Compare commits

...
Author SHA1 Message Date
morluto 726a10dc6e chore(mcp): include current main checks 2026-11-07 23:12:01 +08:00
morluto ac30fbd658 style(test): format CLI help default annotations (#9280) 2026-11-07 23:02:44 +08:00
Nianze Wuandmorluto 5d21bddb18 fix(desktop): read JSON with deeply nested overwritten values (#9277)
* fix(desktop): read JSON with deeply nested overwritten values

* fix(desktop): bound exported JSON construction by view depth

* fix(desktop): preserve deep JSON numeric facts within bounded memory

---------

Co-authored-by: morluto <[email protected]>
2026-11-07 22:57:15 +08:00
morlutoandCursor Agent 2e2f981ccf test: assert absolute-path facts without frozen sentences (#9231)
* test: assert absolute-path facts without frozen sentences

Sibling path tests now require the field path, invalid_format, and regex
expectation, and they record that each negative case ran. NUL rejection
follows the argument that contains the byte instead of searching for the
word NUL.

* test: retain actionable NUL path diagnostics

---------

Co-authored-by: Cursor Agent <[email protected]>
2026-11-07 22:46:17 +08:00
morluto 180f121ba7 fix(mcp): reconcile candidate refresh with current session 2026-11-07 22:46:05 +08:00
Heo Hao f8bb792f15 fix(cli): stop restating schema defaults in help descriptions (#9278)
The help renderer appends "(default: X)" whenever the option field declares a
schema default, so descriptions that also spell the default out print it twice:

  --block-size <value>  Power-of-two block size from 256 to 65536 bytes
                        (default 4096) (default: 4096)

Remove the hand-written default from the eleven options that own a schema
default across inspect-byte-entropy, inspect-devicetree,
inspect-packet-capture, inspect-packet-fields, inspect-signal-decoders and
decode-signal-recording, matching the rule in cliDecimalField's contract that
the domain field owns bounds, defaults and help.

Options without a schema default keep their prose: --match_limit, --line_limit,
--limit and --packet_limit are nullable or optional, so the renderer emits no
annotation and the description is the only place the effective value appears.

Add a boundary test asserting each such option states its default exactly once,
so a restated default cannot come back unnoticed.
2026-11-07 22:38:03 +08:00
morluto b65ff9bb79 test(mcp): bound candidate refresh lifecycle gates 2026-11-01 19:11:33 +08:00
morluto 5628b6ebd2 fix(mcp): clarify provider reselection on reopen 2026-11-01 18:54:40 +08:00
morluto 2f2665b5d2 docs: clarify candidate inspection ownership 2026-11-01 18:52:51 +08:00
morluto 7a76f676ab feat(mcp): refresh provider candidates without switching 2026-11-01 18:52:16 +08:00
morluto 0e424cae8d fix(mcp): report provider candidate observation scope 2026-11-01 18:41:23 +08:00
30 changed files with 1261 additions and 222 deletions

No files matched your search

+2 -2
View File
@@ -310,8 +310,8 @@ entry it does not read. The module reads the central directory itself, as REA
does, and shows `zipfile` only the records it lists, reading each entry in
bounded chunks no further than a byte past its declared size. A JSON file
holding a number JavaScript cannot read exactly, such as an integer above 2^53,
or nested more than 500 levels deep, gets an `error` instead of a view, as in
REA, while text that is not one JSON
or whose retained value nests more than 500 levels deep, gets an `error` instead of a view, as in
REA. Duplicate keys keep their last value before the depth check. Text that is not one JSON
document, such as JSON Lines, stays text. Format detection follows REA's
rules. Files larger than 64 MiB get an `error` instead of a view
and are not read. XML is read with Python's `expat` twice: with namespace
+14 -2
View File
@@ -1012,8 +1012,20 @@ limitations. An unavailable group carries `remediation` when its readiness
owner supplied a repair instruction; `tool_availability` uses that instruction
rather than the failure description.
`binary_session.analysis_provider_candidates` is authoritative for deep-engine
discovery. Each candidate's capability sets, returned with complete detail,
`binary_session.analysis_provider_candidates` reports deep-engine discovery.
`analysis_provider_candidates_observed` is `target_selection` when those facts
were retained from the active target's selection, `status_request` when the
status request sought candidate observations, and `not_applicable` for resource-only
sessions. Selection-time readiness may have changed since the target was opened.
Set `refresh_candidates: true` to inspect candidate readiness and target support
again without replacing the active binding or engine. That does not resolve a
new analysis profile or start a provider client. Later transitions and analysis
calls wait for the inspection to finish, including cleanup. Current-route capabilities
remain those of the bound session. Candidate statuses can still be unknown when
an inspection cannot complete.
`open_binary` with an explicit `provider_id` performs a new selection. Reopening
the same target without `provider_id` reuses its current route. A changed route
can retire the previous engine and its ephemeral edits. Each candidate's capability sets, returned with complete detail,
describe that possible provider;
they do not select it. The top-level `capabilities` describe the current route,
while `analysis_provider_binding` identifies the concrete selected route. These
+5
View File
@@ -277,6 +277,11 @@ const openServer = async (provider, kunaSettings = {}) => {
path: provider.path,
provider_id: provider.id,
});
const session = (await call("binary_session")).value;
assert.equal(
session.analysis_provider_candidates_observed,
"target_selection",
);
} catch (cause) {
// A rejected connection/target still owns its transport and workspace.
await close(cause);
+16 -2
View File
@@ -1304,8 +1304,20 @@ export class BinarySession
options: {
readonly signal?: AbortSignal;
readonly detail?: BinarySessionDetail;
readonly refreshCandidates?: boolean;
} = {},
): Promise<JsonValue> {
return options.refreshCandidates === true
? this.#serialize(() => this.#status(options), { abortPrompts: false })
: this.#status(options);
}
/** Keep an explicit candidate refresh ordered with target transitions. */
#status(options: {
readonly signal?: AbortSignal;
readonly detail?: BinarySessionDetail;
readonly refreshCandidates?: boolean;
}): Promise<JsonValue> {
const resourceRun = this.#activeBundleResources;
const run =
this.#active ??
@@ -1604,9 +1616,11 @@ export class BinarySession
#serialize<T>(
operation: (admittedThrough: number) => Promise<T>,
options: { readonly abortPrompts?: boolean } = {},
): Promise<T> {
for (const controller of this.#promptCompletionControllers)
controller.abort();
if (options.abortPrompts !== false)
for (const controller of this.#promptCompletionControllers)
controller.abort();
this.#transitionPending += 1;
// Later calls wait for this transition and must not be included in its drain.
const admittedThrough = this.#transitionGeneration++;
@@ -167,6 +167,7 @@ it("opens a resource-only bundle without executable identity and keeps snapshot
expect(await session.status()).toMatchObject({
open: true,
session_kind: "bundle_resources",
analysis_provider_candidates_observed: "not_applicable",
resource_root: { canonical_path: opened.value.canonicalPath },
});
const snapshot = session.exportAnalysisSnapshot();
@@ -140,6 +140,7 @@ export interface BinarySessionPort
status(options?: {
readonly signal?: AbortSignal;
readonly detail?: BinarySessionDetail;
readonly refreshCandidates?: boolean;
}): Promise<JsonValue>;
activeTarget(): BinaryTarget | undefined;
/** Read a snapshot with failed descriptors retained for session cleanup retry. */
+15 -1
View File
@@ -47,6 +47,7 @@ interface BinarySessionStatusInput {
readonly cleanupRetry?: AnalysisCleanupRetry;
readonly detail?: BinarySessionDetail;
readonly signal?: AbortSignal;
readonly refreshCandidates?: boolean;
}
/** Project internal provider routing state into the caller-visible session status. */
@@ -66,6 +67,7 @@ export const binarySessionStatus = async ({
cleanupRetry,
detail = "complete",
signal,
refreshCandidates = false,
}: BinarySessionStatusInput): Promise<JsonValue> => {
const routeCapabilities = (
bundleResourceRoot === undefined
@@ -112,7 +114,13 @@ export const binarySessionStatus = async ({
const analysisProviderCandidates =
bundleResourceRoot !== undefined
? []
: (await router.candidateStatuses(route, signal)).map((candidate) =>
: (
await router.candidateStatuses(route, {
...(signal === undefined ? {} : { signal }),
...(target === undefined ? {} : { target }),
refresh: refreshCandidates,
})
).map((candidate) =>
detail === "summary"
? {
provider: providerSummary(candidate.provider),
@@ -146,6 +154,12 @@ export const binarySessionStatus = async ({
capabilities,
analysis_provider_binding: analysisProviderBinding,
analysis_provider_candidates: analysisProviderCandidates,
analysis_provider_candidates_observed:
bundleResourceRoot !== undefined
? "not_applicable"
: route.selection === undefined || refreshCandidates
? "status_request"
: "target_selection",
analysis_run:
runId === undefined
? null
@@ -129,11 +129,26 @@ export class SessionProviderRouter {
);
}
/** Candidate status authoritative for the active or target-free route. */
/** Preserve selection observations unless a caller requests a fresh inspection. */
async candidateStatuses(
route: SessionProviderRoute,
signal?: AbortSignal,
options: {
readonly signal?: AbortSignal;
readonly target?: BinaryTarget;
readonly refresh?: boolean;
} = {},
): Promise<readonly AnalysisProviderCandidateStatus[]> {
const { signal, target, refresh = false } = options;
if (refresh && route.selection !== undefined) {
const current = await this.registry.candidates(
target,
signal === undefined ? {} : { signal },
);
return current.map((candidate) => ({
...candidate,
selected: candidate.provider.id === route.binding?.identity.id,
}));
}
return (
route.selection?.candidates ??
this.registry.candidates(
+6 -8
View File
@@ -266,12 +266,10 @@ export const registerBinaryDiagnosticsCommands = (
// "0x10" or "1e2".
blockSize: cliDecimalField(
inspectByteEntropyInputSchema.shape.block_size,
).describe(
"Power-of-two block size from 256 to 65536 bytes (default 4096)",
),
).describe("Power-of-two block size from 256 to 65536 bytes"),
offset: cliDecimalField(
inspectByteEntropyInputSchema.shape.offset,
).describe("Window start as a decimal file byte offset (default 0)"),
).describe("Window start as a decimal file byte offset"),
length: cliDecimalField(
inspectByteEntropyInputSchema.shape.length,
).describe("Window length in bytes; omit to extend through EOF"),
@@ -429,16 +427,16 @@ export const registerBinaryDiagnosticsCommands = (
options: z.object({
maxFileBytes: cliDecimalField(
inspectDevicetreeInputSchema.shape.max_file_bytes,
).describe("Maximum FDT file size in bytes (default 4194304)"),
).describe("Maximum FDT file size in bytes"),
maxNodes: cliDecimalField(
inspectDevicetreeInputSchema.shape.max_nodes,
).describe("Maximum number of device-tree nodes (default 8192)"),
).describe("Maximum number of device-tree nodes"),
maxProperties: cliDecimalField(
inspectDevicetreeInputSchema.shape.max_properties,
).describe("Maximum number of device-tree properties (default 32768)"),
).describe("Maximum number of device-tree properties"),
maxDepth: cliDecimalField(
inspectDevicetreeInputSchema.shape.max_depth,
).describe("Maximum device-tree nesting depth (default 64)"),
).describe("Maximum device-tree nesting depth"),
...analysisEvidenceDeliveryOptions,
}),
run: ({ args, options }) =>
+1 -1
View File
@@ -123,7 +123,7 @@ describe("CLI JSON path normalization", () => {
expect.objectContaining({
path: ["executable_path"],
code: "invalid_format",
message: expect.stringContaining("absolute local filesystem path"),
format: "regex",
}),
]);
});
+6
View File
@@ -27,6 +27,12 @@ export const binarySessionInputSchema = z.strictObject({
.describe(
"Start with summary (default): it keeps open state, the binding, current-route capabilities (including selected Ghidra checkpoint mode), each provider candidate's availability and target-support status, code and reason, and every tool's availability, reason and remediation; omitted_fields lists what it leaves out. Request complete only when you need unavailable-candidate remediation, provider diagnostics, candidate capabilities or static tool surface, effect and annotation flags.",
),
refresh_candidates: z
.boolean()
.default(false)
.describe(
"Inspect current provider candidates for the active target without replacing or restarting the selected provider. Later transitions and analysis calls wait for inspection finalization. Omitted or false reports the retained selection observations.",
),
expected_package_version: z.string().min(1).optional(),
expected_catalog_digest: z
.string()
+1 -1
View File
@@ -88,7 +88,7 @@ export const SESSION_TOOL_CONTRACTS = [
),
session(
"binary_session",
'Report the active binary target or resource-only app root, provider and capability availability when applicable, client-feature, analysis, and server-identity status without starting analysis. Start with the default summary: after selecting Ghidra, capabilities[].shared_limitations already reports enabled checkpoint or ephemeral mode. Request detail: "complete" only for provider diagnostics, candidate capabilities or static tool flags; omitted_fields lists what summary leaves out.',
'Report the active binary target or resource-only app root, provider and capability availability when applicable, client-feature, analysis, and server-identity status without starting analysis. analysis_provider_candidates_observed distinguishes this status request from retained target-selection observations; selection-time readiness may have changed. Set refresh_candidates true to inspect current candidates without changing the active binding or engine. open_binary with an explicit provider_id reselects and may retire the previous engine and its ephemeral edits; reopening the same target without provider_id reuses its current route. Start with the default summary: after selecting Ghidra, capabilities[].shared_limitations already reports enabled checkpoint or ephemeral mode. Request detail: "complete" only for provider diagnostics, candidate capabilities or static tool flags; omitted_fields lists what summary leaves out.',
binarySessionInputSchema,
),
session(
@@ -556,6 +556,11 @@ export const sessionProviderFacts = z
analysis_run: analysisRun,
analysis_activity: analysisActivity,
provider_operation_health: providerOperationHealth,
analysis_provider_candidates_observed: z
.enum(["status_request", "target_selection", "not_applicable"])
.describe(
"When provider candidate observations were collected: this status request, the retained target selection, or not applicable to a resource-only session.",
),
analysis_provider_binding: z
.object({
provider: providerIdentity,
+1 -1
View File
@@ -51,7 +51,7 @@ const captureInputShape = z.strictObject({
.int()
.nonnegative()
.default(0)
.describe("Skip this many matching packets after filtering; default 0"),
.describe("Skip this many matching packets after filtering"),
match_limit: z
.number()
.int()
+2 -4
View File
@@ -5,9 +5,7 @@ export const inspectPacketFieldsInputSchema = z.strictObject({
kind: z
.enum(["protocols", "fields"])
.default("protocols")
.describe(
"Return protocol registrations or field definitions; defaults to protocols",
),
.describe("Return protocol registrations or field definitions"),
protocol: z
.string()
.min(1)
@@ -30,7 +28,7 @@ export const inspectPacketFieldsInputSchema = z.strictObject({
.nonnegative()
.max(Number.MAX_SAFE_INTEGER)
.default(0)
.describe("Number of matching records to skip; defaults to 0"),
.describe("Number of matching records to skip"),
limit: z
.number()
.int()
+3 -142
View File
@@ -1,3 +1,4 @@
import { PYTHON_JSON_READER } from "./appExportPythonJson.js";
import { PYTHON_XML_READER } from "./appExportPythonXml.js";
import { PYTHON_ZIP_HELPERS } from "./appExportPythonZip.js";
@@ -12,6 +13,7 @@ export const PYTHON_READY_CHECK = `
export const PYTHON_IMPORTS = `from __future__ import annotations
from array import array
import contextlib
import copy
import hashlib
@@ -57,11 +59,6 @@ MAX_PARSED_BYTES = 4 * 1024 * 1024
PARSED_TOO_LARGE = "larger than the 4 MiB limit for JSON and key file views"
MAX_JSON_VIEW_DEPTH = 500
JSON_TOO_DEEP = f"nested deeper than the {MAX_JSON_VIEW_DEPTH} levels a JSON view keeps"
#: A JSON token after any whitespace, by group: an opening, a closing, a comma, a colon, a string's quote, or a number
#: or literal. Strings are read by json's scanner, since a pattern repeating per escape holds memory for each.
JSON_TOKEN = re.compile(r'[ \t\n\r]*(?:([\[{])|([\]}])|(,)|(:)|(")|(-?(?:0|[1-9][0-9]*)(?:\.[0-9]+)?(?:[eE][+-]?[0-9]+)?'
r"|true|false|null))")
JSON_SPACE = re.compile(r"[ \t\n\r]*")
def _respelled(text: str, home: Path) -> str:
@@ -254,143 +251,7 @@ def _keyfile(text: str) -> dict:
${PYTHON_XML_READER}
def _not_json(constant: str):
raise ValueError(f"{constant} is not JSON")
class _Inexact(Exception):
"""A JSON number that REA's state view refuses, since reading it would change it."""
def _digits(text: str) -> str:
return re.split(r"[eE]", text.lstrip("-"))[0].replace(".", "").strip("0") or "0"
def _js_string(value: float) -> str:
"""How JavaScript's String() writes a finite double: an integer below 1e21 in full, any other in shortest form."""
return str(int(value)) if value.is_integer() and abs(value) < 1e21 else repr(value)
def _safe_number(text: str) -> bool:
"""lossless-json's isSafeNumber: an integer spelling within 2^53, or a float spelling that keeps its digits."""
if re.fullmatch(r"-?[0-9]+", text):
return len(text.lstrip("-")) <= 16 and abs(int(text)) <= MAX_SAFE_INTEGER
value = float(text)
if not math.isfinite(value):
return False
written = _js_string(value)
return written == text or _digits(written) == _digits(text)
def _exact_number(text: str):
"""A JSON number read only when reading it changes nothing, as REA's state view rules: its spelling and its
written-back form both safe, and not -0."""
value = float(text)
if (not _safe_number(text) or not math.isfinite(value) or not _safe_number(_js_string(value))
or (value == 0 and math.copysign(1, value) < 0)):
raise _Inexact(text)
return int(text) if re.fullmatch(r"-?[0-9]+", text) else value
_INEXACT = object()
_LEVEL_END = object()
def _number_or_mark(spelling: str):
"""A JSON number read exactly, or a mark for one REA's state view refuses."""
try:
return _exact_number(spelling)
except _Inexact:
return _INEXACT
def _holds_inexact(value) -> bool:
"""Whether a parsed JSON value still holds a marked number once duplicate keys kept their last value; one
iterator per nesting level, so no child list is copied."""
levels = [iter((value,))]
while levels:
item = next(levels[-1], _LEVEL_END)
if item is _LEVEL_END:
levels.pop()
elif item is _INEXACT:
return True
elif isinstance(item, dict):
levels.append(iter(item.values()))
elif isinstance(item, list):
levels.append(iter(item))
return False
def _deeper_than(root, limit: int) -> bool:
"""Whether a parsed JSON value nests deeper than a limit, as REA checks it: a value's depth is the number of
iterators open above it, one per nesting level, so no child list is copied."""
levels = [iter((root,))]
while levels:
item = next(levels[-1], _LEVEL_END)
if item is _LEVEL_END:
levels.pop()
elif len(levels) - 1 > limit:
return True
elif isinstance(item, dict):
levels.append(iter(item.values()))
elif isinstance(item, list):
levels.append(iter(item))
return False
def _is_json(text: str) -> bool:
"""Whether text is one JSON value as REA's state view reads it, checked a token at a time without building it or
recursing: it holds a byte per open container and one string at a time, so text too deep for Python's parser or
too large to parse into a view is still told apart from other text."""
containers, expect, empty, at = bytearray(), "value", False, 0
while True:
token = JSON_TOKEN.match(text, at)
if token is None:
break
kind, at = token.lastindex, token.end()
if kind == 5:
try:
at = json.decoder.scanstring(text, at)[1]
except ValueError:
return False
if kind == 1 and expect == "value":
containers.append(ord(text[at - 1]))
expect, empty = ("value" if text[at - 1] == "[" else "key"), True
# A closing bracket's code is its opening's plus two.
elif kind == 2 and (expect == "after" or empty) and containers and containers[-1] == ord(text[at - 1]) - 2:
containers.pop()
expect, empty = "after", False
elif kind == 3 and expect == "after" and containers:
expect = "value" if containers[-1] == ord("[") else "key"
elif kind == 4 and expect == "colon":
expect = "value"
elif kind >= 5 and (expect == "value" or (kind == 5 and expect == "key")):
expect, empty = ("colon" if expect == "key" else "after"), False
else:
return False
return expect == "after" and not containers and JSON_SPACE.fullmatch(text, at) is not None
def _json_view(text: str, size: int):
"""A JSON text's format and view, or why it has none, as REA reads it; None when it is not JSON. Text over the
parsed-view limit, or nested past Python's recursion limit, is told apart without building its value."""
if size > MAX_PARSED_BYTES:
return {"format": "json", "view": None, "error": PARSED_TOO_LARGE} if _is_json(text) else None
# The whole text must parse, and duplicate keys settle, before an inexact number makes it JSON with an error.
try:
view = json.loads(text, parse_constant=_not_json, parse_float=_number_or_mark, parse_int=_number_or_mark)
except RecursionError:
# Nesting too deep for Python to parse is far deeper than REA's view keeps, once the rest of it is JSON too.
return {"format": "json", "view": None, "error": JSON_TOO_DEEP} if _is_json(text) else None
except ValueError:
return None
if _holds_inexact(view):
return {"format": "json", "view": None,
"error": "a number in it cannot be read exactly, such as an integer above 2^53"}
if _deeper_than(view, MAX_JSON_VIEW_DEPTH):
return {"format": "json", "view": None, "error": JSON_TOO_DEEP}
return {"format": "json", "view": view, "truncated": False}
${PYTHON_JSON_READER}
def _text_view(name: str, text: str, size: int) -> dict:
"""The format REA's state view detects for decoded text of size bytes, and its view; a JSON or key file over the
+231
View File
@@ -0,0 +1,231 @@
/** JSON parsing and bounded retained-view checks for exported desktop environments. */
export const PYTHON_JSON_READER = String.raw`
#: A JSON token after any whitespace, by group: an opening, a closing, a comma, a colon, a string's quote, or a number
#: or literal. Strings are read by json's scanner, since a pattern repeating per escape holds memory for each.
JSON_TOKEN = re.compile(r'[ \t\n\r]*(?:([\[{])|([\]}])|(,)|(:)|(")|(-?(?:0|[1-9][0-9]*)(?:\.[0-9]+)?(?:[eE][+-]?[0-9]+)?'
r"|true|false|null))")
JSON_SPACE = re.compile(r"[ \t\n\r]*")
class _Inexact(Exception):
"""A JSON number that REA's state view refuses, since reading it would change it."""
def _digits(text: str) -> str:
return re.split(r"[eE]", text.lstrip("-"))[0].replace(".", "").strip("0") or "0"
def _js_string(value: float) -> str:
"""How JavaScript's String() writes a finite double: an integer below 1e21 in full, any other in shortest form."""
return str(int(value)) if value.is_integer() and abs(value) < 1e21 else repr(value)
def _safe_number(text: str) -> bool:
"""lossless-json's isSafeNumber: an integer spelling within 2^53, or a float spelling that keeps its digits."""
if re.fullmatch(r"-?[0-9]+", text):
return len(text.lstrip("-")) <= 16 and abs(int(text)) <= MAX_SAFE_INTEGER
value = float(text)
if not math.isfinite(value):
return False
written = _js_string(value)
return written == text or _digits(written) == _digits(text)
def _exact_number(text: str):
"""A JSON number read only when reading it changes nothing, as REA's state view rules: its spelling and its
written-back form both safe, and not -0."""
value = float(text)
if (not _safe_number(text) or not math.isfinite(value) or not _safe_number(_js_string(value))
or (value == 0 and math.copysign(1, value) < 0)):
raise _Inexact(text)
return int(text) if re.fullmatch(r"-?[0-9]+", text) else value
_INEXACT = object()
_DEEP = object()
_DEEP_INEXACT = object()
_LEVEL_END = object()
_INVALID_JSON = object()
JSON_ATOMS = {"true": True, "false": False, "null": None}
def _number_or_mark(spelling: str):
"""A JSON number read exactly, or a mark for one REA's state view refuses."""
try:
return _exact_number(spelling)
except _Inexact:
return _INEXACT
def _holds_inexact(value) -> bool:
"""Whether a parsed JSON value still holds a marked number once duplicate keys kept their last value; one
iterator per nesting level, so no child list is copied."""
levels = [iter((value,))]
while levels:
item = next(levels[-1], _LEVEL_END)
if item is _LEVEL_END:
levels.pop()
elif item is _INEXACT or item is _DEEP_INEXACT:
return True
elif isinstance(item, dict):
levels.append(iter(item.values()))
elif isinstance(item, list):
levels.append(iter(item))
return False
def _deeper_than(root, limit: int) -> bool:
"""Whether a parsed JSON value nests deeper than a limit, as REA checks it: a value's depth is the number of
iterators open above it, one per nesting level, so no child list is copied."""
levels = [iter((root,))]
while levels:
item = next(levels[-1], _LEVEL_END)
if item is _LEVEL_END:
levels.pop()
elif item is _DEEP or item is _DEEP_INEXACT:
return True
elif len(levels) - 1 > limit:
return True
elif isinstance(item, dict):
levels.append(iter(item.values()))
elif isinstance(item, list):
levels.append(iter(item))
return False
def _read_json(text: str, build: bool = True):
"""One JSON value, or _INVALID_JSON, read without recursion so duplicate keys settle before depth is checked.
Deep containers are summarized rather than built; exact-number marks still follow last-key-wins semantics.
With build=False, only syntax is checked: one byte per open container and one decoded string at a time."""
containers, expect, empty, at = bytearray(), "value", False, 0
values, keys, root = [], [], _INVALID_JSON
deep_bad, deep_key_at, bad_keys = bytearray(), array("I"), {}
def keep(value) -> None:
nonlocal root
if not values:
root = value
elif containers[-1] == ord("["):
values[-1].append(value)
else:
values[-1][keys[-1]] = value
def record_deep(bad: bool) -> None:
"""Summarize a finished child; keep only object keys with live inexact values."""
depth = len(deep_bad) - 1
if containers[-1] == ord("["):
deep_bad[-1] |= 2 | bad
return
marked = bad_keys.get(depth)
if bad or marked is not None:
key, _ = json.decoder.scanstring(text, deep_key_at[-1])
if bad:
if marked is None:
marked = key
elif isinstance(marked, str):
if marked != key:
marked = {marked, key}
else:
marked.add(key)
else:
if isinstance(marked, str):
if marked == key:
marked = None
else:
marked.discard(key)
if len(marked) == 1:
marked = next(iter(marked))
if marked is None:
bad_keys.pop(depth, None)
else:
bad_keys[depth] = marked
deep_bad[-1] = 2 | (marked is not None)
else:
deep_bad[-1] |= 2
while True:
token = JSON_TOKEN.match(text, at)
if token is None:
break
kind, at = token.lastindex, token.end()
if kind == 5:
string_at = at
try:
value, at = json.decoder.scanstring(text, at)
except ValueError:
return _INVALID_JSON
if kind == 1 and expect == "value":
if build:
if deep_bad or len(containers) >= MAX_JSON_VIEW_DEPTH:
deep_bad.append(0)
deep_key_at.append(0)
else:
value = [] if text[at - 1] == "[" else {}
keep(value)
values.append(value)
keys.append(None)
containers.append(ord(text[at - 1]))
expect, empty = ("value" if text[at - 1] == "[" else "key"), True
# A closing bracket's code is its opening's plus two.
elif kind == 2 and (expect == "after" or empty) and containers and containers[-1] == ord(text[at - 1]) - 2:
opening = containers.pop()
if build:
if deep_bad:
state = deep_bad.pop()
bad = bool(state & 1)
deep_key_at.pop()
bad_keys.pop(len(deep_bad), None)
if deep_bad:
record_deep(bad)
else:
keep((_DEEP_INEXACT if bad else _DEEP) if state & 2 else ([] if opening == ord("[") else {}))
else:
values.pop()
keys.pop()
expect, empty = "after", False
elif kind == 3 and expect == "after" and containers:
expect = "value" if containers[-1] == ord("[") else "key"
elif kind == 4 and expect == "colon":
expect = "value"
elif kind >= 5 and (expect == "value" or (kind == 5 and expect == "key")):
if build:
if expect == "key":
if deep_bad:
deep_key_at[-1] = string_at
else:
keys[-1] = value
else:
if kind == 6:
value = JSON_ATOMS.get(token[6], _INVALID_JSON)
if value is _INVALID_JSON:
value = _number_or_mark(token[6])
if deep_bad:
record_deep(value is _INEXACT)
else:
keep(value)
expect, empty = ("colon" if expect == "key" else "after"), False
else:
return _INVALID_JSON
if expect != "after" or containers or JSON_SPACE.fullmatch(text, at) is None:
return _INVALID_JSON
return root if build else True
def _json_view(text: str, size: int):
"""A JSON text's format and view, or why it has none, as REA reads it; None when it is not JSON. Text over the
parsed-view limit is told apart without building its value."""
if size > MAX_PARSED_BYTES:
valid = _read_json(text, build=False) is not _INVALID_JSON
return {"format": "json", "view": None, "error": PARSED_TOO_LARGE} if valid else None
# The whole text must parse, and duplicate keys settle, before an inexact number makes it JSON with an error.
view = _read_json(text)
if view is _INVALID_JSON:
return None
if _holds_inexact(view):
return {"format": "json", "view": None,
"error": "a number in it cannot be read exactly, such as an integer above 2^53"}
if _deeper_than(view, MAX_JSON_VIEW_DEPTH):
return {"format": "json", "view": None, "error": JSON_TOO_DEEP}
return {"format": "json", "view": view, "truncated": False}
`;
+20 -1
View File
@@ -21,11 +21,20 @@ const PATHS = [
];
it("matches the host path module for absolute local paths", () => {
const checked: string[] = [];
for (const path of PATHS) {
checked.push(path);
expect(absoluteLocalPathStringSchema.safeParse(path).success).toBe(
isAbsolute(path),
);
}
expect(checked).toEqual(
expect.arrayContaining([
"/absolute/path",
"relative/path",
"file:///absolute/path",
]),
);
expect(absoluteLocalPathStringSchema.safeParse("").success).toBe(false);
expect(
absoluteLocalPathStringSchema.safeParse("/absolute/with\0nul").success,
@@ -40,8 +49,18 @@ it("keeps the Windows absolute-path pattern valid in every client mode", () => {
() => new RegExp(windowsAbsolutePathPattern.source, flags),
flags,
).not.toThrow();
for (const path of PATHS)
const checked: string[] = [];
for (const path of PATHS) {
checked.push(path);
expect(windowsAbsolutePathPattern.test(path), path).toBe(
win32.isAbsolute(path),
);
}
expect(checked).toEqual(
expect.arrayContaining([
"C:\\absolute\\path",
"C:relative/path",
"/absolute/path",
]),
);
});
+1 -3
View File
@@ -24,9 +24,7 @@ export const inspectSignalDecodersInputSchema = z.strictObject({
.nonnegative()
.max(Number.MAX_SAFE_INTEGER)
.default(0)
.describe(
"Number of matching decoders or help lines to skip; defaults to 0",
),
.describe("Number of matching decoders or help lines to skip"),
limit: z
.number()
.int()
+1 -1
View File
@@ -31,7 +31,7 @@ const signalInputShape = z.strictObject({
.int()
.nonnegative()
.default(0)
.describe("Skip this many literal LF-separated output lines; default 0"),
.describe("Skip this many literal LF-separated output lines"),
line_limit: z
.number()
.int()
+1
View File
@@ -103,6 +103,7 @@ export const registerSessionStatusTool = (
status = await session.status({
signal: context.mcpReq.signal,
detail: input.detail,
refreshCandidates: input.refresh_candidates,
});
} catch (cause) {
if (!(cause instanceof AnalysisError)) throw cause;
@@ -255,11 +255,11 @@ cliTest(
code: "invalid_request",
details: {
issues: [
{
expect.objectContaining({
path: ["executable_path"],
reason: "invalid_format",
message: expect.stringContaining("absolute local filesystem path"),
},
expected: "regex",
}),
],
},
});
@@ -4,6 +4,7 @@ import { join } from "node:path";
import { fileURLToPath } from "node:url";
import { promisify } from "node:util";
import { expect, it } from "vitest";
import { z } from "zod";
import { toolContract } from "../../../src/contracts/toolContracts.js";
import { createEvidence, parseEvidence } from "../../../src/domain/evidence.js";
@@ -329,21 +330,84 @@ const pathRequests: readonly [string, Record<string, unknown>][] = [
["start_gdb_session", { inferior_tty: nul }],
];
const nulIssuePaths = (
value: unknown,
path: readonly (string | number)[] = [],
): (string | number)[][] => {
if (typeof value === "string") return value.includes("\0") ? [[...path]] : [];
if (Array.isArray(value))
return value.flatMap((item, index) =>
nulIssuePaths(item, [...path, index]),
);
if (typeof value === "object" && value !== null)
return Object.entries(value).flatMap(([key, item]) =>
nulIssuePaths(item, [...path, key]),
);
return [];
};
it("rejects OS-invalid local inputs through the real advertised MCP contracts", async () => {
const { call } = await connectLocalToolsMcp();
const checked: string[] = [];
for (const [name, arguments_] of pathRequests) {
const response = await call(name, arguments_);
expect(response.isError, name).toBe(true);
expect(JSON.stringify(response.content), name).toContain("NUL");
const paths = nulIssuePaths(arguments_);
expect(paths.length, name).toBeGreaterThan(0);
const error = parseMcpToolError(await call(name, arguments_)).error;
expect(error, name).toMatchObject({
code: "invalid_request",
details: { operation: name },
});
const issues = z
.object({
issues: z.array(
z.object({
path: z.array(z.union([z.string(), z.number()])),
reason: z.string(),
message: z.string(),
expected: z.unknown().optional(),
}),
),
})
.parse(error.details).issues;
expect(issues, name).toEqual(
expect.arrayContaining(
paths.map((path) =>
expect.objectContaining({
path,
reason: "invalid_format",
expected: "regex",
message: expect.stringMatching(/\bNUL\b|null byte|zero byte/i),
}),
),
),
);
checked.push(name);
}
expect(checked).toEqual(
expect.arrayContaining([
"extract_firmware",
"inspect_wasm_artifact",
"demangle_swift",
"capture_electron_scenario",
]),
);
const relative = await call("extract_firmware", {
path: "/tmp/firmware.bin",
output_directory: "relative-output",
});
expect(relative.isError).toBe(true);
expect(JSON.stringify(relative.content)).toContain(
"absolute local filesystem path",
);
expect(parseMcpToolError(relative).error).toMatchObject({
code: "invalid_request",
details: {
operation: "extract_firmware",
issues: [
expect.objectContaining({
path: ["output_directory"],
reason: "invalid_format",
expected: "regex",
}),
],
},
});
}, 60_000);
const ignoredOptions = [
@@ -0,0 +1,87 @@
import { describe, expect, it } from "vitest";
import { createCli } from "../../../src/cli.js";
type CliInstance = ReturnType<typeof createCli>;
const serve = async (
cli: CliInstance,
argv: readonly string[],
): Promise<{ readonly stdout: string; readonly exitCode: number }> => {
let stdout = "";
let exitCode = 0;
await cli.serve([...argv], {
env: {},
exit: (code) => {
exitCode = code;
},
stdout: (text) => {
stdout += text;
},
});
return { stdout, exitCode };
};
/**
* Options whose owning field already declares a schema default. The help
* renderer appends that default, so the description must not restate it.
*/
const singleDefaultOptions: readonly {
readonly command: string;
readonly options: readonly string[];
}[] = [
{ command: "inspect-byte-entropy", options: ["--block-size", "--offset"] },
{
command: "inspect-devicetree",
options: [
"--max-file-bytes",
"--max-nodes",
"--max-properties",
"--max-depth",
],
},
{
command: "decode-signal-recording",
options: ["--line_offset", "--line_limit"],
},
{
command: "inspect-packet-capture",
options: ["--match_offset", "--match_limit"],
},
{
command: "inspect-packet-fields",
options: ["--kind", "--offset"],
},
{ command: "inspect-signal-decoders", options: ["--offset"] },
];
const optionLine = (help: string, option: string): string | undefined => {
const line = help
.split(/\r?\n/u)
.find((candidate) => candidate.trimStart().startsWith(`${option} `));
return line?.trim();
};
describe("CLI help default annotations", () => {
it("states each schema-owned option default exactly once", async () => {
const cli = createCli({});
const repeated: string[] = [];
for (const { command, options } of singleDefaultOptions) {
const { stdout, exitCode } = await serve(cli, [command, "--help"]);
expect(exitCode, command).toBe(0);
for (const option of options) {
const line = optionLine(stdout, option);
if (line === undefined) {
repeated.push(`${command} ${option}: option line missing`);
continue;
}
const statements = line.match(/default/giu)?.length ?? 0;
if (statements !== 1)
repeated.push(
`${command} ${option}: ${statements} default statements`,
);
}
}
expect(repeated).toEqual([]);
});
});
@@ -13,7 +13,7 @@ interface PathInputCase {
readonly field: string;
readonly valid: Readonly<Record<string, unknown>>;
readonly invalid: Readonly<Record<string, unknown>>;
readonly issuePath?: readonly string[];
readonly issuePath: readonly string[];
}
const createMcp = async (
@@ -104,7 +104,13 @@ const inputCases = (
field: string,
valid: Readonly<Record<string, unknown>>,
invalid: Readonly<Record<string, unknown>>,
): PathInputCase => ({ name, field, valid, invalid });
): PathInputCase => ({
name,
field,
valid,
invalid,
issuePath: field.split("."),
});
return [
pathCase("open_binary", "path", { path: absolute }, { path: relative }),
@@ -246,8 +252,7 @@ it("advertises full host-absolute path constraints after SDK schema conversion",
const ajv = new Ajv2020({ strict: false, validateFormats: false });
const absolute = resolve("rea-absolute-input");
const relative = "relative-input";
for (const { name, field, valid, invalid } of [
const schemaCases = [
...inputCases(absolute, relative),
{
name: "inspect_go_binary",
@@ -276,7 +281,9 @@ it("advertises full host-absolute path constraints after SDK schema conversion",
right_bundle_path: absolute,
},
},
]) {
];
const checked: string[] = [];
for (const { name, field, valid, invalid } of schemaCases) {
const schema = listed.tools.find(
(tool) => tool.name === name,
)?.inputSchema;
@@ -288,7 +295,16 @@ it("advertises full host-absolute path constraints after SDK schema conversion",
validate(valid),
`${name}.${field}: ${JSON.stringify(validate.errors)}`,
).toBe(true);
checked.push(`${name}.${field}`);
}
expect(checked).toEqual(
expect.arrayContaining([
"open_binary.path",
"extract_firmware.output_directory",
"inspect_go_binary.path NUL",
"compare_bundles.left_bundle_path empty",
]),
);
expect(firmwareCalls).toEqual([]);
} finally {
await close();
@@ -299,19 +315,39 @@ it("rejects relative paths through in-memory MCP handlers", async () => {
const firmwareCalls: string[] = [];
const { client, close } = await createMcp(firmwareCalls);
try {
for (const { name, field, invalid, issuePath } of inputCases(
resolve("rea-absolute-input"),
"relative-input",
)) {
const cases = inputCases(resolve("rea-absolute-input"), "relative-input");
const checked: string[] = [];
for (const { name, field, invalid, issuePath } of cases) {
const result = await client.callTool({ name, arguments: invalid });
expect(result.isError, `${name}.${field}`).toBe(true);
expect(JSON.stringify(result), `${name}.${field}`).toContain("absolute");
if (issuePath !== undefined)
expect(parseMcpToolError(result).error.details).toMatchObject({
operation: name,
issues: [{ path: issuePath, reason: "invalid_format" }],
});
expect(parseMcpToolError(result).error, `${name}.${field}`).toMatchObject(
{
code: "invalid_request",
details: {
operation: name,
issues: [
expect.objectContaining({
path: issuePath,
reason: "invalid_format",
expected: "regex",
}),
],
},
},
);
checked.push(`${name}.${field}`);
}
expect(checked).toEqual(
expect.arrayContaining([
"open_binary.path",
"open_binary.snapshot_path",
"extract_firmware.path",
"extract_firmware.output_directory",
"inspect_analysis_view.source.path",
"capture_browser_scenario.browser.executable_path",
"capture_electron_scenario.application_root",
]),
);
expect(firmwareCalls).toEqual([]);
} finally {
await close();
@@ -994,6 +994,439 @@ it("preserves a selected Ghidra project's read-only guidance when an edit is req
expect(starts).toEqual([]);
});
it("labels unbound provider observations as selection-time after a repair", async () => {
const directory = await createTestTempDirectory("rea-candidate-observation-");
const target = join(directory, "fixture.hop");
await writeFile(target, "fixture");
let available = false;
const base = candidate("alpha", []);
const session = composeBinarySession(
new AnalysisProviderRegistry([
{
...base,
inspectAvailability: () =>
candidate("alpha", [], available).inspectAvailability(),
},
]),
);
const server = createServer(
{ kind: "session", session },
{ logger: silentLogger },
);
const mcp = new Client({ name: "candidate-scope", version: "1" });
const [clientTransport, serverTransport] =
InMemoryTransport.createLinkedPair();
resources.push(mcp, server, { close: () => session.close() });
await server.connect(serverTransport);
await mcp.connect(clientTransport);
expect(
(await mcp.callTool({ name: "open_binary", arguments: { path: target } }))
.isError,
).toBeFalsy();
available = true;
const stale = toolContract("binary_session").outputSchema.parse(
(await mcp.callTool({ name: "binary_session", arguments: {} }))
.structuredContent,
).result;
expect(stale.analysis_provider_binding).toBeNull();
expect(stale.analysis_provider_candidates_observed).toBe("target_selection");
expect(stale.analysis_provider_candidates[0]?.availability.status).toBe(
"unavailable",
);
const inspected = toolContract("binary_session").outputSchema.parse(
(
await mcp.callTool({
name: "binary_session",
arguments: { refresh_candidates: true },
})
).structuredContent,
).result;
expect(inspected.analysis_provider_binding).toBeNull();
expect(inspected.analysis_provider_candidates_observed).toBe(
"status_request",
);
expect(inspected.analysis_provider_candidates[0]?.availability.status).toBe(
"available",
);
const retained = toolContract("binary_session").outputSchema.parse(
(await mcp.callTool({ name: "binary_session", arguments: {} }))
.structuredContent,
).result;
expect(retained.analysis_provider_candidates[0]?.availability.status).toBe(
"unavailable",
);
expect(
(await mcp.callTool({ name: "open_binary", arguments: { path: target } }))
.isError,
).toBeFalsy();
const reused = toolContract("binary_session").outputSchema.parse(
(await mcp.callTool({ name: "binary_session", arguments: {} }))
.structuredContent,
).result;
expect(reused.analysis_provider_binding).toBeNull();
expect(reused.analysis_provider_candidates[0]?.availability.status).toBe(
"unavailable",
);
expect(
(
await mcp.callTool({
name: "open_binary",
arguments: { path: target, provider_id: "alpha" },
})
).isError,
).toBeFalsy();
const refreshed = toolContract("binary_session").outputSchema.parse(
(await mcp.callTool({ name: "binary_session", arguments: {} }))
.structuredContent,
).result;
expect(refreshed.analysis_provider_candidates_observed).toBe(
"target_selection",
);
expect(refreshed.analysis_provider_binding?.provider.id).toBe("alpha");
});
it("preserves the bound client when a candidate refresh is cancelled during finalization", async () => {
const directory = await createTestTempDirectory(
"rea-candidate-refresh-cancel-",
);
const target = join(directory, "fixture.hop");
await writeFile(target, "fixture");
const starts: string[] = [];
const base = candidate("alpha", starts);
let hold = false;
let enter!: () => void;
let finish!: () => void;
let finalized!: () => void;
const entered = new Promise<void>((resolve) => {
enter = resolve;
});
const gate = new Promise<void>((resolve) => {
finish = resolve;
});
const done = new Promise<void>((resolve) => {
finalized = resolve;
});
const session = composeBinarySession(
new AnalysisProviderRegistry([
{
...base,
inspectAvailability: async ({
signal,
}: AnalysisProviderInspectionOptions = {}) => {
const result = await base.inspectAvailability();
if (!hold) return result;
enter();
await gate;
finalized();
if (signal?.aborted)
throw new AnalysisCancelledError("binary_session", {
cause: signal.reason,
});
return result;
},
},
]),
);
const server = createServer(
{ kind: "session", session },
{ logger: silentLogger },
);
const mcp = new Client({ name: "candidate-refresh-cancel", version: "1" });
const [clientTransport, serverTransport] =
InMemoryTransport.createLinkedPair();
resources.push(mcp, server, { close: () => session.close() });
await server.connect(serverTransport);
await mcp.connect(clientTransport);
let refreshing: ReturnType<typeof mcp.callTool> | undefined;
try {
expect(
(
await mcp.callTool({
name: "open_binary",
arguments: { path: target, provider_id: "alpha" },
})
).isError,
).toBeFalsy();
expect(
(await mcp.callTool({ name: "address_name", arguments: {} })).isError,
).toBeFalsy();
expect(starts).toEqual(["alpha"]);
hold = true;
const controller = new AbortController();
refreshing = mcp.callTool(
{ name: "binary_session", arguments: { refresh_candidates: true } },
{ signal: controller.signal },
);
await awaitAcknowledgement(entered, refreshing, "candidate refresh");
controller.abort();
await expect(refreshing).rejects.toThrow();
finish();
await done;
hold = false;
expect(
structured(await mcp.callTool({ name: "binary_session", arguments: {} }))
.result,
).toMatchObject({
analysis_provider_binding: { provider: { id: "alpha" } },
analysis_provider_candidates_observed: "target_selection",
});
expect(
(await mcp.callTool({ name: "address_name", arguments: {} })).isError,
).toBeFalsy();
expect(starts).toEqual(["alpha"]);
} finally {
hold = false;
finish();
await Promise.allSettled(refreshing === undefined ? [] : [refreshing]);
}
});
it("orders a provider switch after an in-flight candidate refresh", async () => {
const directory = await createTestTempDirectory(
"rea-candidate-refresh-switch-",
);
const target = join(directory, "fixture.hop");
await writeFile(target, "fixture");
const starts: string[] = [];
const alpha = candidate("alpha", starts);
let hold = false;
let enter!: () => void;
let finish!: () => void;
const entered = new Promise<void>((resolve) => {
enter = resolve;
});
const gate = new Promise<void>((resolve) => {
finish = resolve;
});
const session = composeBinarySession(
new AnalysisProviderRegistry([
{
...alpha,
inspectAvailability: async () => {
if (hold) {
hold = false;
enter();
await gate;
}
return alpha.inspectAvailability();
},
},
candidate("beta", starts),
]),
);
const server = createServer(
{ kind: "session", session },
{ logger: silentLogger },
);
const mcp = new Client({ name: "refresh-switch-race", version: "1" });
const [clientTransport, serverTransport] =
InMemoryTransport.createLinkedPair();
resources.push(mcp, server, { close: () => session.close() });
await server.connect(serverTransport);
await mcp.connect(clientTransport);
let refreshing: ReturnType<typeof mcp.callTool> | undefined;
let switching: ReturnType<typeof mcp.callTool> | undefined;
try {
expect(
(
await mcp.callTool({
name: "open_binary",
arguments: { path: target, provider_id: "alpha" },
})
).isError,
).toBeFalsy();
hold = true;
refreshing = mcp.callTool({
name: "binary_session",
arguments: { refresh_candidates: true },
});
await awaitAcknowledgement(entered, refreshing, "candidate refresh");
let switched = false;
switching = mcp
.callTool({
name: "open_binary",
arguments: { path: target, provider_id: "beta" },
})
.then((result) => {
switched = true;
return result;
});
await new Promise<void>((resolve) => setImmediate(resolve));
expect(switched).toBe(false);
finish();
const result = await refreshing;
expect(result.isError).toBeFalsy();
expect(structured(result).result).toMatchObject({
analysis_provider_binding: { provider: { id: "alpha" } },
});
expect((await switching).isError).toBeFalsy();
expect(
structured(await mcp.callTool({ name: "binary_session", arguments: {} }))
.result,
).toMatchObject({
analysis_provider_binding: { provider: { id: "beta" } },
});
} finally {
finish();
await Promise.allSettled(
[refreshing, switching].filter((operation) => operation !== undefined),
);
}
});
it("does not cancel an already admitted prompt lookup for a candidate refresh", async () => {
const directory = await createTestTempDirectory("rea-refresh-prompt-");
const target = join(directory, "fixture.hop");
await writeFile(target, "fixture");
const base = candidate("alpha", []);
let enter!: () => void;
let finish!: () => void;
const entered = new Promise<void>((resolve) => {
enter = resolve;
});
const gate = new Promise<void>((resolve) => {
finish = resolve;
});
const session = composeBinarySession(
new AnalysisProviderRegistry([
{
...base,
createClient: (selected, profile, context) => {
const client = base.createClient(selected, profile, context);
return {
...client,
executeIfReady: async (operation, args, options) => {
enter();
await gate;
if (options?.signal?.aborted)
throw new Error("Prompt lookup was cancelled");
return client.execute(operation, args, options);
},
};
},
},
]),
);
const server = createServer(
{ kind: "session", session },
{ logger: silentLogger },
);
const mcp = new Client({ name: "refresh-prompt", version: "1" });
const [clientTransport, serverTransport] =
InMemoryTransport.createLinkedPair();
resources.push(mcp, server, { close: () => session.close() });
await server.connect(serverTransport);
await mcp.connect(clientTransport);
let lookup: ReturnType<typeof session.executeIfProviderReady> | undefined;
try {
expect(
(
await mcp.callTool({
name: "open_binary",
arguments: { path: target, provider_id: "alpha" },
})
).isError,
).toBeFalsy();
expect(
(await mcp.callTool({ name: "address_name", arguments: {} })).isError,
).toBeFalsy();
lookup = session.executeIfProviderReady("address_name", {});
await awaitAcknowledgement(entered, lookup, "prompt lookup");
expect(
(
await mcp.callTool({
name: "binary_session",
arguments: { refresh_candidates: true },
})
).isError,
).toBeFalsy();
finish();
expect((await lookup)?.ok).toBe(true);
} finally {
finish();
await Promise.allSettled(lookup === undefined ? [] : [lookup]);
}
});
it("retains failed candidate-refresh cleanup while preserving the selected binding", async () => {
const directory = await createTestTempDirectory(
"rea-candidate-refresh-cleanup-",
);
const target = join(directory, "fixture.hop");
await writeFile(target, "fixture");
const starts: string[] = [];
const beta = candidate("beta", starts);
let fail = false;
let release = false;
const failure = new ProviderCleanupError("beta", ["beta-refresh-owner"], {
reason: "controlled cleanup failure",
});
const session = composeBinarySession(
new AnalysisProviderRegistry([
candidate("alpha", starts),
{
...beta,
inspectAvailability: () => {
if (fail) throw failure;
return beta.inspectAvailability();
},
closeOwnedResources: async () => {
if (!fail) return ok(null);
if (!release) return err(failure);
failure.resolveCleanup();
return ok(null);
},
},
]),
);
const server = createServer(
{ kind: "session", session },
{ logger: silentLogger },
);
const mcp = new Client({ name: "candidate-refresh-cleanup", version: "1" });
const [clientTransport, serverTransport] =
InMemoryTransport.createLinkedPair();
resources.push(mcp, server, { close: () => session.close() });
await server.connect(serverTransport);
await mcp.connect(clientTransport);
try {
expect(
(
await mcp.callTool({
name: "open_binary",
arguments: { path: target, provider_id: "alpha" },
})
).isError,
).toBeFalsy();
fail = true;
const failed = parseMcpToolError(
await mcp.callTool({
name: "binary_session",
arguments: { refresh_candidates: true },
}),
);
expect(failed.error).toMatchObject({
code: "cleanup_incomplete",
details: { resources: ["beta-refresh-owner"] },
});
const status = structured(
await mcp.callTool({ name: "binary_session", arguments: {} }),
);
expect(status.result).toMatchObject({
cleanup_pending: true,
analysis_provider_binding: { provider: { id: "alpha" } },
});
release = true;
expect(
(await mcp.callTool({ name: "close_binary", arguments: {} })).isError,
).toBeFalsy();
expect(failure.cleanupIncomplete).toBe(false);
} finally {
release = true;
await session.close();
}
});
it("retains attributed mixed profile cleanup through registered MCP calls and partial retry", async () => {
directory = await createTestTempDirectory("rea-profile-completion-mcp-");
const path = join(directory, "fixture.hop");
+26 -4
View File
@@ -709,8 +709,19 @@ describe("session filesystem path boundaries over MCP", () => {
name: "export_evidence_bundle",
arguments: { path: "relative-bundle.json" },
});
expect(exported.isError, JSON.stringify(exported.content)).toBe(true);
expect(JSON.stringify(exported.content)).toContain("absolute");
expect(parseMcpToolError(exported).error).toMatchObject({
code: "invalid_request",
details: {
operation: "export_evidence_bundle",
issues: [
expect.objectContaining({
path: ["path"],
reason: "invalid_format",
expected: "regex",
}),
],
},
});
const opened = await mcp.callTool({
name: "open_binary",
@@ -721,8 +732,19 @@ describe("session filesystem path boundaries over MCP", () => {
name: "close_binary",
arguments: { snapshot_path: "relative-analysis.json" },
});
expect(closed.isError, JSON.stringify(closed.content)).toBe(true);
expect(JSON.stringify(closed.content)).toContain("absolute");
expect(parseMcpToolError(closed).error).toMatchObject({
code: "invalid_request",
details: {
operation: "close_binary",
issues: [
expect.objectContaining({
path: ["snapshot_path"],
reason: "invalid_format",
expected: "regex",
}),
],
},
});
await mcp.callTool({ name: "close_binary", arguments: {} });
}, 10_000);
});
+22 -14
View File
@@ -383,7 +383,7 @@ it("attributes rejected WASM paths to their selected fields before provider work
},
});
onTestFinished(harness.close);
for (const [input, paths] of [
const relativeCases = [
[{ path: "relative.wasm" }, [["path"]]],
[
{ path: value.artifact.path, glue_paths: ["relative.js"] },
@@ -412,7 +412,9 @@ it("attributes rejected WASM paths to their selected fields before provider work
},
[["path"], ["glue_paths", 0], ["candidate_paths", 0]],
],
] as const) {
] as const;
const checked: string[] = [];
for (const [input, paths] of relativeCases) {
const response = await harness.client.callTool({
name: "inspect_wasm_artifact",
arguments: input,
@@ -423,14 +425,19 @@ it("attributes rejected WASM paths to their selected fields before provider work
.object({ issues: z.array(z.unknown()) })
.parse(failure.details);
expect(issues).toEqual(
paths.map((path) => ({
path,
reason: "invalid_format",
expected: "regex",
message: "must be an absolute local filesystem path",
})),
paths.map((path) =>
expect.objectContaining({
path,
reason: "invalid_format",
expected: "regex",
}),
),
);
checked.push(paths.map((path) => path.join(".")).join("+"));
}
expect(checked).toEqual(
expect.arrayContaining(["path", "glue_paths.0", "candidate_paths.0"]),
);
expect(inspect).not.toHaveBeenCalled();
const valid = await harness.client.callTool({
name: "inspect_wasm_artifact",
@@ -760,12 +767,13 @@ it("rejects malformed paths, refuses WASM summary delivery before provider work
expect(
z.object({ issues: z.array(z.unknown()) }).parse(rejected.details).issues,
).toEqual(
[["path"], ["glue_paths", 0]].map((path) => ({
path,
reason: "invalid_format",
expected: "regex",
message: "must be an absolute local filesystem path",
})),
[["path"], ["glue_paths", 0]].map((path) =>
expect.objectContaining({
path,
reason: "invalid_format",
expected: "regex",
}),
),
);
expect(inspect).not.toHaveBeenCalled();
for (const input of [
@@ -0,0 +1,205 @@
import { execFileSync } from "node:child_process";
import { writeFile } from "node:fs/promises";
import { join } from "node:path";
import { expect, it } from "vitest";
import {
PYTHON_HELPERS,
PYTHON_IMPORTS,
} from "../../../src/domain/desktop/appExportPython.js";
import {
detectStateFormat,
textualView,
} from "../../../src/domain/desktop/stateFormats.js";
import { createTestTempDirectory } from "../../fixtures/temporaryDirectory.js";
const arrays = "[".repeat(1500) + "1" + "]".repeat(1500);
const objects = '{"a":'.repeat(1500) + "1" + "}".repeat(1500);
const cases = [
["discarded arrays", `{"value":${arrays},"value":"done"}`, true],
["discarded objects", `{"value":${objects},"value":"done"}`, true],
["escaped duplicate key", `{"val\\u0075e":${arrays},"value":"done"}`, true],
["retained deep value", `{"value":${arrays}}`, false],
[
"retained deep inexact number",
`{"value":${"[".repeat(600)}9007199254740993${"]".repeat(600)}}`,
false,
],
[
"deep overwritten inexact number",
`{"value":${"[".repeat(600)}{"x":9007199254740993,"x":1}${"]".repeat(600)}}`,
false,
],
[
"deep empty-key overwrite",
`{"value":${"[".repeat(600)}{"":9007199254740993,"":1}${"]".repeat(600)}}`,
false,
],
[
"deep multiple inexact keys with one overwritten",
`{"value":${"[".repeat(600)}{"a":9007199254740993,"b":9007199254740993,"a":1}${"]".repeat(600)}}`,
false,
],
[
"deep multiple inexact keys all overwritten",
`{"value":${"[".repeat(600)}{"a":9007199254740993,"b":9007199254740993,"a":1,"b":2}${"]".repeat(600)}}`,
false,
],
[
"deep inexact key replaced by inexact",
`{"value":${"[".repeat(600)}{"a":9007199254740993,"a":9007199254740995}${"]".repeat(600)}}`,
false,
],
[
"deep escaped duplicate overwritten",
`{"value":${"[".repeat(600)}{"\\u0061":9007199254740993,"a":1}${"]".repeat(600)}}`,
false,
],
[
"discarded deep inexact subtree",
`{"value":${"[".repeat(600)}9007199254740993${"]".repeat(600)},"value":"done"}`,
true,
],
["last admitted depth", "[".repeat(500) + "1" + "]".repeat(500), true],
["empty container at limit", "[".repeat(501) + "]".repeat(501), true],
["empty container over limit", "[".repeat(502) + "]".repeat(502), false],
["trailing comma", `{"value":${arrays},"value":"done",}`, false],
["missing value", `{"value":${arrays},"value":}`, false],
["wrong closing bracket", `{"value":${arrays},"value":"done"]`, false],
] as const;
it.each(cases)("reads JSON with %s", async (_name, text, readable) => {
const root = await createTestTempDirectory("desktop-json-depth-");
const script = join(root, "read.py");
const imports = PYTHON_IMPORTS.split("\n")
.filter((line) => !line.includes("agentenv_protocol"))
.join("\n");
await writeFile(
script,
`${imports}${PYTHON_HELPERS}
text = json.load(sys.stdin)
print(json.dumps(_text_view("state.json", text, len(text.encode("utf-8")))))
`,
);
const output = JSON.parse(
execFileSync(
process.platform === "win32" ? "python" : "python3",
["-I", "-B", "-X", "utf8", script],
{ input: JSON.stringify(text), encoding: "utf8" },
),
) as { format: string; view: unknown; error?: string };
const bytes = Buffer.from(text);
const format = detectStateFormat("state.json", bytes);
if (format !== "json" && format !== "text")
throw new Error(`Unexpected native format ${format}`);
const native = textualView(format, bytes);
expect(format === "json" && native.ok).toBe(readable);
expect(output.format).toBe(format);
if (native.ok) {
expect(output.error).toBeUndefined();
// Compare JSON values on their wire form, independent of object prototypes.
expect(JSON.stringify(output.view)).toBe(JSON.stringify(native.value));
if (
readable &&
(_name.startsWith("discarded") || _name === "escaped duplicate key")
)
expect(output.view).toEqual({ value: "done" });
} else {
expect(output.view).toBeNull();
expect(output.error).toBe(native.error);
}
});
it("bounds construction before refusing retained excessive JSON depth", async () => {
const root = await createTestTempDirectory("desktop-json-construction-");
const script = join(root, "bounded.py");
const imports = PYTHON_IMPORTS.split("\n")
.filter((line) => !line.includes("agentenv_protocol"))
.join("\n");
await writeFile(
script,
`${imports}${PYTHON_HELPERS}
import tracemalloc
text = "[" * 20_000 + "0" + "]" * 20_000
recursion_limit = sys.getrecursionlimit()
tracemalloc.start()
entry = _json_view(text, len(text))
peak = tracemalloc.get_traced_memory()[1]
tracemalloc.stop()
print(json.dumps({"entry": entry, "peak": peak, "recursion_unchanged": sys.getrecursionlimit() == recursion_limit}))
`,
);
const output = JSON.parse(
execFileSync(
process.platform === "win32" ? "python" : "python3",
["-I", "-B", "-X", "utf8", script],
{ encoding: "utf8" },
),
) as { entry: unknown; peak: number; recursion_unchanged: boolean };
expect(output.entry).toEqual({
format: "json",
view: null,
error: "nested deeper than the 500 levels a JSON view keeps",
});
expect(output.recursion_unchanged).toBe(true);
// Python heap only, excluding source construction and interpreter/OS memory.
// A value retained only to refuse it must not allocate every input container.
expect(output.peak).toBeLessThan(1024 * 1024);
});
it("bounds allocations when a very deep value is overwritten", async () => {
const root = await createTestTempDirectory("desktop-json-memory-");
const script = join(root, "read.py");
const imports = PYTHON_IMPORTS.split("\n")
.filter((line) => !line.includes("agentenv_protocol"))
.join("\n");
await writeFile(
script,
`${imports}\nimport tracemalloc\n${PYTHON_HELPERS}
text = '{"value":' + '[' * 200000 + '1' + ']' * 200000 + ',"value":"done"}'
tracemalloc.start()
result = _json_view(text, len(text))
_, peak = tracemalloc.get_traced_memory()
print(json.dumps({"view": result["view"], "peak": peak}))
`,
);
const measured = JSON.parse(
execFileSync(
process.platform === "win32" ? "python" : "python3",
["-I", "-B", script],
{ encoding: "utf8", timeout: 30000 },
),
) as { view: unknown; peak: number };
expect(measured.view).toEqual({ value: "done" });
expect(measured.peak).toBeLessThan(8 * 1024 * 1024);
}, 40000);
it("bounds a wide container already beyond the retained view depth", async () => {
const root = await createTestTempDirectory("desktop-json-wide-depth-");
const script = join(root, "read.py");
const imports = PYTHON_IMPORTS.split("\n")
.filter((line) => !line.includes("agentenv_protocol"))
.join("\n");
await writeFile(
script,
`${imports}\nimport tracemalloc\n${PYTHON_HELPERS}
text = '[' * 501 + ','.join(['"abc"'] * 600000) + ']' * 501
tracemalloc.start()
result = _json_view(text, len(text))
_, peak = tracemalloc.get_traced_memory()
print(json.dumps({"error": result.get("error"), "peak": peak}))
`,
);
const measured = JSON.parse(
execFileSync(
process.platform === "win32" ? "python" : "python3",
["-I", "-B", script],
{ encoding: "utf8", timeout: 30000 },
),
) as { error: string; peak: number };
expect(measured.error).toBe(
"nested deeper than the 500 levels a JSON view keeps",
);
expect(measured.peak).toBeLessThan(8 * 1024 * 1024);
}, 40000);
@@ -34,7 +34,7 @@ const execPython = (
try {
writeFileSync(script, args[codeIndex + 1]!, "utf8");
return execFileSync(
"python3",
process.platform === "win32" ? "python" : "python3",
[
"-X",
"utf8",
@@ -375,15 +375,16 @@ it("checks text for JSON the way REA does, token by token, without parsing it",
"-I",
"-B",
"-c",
`${imports}${PYTHON_HELPERS}\nprint(json.dumps([_is_json(text) for text in json.load(sys.stdin)]))`,
`${imports}${PYTHON_HELPERS}\ntexts = json.load(sys.stdin)\nprint(json.dumps({"syntax": [_read_json(text, build=False) is not _INVALID_JSON for text in texts], "built": [_read_json(text) is not _INVALID_JSON for text in texts]}))`,
],
{ input: JSON.stringify(texts), encoding: "utf8" },
),
) as boolean[];
) as { syntax: boolean[]; built: boolean[] };
const parsed = texts.map((text) => safeParseJson(text).ok);
expect(texts.map((text) => isJsonText(text))).toEqual(parsed);
expect(python).toEqual(parsed);
expect(python.syntax).toEqual(parsed);
expect(python.built).toEqual(parsed);
expect(parsed.filter(Boolean)).toHaveLength(10);
});
@@ -445,9 +446,13 @@ print(json.dumps([_text_view("settings.ini", text, len(text.encode()))["view"]["
const root = await createTestTempDirectory("rea-keyfile-python-");
const script = join(root, "fixture.py");
await writeFile(script, program, "utf8");
const output = execFileSync("python3", ["-I", "-B", "-X", "utf8", script], {
input: JSON.stringify(inputs),
encoding: "utf8",
});
const output = execFileSync(
process.platform === "win32" ? "python" : "python3",
["-I", "-B", "-X", "utf8", script],
{
input: JSON.stringify(inputs),
encoding: "utf8",
},
);
expect(JSON.parse(output)).toEqual(inputs.map(keyfileView));
});