere (a field Ghost
derives from `html`), only with that member's recorded value. An action's
`set` rule sets a scalar to the value the recorded calls left it at.
`set` rule sets a scalar to the value the recorded calls left it at, and its
`toggle` rule flips a boolean, refusing a call on an entity that holds none
there (Miniflux's `PUT /v1/entries/{id}/bookmark` stars and unstars).
Its answer has the status and content type of the plan's recorded response,
with the written entity or its identity rendered in, no body, or the recorded
@@ -756,11 +774,19 @@ sending a `derived` rule's member to an entity that does not hold the value the
recorded call found, a nested clock on an entity holding no such object, and a
rendered answer missing a field the recorded response carried, and an update or delete of an entity the state does not hold answers
the recorded 400, 404 or 410 of the tool or of an item view of the type, if
any. A refused call leaves the state unchanged. Values a write generates are
any. A served delete of a type whose item view has a `deleted_flag` (the
member, `deleted`, `isDeleted` or `is_deleted`, its reads set `true` on a
record the application soft-deleted, as EspoCRM does) keeps the entity until
`data/reset` or `data/add`, within the size `data/add` loads, so a read of it by
id answers its last fields with that flag `true`. A refused call leaves the
state unchanged. Values a write generates are
deterministic: the clock starts at `write_clock` and advances one second per
served write, sequences take one past both the largest held value and the
largest they generated, and UUIDs and tokens derive from the state Evidence
id, tool, field and write count, skipping held values. `data/reset` and
id, tool, field and write count, skipping held values. If `write_clock` is null,
plans needing `clock`, `held-clock`, or `nested-clock` retain their learned rules
but are withheld with `unknown-clock`; other confirmed plans can still be served.
`data/reset` and
`data/add` restart the clock, counter and sequences, so the same calls after a
reset give the same state. Answers carry `state: {"source": "state", "view":
"create" | "update" | "delete", …}` or `"document-write"`; a bulk answer adds
@@ -864,8 +890,9 @@ names a `document` view once, at its initial document's `path`. The `version`,
and the `state_evidence_id` and `types` that `data/get` reports, must match
the served state, so a `data/get` result loads back unchanged. A number without
an exact native value, or whose serialized spelling would not be read as exact
(an integer from 2^53 below 1e21, such as `1e20`, in any spelling), or a
`__proto__` member is refused rather than altered.
(an integer from 2^53 below 1e21, such as `1e20`, in any spelling), is refused
rather than altered. Caller data preserves all own JSON members, including
`__proto__`; unknown members of the state document's structure are refused.
An invalid document is `-32602` naming the first problem by JSON pointer, as in
`/rea_state/entities/3/collection` (also in `error.data.path`); a problem
inside caller data (`fields`, `value`, `types`) is located at that member, so
@@ -974,7 +1001,8 @@ serves replay. The complete tool catalog the bundle serves must also fit the
server's advertised catalog budget before any files are written. Privately parsed JSON is sealed once so authentication
and replay share immutable response trees rather than copying the capture repeatedly.
`--name` replaces the manifest's portable kebab-case name as the env folder and env id
(Windows device names such as `con` and `lpt1` are refused);
(at most 63 characters for its DNS service label; Windows device names and
labels interpreted as IP addresses are refused);
`--package-tarball <path>` copies a packed `rea-agents` tarball into the env
folder and makes the Dockerfile install it instead of the registry package and
remove it in the same layer, which is how a checkout rather than a release is
@@ -992,7 +1020,12 @@ Windows inherits ACLs and REA does not establish owner-only access there. Choose
an already private Windows destination when captured content needs it.
The Dockerfile sets `MCP_HOST=0.0.0.0` and `MCP_PORT=18765`, the port the
framework maps for an MCP server env; the framework's own deploy supplies
framework maps for an MCP server env, and `MCP_ALLOWED_HOSTS` naming the loopback
addresses and the environment name, because the container binds outside loopback
and the server refuses to serve on such a bind without an operator-declared
allowlist: the framework's published port is reached over loopback and its
gateway proxies to the env container by the compose service name (the
environment name). The framework's own deploy supplies
`MCP_HOST`, `MCP_PORT` and `ENVIRONMENT_NAME`, so the served card carries the
env's name. It installs Debian's `python3` because the framework's local
sandbox probes a container's readiness with `python3` inside it, which the slim
@@ -1116,7 +1149,12 @@ global conflict involving an equal pair is not used to normalize either pair; th
A step agrees when the statuses and the normalised bodies are equal. The report gives agreement overall, per probe and
per tool, `agreement_ignoring_order` for steps that differ only in item order, and each step's differences, at most 20,
with a JSON pointer and both values cut short:
with a JSON pointer and both values cut short. Each step reports `differences_total` before that cap,
`differences_omitted`, and `differences_truncated` (true when entries were omitted). Agreement and categories use the
complete comparison. Omitted entries are not retained, so `--full-output` cannot recover them; inspect the original
capture and probe Evidence to investigate beyond the returned pointers.
and [string escape owner](https://github.com/iBotPeaches/Apktool/blob/18b5e99cb56ff9451e8aa55b065dcf5bbd616975/brut.j.yaml/src/main/java/brut/yaml/YamlStringEscapeUtils.java#L55-L116).
- **The manifest is the decoded text**, bounded at 512 KiB and reported as
COMPOSITION["src/composition/<br/>fresh typed provider and session factories<br/>one-shot binding and independent optional loading<br/>static WASM/EVM descriptors: named contracts, factories, retryable lifetimes"]
COMPOSITION["src/composition/<br/>fresh typed provider and session factories<br/>one-shot binding and independent optional loading<br/>typed capability bindings: construction, availability and retryable lifetimes<br/>pure capability declarations feed provider catalogs"]
end
subgraph "Adapters"
@@ -21,14 +21,15 @@ graph TD
subgraph "Application Layer"
APP["src/application/<br/>Shared CLI/MCP workflows<br/>enhanced analysis and investigations"]
SESSION["SessionProviderRouter + BinarySession<br/>one immutable deep binding per target<br/>disjoint auxiliary routing"]
REGISTRY["AnalysisProviderRegistry<br/>sorted deep candidates<br/>deterministic selection, no fallback"]
REGISTRY["AnalysisProviderRegistry<br/>sorted deep candidates<br/>deterministic selection, no fallback<br/>ProviderInspectionScope: prerequisite observation lifetime"]
RECORDS["src/application/investigation/<br/>composed InvestigationRecords + narrow ports<br/>EvidenceLedger and Unknown ownership"]
LIFETIMES["CapabilityScope<br/>host-owned service close capabilities<br/>released peers removed; failed cleanup retained for retry"]
SNAPSHOT["BinarySessionRecords + AnalysisSnapshotCache<br/>binary target/profile snapshots and session file ownership<br/>record coordination and post-commit notifications"]
@@ -52,7 +52,11 @@ observations may omit the coverage field; treat omission as `unknown` rather
than inferring completeness from the array.
Passive DOM comparison checks complete node inventories and complete frame
parent/child structure independently. Frame IDs and sibling enumeration order
parent/child structure independently. Node value lengths are comparable only
when both nonempty snapshots declare the same `node_value_length_unit`.
Missing or different units leave length comparison unknown; independent node
or frame structural differences can still prove change. Empty inventories have
no value lengths to compare. Frame IDs and sibling enumeration order
are capture-local and do not establish a change; duplicate IDs, missing parents,
and cyclic ancestry remain unknown. A complete frame difference can still prove
change when DOM nodes are incomplete, and vice versa. Partial DOM or metadata
@@ -400,7 +404,19 @@ contextual evidence with `timestamp: null`, accompanied by a limitation stating
that it does not establish activity in the selected window. Main-frame scope
enforcement still applies to contextual navigation events. Network redirect
events retain their paired prior-hop response context when the timestamped
next-request event falls inside the window. The pre-arm timeline buffer is bounded
next-request event falls inside the window. A redirect row's optional
`redirect_response` contains the prior response `url`, CDP-reported `status`,
and `scope`. The URL has transport credentials removed; URL and status are
withheld unless its origin is approved. Missing or malformed response URLs have
null scope and add `invalid_protocol_value` timeline coverage; a valid URL with an
unsupported protocol has `unsupported` scope. Outside-policy and unsupported
responses mark the timeline as policy-filtered with `disallowed_origin` or
`unsupported_url`. Missing status remains null rather than zero. A supplied
non-numeric status is also null and records `invalid_protocol_value` coverage;
the other valid response facts remain available. Historical rows may omit this
field; omission does not establish that no prior response existed. Response
headers and bodies are discarded. The row timestamp belongs to the next request,
not an exact prior response time. The pre-arm timeline buffer is bounded
by original CDP message bytes. If clock sampling completes after that budget is
exhausted, the operation returns `payload_limit` instead of a partial window.
Cancellation and clock-request failures retain their own outcomes. Session
@@ -415,7 +431,8 @@ allowance; exhaustion fails the observation explicitly instead of guessing scope
Response metadata parses repeated CDP header field lines independently. The
Link projection retains their declarations in order. `csp.directives` lists the
directives of every CSP policy in order, across field lines and comma-separated
policies, without recording policy boundaries, so a directive name can repeat.
policies, without recording policy boundaries, so a directive name can repeat
across policies. As in CSP3, a repeated name within one policy is ignored.
Single-valued policy fields combine their lines as one comma-separated value,
as RFC 9110 specifies. Each link keeps its target in `href` and
its relationship context in `context_url` with a separate `context_scope`;
@@ -448,7 +465,7 @@ not bound the whole process heap.
REA retains local observation data and removes only URL userinfo credentials, explicitly declared scenario secrets, and values from structured credential/storage fields:
- URL observations preserve the complete local URL, including query values, fragments, duplicate parameters, and their order. Only username/password userinfo is removed; malformed original URL strings remain available with `origin: null`.
- DOM snapshots retain node types, node names, value lengths, and attribute names, but not text or attribute values.
- DOM snapshots retain node types, node names, value lengths, and attribute names, but not text or attribute values.`dom.node_value_length_unit` is `utf16-code-units` for web observations and applies to every node's `node_value_length`; it counts JavaScript string code units, not Unicode code points or UTF-8 bytes. Electron observations declare `utf8-bytes` instead. Historical captures without the unit leave it unknown; reading them does not infer a unit or convert their stored lengths.
- Supported DOM URL attributes are projected separately as resolved destinations. Empty `action` on HTML or XHTML forms and empty `formaction` on HTML or XHTML submit controls resolve to the document URL; other relative URL attributes resolve against the captured document base.
- Accessibility structure and reported roles are retained by default, while names and descriptions require `include_accessibility_text: true`. The browser protocol may omit an individual role; that role remains unknown for comparison. Other observed accessibility differences can still establish a change.
- Network observations retain method, status, MIME type, size, type, initiator stack location, and complete URLs with userinfo credentials removed. Redirects reported by CDP are attached as ordered `redirects` on the one request ID, with the prior request URL, CDP response URL, method/type, response status/MIME/encoded length, request timestamp, and the next `requestWillBeSent` event timestamp. CDP delivers `redirectResponse` with that next request event, so the latter is an event boundary rather than an exact response timestamp. Redirect response headers and bodies are discarded; hops outside the selected origins are discarded with the request chain. Headers are discarded after an allowlisted projection of length/encoding, structured CSP/Link/policy fields, and untrusted agent hints. Cookies and authorization headers are never retained.
@@ -483,7 +500,7 @@ REA retains local observation data and removes only URL userinfo credentials, ex
connection/frame inventories; matching direction/opcode sequences alone do
not prove alignment after drops. Redirect resource types use only shared known metadata:
an omitted type leaves equality unknown rather than proving a modification.
- Console observations with a stack source retain call type, argument types, the CDP `Runtime.Timestamp` in milliseconds since Unix epoch, and source location. An absent or invalid timestamp is `null` and makes `console_events` incomplete; an observed zero remains zero. Argument types are listed by CDP argument position. An absent or non-array argument inventory is `null`, and a malformed argument keeps its position as `unknown`; either makes `console_events` incomplete, while an observed empty inventory remains `[]`. A malformed stack or top stack frame is not replaced by its caller: the console event is excluded, and a request initiator location or WebMCP `registration_source` is `null`; each is reported as `invalid_protocol_value`. This epoch-millisecond clock is separate from the monotonic-second timestamps in `observe-web-session`. When console text capture is selected, already-delivered primitive values are retained verbatim with their original `argument_index`; objects, getters, and remote properties are never expanded. Text capture is `unavailable` when the argument inventory is unknown.
- Console observations with a stack source retain call type, argument types, the CDP `Runtime.Timestamp` in milliseconds since Unix epoch, and source location. An absent or invalid timestamp is `null` and makes `console_events` incomplete; an observed zero remains zero. Argument types are listed by CDP argument position. An absent or non-array argument inventory is `null`, and a malformed argument keeps its position as `unknown`; either makes `console_events` incomplete, while an observed empty inventory remains `[]`. A malformed stack or top stack frame is not replaced by its caller: the console event is excluded, and a request initiator location or WebMCP `registration_source` is `null`; each is reported as `invalid_protocol_value`. Source locations are CDP's zero-based lines and UTF-16 columns. A negative column, which Chrome reports on the first line of some inline scripts, is kept as reported and does not locate a column in the source URL; an invalid line or a non-integer column is `null` (WebMCP `registration_source` is `null`) and is reported as `invalid_protocol_value`. This epoch-millisecond clock is separate from the monotonic-second timestamps in `observe-web-session`. When console text capture is selected, already-delivered primitive values are retained verbatim with their original `argument_index`; objects, getters, and remote properties are never expanded. Text capture is `unavailable` when the argument inventory is unknown.
- WebSocket observations retain direction, opcode, and payload byte length. A missing or malformed CDP opcode is returned as `null`, with opcode-dependent payload byte length and shape also `null`, and marks `websocket_frames` incomplete. Missing or wrong-typed payload data is not treated as empty; its byte length is `null`, and a text payload shape is unknown. Numeric opcode `0` and an observed empty payload remain distinct from missing data. Opcode is nullable in the public inspection schema, so clients must handle unknown protocol values. When shape capture is selected, text messages are classified as text or complete value-free JSON shape; binary bytes, hashes, prefixes, and raw payloads are never retained.
- Storage observations always redact values. Key names, IndexedDB names, and cache names require `include_storage_keys`; stable content fingerprints require the additional `include_storage_fingerprints` selection. Cookie fingerprints cover cookies applicable to the current main-frame URL, including its path, rather than every cookie path on that origin. Other storage remains scoped to the origin. Cache fingerprints hash all decoded response bytes, including bodies above 64 KiB, subject to the [64 MiB per-message CDP limit](#request-boundary). Missing or undecodable cache bodies and partial IndexedDB remote objects make fingerprint coverage incomplete, so identical observations remain `unknown`.
- Unavailable or malformed storage inventories mark `storage_keys` unavailable, even without fingerprints. Comparisons suppress inferred key additions and removals while preserving modifications proven by complete fingerprints on both sides. Previously exported captures are not retroactively corrected; recapture them to obtain these coverage markers.
- A `request` step saves with `save: [{name, json_pointer, secret}]`, from its
own response. A `click` or `press` step saves with `save_response: {method,
path, save}`, from a newly observed matching response while the step waits.
The recorder watches before the action and excludes requests already running,
such as an earlier autosave. Method/path matching does not prove that the UI
caused the request: use a path that distinguishes the intended operation from
background polling. Multiple observed matches before binding fail as ambiguous;
requests starting after this observation window cannot be ruled out. It matches the method and a path on the
start origin as the browser spells it, so a path may name `My List` or
`My%20List` (`{*}` matches one segment, and the query is ignored), and fails
the step unless a 2xx JSON response arrives before its `timeout_ms`; one still
being read when the step times out or is cancelled saves nothing. A pointer
that names no string or number fails the step, so later steps never send a
missing value, as does a response over 4,000,000 characters: a request step
stops reading and aborts its fetch there. A UI step reads a body only when
its decoded size is bounded at 12,000,000 bytes or less: its received size,
or the declared Content-Length of one served from cache, times 1032 for each
gzip or deflate coding, the most DEFLATE expands. Other codings and unknown
or zero sizes are refused before materializing the body; decoded text over
the character limit also fails. The UI action may already have applied.
- `{saved:name}` sends a saved value in a request step's `path`, including its
query, and in a `goto` destination's `url`, percent-encoded; a destination's
structured `query` values are sent as declared. A saved value that makes a
path segment `.` or `..`, alone or with the text beside it (including `%2e`),
fails the step, since the URL would drop it. In a request step's `json` body,
a string that is only `{saved:name}` becomes the saved value with the JSON
type it was saved with, so a saved number id is sent as a number, and
`{saved:name}` inside a longer string is filled as text. A header value or
Basic credentials part sends one as `{"source": "saved", "name"}`. Other
braces, such as `{id}`, are sent as written, and so is a percent-encoded
`%7Bsaved:name%7D` anywhere but a `goto` URL, whose parsed form spells the
braces that way.
- Validation refuses a saved value that no earlier step saves. A later save of
the same name replaces the value. An integer past 2^53, which JSON parsing
would round to another number, and a number that is not finite, such as
`1e400`, negative zero, or a value that underflows to zero, such as `1e-400`, fail the step
that saves it.
- A scenario keeps at most 4,000,000 characters of saved values at once,
counting distinct retained redaction spellings, including trimmed variants,
of values saved as secret. Replacing a binding releases its live value
charge, while secret spellings remain protected for the scenario lifetime.
A save past the budget fails before changing the binding.
- A value saved with `secret: true` is redacted however short, as a declared
secret is, as `[REDACTED:page:<step_id>.<name>]`. The selected source
response is protected before retaining its body or header values, including
when only headers were requested. If that body cannot be read safely, its
body and header values remain unavailable. Before publishing the capture,
the selected source transaction's earlier URL, request headers and request
body are also protected using the admitted values. Unresolved source values
withhold that transaction's affected metadata/content with an explicit
limitation. Before publication, retained events are rechecked
with the final marked-response values, including immediate follow-up requests
and their response bodies, console/page errors, WebSocket text and event URL/name fields. Learned redaction facts are independent of source-identity readiness. Unknown source values withhold only the selected
source transaction. Other saved values, such as ids, stay visible, since
environments model them. The environment build warns about saved values
named like credentials that are not marked secret
(`credential-named-saved-values`).
- Page-derived cookie/local-storage values and constructed Basic credentials
remain redacted only from observations made after they were read/built.
They do not retroactively rewrite earlier facts. The final event projection
borrows only declared and response-marked patterns; promotion of an already
held page value to an explicit response secret preserves its first marker.
Projection state is session-local, references admitted pattern/byte owners,
and is invalidated on admission or response classification changes.
- Screenshots are withheld as `missing` from the start when the scenario plans
any response save marked `secret: true`, page-derived or Basic authentication
header values, or whenever nonempty sensitive values
are already known. Raw pixels cannot be safely text-redacted, so the provider
does not request pixels in those cases. This also protects earlier steps and
partial failures without retaining an image to scrub later. Other requested
artifacts remain available; affected screenshot/aggregate completeness is
partial and is not eligible for equality.
- Saved URLs and expanded request JSON are admitted before building their expanded strings or
serializing. Its allowance is the smaller of Node's string limit and
one sixteenth of current V8 available heap bytes, allowing for the expanded
tree, serialization, Playwright transport and browser-side copies. Size
accounting includes JSON escaping and repeated references; each distinct
saved value's size is measured once per request. An oversized request fails
before browser dispatch; the complete request envelope, including repeated
header values, is checked as well. URL encoding uses a conservative
nine-characters-per-UTF-16-unit allowance before encoding a saved value.
Reduce the payload/repetitions, or raise the parent
Node heap and restart; this is a conservative allocation estimate, not a
measurement or guarantee of browser/process-family memory.
- Evidence records a request step's empty `save` list as absent, so a scenario
that saves nothing records the same parameters and `scenario_sha256` as
before saved values existed, and REA versions without saved values read its
@@ -75,6 +75,13 @@ DOM node indices retain their snapshot positions. Invalid rows are omitted with
completeness diagnostics, so `total_nodes` can exceed the retained node count
and a parent index can refer to an omitted source node.
`dom.node_value_length_unit` is `utf8-bytes`: each node's `node_value_length`
counts the bytes needed to encode its value as UTF-8, not characters. Web
observations instead declare `utf16-code-units`. These existing measurements
are preserved; the unit is reported once for the whole DOM, without retaining
text values. Historical captures without the unit leave it unknown; reading
them does not infer a unit or convert their stored lengths.
Script metadata retains its execution-context frame ID when CDP supplies one and
records admission separately in `frame_scope`. An `authorized` scope refers to
the observed document generation; `unknown` retains the known frame ID without
@@ -140,6 +147,14 @@ window-targeted renderer reload/crash actions, and synthetic `open-url` or
WebContents identities, process metrics, and IPC channel and value-shape metadata.
Payload values are not retained.
For the exact Playwright launch sequence, the active hook removes only REA's
bootstrap arguments from the application-visible `process.argv`. The vector keeps
the selected entry point and every caller argument, in order, including repeated
hook paths, `.asar` filenames, and debug-like options. The private launch handoff
is consumed before application code runs. If the observed sequence cannot be
matched exactly, argv stays unchanged and capture coverage reports `hook_conflict`;
REA does not guess an entry point or remove arguments by filename patterns.
`window_index` selects from Playwright's current Page list at the start of each action. REA resolves that Page to its corresponding BrowserWindow once; the action's `target` uses the BrowserWindow's capture-scoped `window_id`, matching the window inventory. Reordering either enumeration after selection does not retarget the action. Handle cleanup failures retain the completed action records and remain owned for provider cleanup retry.
Run the real fixture verifier with an operator-provided Electron runtime:
@@ -186,6 +201,9 @@ neither result means the requested host operation occurred. The existing
`openExternal`, `openPath`, and `trashItem` blocks reject their promises.
The `trashItem` block prevents that call from reaching the OS trash operation;
it does not intercept other ways the application can remove files.
Process rejection and uncaught-exception events are observed at dispatch without
adding exception handlers. The runtime and application retain their warning,
handler, and fatal-exit policies, including for rejected shell-block promises.
CLI SIGINT/SIGTERM requests cancellation and awaits owned-process cleanup.
Window IDs, URLs, and cached titles come from one main-process Electron snapshot,
so a crashed renderer does not require a successful renderer-side title query.
| Segment data or a segment header beyond the end of the file | Truncation limitation; checksum and SHA-256 `unavailable`; coverage `unknown` |
| File ends before the checksum or the appended digest | That status is `unavailable`, with a limitation; coverage `unknown`, since a signature sector could follow |
| Checksum or SHA-256 mismatch | `invalid`, with a limitation; coverage is unchanged |
| Erased flash after the image that runs past the three slots | Described; no limitation |
| Erased flash that ends before the three slots are complete | Limitation; coverage `unknown`, since a signed image cut there would have lost its sector |
| A signature sector the file cuts short | Limitation; coverage `unknown` |
| Any other bytes after the image | Ranges in `undescribed_trailing`, a limitation, and coverage `unknown` |
| Segment length that is not a multiple of 4 | Limitation: ESP-IDF rejects such an image |
| Segment of 16 MiB or more, except on chips with 32-bit cache addresses | Limitation: ESP-IDF rejects such an image |
ESP32-S3, ESP32-P4, ESP32-C5, ESP32-C61, ESP32-H4 and ESP32-S31 map flash
through 32-bit cache addresses (`SOC_SPI_MEM_SUPPORT_CACHE_32BIT_ADDR_MAP`), so
ESP-IDF sets no 16 MiB segment limit for them. For an unknown chip the limit
is not applied.
Undescribed bytes include other flash contents such as an app partition's
stale data, a Secure Boot v1 signature, and the flash segments that esptool
`--ram-only-header` writes after the checksum without counting them. REA
reports their ranges and does not guess their structure. In the padding, in a
sector without a valid block, in the sector's slack and after the sector, the
first byte that is not erased starts a range to the end of that region. A
slot that is neither a block nor erased is reported whole. Adjacent ranges
are merged.
A whole-flash dump whose partition table REA finds is read as an
[ESP flash image](esp-flash-image.md). Without one, a dump of a chip whose
bootloader is at offset 0, such as ESP32-C3 or ESP32-S3, is recognized as that
bootloader image, and the partition table, apps and data follow as
undescribed bytes.
A dump read from a later flash address, such as an ESP32 dump read from
`0x1000` with `read-flash 0x1000 …` or `merge-bin --target-offset 0x1000`,
opens as its bootloader image. REA says so when the file shows it: the image
at file offset 0 is not an app, its chip's ROM loads the bootloader from a
later address (`0x1000` for ESP32 and S2, `0x2000` for C5, H4, P4 and S31),
and a partition table sits where that origin puts `0x8000` (file offset
`0x7000` for `0x1000`), after the image ends. Then `flash_dump` records
`flash_offset` and
`table_file_offset`, a limitation says to reopen the file with
`format: "esp-flash-image"` and that `flash_offset`, and `inspect_artifact`
lists the same `open_binary` call as a next probe. `flash_dump` is null for
every other image. Evidence that REA recorded before it reported the field
omits it; REA still reads and compares such Evidence, and the omission says
nothing about where the file was read from.
## Limits
An image has at most 16 segments and 3 signature blocks. Decoding reads each
byte once and retains only fixed-size fields. It reads through erased flash to
find the first byte that is not erased, and stops there. There is no
file-size limit.
## Analyzing a segment
Inventory does not disassemble code or build a memory map. `open_binary`
admits neither RISC-V nor Xtensa as a raw-binary architecture, so a segment
cannot yet be opened as a raw image. Each segment's `source_location` and
`load_addr` give the file offset and address to map it at by hand, for
example with the Rizin tool's `om` command.
## Verification
Synthetic fixtures with computed checksums, digests and CRCs cover
recognition, ESP8266 layouts, the ESP32, ESP32-C3, ESP32-C6 and ESP32-P4
memory maps, truncation at each structure, integrity mismatches, erased and
undescribed trailing bytes, signature slots including damaged and unknown
layouts, and the refusals. CLI and stdio MCP tests exercise the same artifact
provider and advertised schemas.
The optional real lane builds ELF files from source, converts them with
esptool and compares REA's facts with `esptool image-info`:
@@ -30,6 +30,18 @@ signal. Confirm resource causes from process output or host/container records;
operation-scoped JVM `OutOfMemoryError` observations retain their separate resource
classification. Automatic kernel or cgroup OOM attribution is not implemented.
After the authenticated startup handshake, rejected inputs (`invalid_request`,
`not_found`, `ambiguous`, an invalid function name, or an address without one
exact file offset) use
`diagnostics.diagnostic_scope: "request"` when their active output checkpoint is
available. They preserve the active request's
`failure_output`, target/profile identity and runtime/log coordinates, while
omitting cumulative launcher stdout/stderr and launch-support output. Startup,
process, protocol and other remote failures retain full session diagnostics.
The Ghidra client still owns its bounded cumulative output; this projection does
not clear it or alter session readiness. Log paths refer to the owned ephemeral
runtime and are removed by normal cleanup.
Provider selection and profile identity follow
[MCP contracts](mcp-contracts.md#identity-and-discovery). One deep
provider stays bound to the target; runtime failure never selects another
@@ -160,6 +172,48 @@ have no file backing. ELF mappings are observed
without independent format-specific load-image verification, as the result's
status and limitations state.
Relocation rows retain Ghidra's `symbol` value, including an empty string or
null when returned by Ghidra. It is a provider observation, not proof of the
runtime-resolved address. Older Evidence may not have recorded this field;
a relocation page requesting it reports that uncertainty.
For large relocation inventories, select MCP `inspect_native_load_image` with
`detail: "summary"`. It returns verification state, observed identity and counts,
while retaining complete Evidence. Use `normalized_result.parent_evidence_id`
with `inspect_analysis_view` and a view such as:
```json
{
"kind": "native-load-image",
"collection": "relocations",
"offset": 0,
"limit": 32,
"fields": ["address", "status", "type"]
}
```
Collections are `source_files`, `mappings`, `relocations`, `entry_points`,
`entry_context`, `checks` and DOL `sections`. Optional
`indices` select original row indices before paging; empty indices select none.
Fields are collection-specific. Relocation defaults omit byte payloads; check defaults omit potentially large `expected`/`observed` inventories. Request those fields explicitly when needed. Other collections default to all fields. Metadata recovery remains counted and retained in complete parent Evidence.
Parent Evidence, collection and `source_index` identify a row, including same-address
relocation alternatives. Null counts or rows mean unrecorded or inapplicable;
they never establish an empty observation. Exhausting a page does not upgrade
unsupported verification or runtime coverage. Pages query immutable retained
facts without reopening the target or repeating Ghidra work.
Each page carries the source's structured `verification_status` independently
of its page coverage.
CLI summary delivery requires a complete Evidence bundle:
```sh
rea inspect-native-load-image <path> --provider ghidra --detail summary --evidence_output /new/load-image.json --json
```
Read selected pages later through `inspect-analysis-view` using the returned
parent ID and an `evidence-file` source pointing to that bundle. Complete
delivery remains the default in both adapters.
An unsuccessful numeric procedure lookup reports the current Ghidra image base
and gives the CLI commands above alongside the corresponding MCP operations.
Repeat any target-format, architecture, and entry-point options used to open the
@@ -251,7 +305,7 @@ host/provider acceptance run with only the host compiler.
`--dll`. These static fixtures contain source-owned encoded instructions; they
are not compiler-produced programs. See the [Windows verification scope](windows-ghidra-p0.md#native-authority-and-target-identity)
for real-provider coverage and exclusions.
These lanes check all 26 read-only operations, target/snapshot/import SHA-256 linkage,
These lanes check all 27 read-only operations, target/snapshot/import SHA-256 linkage,
and project, endpoint, process, and runtime cleanup. The independent native
lane checks DACLs, handle admission, cancellation, and Job Object lifecycle.
`npm run verify:ghidra:windows:package` packs and installs REA into an isolated
| Integer and pointer values | `a0`–`a7`, then the stack in 4-byte-aligned slots | The default integer and stack entries, unchanged |
| `float` | One integer register, or a stack slot | The `fa0`–`fa7` entries and the `float` rule are removed, so a `float` takes the next integer register |
| `double` and 8-byte values | A register pair with the low word in the lower register. Named arguments need no aligned pair. With one register left, the low word goes in `a7` and the high word on the stack | The default `join` rule with stack spill: Ghidra writes `a2:4,a1:4` for a pair and `Stack[0x0]:4,a7:4` for the split |
| Aggregates up to 8 bytes | One or two integer registers, laid out as in memory, also when every member is floating-point | The homogeneous-float-aggregate rule is removed; the `join` rule applies |
| Aggregates over 8 bytes | By reference: the caller passes the address of a copy | The default rule that converts structures and unions of 9 bytes or more to pointers |
| Return values | As the first argument of the same type: `a0`, or `a0` and `a1`. Larger values go to a caller buffer whose address is an implicit first argument in `a0` | The `fa0`/`fa1` return entries and floating-point rules are removed; Ghidra's hidden-return fallback supplies the pointer in `a0` |
| Variadic arguments | The integer convention. A variadic argument with 8-byte alignment and size takes an aligned (even-numbered) register pair, or the stack; after the first stack argument, the rest go on the stack | The default variadic rule (`join` with alignment), unchanged |
| Preserved registers | `s0`–`s11` and `sp`; `gp` and `tp` are not allocated. No floating-point register is preserved | The default preserved list without `fs0`–`fs11`, which join the killed-by-call list |
| Scalars over 8 bytes | By reference (`long double` is 16 bytes), and returned through a caller buffer | A rule before the final join converts any value of 9 bytes or more to a pointer; a return that size already goes to a caller buffer |
Ghidra recovers the arguments of a variadic call from the registers the caller
writes, not from their types. A padding register that still holds an earlier
value can appear as an argument, and arguments after an unwritten register are
omitted.
### ilp32f
Written policy for `EF_RISCV_FLOAT_ABI_SINGLE` on ELF32. REA does not admit it
| `float` | `fa0`–`fa7` while one is free, then the integer convention |
| `double` and other values wider than 32 bits | The integer convention, as in ilp32 |
| Structure of one or two floats (after flattening) | One floating-point register per member when enough are free, otherwise the integer convention |
| Structure of one float and one integer, in either order | One floating-point and one integer register when both are free, otherwise the integer convention |
| Unions | Never flattened; the integer convention |
| Return values | As the first argument: `fa0`, `fa0` and `fa1`, or `fa0` and `a0` |
| Variadic arguments | The integer convention, as in ilp32 |
| Preserved registers | `s0`–`s11`, `sp`, and `fs0`–`fs11` for values of 32 bits or fewer |
The planned prototype limits the default floating-point entries to 4 bytes and
keeps the shared rule that passes values of 9 bytes or more by reference.
Observed on Ghidra 12.1.4, a `double` then takes an integer pair and a structure
of two floats takes `fa0` and `fa1`. Ghidra's model has no rule for a structure
of one float and one integer: it joins `struct { float; int; }` into `a1:a0`
instead of `fa0` and `a0`. The ilp32f policy needs that rule and its own
fixtures before it is admitted.
### ilp32d and lp64d
The double-float lanes use Ghidra's stock `riscv32-fp.cspec` and
`riscv64-fp.cspec` models unchanged. This page does not verify them. They share
the float-and-integer structure gap above.
## ISA declarations
The target reader finds the one `SHT_RISCV_ATTRIBUTES` section and reads
`Tag_RISCV_arch`, such as `rv32i2p1_m2p0_c2p0_zicsr2p0_zifencei2p0`. It
retains the string, the base and the extension names, or an unavailable reason
when the section is malformed, declares other vendors or scopes, or uses a
non-canonical string. A missing section or tag is recorded as absent.
Calling-convention lanes admit an image only when it declares an RV32I base
and only extensions that Ghidra 12.1.4's generic RV32 language implements with
p-code: `m`, `a`, `f`, `d`, `c`, `zicsr`, `zifencei`, and the subset names
toolchains record, `zmmul`, `zaamo`, `zalrsc`, `zca`, `zcf` and `zcd`. Any
other extension is refused. In particular:
- Ghidra 12.1.4 decodes no Zcb, Zcmp or Zcmt instruction.
- Its bit-manipulation, cryptography and packed-SIMD files follow draft
encodings and mark nearly every instruction unimplemented.
- Vendor extensions, such as Espressif's `xesploop`, `xespv` and `xespdsp`,
decode only as custom-opcode placeholders.
ESP-IDF builds ESP32-C2 and C3 images as `rv32imc_zicsr_zifencei` and ESP32-C5,
C6, C61, H2 and H21 images as `rv32imac_zicsr_zifencei_zaamo_zalrsc`, all ilp32.
Those base strings are admitted. Builds that add `_zba_zbb_zbs`,
`_zcb_zcmp_zcmt` or an `xesp` extension are refused.
## Evidence
The lane identity is `elf32-exec-le-double-float-generic-v1` or
`elf64-exec-le-double-float-generic-v1`, and `riscv_elf` retains `elf_class`,
`byte_order`, `type`, and the raw `flags` alongside the read-only booleans
`float_abi_double`, `rvc_declared`, `rve_declared`, and `tso_declared`. Those
fields report declarations the header makes; none of them is evidence of the
original compiler, of instruction conformance, or of a floating-point calling
convention beyond the ABI the pinned language models.
The lane identity is one of the lane ids above, and `riscv_elf` retains
`elf_class`, `byte_order`, `type`, and the raw `flags` alongside the read-only
booleans `float_abi_double`, `rvc_declared`, `rve_declared`, and
`tso_declared`. Soft-float profiles also retain `float_abi: soft`, the declared
`arch` string, and `riscv_calling_convention: ilp32-prototype-v1`. Those fields
report declarations the header and attributes make, and the policy REA applied;
none of them is evidence of the original compiler or of instruction
conformance. Session Evidence states the prototype and how many functions and
definitions use it.
## Verification and limits
@@ -76,9 +268,9 @@ tree ships no host decompiler binary, so `Decompiler/os/<platform>/decompile` an
[installation](installation.md#ghidra).
Run `npm run verify:ghidra:riscv` with an existing Ghidra 12.1.4 installation,
JDK 21+, and a RISC-V-capable Clang, LLD and LLVM readelf. The verifier acquires
no tools and never executes target code. Select installed tools explicitly when
they are not on `PATH`:
JDK 21+, and a RISC-V-capable Clang, LLD, LLVM readelf and LLVM objdump. The
verifier acquires no tools and never executes target code. Select installed
| Gaps between regions | Separate regions; never zero-filled |
| Out-of-order records | Sorted into regions |
| An address written twice with the same value | Accepted and counted in `duplicate_bytes` |
| An address written twice with different values | Fails, naming the address and the line spans of both writes |
| Checksum mismatch, odd hex length, non-hex character, byte count disagreeing with the line | Fails with the line, declared and computed values, and record type |
| Line longer than 521 (Intel) or 514 (S-record) characters, excluding the line ending | Fails before the line is decoded |
| Intel type `06` or higher, `S4`, wrong fixed length for Intel `01`–`05` or `S5`–`S9` | Fails |
| Missing Intel `01` or `S7`–`S9` | Regions are returned; coverage is unknown and the file may be truncated |
| A non-blank record after termination, or a second `S0` | Fails; blank lines are allowed anywhere |
| `S5`/`S6` count different from the data records before it | Fails |
| Termination type not matching the data address width | Reported as a limitation; the declared fact is kept |
| Conflicting repeated `03`/`05` start records | Fails; identical repeats are allowed |
| An Intel data record whose 16-bit offset wraps past `0xFFFF` | Fails: producers disagree on whether the address wraps or continues linearly |
| I16HEX data above `0xFFFFF`, or data crossing the S1/S2/S3 address width | Fails |
Lowercase hex, LF and CRLF line endings, and trailing spaces or tabs are
accepted. A CR not followed by LF is an invalid character. Failures apply to
the whole file and return no partial regions.
## Limits
The file must be at most 128 MiB. Every data byte needs two hex digits, so the
decoder reserves at most half the file size for decoded bytes before reading,
and stores only written addresses. The decoder retains at most 65,536
address-discontiguous record runs before merging them into regions. A run is a
sequence of consecutive records whose addresses continue each other. Both are
fixed resource bounds, not format limits, and fail as limits (`file-size` and
`capacity`) with no runtime override. Configurable ceilings are tracked in
[#6176](https://github.com/morluto/rea/issues/6176). The line-length limits are
the one-byte record count: Intel HEX allows 255 data bytes, while S-record
counts include address and checksum bytes.
Decoded regions are cached only for the reader's lifetime. Each open checks
its admitted carrier identity again. Region streams copy at most 64 KiB per
chunk, so consumers cannot mutate cached bytes and opening a large region does
not copy it in full. Caller cancellation keeps its exact reason, and closing
the reader waits for active region streams to close.
## Analyzing a region
Inventory does not establish the CPU, byte order or memory map, or that a
declared start address is executable. To disassemble a region, extract it and
open the region file with `open_binary` using format `raw-binary`, an
architecture you choose, and `entry_point` set to the region start address;
raw-binary maps file offset 0 at `entry_point`. Each region is opened
separately, so references between regions are not resolved. A caller-selected
Ghidra language and base address for raw-binary imports is tracked in
@@ -43,7 +43,8 @@ rea decompile /absolute/path/to/program main --json
If the registration cannot be read or validated, provider diagnostics retain its
selected `registration_path` and the configuration owner's setting/constraint
facts in `configuration_details.settings`, plus `configuration_remediation`.
facts in `configuration_details.settings`. The availability's `remediation`, and
each operation it makes unavailable, carry the configuration owner's repair.
`configured` means a path was selected, not that it is valid or an IDA connection
was established; failed registrations remain unavailable. Correct
`REA_IDA_MCP_CONFIG` in the next CLI invocation's environment or the MCP server's
@@ -80,6 +81,71 @@ Native MCP 1.1.2 can navigate in the previously focused pseudocode tab when swit
Native tool observations retain actual tool names and complete JSON text. Upstream `structuredContent` can be a shortened preview even while text contains the full result. REA reads the full native representation and rejects missing/malformed text instead of treating previews as complete observations.
### Migrate a legacy attached registration
Both attached profiles remain supported. Migrating replaces the upstream server;
adding `"protocol": "native"` to a legacy Python proxy registration is insufficient.
The native plugin is separate from mrexodia/ida-pro-mcp, and upgrading mrexodia to
its modern tool catalog does not select this profile.
1. Back up the working registration file. Install and start the native plugin
using its [upstream instructions](https://github.com/0xshlomil/ida-free-mcp).
IDA and the plugin are bring-your-own; REA does not install or build them.
Use a build with the navigation correction described above, and restrict its
listener to loopback or apply the documented host firewall restriction.
Stop the legacy endpoint first if both servers would use the same port.
2. Open the same original input binary in IDA and wait for analysis. Keep that
file readable by REA. The native metadata resource must report its recorded
input SHA-256; the resource's `path` identifies the database, not an input
file to substitute. Universal Mach-O inputs still require a thin extraction.
3. Replace the registration, using the native plugin's actual local endpoint.
For example, this legacy registration:
```json
{
"mode": "attached",
"command": "python",
"args": [
"/absolute/path/to/ida_pro_mcp/server.py",
"--ida-rpc",
"http://127.0.0.1:13337"
]
}
```
becomes:
```json
{
"mode": "attached",
"protocol": "native",
"url": "http://127.0.0.1:13337/mcp"
}
```
Keep `REA_IDA_MCP_CONFIG` pointing at the replacement file and close/reopen
the REA binary session to connect with it. If changing the environment's
configuration path, restart the MCP server as described above. Do not reuse
the legacy RPC URL as an MCP HTTP endpoint.
4. Check `binary_session` availability before resuming workflows. Direct
`procedure_callers` becomes unavailable and dossier callers remain unknown;
incoming address xrefs do not establish direct call edges. Native strings
use C-string rendering. Assembly and instruction reads no longer navigate
the GUI; native pseudocode still changes focus and closes existing
pseudocode tabs. IDA Free decompilation may require cloud access and consent.
5. Run the [real-provider verification](#verification) with two distinct
functions and the required second-function marker. Supply independently
known string and xref oracles for those checks. Existing Windows evidence
covers IDA Pro 9.3 with the documented source build and listener change;
it does not establish stock-plugin, Windows Free, or native macOS behavior.
A valid registration or successful doctor check alone does not verify the
connection, target identity, or function switching.
REA continues to verify identity and image base around requests and closes only
its connection, preserving the attached database. Restore the backed-up legacy
registration and its upstream endpoint to return to the legacy profile.
The legacy producer moves the active GUI as a side effect: assembly and raw function instructions jump to the function, and successful pseudocode opens or reuses a pseudocode view. A function dossier does both. Native GUI decompilation also changes the active view. `binary_session` reports these operations with `effects.may_show_ui: true`; headless operations and the other attached operations report `false`.
## Give REA a target for headless analysis
@@ -155,6 +221,12 @@ For guided registration, set `REA_IDA_MCP_CONFIG` before `rea setup --client <cl
## Coverage and evidence
REA admits x86, x86_64, arm and arm64 executables through IDA. Every other CPU
family, including families that REA recognizes later, is refused until REA
admits it explicitly. Raw-binary and GameCube DOL targets are refused; open
them with Ghidra. Admission is not verification: the real workflows under
[Verification](#verification) cover x86_64 and arm64 only.
All profiles support procedure/string inventory and literal or regex search, procedure resolution, pseudocode, assembly, raw function instructions, resolved callees, incoming address references, and a function dossier. Inventories and modern pseudocode/assembly follow upstream pagination. Upstream limits on modern xrefs and callees are retained as explicit truncation limitations. In the modern headless profile, targets that are not proven x86-only use processor-neutral `callgraph` edges instead of the x86 instruction checks in upstream `callees` (mrexodia/ida-pro-mcp#549). REA requests depth zero: the pinned upstream emits the root's outgoing edges without scanning child functions, then REA batches function-name lookup for those addresses. The graph covers recognized function destinations only; imports and other non-function targets remain unknown. A server without `callgraph` returns an empty relationship list with an explicit coverage limitation. Graph and path workflows retain those limitations alongside observed edges. A failed or inconsistent graph reply remains an error. A transport-level truncated preview is rejected rather than reported as complete evidence; the error preserves upstream retrieval guidance.
All profiles expose `list_documents` as the single target admitted for the session. Attached mode verifies its digest against IDA metadata before returning it; headless mode binds the digest-checked private input. `inspect_address_context`, `trace_feature`, and `binary_overview` can therefore resolve that identity without upstream document-listing tools. Segment enumeration is optional: when the selected profile cannot provide it, `binary_overview` still returns document and procedure/string counts and omits `segments` and `segment_count`; omission means unobserved, not empty. The modern headless profile also supports `resolve_containing_procedure`. An address resolves through upstream `lookup_funcs` to the canonical entry of the function that contains it; the body stays unavailable like other IDA identities. When no function contains it, REA reads the database-scoped minimal `survey_binary` and reports `not_in_procedure` for an address inside one of its segments (ends are exclusive) and `outside_segments` otherwise. A missing, truncated or inconsistent survey fails the read instead of guessing a reason. `trace_feature` marks a trace truncated, with a residual unknown for that address, when upstream caps the references it returned. Attached profiles do not declare address containment resolution.
@@ -175,7 +247,7 @@ The verifier requires the resolved procedure in both inventory and procedure sea
The initial real workflows use stdio registrations for the legacy Windows GUI profile and a Windows x64 headless supervisor with IDA 9.3. The headless workflow has also passed on macOS arm64 with IDA Professional 9.1 (build 9.1.250226) against the same upstream commit `c133c3853faa111a9b00ee615c013b720d0c4acd`, using a source-built arm64 Mach-O fixture; Mach-O C symbols carry a leading underscore, so the procedure name is `_rea_fixture_add`. Streamable HTTP is tested against a loopback MCP fixture with the pinned client SDK; a real IDA HTTP-supervisor workflow remains unverified. Linux headless engine operation, other IDA releases, architectures, and modern attached GUI tools require their corresponding real workflows before support is claimed.
Native Windows verification uses IDA Pro 9.3 with the public two-function fixture and Windows `whoami.exe`, using upstream `c9d15d98e8a143fc74f6e14037c48f63889b6d6c` with a loopback-listener change. It covers production CLI/MCP parity, distinct-function switching, identity and connection cleanup. See the [correctness and performance measurements](https://github.com/morluto/rea/blob/main/experiments/analysis-efficiency/ida-native-windows.md). This follow-up did not run Windows IDA Free or native macOS.
Native Windows verification uses IDA Pro 9.3 with the public two-function fixture and Windows `whoami.exe`, using upstream `c9d15d98e8a143fc74f6e14037c48f63889b6d6c` with a loopback-listener change. It covers production CLI/MCP parity, distinct-function switching, identity and connection cleanup. See the [correctness and performance measurements](https://github.com/morluto/rea/blob/a2e2c29b59e8775b0e6aedb09a9eb16adf32646b/experiments/analysis-efficiency/ida-native-windows.md). This follow-up did not run Windows IDA Free or native macOS.
Headless readiness distinguishes an explicit `auto_analysis_ready: false` from
unknown readiness (`null`). The former returns a retryable capability refusal
| One analysis engine | `rea doctor --provider ghidra --json` (or `hopper`, `ida`) |
| One analysis engine | `rea doctor --provider ghidra --json` (or `hopper`, `ida`, `kuna`) |
| One agent registration | `rea doctor --client codex --json` (see [client IDs](#supported-agents)) |
| Installed workflow instructions | `rea doctor --skill --json` |
@@ -383,6 +390,20 @@ A scoped report has `scope.mode: "explicit"`. Its `scope_checks` determine
`environment_healthy` summarizes the full audit. `--target` adds a target check,
but the report remains audit-wide unless a scope option is also supplied.
Provider checks read the terminal running doctor. A client-started server may
receive a different environment, and doctor cannot observe it. When an available
provider relies on a setting that a registration's file does not record, or
records with other text, doctor adds a non-blocking
`registration_environment:<client>` check with `classification: "unknown"`. The
setting names appear in that registration's `provider_settings`. Selecting the
client or an affected provider places the check in `scope_checks`; it never
changes `healthy`, because the client may still pass the setting on. Preview
`rea setup --client <client-id> --skill=false --dry-run` from that terminal and
approve it to record the settings, then restart the client. To see what the
running server can use, call `binary_session` in the client and read
`result.tool_availability`. For Hopper, only an explicit `HOPPER_LAUNCHER_PATH`
is compared.
An invalid `REA_LOG_LEVEL` fails every scope, including one client or the
installed skill, because the executable rejects it before any command. A
selected client also fails for every other setting that MCP startup rejects,
@@ -412,6 +433,13 @@ The public error code varies by caller: `capability_unavailable`,
only a generic startup error. Check this prerequisite before following
provider-specific recovery advice.
`binary_session` still returns status on such a host. Providers whose
readiness check reaches the owned-child launch, such as Ghidra's Java probe,
JADX and Apktool, report themselves unavailable with a reason naming the
missing `ps`. A provider that fails an earlier configuration check, such as
JADX without `REA_JADX_MCP_JAR`, keeps that reason. The other providers keep
their own status. `open_binary` keeps the refusal described above.
## Diagnose, update, and remove
`rea doctor --json` is strictly read-only. `rea update` updates only the npm or Bun global installation that owns the running CLI. Source checkouts and package-runner copies must be updated through the mechanism that owns them; a fresh package-runner invocation can use `npx rea-agents@latest` or `bunx rea-agents@latest`. `rea uninstall` removes only REA-owned agent registrations and skill files; `--purge-data` additionally removes REA cache and state paths.
@@ -429,11 +457,19 @@ and other MCP servers. Purging removes REA's legacy cache and state under
archives under `${XDG_CACHE_HOME:-~/.cache}/rea/hopper-releases`. Other cache
namespaces remain intact. A caller-selected `REA_KUNA_CACHE_DIR` is retained
and reported because REA cannot establish ownership of its entire contents.
Lexically equivalent spellings of that default directory, including a trailing
separator, a `.` segment, or a repeated separator, share its purge result.
The comparison preserves POSIX `..` traversal and literal backslashes.
A client configuration that is malformed, unreadable, or at an unsafe
path stops the operation before anything is removed, and a client that fails
while being updated stops the remaining removals. Uninstall reports each
candidate `configuration_path` with its outcome. A client with multiple files
keeps their individual `actions` alongside its aggregate status and detail,
path stops the operation before anything is removed when that client's
product applicability is established or unresolved. Shared registration
storage such as `~/.agents/mcp.json` whose product cannot be established
(for example when FreeBuff and ZCode discovery directories are absent) is
retained unchanged and does not block other REA removals, including cache
and state deletion with `--purge-data`. A client that fails while being
updated stops the remaining removals. Uninstall reports each candidate
`configuration_path` with its outcome. A client with multiple files keeps
their individual `actions` alongside its aggregate status and detail,
including earlier removals when a later file fails. The paths name
the discovered configuration candidates, not resolved physical file identities.
For JSON clients, uninstall removes every repeated `rea` member in the effective
@@ -495,6 +531,10 @@ Do not rename `index.cjs` to `index.mjs` or disable protection to make startup p
### Update the owning installation
Read-only metadata and version checks time out after 10 seconds. This does not
interrupt a package installation already in progress. If a probe cannot be
stopped, the updater reports a cleanup failure.
`rea update` installs the exact resolved
release into the npm prefix or Bun global directory that owns the running
package, then checks the new executable's version before reporting success.
@@ -525,8 +565,8 @@ non-TTY, and JSON modes without applying configuration changes.
## Released package and main
Repository main documents its current code and generated catalog. `@latest`
selects the npm release, and persistent MCP registrations are pinned to the
version that performed setup. Installing newer instructions does not update a
selects the npm release, and managed MCP registrations bind the installation
that performed setup. Installing newer instructions does not update a
running server or its registration.
The npm release checked on **2026-10-11** was **6.4.0**, published from
@@ -640,7 +680,8 @@ Hopper is separate commercial software with its own license. Its free demo has
vendor-defined limits, and a paid license is optional. REA reuses any detected
installation and preserves Hopper during uninstall.
The automated Hopper installer supports macOS 12+ and x86_64 hosts running
The automated Hopper installer supports macOS 12+, subject to the minimum
macOS version of the current Hopper release, and x86_64 hosts running
Ubuntu 24.04+, Fedora 41+, Nobara 44+, Arch Linux, CachyOS, or Omarchy. The
official Linux packages are x86_64-only, so setup does not offer Hopper on other
Linux architectures; configure [Ghidra](#ghidra) there instead. REA itself runs on any
@@ -649,8 +690,33 @@ an existing Hopper launcher is selected through `HOPPER_LAUNCHER_PATH`. Ghidra
and IDA have their own provider-specific host requirements; Windows Ghidra uses
the [experimental P0 boundary](windows-ghidra-p0.md).
Bazzite uses RPM packages, but REA does not install Hopper automatically on it.
Setup skips the Hopper offer because Bazzite is outside the supported host list,
regardless of whether Xvfb is installed. Bazzite documents
[containers for installing Linux packages](https://docs.bazzite.gg/Installing_and_Managing_Software/Containers/).
If using a container, check its distribution and architecture against the list
above. Install REA and Hopper in that container, give it access to the files you
want to analyze, and configure your agent to launch REA there. Verify installation
and analysis before relying on that workflow.
`rea doctor` checks health; it does not offer to install Hopper. Run `rea setup`
for installation. Setup cannot prompt when its output is redirected or during a
container image build. To approve Hopper installation without prompts, run
`rea setup --yes --install-hopper`.
If a Hopper request fails on macOS or Linux, the setup result's `download`
field names the request (`release_feed`, `package` or `public_key`), its `url`,
and the observed outcome `kind`: `http_status` with `httpStatus`, `timeout`,
`network`, `response_too_large`, or `url_refused`. The remediation names the
same URL and outcome. HTTP 404 or 410 means the file is missing, so an
immediate retry is unlikely to help; 5xx, 408 and 429 indicate a temporary
service failure. The status does not establish the underlying cause or guarantee
that retrying will succeed. Doctor and Linux launches report the same request
and status when publisher verification cannot download the current package.
Hopper metadata and package downloads require HTTPS throughout the redirect
chain; a redirect to HTTP is refused before that destination is requested.
chain; a redirect to HTTP or to a URL carrying credentials is refused before
that destination is requested, and those credentials are not reported.
On macOS, approved setup downloads the official DMG, checks its published size
and digest, validates the application bundle, and atomically installs it to
@@ -658,6 +724,15 @@ and digest, validates the application bundle, and atomically installs it to
operator can choose its demo mode or activate an existing license. Homebrew and
administrator access are not used.
Each Hopper release declares its own minimum macOS version, so the setup plan
cannot confirm support before approval. Setup compares this Mac's version with
the release metadata's `deployment_target` before downloading the DMG, and with
the signed application's `LSMinimumSystemVersion` before copying it. If this Mac
is older, setup fails with `incompatible_host_version`, names both versions, and
installs nothing. Update macOS, or install a Hopper release that supports this
macOS version yourself and set `HOPPER_LAUNCHER_PATH`. An existing Hopper
installation is not checked against the current release's minimum.
The Hopper launcher action used by REA creates a document from an executable;
its supported command-line interface does not attach to an already-open
document. While another live REA session owns the same target and loader
@@ -698,6 +773,12 @@ Rotating signing subkeys remain bound to that primary identity.
The feed row itself is not PGP-signed; doctor labels its display version as
`reported_release_version`, separately from the authenticated launcher identity.
Even a correctly signed package can contain a launcher whose hash disagrees with
the feed. REA stops before installing that package. When reporting this error,
include the release version, package format and both hashes from the diagnostic.
Each format needs its own check: a matching Ubuntu DEB does not establish that
the Fedora RPM matches.
The approval plan includes the caller-selected cache location. Approved setup
retains the signed archive, public key and release metadata under
Check that preparation completed successfully before selecting the project.
Preparation time, disk copy/fingerprinting, engine startup and client deadlines
are separate costs; reusing a project avoids import/analysis, not all startup.
Use the [heap/CPU controls](#ghidra-heap-and-cpu-controls) appropriate to the target.
### Ghidra import language and analysis seeds
Ghidra detects the language from the executable header. To set it (for example, Borland C++ PE images):
@@ -1310,6 +1483,33 @@ REA passes `-processor`/`-cspec`, checks both against the installation's `.ldefs
Language definitions must be well-formed XML encoded as UTF-8 or BOM-marked UTF-16, with any encoding declaration matching the bytes. Literal Unicode replacement characters (U+FFFD) are accepted; malformed source bytes are rejected.
Ghidra imports little-endian ARM ELF files, including Cortex-M
(`arm-none-eabi`) firmware, as `ARM:LE:32:v8`. That language defaults to ARM
state, and its ELF loader marks code as Thumb only where `$t` mapping symbols
or odd function-symbol addresses say so. Cortex-M cores run only Thumb code, so
the gap matters mainly for code without those markers, such as stripped images
and handlers reached only through the vector table. REA has no M-profile ARM
lane. As an unverified interim, select an M-profile language, which decodes in
Thumb state by default. For ARMv6-M, ARMv7-M and ARMv7E-M cores:
```bash
export REA_GHIDRA_LANGUAGE_ID="ARM:LE:32:Cortex"
```
For ARMv8-M cores, use `ARM:LE:32:v8-m` instead. REA has not verified either
language for Cortex-M analysis. While the variable is set, it applies to every
Ghidra import in the REA server:
- raw-binary, DOS, GameCube DOL and PSP opens are refused, as are opens in
lanes that reject a different language, such as RISC-V;
- REA asks Ghidra to import everything else in the selected ARM language,
including x86, x86-64, arm64, PowerPC and Hexagon ELF files and PE and
Mach-O files, because it does not check that the override's processor
matches the target.
Restart REA without the variable before you analyze other targets. A future
M-profile lane may refuse it; remove the setting then.
To seed code that auto-analysis cannot reach (computed jumps, pointers in data):
```bash
@@ -1351,6 +1551,21 @@ current setting. For Windows PE resource analysis, the exact name is
`WindowsResourceReference`. Disabling an analyzer skips the facts it would have
produced.
Ghidra's Objective-C Message Analyzer can log
`Required Objective-C data type not found in data type manager` /
`Try adding libobjc.dylib` when a Mach-O links Objective-C runtime symbols but
defines no class metadata (thin Cocoa glue in a C++ binary is a common case).
That text is non-fatal Ghidra analyzer output: auto-analysis continues, and the
Run current setup to migrate recognized older `npx`/`bunx` registrations, with
backups, then restart the client. Unrelated or unidentified registrations remain
untouched. Setup does not prewarm a separate package-runner command.
Persistent registrations should use one exact package version. `rea setup`
writes the same package-runner shape, or `bunx rea-agents@<version> mcp` when
setup runs through `bunx`, and pins it to the exact version that performed
setup. Run current setup to refresh an older registration, then restart the
client.
Package runners key their package cache by the requested spec, so the entry
installed for the `@latest` invocation that ran setup cannot serve the pinned
registration. Setup therefore materializes the pinned entry once, best-effort
with a bounded timeout, before finishing. When that step cannot run, for
example offline, the first client start performs the one-time install inside
the client's MCP startup deadline and may need one more restart.
If setup itself runs from a package-runner cache, the registration depends on
that cache entry and the selected runtime remaining installed. Removing either
makes startup fail; it does not download or resolve another package. Restore the
same installation and rerun setup. Changing package requests, cache roots, or
runtimes can create a different installation identity. If the old entry cannot be
identified, restore it first or manually remove the confirmed REA registration
before running setup again. A global installation provides a more durable path.
## Package runners and global installation
A package-runner invocation can use a REA version installed in the current project. To
request the latest release explicitly, use:
Packagerunners can select project-local packages and read project configuration,
even for explicit version requests. Acquire REA from a trusted working directory,
outside the target project, and pass absolute target paths to analysis commands.
REA cannot validate package acquisition that happens before its entrypoint starts.
To request the latest release explicitly, use:
```bash
npx rea-agents@latest setup
@@ -1512,16 +1813,14 @@ For an intentional rollback, make the package request explicit:
npm exec --yes --package=rea-agents@2.4.0 -- rea setup
```
Setup continues to pin persistent MCP registrations to the exact version that
performed setup. Running current setup later migrates unversioned or older
managed registrations through the normal reviewed setup transaction.
Setup binds persistent MCP registrations to the installation that performed setup.
Running current setup later migrates recognized legacy package-runner registrations
through the normal reviewed setup transaction.
With Bun, `bunx rea-agents setup` (or `bunx rea-agents@latest setup`) is the
equivalent entry point. Setup run through `npx` registers
`npx -y rea-agents@<version> mcp`; setup run through `bunx` or `bun x` registers
`bunx rea-agents@<version> mcp`. On Windows, setup wraps either command in
`cmd.exe /d /c` so clients can start the runner's command shim. Neither
registration points into the runner's temporary package cache.
equivalent acquisition entry point. Managed registrations use the selected Bun
executable and REA entry script directly. Windows registrations likewise use
absolute runtime and script paths, without a package-runner command shim.
REA supports Node.js 22.x (>=22.23.4), 24.x (>=24.22.1), and 26.12.1+. Node.js 23, 25, earlier 26.x releases, and prereleases are unsupported. It uses the npm already paired with that runtime and never upgrades Node.js, npm, or Homebrew.
@@ -1555,8 +1854,9 @@ Review the plan, approve the selected client changes, then restart those clients
This registers the installed entrypoint directly; installing the global package
alone does not replace existing `npx` registrations. Each client still has its own
REA process and analysis state. Direct registrations follow the installed CLI;
use `rea update` to update it. Running setup through `npx` again selects the pinned
package-runner command. Do not register paths inside npm's temporary package cache.
use `rea update` to update it. Running setup through `npx` again selects the
installation that executes setup, potentially in npm's temporary package cache;
retain that cache entry for as long as the registration uses it.
@@ -63,7 +63,7 @@ All four preserve target-file bytes. `open_jeb_project` selects or creates an en
- The engine caps `list_units` pages at 100; a full page is reported as `coverage: "partial"` because more units may exist. An empty remainder is `complete` only when fewer units than requested were returned.
- When `decompile_jeb_item` omits `unit_path`, the engine implicitly selects the project's first code unit; REA reports `unit_path: null` and a limitation instead of guessing the effective unit.
- `item_address` is a JEB code-item reference, such as a value returned by the engine's `list_code_items`. Preserve that representation and select its code unit explicitly. For the audited CE SQLite unit, the exported method reference was `sqlite3_libversion_number`; using its raw ELF symbol address did not produce a successful decompilation. File offsets, ELF virtual addresses and engine item references have different semantics. REA currently exposes unit listing, while code-item discovery remains an upstream JEB operation.
- Engine refusals (`success: false` envelopes) are preserved as unsupported-target errors carrying the engine's own message, for example a nonexistent item address.
- Valid engine refusals (`success: false` envelopes) return `execution_failure` with the engine's message when available. Diagnostics preserve the original envelope, including any recovery advice, and the inputs of a refused primary request. If client metadata fails after a primary reply, that reply and the caller's inputs remain in partial Evidence. REA does not infer an unsupported target from a refusal, invoke a recovery tool, or retry the request.
- Successfully parsed client information is reused for the lifetime of the MCP connection, including when the engine omits its version. Explicit client inspection and availability probes refresh it.
- An unreachable endpoint is a provider-availability failure with recovery guidance; REA does not retry or spawn the engine.
[Kuna](https://github.com/Noelo-Lab/kuna) is an Apache-2.0 decompiler distributed as one `kuna` executable. REA uses a `kuna` you supply. It does not download, install, update or bundle Kuna. REA verified this provider with Kuna 1.833; it accepts other releases and reports them as unverified in `rea doctor` and in each analysis profile.
[Kuna](https://github.com/Noelo-Lab/kuna) is an Apache-2.0 decompiler distributed as one `kuna` executable. For a new installation, use Kuna 1.989, REA's recommended verified release. REA uses a `kuna` you supply; it does not download, install, update or bundle Kuna. REA accepts other releases; `rea doctor` and analysis Evidence report them as unverified. Verification compares the reported version; it does not authenticate the executable. Analysis profiles retain the observed version and executable digest, independently of REA's current verified-release policy. The 1.989 real-provider checks ran on macOS arm64; Linux and other hosts have not been verified with that release.
Kuna is static: it reads the selected image and never executes it.
This profile change makes earlier Kuna analysis snapshots and resumable directory
exports incompatible, even with the same Kuna executable. Keep the old files:
their recorded Evidence remains historical. To continue analysis, open the
binary without the old snapshot and save a new one, or choose a new export
directory. Existing snapshot Evidence can be extracted from its
`evidence_bundle` JSON member and imported as a separate Evidence bundle; that
does not restore cached workflow entries or resume the old directory export.
The CLI command validates the bundle in a one-shot session. To use the records
in an MCP session, call `import_evidence_bundle` with the extracted file path.
This validates the embedded bundle, not the rest of the snapshot container.
## Configure
```bash
@@ -30,6 +48,10 @@ completes. Cleanup retries keep both the observations and the exact file owner.
| `REA_KUNA_TIMEOUT_MS` | 120 s | Deadline for one single-function decompile, and for the inventory run REA makes when the export ends without one. |
| `REA_KUNA_JOBS` | 1 | `decompile-project --jobs`. 1 is Kuna's reference answer; more jobs finish sooner but lose callee-first prototypes, and change about 40–60% of function bodies on the measured binaries. |
Pseudocode read from a parallel project export includes a prototype-context
limitation in its Evidence, including in function dossiers and composed
workflows. Early single-function results have their own source limitation.
A single-function decompile, inventory, string, graph or xref run that exceeds its deadline fails as `provider_timeout`, naming the setting that owns the deadline and that raising it takes a restart. A project export that exceeds `REA_KUNA_PROJECT_TIMEOUT_MS` is an export failure instead: requests that need the export report `capability_unavailable` with the export's own retry guidance.
Windows is not supported yet. Supported targets are ELF and thin Mach-O executables for x86-64, AArch64 and ARM; PE and universal Mach-O are rejected with a typed reason.
@@ -38,7 +60,7 @@ Windows is not supported yet. Supported targets are ELF and thin Mach-O executab
| `list_procedures`, `search_procedures`, `procedure_address` | Kuna's callable inventory (`kuna functions --json`), including import slots it does not decompile |
| `list_procedures`, `search_procedures`, `procedure_address` | Kuna's callable inventory (`kuna functions --json`), including import slots but not [Mach-O data](#mach-o-data-entries) |
| `address_name` | Function entries, then the data labels Kuna's export assigns |
| `procedure_pseudo_code`, `batch_decompile` | The project export, or a single-function run while the export builds |
| `procedure_assembly` | The project export's labeled listing |
@@ -47,6 +69,8 @@ Windows is not supported yet. Supported targets are ELF and thin Mach-O executab
| `analyze_function` | Pseudocode and assembly from the export, plus call relationships; typed references, comments, strings and CFG remain unavailable |
Kuna's inventory gives some entries aliases, such as ELF versioned-symbol names and generated `sub_` names. Name selectors accept an alias that names one entry and list every entry a shared alias names. `search_procedures` matches aliases too: each entry appears once under its primary name, with the matching aliases in `matched_aliases`. `list_procedures` shows primary names only.
Typed procedure references, basic blocks, raw-memory tools, annotations and type recovery are not available from this provider. Use another provider for those operations.
## How answers are produced
@@ -62,7 +86,7 @@ The first text or assembly request in a session looks for this target's export i
- Assembly and data names wait for the export; progress notifications report the wait every ten seconds. A session does not rebuild a failed export: requests that need it fail with `retryable: false`, the export's failure, its captured Kuna output and what to change before reopening the target. A setting such as `REA_KUNA_PROJECT_TIMEOUT_MS` or `REA_KUNA_JOBS` is read when REA starts, so changing it also needs a restart. Pseudocode keeps using single-function runs.
- A single-function answer says so in its limitations. It lacks the export's whole-program prototypes, so a later request can return different text for the same function.
Pseudocode, assembly, `analyze_function` and data-name answers record where their text came from in `raw_result.source`: `single-function`, or `project-export` with the cache directory. `procedure_pseudo_code` and `procedure_assembly` answers from the export also name the file, byte span and the file's SHA-256. Combined `analyze_function` answers retain both facet records in `raw_result.source.files`, each with its file, digest and byte span. Both texts come from one export; losing that project retries the complete report. Listings and procedure lookups carry no `raw_result`. Agents can open or search the export's `.c` and `.asm` there directly when that is cheaper than calling a tool per function. Text read that way is target-derived, like any decompiler output, but it bypasses the untrusted-content marker that REA's tool results carry.
Pseudocode, assembly, `analyze_function` and data-name answers record where their text came from in `raw_result.source`: `single-function`, or `project-export` with the cache directory. `procedure_pseudo_code` and `procedure_assembly` answers from the export also name the file, byte span and the file's SHA-256. Combined `analyze_function` answers retain both facet records in `raw_result.source.files`, each with its file, digest and byte span. Both texts come from one export; losing that project retries the complete report. Listings and procedure lookups carry no `raw_result`; omitted [Mach-O data entries](#mach-o-data-entries) are reported in their limitations. Agents can open or search the export's `.c` and `.asm` there directly when that is cheaper than calling a tool per function. Text read that way is target-derived, like any decompiler output, but it bypasses the untrusted-content marker that REA's tool results carry.
The background export belongs to the session. `close_binary`, or the end of a CLI command, cancels it unless another session still waits for it. A one-shot CLI decompile therefore answers from a single-function run; any request that needs the export, such as `procedure_assembly` or `analyze_function`, keeps it.
@@ -72,6 +96,17 @@ Regex procedure and string search uses ECMAScript Unicode matching in a cancella
Single-function subprocess output and each cached source facet share a 16 MiB buffered-text capacity. Cached spans are checked before allocation; larger text remains in the local export and the error identifies the file and range. `analyze_function` with `detail: "complete"` (the CLI default) returns the complete dossier inline, which can still be large; later retained views do not remove that first transfer. The MCP default retains the dossier and returns its summary, so page the facets you need through `inspect_analysis_view`, or prefer the individual text tools for a focused question.
## Mach-O data entries
Kuna's inventory also lists Mach-O exports that are data, such as globals in `__DATA,__data`, constants in `__TEXT,__const` and `__mh_execute_header`. REA classifies each entry with the target's own section table, the same load-command facts `inspect_macho` reports, read from the leading bytes kept while the session's private copy is hashed:
- A section with `S_ATTR_PURE_INSTRUCTIONS` or `S_ATTR_SOME_INSTRUCTIONS`, or of type `S_SYMBOL_STUBS`, holds code; the entry stays.
- A lazy or non-lazy symbol-pointer section (`__got`, `__la_symbol_ptr`) holds import slots; the entry stays.
- Any other section, or the header and load commands before the first section, holds data. Listings and searches omit the entry and say how many they omitted in each region. Pseudocode, assembly, `analyze_function` and `procedure_address` refuse it by name or address, and call queries refuse its name, naming its region; call queries by exact address stay validated by Kuna's graph. `address_name` and `xrefs` still accept it.
- An address outside every section and the header stays, with a limitation saying REA cannot classify it. If the section table cannot be read, every entry stays and listings say why. Answers about one such entry, including pseudocode, assembly, `analyze_function`, `procedure_address`, call queries by name and `address_name`, carry the same limitation; code and import slots carry none.
A section's attributes are the format's declaration, not proof of what runs there, and a kept entry is not proven to be a function. ELF inventories are kept as Kuna reports them. On a Mach-O target, the first listing or lookup establishes the private copy that Kuna commands also use, once per session, even when a warm export answers it.
## Strings
`list_strings` and `search_strings` share one lazy `kuna strings --json --encoding all --min-length 5 --termination any --no-xrefs` scan of the digest-checked private input. Skipping references avoids program lifting, a call graph and a C export. REA validates the input, scan policy, every row and the total before caching any facts. Encoding, character count, byte length and terminator facts must agree; supplementary UTF-8 characters count as Unicode scalars. Malformed output is an error, not an empty search. REA parses the JSON as Kuna writes it instead of buffering it, so the scan has no fixed size: Kuna writes about 290 bytes per row, 31 MB for a 20 MB Ghostty binary and 179 MB (611k rows) for an arm64 Firefox `XUL`. Rows are validated one at a time, and the first contradiction stops Kuna. Text REA does not keep is dropped unassembled; kept text and retained rows are admitted against the REA process's live heap headroom as they arrive; a refusal is a `memory` resource constraint with `retryable: false`, and closing other targets or starting REA with a larger Node.js heap makes room. String scans use the whole-program `REA_KUNA_PROJECT_TIMEOUT_MS` deadline; `REA_KUNA_TIMEOUT_MS` applies to single-function decompilation. The parsed inventory belongs to one selected target/executable profile and expires on session close; Kuna diagnostics beyond their 8 MiB buffer are retained as a failure until close, while other failures are not cached.
@@ -125,7 +160,19 @@ REA writes Kuna's exports only inside `REA_KUNA_CACHE_DIR`:
- `trash/` holds retired builds and evicted exports awaiting removal.
- `home/` is Kuna's HOME.
Builds publish by one rename, so a reader sees a complete export or none. If later staging retirement fails, the typed cleanup error retains the completed export's key, directory and manifest in `published_export`. The session adopts that verified project; later queries reuse its immutable files while `project_export.cleanup_incomplete` reports the retained owner's live state. `close_binary` retries only the remaining cleanup, and a successful retry resolves prior cleanup errors. A failed export followed by failed staging cleanup retains both failures; it does not invent a publication receipt. A session keeps the export it uses from eviction only within its own REA process. If another REA process sharing the cache evicts it, the session rebuilds it; meanwhile pseudocode falls back to single-function runs. A damaged or edited export is removed and rebuilt. Staging moves to `trash/` for removal only after REA proves its Kuna process stopped. Staging of a process it cannot prove stopped is never swept; REA keeps it owned, retries the stop on `close_binary`, and reports it until the stop is proven. Failed file removal also stays bound to its export and is retried by `close_binary`; closing does not report success while that cleanup remains. `rea doctor --provider kuna` reports the cache directory without creating it.
The cache key includes the selected launcher and resolved executable paths, plus
the target's native basename: that name determines the export filenames and
appears in generated source. REA accepts one current manifest layout. Older
exports are rebuilt on demand and retired through normal owned-cache cleanup;
no manual deletion or migration is needed. Manifests must name the producer's
exact flat file layout, with matching basenames and producing inputs. Replaced
entry, project or manifest symlinks are refused before their targets are read;
an unverified `projects/` parent grants no removal authority. Warm reuse checks
the manifest's file identity as well as the export files. These refusal and
migration cases are verified with controlled Kuna-format exports; they do not
establish behavior of an independently installed Kuna release.
Builds publish by one rename, so a reader sees a complete export or none. If later staging retirement fails, the typed cleanup error retains the completed export's key, directory and manifest in `published_export`. The session adopts that verified project; later queries reuse its immutable files while `project_export.cleanup_incomplete` reports the retained owner's live state. `close_binary` retries only the remaining cleanup, and a successful retry resolves prior cleanup errors. A failed export followed by failed staging cleanup retains both failures; it does not invent a publication receipt. A session keeps the export it uses from eviction only within its own REA process. If another REA process sharing the cache evicts it, the session rebuilds it; meanwhile pseudocode falls back to single-function runs. A damaged or edited export is removed and rebuilt. REA retains the file state verified with each digest, checks it before warm-cache reuse, and checks open descriptors before and after reading text. Assembly indexes are checked against the same state; a changed file makes the whole report retry rather than mixing facets from different exports. Staging moves to `trash/` for removal only after REA proves its Kuna process stopped. Staging of a process it cannot prove stopped is never swept; REA keeps it owned, retries the stop on `close_binary`, and reports it until the stop is proven; that MCP error names the same-connection retry as `recovery.kind=retry_cleanup` ([retained cleanup retry](mcp-contracts.md#retained-cleanup-retry)). Failed file removal also stays bound to its export and is retried by `close_binary`; closing does not report success while that cleanup remains. `rea doctor --provider kuna` reports the cache directory without creating it.
and class-null declarations are supported. `i8` and `u8` values are decimal
strings to avoid JSON integer precision loss. Floating-point negative zero,
NaN and infinities are the strings `-0`, `NaN`, `Infinity` or `-Infinity`;
`raw_hex` preserves their original bits. UTF-16 code units, including embedded
NULs and unpaired surrogates, are preserved.
`constant: null` means neither a Constant declaration nor `HasDefault` was
observed. Historical Evidence can omit the field entirely; omission does not
prove that the artifact has no default. Missing or duplicate declarations,
inconsistent flags, malformed values and unsupported element types return
`state: "unavailable"`, a limitation and partial coverage while retaining other
member facts. This decoding shares the existing metadata representation budget
and only decodes blobs for the selected fields.
Retained `managed-members` field pages can select `fields: ["constant"]` while
keeping original field identity and source offsets. These values are metadata
declarations, including enum members' encoded underlying constants, not observed
runtime state or inferred enum-name mappings. Parameter/property defaults are
not exposed by this field-focused reader. Member comparison still covers field
names, flags and signatures; it does not compare constant defaults.
The encoding follows [ECMA-335, Partition II, sections 22.9 and 23.1.16](https://ecma-international.org/publications-and-standards/standards/ecma-335/).
### Metadata overview and retained rows
The existing CLI and omitted/complete MCP detail return complete metadata
@@ -120,6 +152,50 @@ literal pool's own coverage. Older Evidence that never examined exported types
or user strings reports a null count and refuses those pages rather than
returning an empty one.
To find specific retained members without returning their IL anchors, select
exact names and defining types, then project columns:
| `compare_application_versions`, `compare_javascript_export_shapes` | `left`, `right`; version comparison also accepts both native-observation arrays |
Keep the connection open, let the reported condition settle or resolve it, call
`close_binary` without `snapshot_path`, and check the result. The retry does not
rerun the failed analysis and can report `cleanup_incomplete` again.
`retryable` still describes the failed operation and stays `false`. A failed
`snapshot_path` save returns before retirement, keeps the target open and has
no `retry_cleanup` hint. `open_binary`, analysis tools and one-shot CLI commands
do not report this hint either, even when their cleanup remains pending.
## Initial target at startup
MCP normally starts without a target; use `open_binary` to select one during the
session. To open a target before serving MCP, set `HOPPER_TARGET_PATH` to the
selected path in REA's launch environment. `HOPPER_TARGET_KIND` accepts
`executable` (the default) or `database`. Despite the names, these are initial
session settings; the configured provider still determines how the target is
opened. If the initial open fails, REA reports the failure and exits with code 1
without serving MCP. Shutdown during startup cancels and cleans up the open as
described below.
The source-checkout helper `node scripts/print-mcp-config.mjs /path/to/target`
prints a registration containing these settings; `--kind database` selects the
database kind. The helper is not part of the installed npm package. `rea setup`
does not copy these one-shot target settings into generated registrations.
## Transports
The eager stdio runtime owns SIGINT, SIGTERM and stdin EOF before opening an
@@ -775,7 +1143,12 @@ the bound address is local; names are added only after classifying the actual IP
On such a bind, requests whose `Host` header is outside the SDK loopback allowlist plus
the configured name and the bound address are refused with `403`, and a browser
`Origin` outside that allowlist is refused the same way (a missing `Origin`
passes); a non-loopback bind relaxes both checks and logs the relaxation. A reply sent before a request body was read (a refused `Host` or `Origin`, an unknown route, the startup `503`) closes the connection after the reply, so an unread upload cannot keep a socket open. Request targets are taken verbatim as origin-form paths (`//host/path` is a path, not another origin); an absolute-form target, a target containing a raw fragment delimiter (`#`), or one whose path URL parsing would rewrite (dot segments such as `/public/../mcp`), is refused with `400`, and a method Fetch does not support with `405`. Percent-encoded `%23` remains valid query data. Only an `initialize` object can replace an open session. A sessionless array containing initialize, including a one-element array, is refused with HTTP `400` before session creation or handover. Scoped
passes). A non-loopback bind requires `MCP_ALLOWED_HOSTS`, or
`--allowed-hosts` for `rea mcp --http`, listing every client host name or IP
address. Configuration validates and canonicalizes complete tokens once;
URLs, ports and empty entries are refused. Both Host and Origin checks remain
active on that declared list. The announced URL uses the bound hostname if
admitted, otherwise the first declared hostname, with the actual bound port. A reply sent before a request body was read (a refused `Host` or `Origin`, an unknown route, the startup `503`) closes the connection after the reply, so an unread upload cannot keep a socket open. Request targets are taken verbatim as origin-form paths (`//host/path` is a path, not another origin); an absolute-form target, a target containing a raw fragment delimiter (`#`), or one whose path URL parsing would rewrite (dot segments such as `/public/../mcp`), is refused with `400`, and a method Fetch does not support with `405`. Percent-encoded `%23` remains valid query data. Only an `initialize` object can replace an open session. A sessionless array containing initialize, including a one-element array, is refused with HTTP `400` before session creation or handover. Scoped
IPv6 literals (`fe80::1%eth0`) are refused before binding, because no URL can
name them. Listener errors reach the runtime's connection-loss handler only
after a successful bind; a bind failure prints the listen-failure message alone. Request bodies are bounded by the SDK's
@@ -926,6 +1299,15 @@ mapping. A successful start-address mapping does not establish that an entire
range is contiguous in the original file, or that its bytes equal analysis
memory after loader fixups. Uninitialized, missing, and ambiguous mappings fail
explicitly; preserve the reported failure rather than guessing an offset.
For Ghidra, such a well-formed address returns `invalid_request` whose message
states the bridge's reason and whose remediation points to
`inspect_native_load_image`. It carries no `inspect_input_contract` recovery.
`details.issues[0].observed` keeps the caller's exact `address` and a
`file_mapping` outcome: `unmapped` (outside every memory block),
`uninitialized`, `absent` (no authoritative target-file source), `ambiguous`,
or `unrepresentable` (beyond the exact JSON integer range). Initialization, not
the loaded flag, decides whether source bytes can exist, so an initialized
file-backed overlay still maps.
For constructor-populated pointer or resource slots, a static zero is not proof
of a later null value. Inspect the slot's `xrefs` and the writer's assembly or
@@ -958,6 +1340,12 @@ A successful MCP initialize handshake establishes the REA connection.
that Ghidra has finished importing the target. The first Ghidra-backed query,
such as `binary_overview`, starts the engine and waits for import, default
auto-analysis, bridge connection, and health readiness before returning analysis.
With `REA_GHIDRA_PROJECT_PATH`, shared CLI/MCP configuration instead selects an
| Firmware extraction and analysis | `firmware,session,direct` | extract firmware regions, open the extracted image, inspect it with the proxied disassembly tools |
| Web investigation | `browser,pcap,session,javascript-runtime` | page, bundle and session inspection, packet captures, JS runtime observation, evidence workflow |
| Android application analysis | `android,adb,apktool,session` | class and method inspection, device bridge, resource decoding; add `frida` for dynamic instrumentation, `jeb` for JEB projects, `flutter` for Flutter builds |
| Native macOS or iOS binary | `native,artifact,session` | Mach-O, signature and property-list inspection plus container extraction |
| Managed .NET assembly | `managed,session` | managed member comparison and reconstruction against the evidence workflow |
| Electron or browser-hosted JavaScript application | `browser,electron,application,javascript-runtime,session` | runtime targets, live page inspection, bundle and capture tools, feature tracing, JavaScript semantics and reconstruction obligations |
| Evidence bookkeeping only | `session` | open and close binaries, bundles, comparisons, unknown tracking |
Scopes are additive and start narrow: begin with the families the workflow
names, then add one when a call fails with the SDK's `Tool ... disabled`
error. The error names the disabled tool; `tools.families` in the product
catalog maps it back to the family that would re-enable it, and the server
restarts with the wider scope through the same setup flag.
### Output schemas
Self-contained output schemas advertise a content-bound `$id`, including their
@@ -1315,21 +1816,39 @@ identity. Explicit schema IDs and relative external reference bases are
preserved. This keeps every tool and validation rule in discovery and does not
change the SDK's catalog invalidation or availability checks.
The server identity's tool-schema digest includes both the canonical output
contracts and the selected output-schema profile. `full` and `off` therefore
have distinct identities even though `off` keeps the same canonical output
validation on the server.
The server identity's tool-schema digest includes the selected output-schema
profile and the advertised form that profile produces. `full`, `compact`, and
`off` therefore have distinct identities. The `compact` digest covers the
bounded schemas clients receive; `off` keeps the canonical output contracts in
the digest while advertising nothing.
Some stdio hosts reject a complete `tools/list` frame before they load any
tools. The opt-in `REA_MCP_OUTPUT_SCHEMA_PROFILE=off` setting omits `outputSchema`
from the advertised catalog for hosts with a small frame limit (the measured
pi-web-ui failure and catalog breakdown are in [#2512](https://github.com/morluto/rea/issues/2512)).
The default `full` profile is unchanged. With `off`, REA validates successful
structured results against the same named canonical tool contract before
delivery; it still returns complete `structuredContent` and text results.
Only client-side schema discovery and validation are omitted, so hosts using
this profile should rely on REA's validation and retain their usual response
size limits.
tools. The opt-in `REA_MCP_OUTPUT_SCHEMA_PROFILE=compact` setting bounds every
advertised output schema to 4,096 bytes; `off` omits `outputSchema` entirely
(the measured pi-web-ui failure and catalog breakdown are in
[#2512](https://github.com/morluto/rea/issues/2512)). The default `full` profile
is unchanged. Both settings keep the same named canonical tool contract for
server-side result validation, and both still return complete
`structuredContent` and text results.
`compact` keeps the complete validation structure when it fits the budget.
Otherwise it advertises the root property shape, first-level reference-resolved
types, first-level guidance, and the root's `required`/`additionalProperties`
constraints, with nested validation detail elided. Local references are inlined
within the budget, so the advertised form is self-contained and omits the
canonical contract's content-bound `$id`. The projection only widens: a valid
canonical result still satisfies the advertised schema, but a client validating
against it can accept a result the canonical contract rejects. REA still
validates every result before delivery. On catalogs large enough that full
input schemas alone approach the frame limit, keep the input profile at its
`compact` default as well; with full input schemas the advertisement is
dominated by the input side.
The full input profile has no one-megabyte catalog-frame guarantee, including
when output schemas are omitted.
Only client-side schema discovery and validation are reduced or omitted, so
hosts using these profiles should rely on REA's validation and retain their
object ID and member position on the outlet observation. NIB hierarchy uses
recognized child fields, reverse `NSSuperview` links and window content links.
AppKit registry pairs with a resolvable `NSRoot` owner establish top-level
membership; other registry associations alone do not establish visual
parentage and leave hierarchy coverage partial. Known non-view cell ownership
and control-view backreferences remain raw facts. Conflicting explicit parents
remain unknown rather than choosing an archive-order winner. Retained objects
keep their decoded fields; uncertain registry diagnostics retain the source
registry, child and value IDs. Missing, malformed or repeated registry
declarations also qualify hierarchy coverage.
`NSString` objects with the [recognized `NS.bytes` representation](https://github.com/swiftlang/swift-corelibs-foundation/blob/main/Sources/Foundation/NSString.swift#L201-L215) are decoded
as UTF-8 without substituting malformed bytes. A failed conversion retains
the source string object ID and `nibValues.NS.bytes` base64 data; dependent
@@ -727,6 +858,11 @@ complete observation.
files and universal slices before metadata decoding. CPU subtypes and
capability bits remain source facts; an `arm64` request may select an
`arm64e` image under the existing slice preference policy.
Objective-C class, protocol and category lists require eight-byte entries;
Swift type and conformance sections require four-byte entries. Misaligned
section lengths and unsupported non-PowerPC big-endian CPU declarations return
`unsupported_target` with the selected path and precise reason. Recovery
selects a well-formed or supported artifact and reopens that target.
The byte reader holds the whole selected file in memory, so it admits files up
to 64 MiB / 67108864 bytes by default. Select a larger budget with
`max_input_bytes` (CLI `--max-input-bytes`), at most 1 GiB / 1073741824
@@ -821,7 +957,11 @@ complete observation.
and `__objc_catlist` categories are also decoded. Categories record the
extended class, local or external, plus
their methods, protocols and properties; category methods appear as
implementations with a `category`. Swift field-offset globals that are only
implementations with a `category`. A category whose class reference is null,
or is bound to a symbol other than `_OBJC_CLASS_$_`, such as the `_$s…CN`
metadata symbol of another module's Swift class, keeps a null class, and
`objc_properties_categories` coverage names that reference. Swift
field-offset globals that are only
initialized at runtime stay unresolved. Generic, resilient, async and
coroutine tables, and inherited overrides, have explicit unsupported or
partial coverage. Other providers retain the
@@ -869,7 +1009,9 @@ with a path-specific gap and incomplete capture coverage. Null alone does not
establish that an accessibility value was successfully read or absent.
An initial PID or window mismatch is `invalid_request` on `pid` or
`window_id` and leaves provider availability intact. An unavailable executable
`window_id` and leaves provider availability intact. A selected window that
exists off screen is not a mismatch: selection succeeds and the permission and
capture checks report their own outcome. An unavailable executable
URL or unresolved executable path leaves identity unknown and reports
`capability_unavailable`. Identity or window loss
after the initial capture remains an observation failure with collected facts.
@@ -916,15 +1058,22 @@ missing LLDB. `duration_ms` controls the later target observation window.
rea observe-native-calls /Applications/Example.app '{"breakpoints":[{"kind":"objc-method","class_name":"NSURLSession","selector":"dataTaskWithRequest:completionHandler:"},{"kind":"function","name":"open","module":"libsystem_kernel.dylib"}],"duration_ms":5000,"backtrace_frames":4}' --json
```
Function names select LLDB runtime symbols. On Darwin, a C function reported as
`_main` by `inspect_macho` or `nm` can resolve as `main` in LLDB. REA preserves the
exact selected name; it does not remove underscores or translate static names.
Use `{"kind":"function","name":"main","module":"Example"}` to limit that
runtime symbol to the executable image named `Example`, then inspect the reported
locations and image paths. Omitting `module` deliberately matches every loaded
image, including system libraries. A match outside the executable is valid for
that request and does not by itself make coverage partial. This C naming example
does not establish a translation rule for Swift, C++ or other mangled symbols.
Function names select LLDB runtime symbols. On Darwin, the Mach-O symbol table
spells language-level symbols with one leading underscore, which LLDB and `dlsym`
omit: `_main` is `main`, C++ `__ZN3rea3addEii` is `_ZN3rea3addEii`, and Swift
`_$s…` is `$s…`. Names without that underscore, such as `-[Class selector]`, are
unchanged. `inspect_macho` keeps each export and import `name` exactly as the
symbol table spells it and derives its `runtime_name`; pass `runtime_name` as the
breakpoint `name`. `observe_native_calls` preserves the exact selected name; it
does not remove underscores or translate static names, so `_main` selects a
different runtime symbol. Use `{"kind":"function","name":"main","module":"Example"}`
to limit that runtime symbol to the executable image named `Example`, then inspect
the reported locations and image paths. Omitting `module` deliberately matches
every loaded image, including system libraries. A match outside the executable is
valid for that request and does not by itself make coverage partial. Event and
location `symbol` values are LLDB display names, which are demangled for C++ and
Swift; join them to `inspect_macho` through `breakpoint_index` and the request
`name`, or through `file_address` and the export `address`.
Each event is an observation: thread, image, symbol, load and file address,
the raw integer argument registers, optional caller frames and, for an
@@ -995,7 +1144,7 @@ is present. A denied trace reports the `debugger-attach-denied` reason.
## Provider and verification boundaries
Install a Ghidra 12.1.x release and the 64-bit full JDK it declares, then configure REA to
Install a Ghidra 12.x (>=12.1) release and the 64-bit full JDK it declares, then configure REA to
use them. Ghidra analysis supports Linux x64/arm64 and macOS x64/arm64; each installation requires
the matching native decompiler. Experimental Windows x64 P0 admits native
x86, x86-64, ARMNT (Thumb-2) and ARM64 PE applications and DLLs, plus explicitly selected
@@ -1011,11 +1160,55 @@ unpaired Unicode surrogate; a rejection identifies the field and UTF-16 index
and leaves every annotation unchanged. CRLF, supplementary Unicode characters,
and combining characters are preserved. Windows P0 remains read-only. Ghidra has no
GUI authority, and REA never falls back automatically to Hopper.
For a before-and-after `compare_functions` in MCP, call `analyze_function`
before and after the edit while the session remains open; the annotation Evidence
itself is not an `analyze_function` input. If comparing after close, export the
needed Evidence before closing and import it into the later session. A one-shot
CLI annotation returns its edited dossier before its temporary session is
discarded; a later CLI invocation with the same fresh-import settings starts from
the original executable. Comparison preserves correspondence across renames
when authenticated artifact identity, architecture and provider-owned mapping
and slice commitments establish the same canonical entry. The result reports
`matched`, `method: "same-entry"` and `entry_address`; name and exact-text changes
remain separate dimensions. Equal numeric addresses without applicable coordinate
commitments do not establish same-entry correspondence. Inspect the retained
source Evidence and coverage when correspondence remains ambiguous.
To aggregate already-retained comparison results, pass their exact IDs to
and decoding examined the object sources; hierarchy-only and connection-only
unknowns do not make that facet partial. Asset-catalog resource-key joins use
this source coverage together with document and object projection coverage.
Incomplete hierarchy or synthetic hierarchy-node limits alone do not qualify
otherwise complete resource-key joins; omitted archives or resource objects do.
Archive omission reasons are counted independently of the bounded path samples.
A UID outside the exact supported range reports `unsupported_uid_exact_range`,
not budget exhaustion. Mixed source failures report `multiple_archive_decode_causes`;
limitations retain omission counts by cause even when individual paths exceed
the reporting allowance.
UI dispatch joins use the authored receiver's class independently of its
presentation kind (controller, view, control or another object). An unresolved
Loaded 100 of 4014 files, more files were not shown because too many files have changed in this diff.
Show more
Reference in new issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Cursor Agent
