15 Commits
Author SHA1 Message Date
czcdr 95803a6833 docs: date the 2.0.9 entry for its release 2026-10-18 15:35:07 +08:00
czcdr 7b02aa95b6 setup: strip the BOM from the records setup itself reads (2.0.9)
2.0.9 taught the shim to drop a leading U+FEFF but left setup.mjs parsing raw.
Both programs read the same mirror.json and bundle.json, so a BOM made one file
mean different things: the shim found the checkout, setup did not — and a
mirrored setup re-mirrored the previous manifest and provisioned from it.
bundle.json was read as "nothing is provisioned" for the same reason.

Both sides now use the same readJson(). verify-branch.ps1 and CI fail on a bare
JSON.parse(readFileSync(...)) in either file, and two tests cover a BOM-ed
mirror.json and a BOM-ed bundle.json; both fail without the strip (checked).
2026-10-18 15:16:40 +08:00
czcdr ebb912ace2 ci: keep the README check's inline script quote-free
The step runs its JavaScript inside `node -e "..."`, so a double quote in a
comment ended the shell string early: bash then ran the tail as commands and node
saw a truncated script ("Unexpected end of input"). The step failed on both
runners while the same comparison passed locally in verify-branch.ps1, which is
exactly the split that hides a broken gate.

The comment is quote-free now, and the extracted script runs clean from the repo
root (`readme range ok`).
2026-10-18 15:03:37 +08:00
czcdr 3f413e1989 shim: settle the re-exec only when it fails, and tolerate BOM-ed JSON (2.0.9)
Review of 2.0.9 found that the fallback still could not fire in the case it was written for, and that a hand-edited mirror.json behaved as if absent.

- The re-exec promise resolved true at the `spawn` event, which fires before the child has read anything. A later "it actually failed" could no longer be seen, so a non-zero exit with the source gone fell through to `resolve(false)`, was ignored, and the process ended 0 - silently, with the message never printed. It now resolves ONLY on failure: while the child lives this process is parked and the child's exit is the exit. Measured before/after with a checkout launcher that deletes itself and exits 3: exit 0 (no output) -> exit 78 with "could not run ... running the mirrored copy instead".
- That also covers the realistic upgrade race, which is not a spawn failure at all: node starts, finds its file replaced, and exits 1 with MODULE_NOT_FOUND. A non-zero exit whose source no longer exists now falls back instead of being passed through.
- JSON reads strip a leading U+FEFF. JSON.parse rejects it, and every caller treats a parse failure as "no record", so a BOM (which Windows editors write) silently disabled the mirror check and made bundle.json look unprovisioned. Measured with PowerShell-written fixtures.

Docs/CI: the README pin example names the version this tree publishes (#v2.0.9); both verify-branch.ps1 and validate.yml now fail when the pin does not equal package.json's version, and the README range check moved into the same step. A mutation (pin reverted to v2.0.8) fails the check and was restored.

Tests: 69 (66 pass + 3 skipped); the new one deletes the checkout launcher mid-flight and asserts the fallback. verify-branch 41 checks green, audit clean, host-compat 10 covered / 5 outside. Version 2.0.9, NOT published: no tag, no release, no npm.
2026-10-18 15:01:52 +08:00
czcdr edf3968319 shim: make the mirror fallback real, repair the mojibake, and close the doc gaps (2.0.9)
Two problems blocked publishing 2.0.8's follow-up, both found by an external review:

- reexecFromCheckout announced a fallback it did not perform. On a spawn error it printed "continuing with the mirrored copy" and returned; main() had already returned, no exit event followed, the loop drained, and the row exited 0 with no tools and no message. It now resolves a boolean: only a started child takes over the exit code, and a failure falls through to the local launch path with an honest message.
- bin/rea-shim.mjs carried 16 mojibake dashes (U+9225 + "?") from a GBK edit in 2.0.7, two of them in user-visible stderr ("first run - provisioning started", "nothing is running right now - read the log"). All restored to an em dash and the space the mis-decode had eaten.

A checkout copy no longer hops into the plugin a bundle's mirror.json names: only a mirror re-execs, a checkout (it ships cordis.patch.yml) never does. That was a real footgun - a developer running the checkout would silently test the installed plugin. The re-exec's comment now matches the code, and the mirror tests run the mirrored copy instead of the checkout.

Docs and CI: the README pin example moves from v2.0.0 (the most bug-prone release) to v2.0.8; the READMEs state the Windows limitation that a forced kill of the shim leaves the delegated REA running; docs/bundle-design.md gains the 2.0.7 launch mode and the 2.0.8 self-healing mirror, including mirror.json's shimSha256/pluginVersion/dshHome; and CI now checks that both READMEs quote engines.dsh verbatim, which previously only verify-branch.ps1 did. systemReaCandidates dedupes on the platform argument rather than process.platform.

Tests: 68 (65 pass + 3 skipped), one new for the checkout guard. verify-branch 39 checks green, audit clean, host-compat 10 covered / 5 outside. Version is 2.0.9 and NOT published: no tag, no release, no npm.
2026-10-18 14:47:44 +08:00
czcdr e3b2a064af docs: keep the declared range in the READMEs, and fix the home fallback (2.0.8)
An independent audit caught three non-blocking inconsistencies in the 2.0.8 tree:

- Both READMEs still quoted engines.dsh as ">=0.1.7-rc.2 <0.3.0" after the declaration moved to "<0.3.0-0" — the stale form the range's own negative test proves is wrong (it admits 0.3.0 prereleases). verify-branch now compares the sentence with package.json, so the drift cannot come back silently; a mutation (reverting one README) fails the check and the file was restored.
- CHANGELOG 2.0.8 still claimed 62 tests (59+3); the tree has 67 (64+3), seven of them new in this release.
- check-updates.ps1 computed $homeEarly only inside the "no REA_BUNDLE_ROOT" branch, so with the override set it stayed $null and Join-Path threw, degrading the profile section to "no profile". It is resolved unconditionally now and the profile lookup uses it.

Verified: 67 tests (64 pass + 3 skipped), audit clean, verify-branch ALL CHECKS PASSED with the new README check, and the home/profile block re-run with REA_BUNDLE_ROOT set resolves DSH_HOME and the real profile directory.
2026-10-18 14:17:07 +08:00
czcdr 3b57a9d1ac shim: find the real DSH home when the harness has scrubbed the environment (2.0.8)
The duplicate-registration check reads <DSH_HOME>/cordis.patch.yml, but an MCP child has every DSH_* variable stripped, so a mirrored copy with a custom home read ~/.dsh instead: the wrong file, and a conflict it could never report. setup.mjs now records the home it ran under in mirror.json (dshHome), and the shim prefers the environment, then that record, then ~/.dsh. The mirrored copy's own bundle root was never affected - it comes from its location - so this only changes which patch file the warning reads.

check-updates.ps1 had a second home algorithm in the same script: its profile section hardcoded $env:USERPROFILE, so on a custom DSH_HOME it reported "no profile found" while the rest of the run had resolved the home correctly. It now reuses the same rule.

Test: a bundle whose mirror.json names a home containing a foreign serverName: rea, run with DSH_HOME="" (the scrubbed shape), must report duplicateRegistration with that file; 67 tests total (64 pass + 3 skipped).
2026-10-18 14:09:16 +08:00
czcdr 5c22588001 shim: refresh a stale mirrored launcher from the installed plugin (2.0.8)
The row launches <bundle>/bin/rea-shim.mjs, a copy made by setup. A market upgrade replaces the package and runs no setup, so that copy kept executing the code it was made from: every fix in a new plugin version reached a user only if they happened to re-run setup, and an old copy keeps working, which is why nobody notices.

From 2.0.8 the copy compares itself with <mirror.json.pluginRoot>/bin/rea-shim.mjs and re-execs that file when the two differ: one hop, guarded by REA_BUNDLE_SHIM_REEXEC, and only mirror -> checkout (the checkout finds itself and does nothing). --shim-status reports mirror.{pluginRoot,mirroredAt,stale,reason}; setup records shimSha256 and pluginVersion; check-updates.ps1 reports a lagging mirror and says to re-run setup.

A version older than this one cannot repair its own upgrades - the copy has no such code - so upgrading from 2.0.0..2.0.7 still needs one setup run. The README says so in both languages.

Tests: 66 (63 pass + 3 skipped). Four new ones cover the re-exec, the guard, the stale report and the identical case.
2026-10-18 13:12:21 +08:00
czcdr 0e3aa428bf shim: stop the launch probe from recursing or leaking its tree (2.0.8)
2.0.7's probe spawned a candidate with the plain environment, so a `rea` on PATH that wraps this shim made the probe recurse: probe -> wrapper -> a fresh shim -> probe again. Measured 83+ concurrent node processes, still respawning after the parent was killed. The guard only covered delegation, not probing; the probe now runs with REA_BUNDLE_DELEGATED=1 and REA_BUNDLE_AUTO_PROVISION=0, so the nested run takes the bundle path and returns (0.2 s, no extra processes).

The timeout path leaked a grandchild: spawnSync's timeout reaps the child it started, so the following taskkill /T could not find the parent, and a hanging candidate left a live REA per attempt. The probe is now an async spawn with its own timer, which kills the tree while the parent is still alive (measured: 8.3 s, rejected ["timeout"], no leftover process).

Also: the Windows signal handlers are removed rather than kept as dead code (child.kill is TerminateProcess there, and a console Ctrl+C reaches the child directly), with the platform limitation stated where it matters; comspec() is validated and falls back to %SystemRoot%\System32\cmd.exe so a broken ComSpec no longer reads as "no REA"; and a directory named like the binary is reported as "not a file" instead of "does not exist".

Tests: 62 (59 pass + 3 skipped). Two of the new ones reproduce the recursion and the ComSpec case.
2026-10-18 13:10:03 +08:00
czcdr 37d2adb17a shim: run the REA a machine already has, and tighten the host declaration (2.0.7)
Launch decision at startup: a REA on PATH that answers --version and is not older than the provisioned build wins, and the bundle is never provisioned (the 1.x contract, reachable from the 2.x row). REA_BUNDLE_LAUNCH=auto|system|bundle overrides it; REA_BUNDLE_SYSTEM_REA names a binary off PATH; --shim-status reports launch.{mode,reason,unusable,systemRea,rejected}. A delegated child carries REA_BUNDLE_DELEGATED=1 so a wrapper cannot make the shim delegate to itself.

Two independent reviews found real defects in the first cut, all fixed here: the cmd line quoted the arguments, so a .cmd's %* received "mcp"; a 3 s probe budget against a measured 2.4 s --version turned detection into a coin flip and leaked a live REA per timeout; SIGINT forwarding on Windows killed only the cmd.exe wrapper. Also: status reported 0 for a pinned-but-unusable system launch, existsSync accepted directories, PATH entries were not normalised, and fail() suggested a setup command without --root.

The engines.dsh ceiling is now <0.3.0-0: with includePrerelease, <0.3.0 admitted 0.3.0-alpha.1 and 0.3.0-rc.1, which nobody audited. host-compat's tests pin the release list literally and prove the leak check has teeth; the evaluator accepts +build metadata and rejects v-prefixes and zero-padded identifiers as semver does.

Environment independence: every shim call in the suite pins REA_BUNDLE_LAUNCH, setup's own shim probe pins it too, and both CI jobs set it at job level. Bootstrap now honours REA_BUNDLE_ROOT like the other three readers.
2026-10-18 12:57:59 +08:00
czcdr a6ee476bfd compat: declare the DSH host range the market reads (2.0.6)
engines.dsh = '>=0.1.7-rc.2 <0.3.0'. The market's deriveHostCompatibility reads only engines.dsh (top-level or dsh.engines) and peers on @deepseek-ai/dsh*; with neither it reports 'unknown', which is what the card has shown until now. Peer declarations are deliberately NOT used: @deepseek-ai/dsh-mcp-client is published on npm at 0.0.1-rc.1 while the host ships 0.2.x, so a peer range would resolve to a version the host does not have.

The range is checked, not guessed: scripts/host-compat.mjs carries the dsh release list (0.1.7-rc.2 .. 0.2.1-rc.2) and fails when the declaration stops covering one of them, or starts admitting 0.3.0. The row surface was compared at dsh-v0.1.7-rc.2 and dsh-v0.2.1-rc.2: packages/mcp/mcp-client exposes the same config keys, and dsh.bundle.patch exists at both.

Upstream rea-agents 6.6.0 is released on GitHub but npm latest is still 6.5.0, so the pin stays at 6.5.0; the bump is scheduled for 2026-10-19.
2026-10-18 12:31:27 +08:00
czcdr d0e2ffd7a6 package: declare the entry file the market's update check requires (2.0.5)
dshmarket's update path asks hasLoadableEntry whether the new build has an entry file that exists (main/exports, else index.js) and rolls back when it does not. This package declared neither, and the package it mounts (@deepseek-ai/dsh-mcp-client) ships inside the harness build rather than in node_modules, so every location the check probes was empty: 2.0.1 -> 2.0.4 failed three times with 'updated build has no loadable entry' on a healthy profile, and the rollback was reported unverifiable for the same reason.

index.mjs is a real but deliberately empty cordis plugin: the working row is still inserted by cordis.patch.yml, so loading it must do nothing (a second serverName: rea would be refused). Two tests, verify-branch and the CI manifest/payload checks now pin both the pointer and its presence in files[].
2026-10-18 11:53:24 +08:00
czcdr 27d3f1106c shim: guard the stale-lock removal and name the duplicate REA registration (2.0.4) 2026-10-17 13:03:15 +08:00
czcdr f6afdea83d shim/setup: fix the wedging stale lock, drop the 1.x examples, and close the second audit
Each item was reproduced before it was changed, and every fix has a test that
fails when it is reverted (18 reverse checks; two of them on WSL because the
code path only exists on POSIX).

- A lock older than the freshness window was never removed, so the exclusive
  create below it failed with EEXIST forever: after a killed download the shim
  never provisioned again — and said nothing, because main() printed no message
  for "in-flight". Reproduced with a 7-hour dead lock (no run started, lock
  untouched, no output) against a 5-minute control lock that behaved correctly.
  An expired lock, or one whose holder is gone, is now cleared; "in-flight",
  "spawn-failed" and "unwritable" each say what they are.
- examples/ still held the 1.x rows (`command: rea`, `cmd.exe /c rea`) and shipped
  inside the 2.x npm package, where copying one bypasses the bundle row entirely.
  Removed from main (v1 keeps them, where they are correct) together with the
  package.json files[] entry and the three validate.yml checks that named them;
  the payload is 19 files now.
- `--offline` was a promise the npm step ignored: provisionReaAgents ran
  `npm install` regardless. It now refuses that step when the vendored copy is
  unusable, and still accepts one that is.
- That npm step is the only one that cannot use this bundle's own route/proxy
  resolution; its failure now says so, and both READMEs name HTTPS_PROXY /
  NPM_CONFIG_REGISTRY.
- POSIX extraction had exactly one tool per format while Ghidra is a zip on every
  platform, so a bare image without unzip failed for certain. zip now falls back
  unzip -> python3 -m zipfile -> the provisioned JDK's jar, tarballs tar ->
  python3, and when none is present the log names the tools it tried.
- The platform-tools arm64 comments were wrong (Google ships x86_64 binaries for
  both linux and darwin): corrected, so an arm64 host gets an accurate
  optional-component report instead of an implied native adb.
- docs/bundle-design.md caught up with the implementation: the row resolves
  <DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs at load time, and the vendored
  CLI lives under vendor/node_modules/rea-agents.
- release.yml's publish job now requires the version's tag and its release asset,
  so a workflow_dispatch run cannot publish to npm while the market's fallback
  tarball link is missing.
- Numeric environment overrides are sanitised instead of becoming NaN ("gave up
  after NaN attempt(s)"), and the effective limits are logged; the PowerShell
  route's docstring now says it is the primary Windows route; five redundant
  dynamic `node:fs` imports are gone.

37 unit tests (Windows 35 pass + 2 platform skips; WSL 36 pass + 1), audit
clean, verify-branch ALL CHECKS PASSED, and the real machine re-provisioned with
the mirror byte-identical to this tree (REA 6.5.0, 207 tools, a live tool call
reporting path_source: environment).
2026-10-17 00:04:04 +08:00
czcdr 975ac5aac8 setup/shim: fix eleven verified defects, and audit the gaps that hid them
Every item below was confirmed in the code (or reproduced) before it was changed,
and each now has a regression test that fails when its fix is reverted.

- `--only` rewrote bundle.json from this run's results alone, erasing every other
  component's record: the shim stopped exporting JAVA_HOME/GHIDRA_INSTALL_DIR and
  the next full run re-extracted them. Untouched records are carried over, and
  only while their recorded path still exists.
- `--only` skipped the requires check entirely, so `--only apktool` resolved
  {jvm} to a vendor/jdk that only a full run creates and failed on a machine whose
  JDK was recorded next to it.
- A complete .part was not treated as finished work: it logged "already complete"
  and then asked for `Range: bytes=<size>-` (416, four attempts, reported failed).
  It is verified and renamed now.
- The Windows PowerShell route deleted the partial file before every attempt, and
  the Node fallback reused a stale `have`, appending a full 200 response onto the
  bytes a failed route had already written. The route resumes from what is on
  disk, and the fallback re-reads the file's real length.
- POSIX wrappers landed 0644 while REA spawns them without a shell (EACCES); a
  POSIX entry marked `executable` is now set 0755.
- apktool's wrapper was pinned to `master` with a hard checksum, so one upstream
  commit was enough to break every new install. Pinned to the immutable v3.0.3
  tag, with the hashes measured from it.
- `--reuse` adopted a JDK because `javac` existed and a Ghidra because
  `analyzeHeadless` existed. A candidate must now pass the component's own probe,
  and a `file` probe finally honours `expect.contains`.
- A malformed bundle.json failed permanently: setup could not get far enough to
  replace the file it could not parse, and nothing started setup. Setup tolerates
  it, and the shim provisions for `malformed` as well as `missing`.
- A failed run kept a fresh lock that made the shim claim "provisioning started in
  the background" for six hours while nothing ran. The lock records its state and
  pid, and a dead holder no longer blocks a retry.
- A successful `--json` run never released the lock, because the removal sat
  inside the non-`--json` branch.
- The shim's fallback REA path omitted `node_modules`, so an older bundle.json
  without paths.reaEntry reported a working install as gone.
- bootstrap.ps1 threw on a non-numeric `node -v` (an int cast under
  ErrorActionPreference=Stop) instead of falling through to the portable download,
  and could exit 0 when `&` failed; it reads REA's engines from the manifest pin
  instead of a second hardcoded copy, and -DryRun no longer creates directories
  while claiming nothing changed.
- check-updates.ps1 looked for REA in the npm global root and for Ghidra in a
  user-level variable, which is the 1.x story: on a 2.x machine the REA section
  pointed at the wrong object, the doctor self-check was skipped entirely, and the
  bundle drift check silently skipped unrecorded pins while printing "all pins
  match". It reads bundle.json first now.

audit-manifest.mjs gained four checks: manifest components vs INSTALL_ORDER,
pins.nodeEngines vs package.json engines, the REA pin probe against the next
patch/minor, and the POSIX wrapper exec flags. validate.yml gained a
windows-latest job, because the PowerShell download route, the zip extraction and
the bootstrap were Windows-only code that no CI ever ran.

33 unit tests (32 pass, 1 platform skip), audit clean, on Windows and WSL.
2026-10-16 23:26:08 +08:00
21 changed files with 4091 additions and 296 deletions

No files matched your search

+19
View File
@@ -78,5 +78,24 @@ jobs:
npm install -g 'npm@^11'
npm --version
- name: The version being published is tagged and carries its release asset
# workflow_dispatch can start this job with no Release behind it, and then npm
# and the GitHub tarball drift apart: the market's fallback link points at the
# release asset, so publishing a version whose asset does not exist ships a 404
# to half the install paths.
env:
GH_TOKEN: ${{ github.token }}
run: |
ver="v$(node -p "require('./package.json').version")"
if ! git ls-remote --tags --exit-code origin "refs/tags/$ver" >/dev/null 2>&1; then
echo "FAIL: no tag $ver on origin — publish the GitHub Release first"
exit 1
fi
if ! gh release view "$ver" --json assets --jq '.assets[].name' | grep -qx 'dsh-plugin-rea.tgz'; then
echo "FAIL: release $ver has no dsh-plugin-rea.tgz asset — the market link would 404"
exit 1
fi
echo "$ver is tagged and carries the version-free tarball"
- name: Publish to npm via trusted publishing
run: npm publish --provenance --access public
+98 -10
View File
@@ -11,6 +11,11 @@ permissions:
jobs:
validate:
runs-on: ubuntu-latest
# The shim's launch decision reads this machine's REA. No runner has one, so
# the default `auto` would happen to take the bundle path today — pin it so
# that stays true if an image ever ships REA.
env:
REA_BUNDLE_LAUNCH: bundle
steps:
- uses: actions/checkout@v4
@@ -25,13 +30,74 @@ jobs:
const fail = (m) => { console.error('FAIL: ' + m); process.exit(1) };
if (!p.dsh || !p.dsh.bundle || !p.dsh.bundle.patch) fail('package.json is missing dsh.bundle.patch');
if (p.private) fail('package.json must not be private to be publishable');
for (const f of ['cordis.patch.yml', 'README.md', 'LICENSE', 'check-updates.ps1']) {
for (const f of ['cordis.patch.yml', 'index.mjs', 'README.md', 'LICENSE', 'check-updates.ps1']) {
if (!p.files.includes(f)) fail('files[] must ship ' + f);
}
if (!require('fs').existsSync(p.dsh.bundle.patch)) fail('dsh.bundle.patch target does not exist');
// The market's update path rolls a build back when package.json points
// at no existing entry (dshmarket's hasLoadableEntry); this package
// shipped no main at all until 2.0.5, so every update of a healthy
// profile failed with "updated build has no loadable entry".
if (typeof p.main !== 'string') fail('package.json must declare main');
if (!require('fs').existsSync(p.main)) fail('main points at no shipped file: ' + p.main);
// engines.dsh is what the market reads to say "compatible" instead of
// "unknown"; without it the plugin is reported as undeclared.
if (!p.engines || typeof p.engines.dsh !== 'string') fail('package.json must declare engines.dsh');
console.log('manifest ok');
"
- name: Declared DSH range covers every released host
# A wrong range is silent on both sides: unparseable reads as "undeclared"
# and a drifted floor drops hosts the README still lists.
run: node scripts/host-compat.mjs
- name: The READMEs quote the declared DSH range verbatim
# The same comparison verify-branch.ps1 makes locally. Without it here, a
# stale quote survives every CI signal: the JSON, host-compat and the
# audit all agree while the sentence a human reads names another ceiling
# (that is how `<0.3.0` outlived the move to `<0.3.0-0`).
run: |
node -e "
const p = require('./package.json');
const { readFileSync } = require('fs');
const fail = (m) => { console.error('FAIL: ' + m); process.exit(1) };
for (const doc of ['README.md', 'README.zh-CN.md']) {
const text = readFileSync(doc, 'utf8');
if (!text.includes(p.engines.dsh)) fail(doc + ' does not quote engines.dsh (' + p.engines.dsh + ')');
// The pin example is copy-pasteable: it must name the version this
// tree publishes, or the reader gets a ref that does not exist.
if (!text.includes('dsh-plugin-rea#v' + p.version)) fail(doc + ' does not pin dsh-plugin-rea#v' + p.version);
}
console.log('readme range ok');
"
- name: Both JSON readers tolerate a BOM
# verify-branch.ps1 runs the same check locally. bin/rea-shim.mjs and
# scripts/setup.mjs read the same mirror.json/bundle.json, so a strip on
# one side only makes the same file say "there is a checkout" to the shim
# and "there is none" to setup — which shipped once: setup re-mirrored the
# previous manifest. A bare JSON.parse(readFileSync(...)) is that bug.
run: |
node - <<'JS'
const { readFileSync } = require('node:fs');
const blind = [];
for (const file of ['bin/rea-shim.mjs', 'scripts/setup.mjs']) {
const text = readFileSync(file, 'utf8');
for (const m of text.matchAll(/JSON\.parse\(\s*readFileSync\(/g)) {
// readJson() itself is the strip; anything else must not parse a file raw.
if (!/replace\(/.test(text.slice(m.index, m.index + 200))) {
blind.push(file + ':' + text.slice(0, m.index).split('\n').length);
}
}
if (!text.includes('replace(/^\\uFEFF/')) blind.push(file + ': no U+FEFF strip');
}
if (blind.length) {
console.error('FAIL: BOM-blind JSON read at ' + blind.join(', '));
process.exit(1);
}
console.log('bom-blind reads: none');
JS
- name: JSON files parse
run: |
node -e "
@@ -44,16 +110,15 @@ jobs:
- name: The npm payload ships every market-facing file
run: |
# Regression guard for files[]: the market reads screenshots.json and the
# images next to package.json, and a user copies examples/ by hand — none of
# that is worth anything if npm pack drops it.
# images next to package.json — none of that is worth anything if npm pack
# drops it.
node -e "
const { execSync } = require('child_process');
const meta = JSON.parse(execSync('npm pack --dry-run --json', { encoding: 'utf8' }))[0];
const files = meta.files.map((f) => f.path);
const need = [
'cordis.patch.yml', 'README.md', 'README.zh-CN.md', 'LICENSE', 'CHANGELOG.md', 'check-updates.ps1',
'cordis.patch.yml', 'index.mjs', 'README.md', 'README.zh-CN.md', 'LICENSE', 'CHANGELOG.md', 'check-updates.ps1',
'screenshots.json', 'assets/screenshot-1.png', 'assets/screenshot-2.png',
'examples/cordis.patch.posix.yml', 'examples/cordis.patch.with-env.yml',
];
const missing = need.filter((f) => !files.includes(f));
if (missing.length) {
@@ -77,7 +142,7 @@ jobs:
'tag:yaml.org,2002:js',
lambda loader, node: loader.construct_scalar(node),
)
files = ['cordis.patch.yml', 'docs/marketplace-entry.yml'] + sorted(glob.glob('examples/*.yml')) + sorted(glob.glob('.github/workflows/*.yml'))
files = ['cordis.patch.yml', 'docs/marketplace-entry.yml'] + sorted(glob.glob('.github/workflows/*.yml'))
for f in files:
with open(f, encoding='utf-8') as fh:
yaml.safe_load(fh)
@@ -87,10 +152,9 @@ jobs:
- name: The shipped patches stay machine-independent
run: |
set -e
# The row DSH loads, plus every example a reader is told to copy: none of
# them may carry a real machine path. `<placeholders in angle brackets>`
# are the intended form.
targets=$(printf '%s\n' cordis.patch.yml examples/*.yml)
# The row DSH loads may carry no real machine path; `<placeholders in angle
# brackets>` are the intended form.
targets=$(printf '%s\n' cordis.patch.yml)
if grep -nE '[A-Za-z]:\\' $targets; then
echo "FAIL: a shipped patch contains an absolute Windows path"
exit 1
@@ -163,3 +227,27 @@ jobs:
# Checksum coverage, unresolved tokens, duplicate JSON keys, and the
# patch-wiring agreement between the row, the shim and setup.mjs.
run: npm run bundle:audit
windows:
# The PowerShell download route, the tar/Expand-Archive extraction path and
# scripts/rea-bundle-bootstrap.ps1 are Windows-only code, and an ubuntu-only
# matrix never executed any of them — they were verified by hand, if at all.
# The tests are platform-sensitive on purpose: each platform exercises the
# download route it actually chooses.
runs-on: windows-latest
# Same pin as the ubuntu job: the launch decision must not depend on what the
# runner image happens to have installed.
env:
REA_BUNDLE_LAUNCH: bundle
steps:
- uses: actions/checkout@v4
- uses: actions/setup-node@v4
with:
node-version: '22'
- name: Unit tests for the pure logic (Windows)
run: npm run bundle:test
- name: Bundle manifest audit (Windows)
run: npm run bundle:audit
+314
View File
@@ -2,6 +2,320 @@
本插件遵循语义化版本。版本号变化说明「配置组合」这层的变化,与 REA 自身的版本无关。
## 2.0.9 — 2026-10-18
**2.0.8 的两处发布前问题(假回退、乱码)与若干文档缺口;不改正常路径,只让失败路径说真话。**
### 修复
- **`reexecFromCheckout` 的「回退」是假话**:spawn 失败时只打印「继续用镜像副本」,函数却返回了,`main()` 已经
return —— 没有 `exit` 事件、事件循环排空、进程以 **exit 0 静默退出**,行看起来启动成功却没有任何工具。现在这个
Promise **只在没接管时才 resolve**:子进程活着时本进程等待、由子进程的退出码决定;失败则 `main()` 继续用本地副本
启动,并如实说明在回退。
- **回退覆盖了升级竞态的主形态**:`spawn` 的目标是 `process.execPath`,spawn 级失败几乎只剩资源耗尽;真正常见的是
「哈希比对通过 → node 启动 → 文件已被换掉」→ node 以 `MODULE_NOT_FOUND` 退 1。现在**退出码非 0 且源文件已不存在**
时同样回退到本地副本(窗口毫秒级,代价是一次 `stat`),不再把竞态变成一行死掉的 MCP。
- **乱码 `鈥?` 混进用户可见输出**:2.0.7 在 GBK 编辑器里碰过 `bin/rea-shim.mjs`,16 处破折号被写成 `鈥?`,其中两条是
stderr 消息(首次自动供给、上次供给失败)—— 每个用户都会看到。全部还原为 `— `(含被吃掉的空格)。
- **开发时可能测到「已安装插件」而不是自己的改动**:checkout 副本会顺着 bundle 的 `mirror.json` 转发到插件目录。
现在 checkout(带 `cordis.patch.yml`)永不转发,只有镜像副本会;注释也改成与行为一致。
- **带 BOM 的 `mirror.json` / `bundle.json` 被当成「没有记录」**:`JSON.parse` 拒绝前导 BOM,而每个调用方都把解析失败
读成「无记录」。Windows 编辑器保存的 JSON 就会带 BOM,表现是镜像自愈静默不生效、bundle 被当成未供给。`bin/rea-shim.mjs`
与 `scripts/setup.mjs` 现在都经各自的 `readJson()` 去掉 `U+FEFF` 再解析。**两个文件读的是同一份记录,所以缺一不可**:
此前只有 shim 剥了 BOM,带 BOM 的 `mirror.json` 在 setup 眼里仍是「没有 checkout」,镜像 layout 下的 setup 于是把上一版
manifest 又镜了一遍并从它供给;`bundle.json` 同理被读成未供给。`verify-branch.ps1` 与 CI 都加了守卫:这两个文件里
出现裸的 `JSON.parse(readFileSync(...))` 即失败。
### 文档 / CI
- 两份 README 的钉版本示例改为**当前版本** `#v2.0.9`(2.0.8 那一轮已从 `#v2.0.0` 提到 `#v2.0.8`,版本号一动就得跟上,
否则读者拿到的 ref 不存在);README 的「机器上已有 REA」一节补上 Windows 限制:**强制结束 shim 不会带走被委派的 REA**。
- `docs/bundle-design.md`(随包发布)补上 2.0.7 的 launch mode 与 2.0.8 的镜像自愈,含 `mirror.json` 的
`shimSha256` / `pluginVersion` / `dshHome` 三个字段与控制流图。
- CI 新增两条:README 必须逐字引用 `engines.dsh`;README 的钉版本示例必须等于 `package.json` 的 version(`verify-branch.ps1`
同步加了第二条,避免同一个漂移第三次发生)。
- `systemReaCandidates` 的去重键改用传入的 `platform` 参数(测试缝一致,生产无变化)。
### 验证
- 单测 **71 条**(68 过 + 3 平台跳过;新增「checkout 副本不转发」「文件中途消失 → 回退」「带 BOM 的 `mirror.json`」
「带 BOM 的 `bundle.json`」四条);镜像转发测试改为真的运行镜像副本。
- `verify-branch.ps1` 44 项全绿;`audit-manifest` clean;`host-compat` 10 covered / 5 outside。
## 2.0.8 — 2026-10-18
**2.0.7 的启动检测在两种边界上会失控:一个包着本 shim 的 `rea` 会让探针递归成进程链;超时探针会漏掉一个孙进程。两条都已修,并补了复现测试。**
### 修复
- **重复注册检测在自定义 `DSH_HOME` 下失效**:harness 会从 MCP 子进程里抹掉所有 `DSH_*`,镜像副本因此读
`~/.dsh` —— 那是错的文件,上游写进真实 home 的第二条 `serverName: rea` 永远报不出来。`setup.mjs` 现在把运行时
的 home 记进 `mirror.json` 的 `dshHome`,shim 优先读环境、其次读这条记录。`check-updates.ps1` 的 profile 段也
改用同一条 home 规则(此前硬编码 `$env:USERPROFILE`,自定义 home 时报「找不到 profile」)。
- **镜像启动器不会自己更新**:行执行的是 `<DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs` 这份**副本**,而市场升级
只换包、不跑 setup —— 副本会一直跑它被复制时的代码,于是新版本里的修复对老用户静默失效(旧副本照常工作,最难发现)。
现在镜像启动时会与 `mirror.json` 记录的插件目录比对哈希,不同就转发执行插件里的那一份(一跳,带
`REA_BUNDLE_SHIM_REEXEC` 守卫,且只从镜像指向插件,不会反向)。`--shim-status` 新增 `mirror.stale`,
`check-updates.ps1` 也会报「镜像落后,重跑 setup」;`mirror.json` 增加 `shimSha256` 与 `pluginVersion` 便于追溯。
**2.0.8 之前的版本自身无法完成这一步**(副本里没有这段代码),从那些版本升级仍需手动跑一次 setup。
- **探针递归**:`probeRea` 派生候选时没带 `REA_BUNDLE_DELEGATED=1`,而防自委派的守卫只在**委派**前生效。若 PATH 上是「调用本 shim 的 wrapper」,
探针 → wrapper → 新 shim → 再探针……实测并发 node 峰值 83+、杀掉后仍自我再生。现在探针环境同时带 `REA_BUNDLE_DELEGATED=1` 与
`REA_BUNDLE_AUTO_PROVISION=0`:嵌套运行直接走 bundle 并返回,实测 0.2 秒、零额外进程。
- **超时探针泄漏孙进程**:`spawnSync` 的超时只杀它自己启动的 cmd.exe,而 `taskkill /T` 在 `spawnSync` 回收之后调用已找不到父进程 ——
每次超时留一个 REA/ping 在跑。探针改为异步 `spawn` + 自己的计时器:超时发生在父进程仍存活时,`taskkill /T /F` 能真正清掉整棵树。
实测 8.3 秒放弃、`rejected=["timeout"]`、无残留进程。
- **Windows 信号处理器是死代码**:`child.kill("SIGTERM")` 在 Windows 上是 TerminateProcess,不可捕获;控制台的 Ctrl+C 又直达子进程。
该分支已删除,并如实写明限制:**在 Windows 上强制结束 shim 不会带走被委派的 REA**(Node 没有 job object 可用)。
- **`comspec()` 不校验**:`ComSpec` 指向不存在的文件时,正常的 `.cmd` REA 被判 `spawn-failed`(`auto` 静默退回 bundle,`system` 直接 78)。
现在校验为常规文件,否则回落到 `%SystemRoot%\System32\cmd.exe`;实测 `ComSpec` 指向不存在路径仍能识别 6.9.0。
- **文案**:显式路径是目录时报「does not exist」,改为「is not a file」。
### 验证
- 单测 **67 条**(2.0.7 是 60 条,本轮 +7:wrapper 递归必须被判为 `exit 78` 且 20 秒内收敛;`ComSpec` 失效仍识别 `.cmd` REA;
镜像自愈的转发/守卫/`stale` 上报/同内容不转发;`DSH_HOME` 被抹时从 `mirror.json` 找到真实 home 并报出重复注册):
Windows 64 过 + 3 平台跳过。
- `verify-branch.ps1` 全绿(39 项),含「两份 README 必须写出 package.json 里的 `engines.dsh` 原文」;`audit-manifest` clean;
`host-compat` 10 covered / 5 correctly outside。
- 手工复现三例:wrapper 递归 0.2 秒收敛、0 残留进程;挂起候选 8.3 秒放弃且无 `PING` 残留;`ComSpec` 指向空路径仍识别系统 REA。
## 2.0.7 — 2026-10-18
**机器上已有可用的 REA 时,2.x 直接按 1.x 的方式跑它(环境检测 + 委派),不再供给;同时把 2.0.6 的宿主声明收紧到真的能证明的范围。**
### 新增:环境检测 → 跑 1.x 的那条路
- **启动时判断该跑谁**:`rea` 在 PATH 上且能应答 `--version`,就用它(`rea mcp`,用户自己的 Node 与 PATH);否则走供给的
bundle。这正是 1.x 的契约,现在从 2.x 的行里也能到。
- **版本优先级**:系统 REA 比供给的旧就不抢(6.4.0 是 170 个工具、6.5.0 是 207 个,静默回退工具数是真回归)。相等或更新
才切;预发布/无法比较时保守留 bundle。
- **覆盖**:`REA_BUNDLE_LAUNCH=auto|system|bundle`(默认 auto)、`REA_BUNDLE_SYSTEM_REA=<path>` 点名不在 PATH 上的二进制。
`system` 在找不到时**报错退出**,而不是替用户去下 800 MB。
- **`--shim-status` 报告决定**:`launch.{preference,mode,reason,unusable,systemRea,rejected}`;`--shim-print-env` 在 system
模式下打印 `{launch,rea,version}`。
- **防自委派**:被委派的子进程带 `REA_BUNDLE_DELEGATED=1`,一个调用本 shim 的 wrapper 不会让委派无限递归。
### 修复(两轮独立复核,逐条可复现)
- **委派参数被加引号**:`cmd /c` 下 `.cmd` 的 `%*` 收到的是 `"mcp"` 而不是 `mcp` —— 探针能过、委派是坏的。现在只给路径加
引号,参数原样传(行固定传 `mcp`;需要 shell 引号的参数本就不在这条路上)。
- **探测预算 3s 与真机同量级**:实测经 `.cmd` 的 `rea --version` 为 2.4s(并发 2.8–3.0s),3s 预算等于掷硬币。改为 8s;超时
用 `taskkill /T /F` 杀进程树,不再只杀 cmd.exe 而留下一个占着管道的 REA。
- **Windows 信号转发是错的**:`child.kill("SIGINT")` 只 TerminateProcess(cmd.exe),里面的 REA 继续跑;现在 Windows 走
`taskkill /T`,POSIX 不重复转发(stdio inherit 已由进程组投递)。
- **`--shim-status` 退出码**:`system` 被点名但不可用时曾返回 0(bundle 恰好已供给),实际启动却 78;现在按「这一行能不能
起来」返回,并新增 `launch.unusable` 字段。
- **候选判定**:`existsSync` 换成 `statSync().isFile()`(名为 `rea.cmd` 的目录曾被当成坏安装);PATH 项去引号、跳过相对项;
显式路径被拒时理由如实报告。
- **`fail()` 的补救命令补上 `--root`**:shim 用 `REA_BUNDLE_ROOT` 解析出的根与 setup 的默认根可以不同,照着旧提示执行会供给
到另一个根。
- **`engine.dsh` 上限收紧**:`<0.3.0` 在 includePrerelease 语义下会放行 `0.3.0-alpha.1`/`0.3.0-rc.1`;改为 `<0.3.0-0`,并用
市场自己的 `satisfiesRange` 逐版本核对。
- **`host-compat.mjs` 的测试加固**:发布清单逐项钉住;新增「低地板必须被判为泄漏」的反向对照(原实现把 `leaked` 改成 `[]`
仍然全绿);`+build` 元数据按 semver 接受、`v` 前缀与零填充预发布按 semver 拒绝。
- **环境独立性**:`setup.mjs` 自检的 shim 步骤钉 `bundle` 模式(否则机器上更新的 REA 会冒充本次供给的产物);CI 两个 job
各加 job 级 `REA_BUNDLE_LAUNCH=bundle`;测试里的每个 shim 调用点都钉住模式,测试不再随机器变。
- **`rea-bundle-bootstrap.ps1` 认 `REA_BUNDLE_ROOT`**:四份代码里只有它忽略这个覆盖,导致 bootstrap 把 800 MB 装到
`<DSH_HOME>/plugins/rea-bundle`,而 shim/setup/check-updates 去 `REA_BUNDLE_ROOT` 找,看到空 bundle 再起一次。
- 中文 README 的「更新到 2.0.4」措辞改正为「更新到 2.0.5」。
### 验证
- 单测 **60 条**(新增 12 条:6 条启动模式 + 1 条清单钉住 + 1 条泄漏反向对照 + 1 条 semver 容差 + 3 条 PATH/空格路径/委派退出码),
Windows 57 过 + 3 平台跳过。
- `verify-branch.ps1` 全绿;`audit-manifest.mjs` clean;`host-compat.mjs` 10 covered / 5 correctly outside。
- 真机:全局 `rea` 6.4.0 与供给的 6.5.0 并存时 `auto` 留 bundle(不降工具数),`REA_BUNDLE_LAUNCH=system` 才切过去。
## 2.0.6 — 2026-10-18
**声明宿主版本:`engines.dsh` 覆盖 `0.1.7-rc.2` … `0.2.1-rc.2` 全部发布版本,市场卡片因此显示「兼容」而不是「未声明」。**
### 兼容声明
- **`engines.dsh: ">=0.1.7-rc.2 <0.3.0"`**:市场侧的 `deriveHostCompatibility` 只认两类声明 —— `engines.dsh`
与指向 `@deepseek-ai/dsh*` 的 `peerDependencies`;两者都没有就是 `unknown`(卡片「兼容性未知」)。
这里只声明前者,刻意不声明 peer:`@deepseek-ai/dsh-mcp-client` 在 npm 上只有 `0.0.1-rc.1`,
而宿主随包内置的是 0.2.x,声明 peer 会让 pnpm 解析出一个与宿主不一致的版本,反而制造假冲突。
- **范围来自核对,不是估计**:`scripts/host-compat.mjs` 内置 dsh 发布列表(0.1.7-rc.2 至 0.2.1-rc.2 共 10 个),
逐个验证是否被声明覆盖,并反向确认 `0.3.0` 与 `0.1.7-rc.1`、`0.1.7-alpha.2` 落在范围之外。
- **跨版本比对过的东西**:`packages/mcp/mcp-client` 在 `dsh-v0.1.7-rc.2` 与 `dsh-v0.2.1-rc.2` 两个 tag 上,
配置面与发布的行一致(`serverName`/`transport`/`command`/`args`/`env`/`failOnStartupError` 均存在且同名),
且两个 tag 都有 `dsh.bundle.patch`。范围声明的是「核对过的宿主」,不是「应该能用」。
### 上游(只更新仓库,暂不带入)
- 上游已发布 **rea-agents 6.6.0**(GitHub release `rea-agents-6.6.0`,2026-10-18);npm `latest` 仍是 **6.5.0**,
所以本包的钉(`manifest/bundle.manifest.json` 的 `pins.reaAgents` 与 `components.reaAgents.sources.npm.version`)
**保持 6.5.0 不动**。已核 6.6.0 的 `engines.node`(`^22.19.0 || ^24.11.0 || >=26.0.0`)与 `bin`
(`scripts/rea.mjs`)均未变化,`optionalDependencies.frida` 为 17.23.2。
- 计划:**2026-10-19** 随下一版把钉升到 6.6.0 并复跑真机;届时工具数、探针正则与文档一并更新。
### 修复 / 工程
- 新增 `scripts/host-compat.mjs`(纯函数 + CLI):semver 预发布排序、`>=floor <ceiling` 判定、发布列表覆盖检查。
- `bundle:host-compat` npm script;`verify-branch.ps1` 新增 `engines.dsh` 存在性与覆盖率两项;
`validate.yml` 新增「Declared DSH range covers every released host」步骤,清单检查也要求 `engines.dsh`。
### 验证
- 单测 **46 条**(新增 4 条:声明可解析、10 个发布版本全覆盖且 3 个未核对版本被排除、反向对照能抓出
`>=0.2.0` 这种看似合理却漏掉整个 0.2.0-rc 线的写法、预发布排序符合 semver 而非字符串序):Windows 43 过 + 3 平台跳过。
- `node scripts/host-compat.mjs`:10 个发布版本 covered、3 个 correctly outside。
- `verify-branch.ps1` 全绿;`audit-manifest.mjs` clean(两份 README 结构仍平行)。
## 2.0.5 — 2026-10-18
**市场更新失败已定位并修复:清单里没有 `main`,市场因此在每次更新时判「新构建没有可加载入口」并回滚。**
### 修复
- **应用内市场更新到 2.0.4 失败**:`dshmarket` 的更新路径用 `hasLoadableEntry` 判断新构建能不能加载
(实现是 `entryArtifactExists`:读 `main`/`exports`,两者都没有就回落 `index.js`)。本包是纯 bundle
(只有 `dsh.bundle.patch`),从未声明 `main`;而它挂载的 `@deepseek-ai/dsh-mcp-client` 属于 harness 自身
构建、不在 `node_modules` 里,于是该检查要看的 profile / 包内 / workspace 三个位置全空。结果:2.0.1 → 2.0.4
的更新连续三次报「更新后缺少入口文件……且未能验证恢复原版本文件」,而回滚其实成功了 —— 旧版同样没有入口,
所以回滚后的复检依然为 false。现已声明 `main: ./index.mjs`,并让该文件进入 `files[]`。
- **`index.mjs` 是真空插件,不是占位**:导出 `name` 与空 `apply()`,不注册行、不起进程、不写状态;真正起作用的
行仍由 `cordis.patch.yml` 插入。若 harness 某天真把它当插件加载,行为仍是「什么都不做」,因此不会造出第二条
`serverName: rea`。
### 文档
- 两份 README 新增「在市场里更新 / Updating from the market」小节:说明该报错来自市场侧的入口判定、与 tarball
完整性无关、profile 未受损,以及 2.0.5 起的做法。
### 验证
- 用 `dshmarket` 自身的 `hasLoadableEntry` 实测:把本包按 profile 布局模拟安装,修复前 `false`、修复后 `true`;
对仓库目录 `entryArtifactExists` 修复后为 `true`。
- 单测 42 条(新增 2 条:清单声明的入口必须存在且被 `files[]` 收录;入口可 `import`、导出的 `name` 与包名一致、
`apply()` 返回 `undefined`):Windows 39 过 + 3 平台跳过。
- `verify-branch.ps1` 新增 `main` 存在性与 `files[]` 收录检查;`validate.yml` 的清单检查与 npm 载荷检查加了同一条守卫。
- `npm pack --dry-run`:20 个文件,含 `index.mjs`。
## 2.0.4 — 2026-10-17
**两处已核实的缺陷:只读 bundle 目录下 shim 启动即崩;与上游 `rea setup --client dsh` 的同名注册冲突现在会被明确提示。**
### 修复
- **只读 bundle 根目录 + 陈旧 lock → shim 未捕获 EACCES 崩溃**:2.0.3 把删锁移出了 `if (fresh)`,却没包
try/catch。目录只读(`chmod 555`、系统级安装目录)且锁已陈旧时,`rmSync(lock)` 抛 `EACCES`,shim 以
未捕获异常退出(exit 1 + 堆栈)—— 比原来的静默更吵,但同样起不来。沙箱已复现(WSL `chmod 555`)。
现在删锁失败返回 `unwritable`,走既有的「无法写入,请手动跑 setup」路径(exit 78,行不崩)。
- **与上游 `rea setup --client dsh` 的同名注册冲突**:该命令往 `<DSH_HOME>/cordis.patch.yml` 写入
`id: rea-mcp`、`serverName: rea`,而本包是 `id: mcp-rea`、同一个 `serverName`;DSH 对同名条目只保留
一个(后加载者报 duplicate server name)。已按上游 6.5.0 实测生成该文件确认。shim 现在启动时检测同
文件的 `serverName: rea`(id 不是 `mcp-rea`)并向 stderr 提示「二选一」,`--shim-status` 也会在
`duplicateRegistration` 下报告。
### 文档
- 两份 README 新增「只保留一条 REA 注册」小节:与 `rea setup --client dsh` 二选一,并说明两条路线的
取舍(零依赖供给 vs 上游官方纯配置注册)。
- v1 分支 README 顶部加横幅:指向 2.x(`main`,默认推荐)与上游官方注册。
### 验证
- 单测 40 条:Windows 37 过 + 3 平台跳过;WSL 39 过 + 1 跳过(本轮新增 3 条,其中只读目录一条仅 POSIX)。
- 反向验证:回退删锁的 try/catch → 只读目录测试在 WSL 必红(EACCES/exit 1);回退冲突检测 → 重复注册
测试必红。
- `verify-branch.ps1` 全绿。
## 2.0.3 — 2026-10-16
**修复 11 处已核实的缺陷(每条都先确认存在、并配了"回退即变红"的回归测试);上游 REA 钉子从 6.4.0 升到 6.5.0。**
### 上游
- `pins.reaAgents` 与 `components.reaAgents.sources.npm.version` → **6.5.0**:engines 与 bin 入口未变
(`^22.19.0 || ^24.11.0 || >=26.0.0`、`scripts/rea.mjs`),工具数 170 → **207**(运行时 `rea doctor --json` 实测;上游源码静态清点为 206,多出的 1 个注册在 contracts 之外。新增:Frida 9、
GDB 4、Rizin/Cutter 6、pcap/packet 4、flutter 2、sqlite/rust/go/wasm/electron/harmony/hopper 各 1 等)。
新增的 `optionalDependencies.frida` 在 `--ignore-scripts` 下不会执行安装脚本。REA 探针正则收紧为
`^6\.5\.0(\s|$)`,审计新增"下一个 patch/minor 不得通过"的断言。
- Ghidra 12.1.4 / apktool 3.0.3 / jadx-headless-mcp 0.7.1 **经核实仍是当前最新**;Node 24.22.0 虽已发布,
但晚于本插件的实测矩阵,故保持 24.21.0(engines 完全满足)。
### 修复
- **`--only` 抹掉其它组件的记录**:bundle.json 由本轮结果整份重写,未点名组件的 `paths.jdk/ghidra` 变
`null`,shim 不再导出 `JAVA_HOME`/`GHIDRA_INSTALL_DIR`——原生分析静默失效,下次完整跑还得重新解压
750 MB。未点名的组件现在按旧记录结转(且只在记录的路径仍然存在时),缺依赖时优先用旧记录/机器/运行
setup 的解释器来满足,而不是整段跳过检查。
- **完整的 `.part` 不被当作已完成**:只打一行日志,然后发 `Range: bytes=<全长>-`(416;四次重试后把
已经下好的文件报成失败)。现在校验通过即改名收下。
- **Windows 下载主路线没有续传**:PowerShell 路线每轮先删 `.part`;失败后 Node 兜底仍用进本轮前的
`have`,以 append 把 200 全量接在半成品后面(长度必然超标,只能重下)。两条路线现在都按盘上实际
长度发 Range 续传。
- **POSIX 的 apktool wrapper 没有执行位**(copyFileSync 落 0644,而 REA 不加 shell 直接 spawn → EACCES):
manifest 里标了 `executable` 的条目落盘时 chmod 0755。
- **apktool wrapper 钉在 `master` 分支 + 硬校验 sha256**:上游对 wrapper 的任何一次提交都会让全网新装
hash mismatch。改钉不可变的 `v3.0.3` tag,哈希按该 tag 重新实测。
- **`--reuse` 不校验版本**:JDK 只看 `javac` 存在、Ghidra 只看 `analyzeHeadless` 存在(JDK 8/11、Ghidra 11
照收)。现在候选必须通过该组件自己的 probe;同时修好"文件型 probe 从不检查 `expect`"(Ghidra 的
`application.version=` 断言此前是死配置)。
- **`bundle.json` 损坏即永久失败**:setup 读不了它、也没人启动 setup,用户只能手删文件。现在 setup 容忍
它(并按需重写),shim 对 `malformed` 与 `missing` 一样触发供给。
- **失败后 shim 谎报 6 小时**:lock 现在记录状态与 pid;失败态如实报告"上一次失败了",持有者已消失的
lock 不再阻塞重试。
- **成功的 `--json` 轮次从不删 lock**:删除语句原先在非 `--json` 分支里,--json 成功也留下 lock。
- **shim 兜底 REA 路径少了 `node_modules`**:旧 bundle.json 缺 `paths.reaEntry` 时会把可用安装判成"CLI 丢了"。
- **`scripts/rea-bundle-bootstrap.ps1`**:`node -v` 输出非数字时 `[int]` 强转在 `$ErrorActionPreference='Stop'`
下会炸掉整个 bootstrap(而它存在的意义就是这条路径);`&` 调用失败时 `$LASTEXITCODE` 可能为 `$null` →
`exit $null` = 假成功;engines 改从 manifest 的 `pins.nodeEngines` 读(不再抄第二份);`-DryRun` 不再创建
目录(它承诺"什么都没改")。
- **`check-updates.ps1` 的 1.x 残留**:REA 节默认找 npm 全局包(2.x 是 vendored)、Ghidra 节读用户级
`GHIDRA_INSTALL_DIR`、`rea doctor` 拿不到 vendored CLI(整段自检被静默跳过)、一体包对账对"未记录版本"
的组件静默 `continue` 却打印"全部一致"。现在一律先读 `bundle.json`,并如实报告无法对账的项。
### 审计与 CI
- `audit-manifest.mjs` 新增 4 项:manifest 组件 ⇄ `INSTALL_ORDER` 对账、`pins.nodeEngines` ⇄ `package.json
engines`、REA pin 探针不得放行下一个 patch/minor、POSIX wrapper 必须标 `executable`。
- `validate.yml` 新增 `windows-latest` job:PowerShell 下载路线、zip 解压与 bootstrap 此前是零 CI 覆盖的
Windows 专属代码。
- 单测 33 条(32 过 + 1 条按平台跳过);`--reuse adopts a compatible node` 在 engines 之外的宿主上改为显式
skip(此前在 Node 20 上必红,而 CI 钉 22 看不出来)。
### 第二轮修复(同一份**更新后**的清单,2026-10-16 晚)
- 🔴 **过期 lock 永久卡死自动供给(2.0.0–2.0.2 三版全带)**:`maybeAutoProvision()` 只在 lock 新鲜时
处理它,过期的 lock 既不清除也不报告 → 落到独占创建 → `EEXIST` → 返回 `in-flight`,而 `main()` 对
`in-flight` **一个字都不打印**。于是「下载被 kill / 断电」这一常态,在 6 小时后再启动就变成**永久不再
自动供给且完全静默**。已按清单复现(7 小时死 lock:不启动、lock 原样、无任何提示;5 分钟 lock 对照组
正常删锁重启),修法:**不新鲜或持有者已死就一律清锁**(这正是注释里"只在新鲜时尊重它"的本意),
并给 `in-flight` / `spawn-failed` / `unwritable` 补上如实提示。
- 🔴 **`examples/` 两个文件是 1.x 内容,却随 2.x 进了 npm 包**:`cordis.patch.posix.yml` 教人把整行换成
`command: rea`、`cordis.patch.with-env.yml` 给的是 `cmd.exe /c rea` 的 1.x 行——对 2.x 全错(2.x 的行是
跨平台的 `process.execPath` + bundle shim,照抄会整行绕过 bundle)。两份都已是孤儿(README 不再引用),
故从 `main` **删除**,并同步 `package.json files[]` 与 `validate.yml` 的载荷/YAML/机器路径三项检查
(v1 分支保留它们,在 1.x 语境里是对的)。npm 载荷由此 21 → 19 个文件。
- 🟡 **`--offline` 管不住 `reaAgents`**:`provisionReaAgents()` 里 0 处 offline 判断,`npm install` 永远
联网——与「`--offline` is a promise, not a hint」直接矛盾。现在 `--offline` 且内置副本不可用时**直接
FAIL** 并说明原因(而可用的内置副本仍照常跳过)。
- 🟡 **npm 那一步没有代理/镜像兜底**:压缩包有 direct→proxy→PowerShell 三路,npm 只读环境变量。失败时
现在会打印这句提示,两份 README 的能力/注意里也写明 `HTTPS_PROXY` / `NPM_CONFIG_REGISTRY`。
- 🟡 **POSIX 缺 `unzip`/`xz` 必现失败**:Ghidra 全平台只有 zip,而 POSIX 解压原先只有 `unzip` 一条路。
现在 zip 依次尝试 `unzip` → `python3 -m zipfile` → **已供给 JDK 自带的 `jar`**(JDK 在 Ghidra 之前供给,
等于保底),tar 依次 `tar` → `python3`;三者皆无时打印明确的「no zip extractor available」而不是干巴巴的
「extraction failed」。
- 🟡 **manifest 的 arm64 注释是错的**:Google 的 linux/darwin platform-tools 都是 x86_64 二进制,
「The same archive serves x86-64 and arm64」不成立。三个条目改成如实描述(macOS 需 Rosetta 2、
linux-arm64 上 probe 预期失败并被报为可选组件不可用)。
- 🟡 **`docs/bundle-design.md` 与实现漂移**:§3 的「row points at `<plugin>/bin/rea-shim.mjs`」改成
「按 `DSH_HOME` 算出 `<DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs`(镜像副本)」;§7 的
`vendor/rea-agents` 改成 `vendor/node_modules/rea-agents`。
- 🟢 **`release.yml` 的 `workflow_dispatch` 可绕过 Release 直接发 npm**:publish job 现在先断言
`v<package.json version>` 的 tag 在 origin 上存在、且该 Release 挂着版本无关的 `dsh-plugin-rea.tgz`
资产,否则拒绝发布(否则 npm 与市场兜底链接会岔开)。
- 🟢 **数值环境变量会变成 NaN**:`REA_BUNDLE_FETCH_RETRIES=abc` → `Math.max(1, NaN)` → 报
「gave up after NaN attempt(s)」;NaN 的 socket 超时等价于 0(立即超时)。现在非正有限值一律回落到默认,
且启动日志新增一行 `limits connect=… idle=… retries=…`,让覆盖生效可见。
- 🟢 注释与整洁:`downloadViaPowerShell` 的文档说自己是 fallback,实际在 Windows 上是主路线(已改口径);
去掉 setup.mjs 里 3 处对已静态导入成员的重复 `await import("node:fs")`(`readdirSync` 补进静态导入)。
## 2.0.2 — 2026-10-15
**仅文档,无行为变更。**
+118 -8
View File
@@ -82,6 +82,10 @@ Both lines ship the **same patch shape** — one `insert` row of `@deepseek-ai/d
same `mcp__rea__*` surface; they differ in who prepares the runtime. Moving from 1.x to 2.x takes two
commands: see [Upgrading from 1.x](#upgrading-from-1x).
Upstream also registers itself with dsh now (`rea setup --client dsh`); that route is essentially the
1.x model maintained upstream — configuration only, your own Node/REA/Ghidra. This bundle is the
zero-dependency route. Pick one of the three, and never keep two `serverName: rea` rows at once.
## Install
**Nothing to install first.** On a bare Windows x64 machine:
@@ -96,7 +100,7 @@ releases, checksum-verified). Then install the plugin for your profile:
```
dsh plugin --profile <profile> add dsh-plugin-rea
# or pin this release instead of tracking npm latest:
dsh plugin --profile <profile> add github:czcdr/dsh-plugin-rea#v2.0.0
dsh plugin --profile <profile> add github:czcdr/dsh-plugin-rea#v2.0.9
```
> **On an Electron host (the DeepSeek Harness desktop app) the profile is owned by the application.**
@@ -110,6 +114,10 @@ If Node is already installed, you can skip the bootstrap and run provisioning di
node scripts/setup.mjs --reuse # or --dry-run to see the plan first
```
Node, when you provide it instead of letting setup fetch one, must satisfy REA's engines —
`^22.19.0 || ^24.11.0 || >=26.0.0` (check with `node -v`). A lower Node does not fail at install
time; the MCP row fails later, which reads as silence.
Installing the plugin **first** also works, but it takes two steps. The row resolves to
`<DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs`, and that file is created by setup — so until the first
provisioning run the row exits with `MODULE_NOT_FOUND`, and nothing is there to start a background
@@ -117,6 +125,46 @@ provisioning (the harness itself is unaffected: `failOnStartupError: false` is d
`node scripts/setup.mjs` once; from then on the shim is in place, and every later missing state is picked up
by the shim itself, which provisions in the background.
### Which DSH versions
`package.json` declares `engines.dsh: ">=0.1.7-rc.2 <0.3.0-0"`, and that declaration is the answer the plugin market
reads: with it the card says **compatible**; without it the market can only say *unknown*. The range is not a guess —
it is the dsh release list this plugin is checked against, and `node scripts/host-compat.mjs` fails the build when the
two disagree:
| Host | Covered |
|---|---|
| `0.1.7-rc.2` … `0.2.1-rc.2` — every release in between, alpha and rc | yes |
| `0.3.0` and later | no — a new minor line is a host nobody updated this row for yet |
What was actually compared across that range is the row provider: `packages/mcp/mcp-client` at `dsh-v0.1.7-rc.2` and at
`dsh-v0.2.1-rc.2` exposes the same config surface the shipped row uses (`serverName`, `transport`, `command`, `args`,
`env`, `failOnStartupError`), and `dsh.bundle.patch` exists at both tags. A host outside the range can still work; it is
simply not claimed.
### A REA you already have is used as-is
2.x exists to supply what a machine is missing. If REA is already installed — `rea` on PATH, the 1.x setup — the row
runs **that**, exactly as the 1.x line did, and nothing is provisioned:
| What the machine has | What the row runs |
|---|---|
| `rea` on PATH, at least as new as the provisioned build | your REA (`rea mcp`), with your Node and your PATH |
| `rea` on PATH but older than the provisioned build | the provisioned bundle — a stale global CLI would take the tool count backwards (6.4.0 ships 170 tools, 6.5.0 ships 207) |
| no `rea` on PATH | the provisioned bundle, provisioning itself on first use |
- `REA_BUNDLE_LAUNCH=auto|system|bundle` overrides the decision (`auto` is the default); `REA_BUNDLE_SYSTEM_REA=<path>`
names a binary that is not on PATH.
- The probe is the CLI's own `--version`, bounded at 8 s: a `.cmd` shim on Windows takes ~2.4 s to answer, so the check
runs once per start and only when a `rea` actually exists.
- `--shim-status` prints the decision — `launch.mode`, `launch.reason`, `launch.systemRea` — and its `launch.rejected`
list says why a candidate was skipped (`"exit 7"`, `"inside the bundle"`, `"timeout"`).
- A delegated child carries `REA_BUNDLE_DELEGATED=1`, so a wrapper that calls this shim cannot make it delegate to
itself forever.
- **On Windows a forced kill of the shim does not reap the delegated REA.** There is no job object here: Task Manager
or a supervisor terminating the shim leaves your `rea` running with the row's pipes (a console Ctrl+C reaches the
whole process group and does stop it). On POSIX the child is in the same group and dies with the terminal.
## Configure — nothing
The shipped row carries **no machine path**:
@@ -168,6 +216,27 @@ To pin a different launch command on this machine (for example your own globally
the row from the profile's own `cordis.patch.yml` — that file is machine-local and never published. Only one
row may use `serverName: rea` in a profile.
### One REA registration only
Upstream REA 6.5.0 can register itself with DSH:
```
rea setup --client dsh
```
That writes a row with `id: rea-mcp` and `serverName: rea` into `<DSH_HOME>/cordis.patch.yml`. This
bundle's row is `id: mcp-rea` with the same `serverName`, and DSH rejects the later of two entries
that share a server name — so which REA row survives depends on load order. **Use one, not both.**
Which one? Upstream's is the configuration-only route: it expects Node on REA's engines,
`rea-agents` reachable, and a Ghidra/JDK you provisioned yourself. This bundle is the
zero-dependency route: it fetches Node, REA, JDK and Ghidra on first run. The patch shape and the
`mcp__rea__*` surface are the same either way.
The shim reports the conflict on startup — `setup.mjs` records the DSH home it ran under in
`mirror.json`, and the mirrored copy reads that when the harness has stripped `DSH_*` from its
environment; `node bin/rea-shim.mjs --shim-status` lists it under `duplicateRegistration`.
## Verify
Ask for a `mcp__rea__*` call — `binary_session` reports the engine checks
@@ -179,11 +248,11 @@ exact shim launch the MCP row will use.
## Verified against
Author's machine, 2026-10-13 / 14:
Author's machine, 2026-10-13 / 14 / 16:
| Component | Version |
|---|---|
| REA (`rea-agents`) | **6.4.0** (170 tools) |
| REA (`rea-agents`) | **6.5.0** (207 tools) |
| DeepSeek Harness | desktop runtime **0.2.0-rc.2**, with its bundled dsh CLI 0.1.7-rc.2 |
| `@deepseek-ai/dsh-mcp-client` (row provider) | bundled by the harness |
| Ghidra | 12.1.4 |
@@ -192,7 +261,7 @@ Author's machine, 2026-10-13 / 14:
| Host | Windows 11 x64 (build 26200) |
Exercised live: a clean `setup.mjs --reuse` run reached `setup complete`; an MCP `initialize` handshake
**through the shim with no `rea` on PATH** answered `serverInfo {name: rea, version: 6.4.0}`; and the whole
**through the shim with no `rea` on PATH** answered `serverInfo {name: rea, version: 6.5.0}`; and the whole
chain ran under the real desktop host — its own Electron binary launched
`<DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs mcp`, the shim started the **vendored** REA, and a live tool
call reported `path_source: environment` for a path only that machine's `bundle.json` knows.
@@ -203,22 +272,57 @@ macOS/Linux.
## Scope and caveats
- REA exposes 170 tools at 6.4.0, so every session's tool list grows.
- REA exposes 207 tools at 6.5.0, so every session's tool list grows.
- Provisioning is honest about checksums: where a vendor publishes one it is pinned and verified; where none
exists (jadx-headless-mcp, the apktool wrapper, Android platform-tools, the rolling "latest" URLs) the
hash is **measured once and recorded**, and later drift is reported rather than silently accepted.
`bundle.json` lists which components are in that state under `unverifiedChecksums`.
- Downloads are slow on some networks (measured 0.35–0.54 MiB/s from the GitHub release CDN here, so Ghidra
alone takes ~25–45 min). Fetching is resumable: the cache is kept and re-verified, and a killed run
continues instead of restarting.
alone takes ~25–45 min). Fetching is resumable on both routes: a partial `.part` is continued with a Range
request, and a `.part` that already holds the whole file is verified and kept instead of re-fetched.
- Some REA tools launch processes, write files and access the network — DSH's permission policy applies.
Analyse only artifacts you are authorized to inspect.
- The MCP child and the engines it launches are ordinary processes running as **you**. DSH's tool
approvals govern what the model asks for, not what that child may read, write or send on its own;
treat the plugin like any third-party code you install.
- On Linux, upstream's process-ownership checks need a **procps-compatible `ps`** on PATH; the
Ghidra/JADX and firmware paths use it. A minimal image without it fails those tools, not the install.
- `scripts/setup.mjs` runs `npm install` for the REA CLI with `--ignore-scripts`: the published tarball
carries no `node_modules`, so a dependency closure must be resolved, and package lifecycle scripts are
skipped on purpose during provisioning.
skipped on purpose during provisioning. That step is also the **only** one that cannot use this bundle's
own route/proxy resolution — npm reads `HTTPS_PROXY`/`HTTP_PROXY` and `NPM_CONFIG_REGISTRY` from the
environment, so on a network that needs a mirror, set one of those before provisioning
(`NPM_CONFIG_REGISTRY=https://registry.npmmirror.com`). With `--offline` the step is refused outright
instead of quietly going online.
- On POSIX, extraction falls back when the usual tools are missing: a zip is opened with `unzip`, then
`python3 -m zipfile`, then the provisioned JDK's `jar`; a tarball with `tar`, then `python3`. A minimal
image that has none of them fails with an explicit "no zip extractor available" line rather than a bare
"extraction failed". This matters because Ghidra is a zip on every platform and `unzip` is not installed
everywhere by default.
## Updating — three independent things
### Updating from the market
Update it where you installed it. Updating **to 2.0.5 or later** is routine; updating an installed 2.0.1–2.0.4
*through the in-app market* failed with this, three times on a healthy profile:
> dsh-plugin-rea 更新后缺少入口文件(package.json 的 main/exports 指向的文件不存在),且未能验证恢复原版本文件
That message is not about a broken tarball. On every update the market asks whether the new build's
`package.json` points at an entry file that exists (`hasLoadableEntry` in `dshmarket`), and rolls the build back
when the answer is false. A bundle may legitimately own no entry — but only if a package it *mounts* has one, and
the package this bundle mounts (`@deepseek-ai/dsh-mcp-client`) ships inside the harness build rather than in
`node_modules`, so every location that test probes comes back empty. The 1.x and 2.0.x manifests also declared
no `main` at all, so the answer was false even though the profile was working; the rollback was then reported as
unverifiable for the same reason. Nothing in the profile was actually damaged — the update simply did not apply.
2.0.5 declares `main: ./index.mjs`. That file is a real, deliberately empty Cordis plugin: it registers nothing,
spawns nothing, touches no state — the working row is still inserted by `cordis.patch.yml` — and it exists so the
question "can this package load?" has an honest answer, and so `import 'dsh-plugin-rea'` resolves. On an older
version, update to `>=2.0.5`; if the market refuses before that version exists, install `[email protected]`
(Desktop: through the plugin manager) instead of updating in place.
### Upgrading from 1.x
Install 2.x, then run provisioning **once** before restarting dsh:
@@ -236,6 +340,12 @@ back on the next start.
- **This plugin** never updates itself. Installed from a market? Update it there. Installed from a `file:`
checkout? Overwrite the directory and reinstall.
- **The row runs a COPY of this plugin's launcher.** `<DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs` is a mirror
made by setup, so a market upgrade replaces the package but not the file the row executes — until setup runs
again. From **2.0.8** the mirror notices that the installed plugin ships a different launcher and re-execs it
(one hop, guarded by `REA_BUNDLE_SHIM_REEXEC`), so upgrades carry. On an older version the mirror stays behind
silently; `--shim-status` now reports `mirror.stale`, and `check-updates.ps1` says to re-run setup. If you are
upgrading from 2.0.0–2.0.7, run `node scripts/setup.mjs` once after the upgrade.
- **Upgrading from `2.0.0-bundle.1`**: that release mirrored the shim and `setup.mjs` but not the manifest,
and recorded nothing about the plugin checkout, so a bundle it provisioned cannot repair itself — its
background provisioning dies on the missing manifest, invisibly, and its own lock suppresses retries for
+107 -7
View File
@@ -77,6 +77,9 @@ patch 里不含任何机器路径**。
`mcp__rea__*` 工具面;区别只在"谁准备运行时"。从 1.x 到 2.x 只要两条命令,见
[从 1.x 升级](#从-1x-升级)。
上游现在也能自己注册进 dsh(`rea setup --client dsh`):那基本就是上游维护的 1.x 模式 —— 只写配置,
Node/REA/Ghidra 由你准备。本包是零依赖路线。三条路选一条,不要同时留两行 `serverName: rea`。
## 安装
**前置什么都不用装。** 在干净的 Windows x64 机器上:
@@ -91,7 +94,7 @@ profile 安装插件:
```
dsh plugin --profile <profile> add dsh-plugin-rea
# 或钉住这个版本,而不是跟着 npm latest 走:
dsh plugin --profile <profile> add github:czcdr/dsh-plugin-rea#v2.0.0
dsh plugin --profile <profile> add github:czcdr/dsh-plugin-rea#v2.0.9
```
> **Electron 宿主(DeepSeek Harness 桌面版)的 profile 由应用独占。** 那种机器上 `dsh plugin --profile desktop …`
@@ -104,11 +107,52 @@ dsh plugin --profile <profile> add github:czcdr/dsh-plugin-rea#v2.0.0
node scripts/setup.mjs --reuse # 或先 --dry-run 看计划
```
如果 Node 由你自备(而不是让 setup 去取),它必须满足 REA 的 engines:
`^22.19.0 || ^24.11.0 || >=26.0.0`(用 `node -v` 核对)。版本偏低不会在安装时失败,而是到 MCP 行
启动时才失败 —— 表现就是"没有反应"。
反过来**先装插件**也可以,但要分两步:发布的行指向 `<DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs`,
而那个文件**只有 setup 会创建**——所以在第一次供给之前,该行会以 `MODULE_NOT_FOUND` 退出,也就没有
任何东西能去启动后台供给(harness 本身不受影响:`failOnStartupError: false` 是刻意的)。先手动跑一次
`node scripts/setup.mjs`;此后 shim 就位,之后每一次「状态缺失」都由它自己发现并在后台补供给。
### 支持哪些 DSH 版本
`package.json` 里声明了 `engines.dsh: ">=0.1.7-rc.2 <0.3.0-0"`,这就是插件市场读的那句话:有它,卡片显示
**兼容**;没有它,市场只能说*未声明*。这个范围不是猜的——它对应本插件实际核对过的 dsh 发布列表,
`node scripts/host-compat.mjs` 会在两者不一致时让构建失败:
| 宿主版本 | 是否覆盖 |
|---|---|
| `0.1.7-rc.2` … `0.2.1-rc.2`(中间的 alpha 与 rc 全部在内) | 是 |
| `0.3.0` 及以后 | 否 —— 新的小版本线是还没为它核对过这一行的宿主 |
真正被比对过的是「行提供方」:`packages/mcp/mcp-client` 在 `dsh-v0.1.7-rc.2` 与 `dsh-v0.2.1-rc.2` 两个 tag 上,
暴露的配置面与发布的行完全一致(`serverName`、`transport`、`command`、`args`、`env`、`failOnStartupError`),
并且两个 tag 都有 `dsh.bundle.patch`。范围之外的宿主可能照样能用,只是没有替它声明。
### 机器上已有 REA 就直接用它
2.x 存在的意义是补上机器缺的东西。如果 REA 已经装好——`rea` 在 PATH 上,也就是 1.x 的形态——这一行就**直接跑它**,
和 1.x 一模一样,不再做任何供给:
| 机器上的情况 | 这一行跑什么 |
|---|---|
| `rea` 在 PATH 上,且不比供给的版本旧 | 你自己那份(`rea mcp`),用你的 Node 和你的 PATH |
| `rea` 在 PATH 上,但比供给的版本旧 | 供给出来的 bundle —— 旧的全局 CLI 会把工具数往回退(6.4.0 是 170 个,6.5.0 是 207 个) |
| PATH 上没有 `rea` | 供给出来的 bundle,首次使用时自行供给 |
- `REA_BUNDLE_LAUNCH=auto|system|bundle` 覆盖这个判断(默认 `auto`);`REA_BUNDLE_SYSTEM_REA=<path>` 用来点名一个
不在 PATH 上的可执行文件。
- 探测用的是 CLI 自己的 `--version`,上限 8 秒:Windows 上的 `.cmd` 垫片要 ~2.4 秒才答得出来,所以这个检查每次启动
只做一次,而且只在真的存在 `rea` 时才做。
- `--shim-status` 会把判断结果打出来 —— `launch.mode`、`launch.reason`、`launch.systemRea`,`launch.rejected` 里
写的是每个候选被跳过的原因(`"exit 7"`、`"inside the bundle"`、`"timeout"`)。
- 被委派的子进程带着 `REA_BUNDLE_DELEGATED=1`,所以一个调用本 shim 的 wrapper 不会让它无限委派给自己。
- **Windows 上强制结束 shim 不会带走被委派的 REA**:这里没有 job object,任务管理器或 supervisor 终止 shim 时,
你自己的 `rea` 会继续占着这一行的管道(控制台 Ctrl+C 会发给整个进程组,那种情况下会一起停)。POSIX 上子进程
在同一进程组,随终端一起退出。
## 配置 —— 不需要
发布的那一行**不含任何机器路径**:
@@ -159,6 +203,25 @@ node bin/rea-shim.mjs --shim-print-env # MCP 子进程将看到的环境变量
`cordis.patch.yml` 按 id 覆盖即可 —— 那个文件是机器本地的,永不发布。一个 profile 里只允许一行使用
`serverName: rea`。
### 只保留一条 REA 注册
上游 REA 6.5.0 起可以自己注册进 DSH:
```
rea setup --client dsh
```
它会把一行 `id: rea-mcp`、`serverName: rea` 写进 `<DSH_HOME>/cordis.patch.yml`。本包的行是
`id: mcp-rea`、同样 `serverName: rea`,而 DSH 对同名条目只保留一个:后加载的那行会以
duplicate server name 报错并加载失败,最终留下哪一行取决于加载顺序。**二选一,不要同时用。**
选哪个?上游那条是纯配置路线:需要 Node 满足 REA engines、`rea-agents` 可达,Ghidra/JDK 由你
自己准备。本包是零依赖路线:首次运行自动取 Node、REA、JDK、Ghidra。两者的 patch 形态与
`mcp__rea__*` 工具面相同。
shim 会在启动时提示冲突 —— `setup.mjs` 会把运行时的 DSH home 记进 `mirror.json`,镜像副本在 harness 抹掉
`DSH_*` 之后就用这条记录去找那份文件;`node bin/rea-shim.mjs --shim-status` 会把冲突列在 `duplicateRegistration` 下。
## 验证
随便发一个 `mcp__rea__*` 调用 —— `binary_session` 会报引擎各项检查
@@ -170,11 +233,11 @@ shim 启动**。
## 实测环境
作者机器,2026-10-13 / 14:
作者机器,2026-10-13 / 14 / 16:
| 组件 | 版本 |
|---|---|
| REA(`rea-agents`) | **6.4.0**(170 个工具) |
| REA(`rea-agents`) | **6.5.0**(207 个工具) |
| DeepSeek Harness | desktop runtime **0.2.0-rc.2**,内含 dsh CLI 0.1.7-rc.2 |
| `@deepseek-ai/dsh-mcp-client`(行 provider) | 随 harness 打包 |
| Ghidra | 12.1.4 |
@@ -183,7 +246,7 @@ shim 启动**。
| 宿主 | Windows 11 x64(build 26200) |
实测跑过:一次干净的 `setup.mjs --reuse` 走到 `setup complete`;一次**不经 PATH 上的 `rea`、完全经由
shim** 的 MCP `initialize` 握手返回 `serverInfo {name: rea, version: 6.4.0}`;以及**真实桌面宿主下的完整链路** ——
shim** 的 MCP `initialize` 握手返回 `serverInfo {name: rea, version: 6.5.0}`;以及**真实桌面宿主下的完整链路** ——
宿主自己的 Electron 二进制启动 `<DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs mcp`,shim 拉起 **bundle 内置**
的 REA,随后一次真实工具调用报出 `path_source: environment`,而那个路径只有该机的 `bundle.json` 知道。
@@ -192,21 +255,53 @@ shim** 的 MCP `initialize` 握手返回 `serverInfo {name: rea, version: 6.4.0}
## 边界与注意事项
- REA 在 6.4.0 有 170 个工具,每个会话的工具列表都会变长。
- REA 在 6.5.0 有 207 个工具,每个会话的工具列表都会变长。
- 校验和对得起良心:厂商发布了的就固定并验证;没发布的(jadx-headless-mcp、apktool 包装脚本、
Android platform-tools、滚动的 `latest` 地址)则**实测一次并记录**,之后若漂移是**报告**而不是默默接受。
`bundle.json` 的 `unverifiedChecksums` 会列出处于该状态的组件。
- 某些网络下很慢(这里实测 GitHub release CDN 只有 0.35–0.54 MiB/s,光 Ghidra 就要 25–45 分钟)。
下载是**可续传**的:缓存保留并重新校验,被中断的一轮会继续而不是从头再来。
下载是**可续传**的:两条路线都按盘上已有的字节发 Range 继续;若 `.part` 恰好已经完整,会校验后直接收下
而不是重下。
- **代理**:Node 的全局 `fetch` 不读系统代理,而 .NET/PowerShell 会读。在开着 WinINET 代理的机器上,
同一个地址 `Invoke-WebRequest` 通、Node 却 `ECONNRESET`/超时。setup 自己解析代理
(环境变量优先,Windows 上回退读注册表)并用 CONNECT 隧道,`NO_PROXY` 也照常生效。
- REA 的部分工具会启动进程、写文件、访问网络 —— DSH 的权限策略照常适用。只分析你有权检视的样本。
- MCP 子进程与它拉起的引擎都是以**你的身份**运行的普通进程。DSH 的工具审批管的是模型发起的调用,
管不到这个子进程自己去读、写或发送什么;请把插件当作你安装的第三方代码来对待。
- Linux 上,上游的进程归属检查依赖 PATH 上**兼容 procps 的 `ps`**;Ghidra/JADX 与固件工具会用到它。
极简镜像里缺它只会让这些工具失败,不影响整体安装。
- `scripts/setup.mjs` 为 REA CLI 跑的是带 `--ignore-scripts` 的 `npm install`:官方 tarball 不带
`node_modules`,必须解析依赖闭包;而供给过程中**刻意跳过包的 lifecycle 脚本**。
`node_modules`,必须解析依赖闭包;而供给过程中**刻意跳过包的 lifecycle 脚本**。这一步也是**唯一**用不上
本插件自己的路由/代理回退的环节 —— npm 只从环境读 `HTTPS_PROXY`/`HTTP_PROXY` 与
`NPM_CONFIG_REGISTRY`,所以网络需要镜像时请先设其中一个(例如
`NPM_CONFIG_REGISTRY=https://registry.npmmirror.com`)。带 `--offline` 时这一步会被**直接拒绝**,
而不是悄悄联网。
- POSIX 上解压有兜底:zip 依次尝试 `unzip` → `python3 -m zipfile` → 已供给 JDK 自带的 `jar`;tar 包依次
尝试 `tar` → `python3`。三者都没有的最小镜像会得到一句明确的「no zip extractor available」,而不是
干巴巴的「extraction failed」。Ghidra 在所有平台都只有 zip,而 `unzip` 并非每个发行版默认安装,
所以这条不是理论问题。
## 更新 —— 三件互相独立的事
### 在市场里更新
在哪儿装的就在哪儿更新。更新到 **2.0.5 及以上**是常规操作;而把已装的 2.0.1–2.0.4 通过应用内市场**更新到
2.0.5**,在健康的 profile 上连续失败了三次:
> dsh-plugin-rea 更新后缺少入口文件(package.json 的 main/exports 指向的文件不存在),且未能验证恢复原版本文件
这条消息与 tarball 是否完整无关。市场在每次更新时都会问一句:新构建的 `package.json` 是否指向一个真实存在的
入口文件(`dshmarket` 里的 `hasLoadableEntry`);答案是「否」,就把这次更新判为坏构建并回滚。bundle 本来允许
自己没有入口 —— 前提是它**挂载**的某个包有入口;而本包挂载的 `@deepseek-ai/dsh-mcp-client` 属于 harness 自身
构建,不在 `node_modules` 里,于是这项检查要看的每个位置都是空的。1.x 与 2.0.x 的清单又根本没有声明 `main`,
所以答案必然是「否」,哪怕 profile 一直是好的;紧接着的回滚也因此被报成「无法验证」。实际上 profile 没有被
破坏,只是这次更新没有生效。
2.0.5 声明了 `main: ./index.mjs`。那是一个真实存在、刻意留空的 Cordis 插件:不注册任何行、不启动任何进程、
不碰任何状态(真正起作用的行仍由 `cordis.patch.yml` 插入),它的作用只是让「这个包能不能加载」有一个诚实的
答案,并让 `import 'dsh-plugin-rea'` 能解析。还停在旧版本时,更新到 `>=2.0.5` 即可;若市场在 2.0.5 之前就拒绝,
就直接安装 `[email protected]`(桌面端:在插件管理器里装),不要原地更新。
### 从 1.x 升级
装上 2.x 后,**先跑一次供给**,再重启 dsh:
@@ -222,6 +317,11 @@ node scripts/setup.mjs # 或:powershell -File scripts\rea-bundle-boot
工具就回来了。
- **本插件**从不自更新。从市场装的就在市场更新;从 `file:` 检出的就覆盖目录重装。
- **行跑的是本插件启动器的「镜像副本」**:`<DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs` 是 setup 复制过去的,
所以市场升级只换包、不换行执行的那个文件 —— 直到再跑一次 setup。从 **2.0.8** 起,镜像会发现自己落后于已安装的
插件,并转发执行插件里的那一份(一跳,带 `REA_BUNDLE_SHIM_REEXEC` 守卫),升级因此能生效。更早的版本里镜像会
静默地停在旧代码;`--shim-status` 现在会报 `mirror.stale`,`check-updates.ps1` 也会提示重跑 setup。**从
2.0.0–2.0.7 升级的话,升级后跑一次 `node scripts/setup.mjs`。**
- **从 `2.0.0-bundle.1` 升级**:那一版镜像了 shim 与 `setup.mjs`,却漏了 manifest,也没有留下插件 checkout
的记录,所以它供给出来的 bundle **无法自愈** —— 后台供给会因缺 manifest 而崩溃且完全不可见,而它自己写下的
lock 又会让重试被抑制六小时。请**从插件目录**重跑一次 `node scripts/setup.mjs`:它会重新镜像三个文件,
+2 -2
View File
@@ -84,7 +84,7 @@ Draw-Text $g '第三方适配 · 自身不做逆向分析 · 自供给 Node/REA/
$rowY = 180; $boxH = 158; $boxW = 250
Draw-Box $g 60 $rowY $boxW $boxH $([System.Drawing.Color]::White) $line
Draw-Text $g 'DeepSeek Harness 会话' $fBox $ink 80 ($rowY + 18)
Draw-Text $g "mcp__rea__* 工具`n(6.4.0 暴露 170+)`n工具随会话注入`n按 DSH 权限策略执行" $fBody $muted 80 ($rowY + 50)
Draw-Text $g "mcp__rea__* 工具`n(6.5.0 暴露 207)`n工具随会话注入`n按 DSH 权限策略执行" $fBody $muted 80 ($rowY + 50)
Draw-Box $g 360 $rowY $boxW $boxH $([System.Drawing.Color]::FromArgb(255,239,246,255)) $accent
Draw-Text $g '官方 MCP 客户端行' $fBox $accent 380 ($rowY + 18)
@@ -112,7 +112,7 @@ Draw-Line $g 358 446 150 446 $muted
Draw-Arrow $g 150 446 150 342 $muted
Draw-Text $g '上游 REA(MIT):github.com/morluto/rea 本插件:github.com/czcdr/dsh-plugin-rea' $fSmall $muted 60 620
Draw-Text $g '实测 REA 6.4.0 + DSH 桌面运行时 0.2.0-rc.2 + Ghidra 12.1.4 + Node 24.21.0(Windows 11 x64,2026-10-14)' $fSmall $muted 60 646
Draw-Text $g '实测 REA 6.5.0 + DSH 桌面运行时 0.2.0-rc.2 + Ghidra 12.1.4 + Node 24.21.0(Windows 11 x64,2026-10-16)' $fSmall $muted 60 646
Draw-Text $g 'Ghidra 为 Apache-2.0(NSA);REA 为 MIT。本仓库不分发上游代码。' $fSmall $muted 60 672
Save-Png $bmp $g (Join-Path $OutDir 'screenshot-1.png')
Binary file not shown.

Before

Width:  |  Height:  |  Size: 67 KiB

After

Width:  |  Height:  |  Size: 66 KiB

+884 -28
View File
File diff suppressed because it is too large. Load diff
+123 -20
View File
@@ -69,18 +69,53 @@ function Test-NodeEngines([string]$node, [string]$engines) {
}
# ------------------------------------------------------- 定位 REA 与 profile
if (-not $ReaPackageDir) {
$npmRoot = $null
if (Get-Command npm -ErrorAction SilentlyContinue) {
try { $npmRoot = (& npm root -g 2>$null | Select-Object -First 1).Trim() } catch { }
}
$ReaPackageDir = if ($npmRoot) { Join-Path $npmRoot 'rea-agents' } else { Join-Path $env:APPDATA 'npm\node_modules\rea-agents' }
# 2.x 的 REA 在 <bundleRoot>\vendor\node_modules\rea-agents,不在 npm 全局。先读
# bundle.json:把全局路径当默认是 1.x 的答案,在 2.x 机器上会让这一节指向错误的
# 对象("全局包丢了"),并给出错误的升级动作(npm i -g)。
$bundleRootEarly = $env:REA_BUNDLE_ROOT
# Resolve the home unconditionally: the profile section below needs it whether or
# not REA_BUNDLE_ROOT was set, and computing it only inside the fallback branch
# left it $null — `Join-Path $null 'profiles'` throws, so the section degraded to
# "no profile" on exactly the machines that use the override.
$homeEarly = $env:DSH_HOME
if (-not $homeEarly) { $homeEarly = Join-Path $HOME '.dsh' }
if (-not $bundleRootEarly) {
$bundleRootEarly = Join-Path $homeEarly 'plugins\rea-bundle'
}
$bundleState = $null
$bundleJsonEarly = Join-Path $bundleRootEarly 'bundle.json'
if (Test-Path $bundleJsonEarly) {
try { $bundleState = Get-Content $bundleJsonEarly -Raw -Encoding UTF8 | ConvertFrom-Json } catch { $bundleState = $null }
}
$bundleReaPackageDir = $null
$bundleReaEntry = $null
if ($bundleState -and $bundleState.paths.reaEntry) {
$bundleReaEntry = [string]$bundleState.paths.reaEntry
# <vendor>\node_modules\rea-agents\scripts\rea.mjs -> <vendor>\node_modules\rea-agents
$bundleReaPackageDir = Split-Path -Parent (Split-Path -Parent $bundleReaEntry)
}
if (-not $ReaPackageDir) {
if ($bundleReaPackageDir -and (Test-Path (Join-Path $bundleReaPackageDir 'package.json'))) {
$ReaPackageDir = $bundleReaPackageDir
} else {
$npmRoot = $null
if (Get-Command npm -ErrorAction SilentlyContinue) {
try { $npmRoot = (& npm root -g 2>$null | Select-Object -First 1).Trim() } catch { }
}
$ReaPackageDir = if ($npmRoot) { Join-Path $npmRoot 'rea-agents' } else { Join-Path $env:APPDATA 'npm\node_modules\rea-agents' }
}
}
$reaFromBundle = [bool]($bundleReaPackageDir -and $ReaPackageDir -eq $bundleReaPackageDir)
if (-not $ProfileDir) {
if ($env:DSH_PROFILE_DIR) { $ProfileDir = $env:DSH_PROFILE_DIR }
else {
$profiles = @(Get-ChildItem (Join-Path $env:USERPROFILE '.dsh\profiles') -Directory -ErrorAction SilentlyContinue)
$ProfileDir = if ($profiles.Count -gt 0) { $profiles[0].FullName } else { Join-Path $env:USERPROFILE '.dsh\profiles\desktop' }
# Same home rule the bundle uses everywhere else: DSH_HOME, then ~/.dsh.
# Hardcoding $env:USERPROFILE here was a second algorithm in the same script,
# so on a custom DSH_HOME the profile section reported "no profile found"
# while the rest of the run had already resolved that home correctly.
$profiles = @(Get-ChildItem (Join-Path $homeEarly 'profiles') -Directory -ErrorAction SilentlyContinue)
$ProfileDir = if ($profiles.Count -gt 0) { $profiles[0].FullName } else { Join-Path $homeEarly 'profiles\desktop' }
}
}
@@ -94,11 +129,21 @@ if (Test-Path $pkgFile) {
$installed = $pkg.version; $engines = $pkg.engines.node
$binPath = ($pkg.bin.PSObject.Properties | Select-Object -First 1).Value
if ($binPath) { $reaEntry = Join-Path $ReaPackageDir ($binPath -replace '/', '\') }
Ok "本地已安装: $installed"
Info "入口(bin): $binPath ← 插件通过 PATH 上的 rea 调用它,跨版本稳定"
if ($reaFromBundle) {
Ok "一体包内置(vendored): $installed"
Info "位置: $ReaPackageDir"
Info "入口(bin): $binPath"
} else {
Ok "本地已安装: $installed"
Info "入口(bin): $binPath ← 插件通过 PATH 上的 rea 调用它,跨版本稳定"
}
if ($engines) { Info "Node 要求: $engines" }
} catch { Bad "package.json 解析失败: $($_.Exception.Message)" }
} else { Bad "找不到 $pkgFile —— REA 全局包丢了?重装: npm i -g rea-agents" }
} else {
if ($reaFromBundle) { Bad "找不到 $pkgFile —— 内置副本不完整?重跑 setup.mjs" }
elseif ($bundleReaPackageDir) { Bad "找不到 $pkgFile —— 一体包内置副本也缺($bundleReaPackageDir);记录里没有 reaEntry" }
else { Bad "找不到 $pkgFile —— REA 全局包丢了?重装: npm i -g rea-agents" }
}
$latest = $null
foreach ($r in @($Registry, 'https://registry.npmjs.org')) {
@@ -110,7 +155,11 @@ if ($installed -and $latest) {
$vi = Get-NumericVersion $installed; $vl = Get-NumericVersion $latest
if ($vi -and $vl -and $vl -gt $vi) {
Warn "有新版本:$installed → $latest"
$actions.Add("升级上游 REA: npm i -g rea-agents@$latest (只更新 REA;本插件不会自动更新——从市场装的在市场里更新,file: 装的覆盖目录后重装)")
if ($reaFromBundle) {
$actions.Add("升级一体包内置 REA:把 manifest 的 pins.reaAgents / components.reaAgents.sources.npm.version 改成 $latest,然后重跑 node scripts\setup.mjs(内置副本不会自己更新)")
} else {
$actions.Add("升级上游 REA: npm i -g rea-agents@$latest (只更新 REA;本插件不会自动更新——从市场装的在市场里更新,file: 装的覆盖目录后重装)")
}
} elseif ($vi -and $vl -and $vl -eq $vi -and $installed -ne $latest) {
Warn "同版本号但通道不同:$installed → $latest(预发布/正式)"
} else { Ok '已是最新版本' }
@@ -139,7 +188,14 @@ if (Get-Command node -ErrorAction SilentlyContinue) {
# ------------------------------------------------------------ Ghidra 引擎
Head 'Ghidra 引擎'
$gd = [Environment]::GetEnvironmentVariable('GHIDRA_INSTALL_DIR', 'User')
# bundle.json 是 2.x 的真源:shim 只在 MCP 子进程里设 GHIDRA_INSTALL_DIR,用户级
# 变量是 1.x 的写法,没手工设过的机器会被误报"未设置"。
$gd = $null
if ($bundleState -and $bundleState.paths.ghidra) {
$bundleGhidra = [string]$bundleState.paths.ghidra
if (Test-Path $bundleGhidra) { $gd = $bundleGhidra }
}
if (-not $gd) { $gd = [Environment]::GetEnvironmentVariable('GHIDRA_INSTALL_DIR', 'User') }
if (-not $gd) { $gd = $env:GHIDRA_INSTALL_DIR }
if ($gd -and (Test-Path $gd)) {
Ok "GHIDRA_INSTALL_DIR = $gd"
@@ -162,11 +218,28 @@ try {
# --------------------------------------------------------- REA 引擎自检
Head 'REA 引擎自检(rea doctor)'
$env:GHIDRA_INSTALL_DIR = $gd
# 优先取 .cmd:发布出去的那一行是 `cmd.exe /c rea mcp`(走 cmd 的 PATHEXT,命中 rea.cmd),
# 而 PowerShell 的 Get-Command rea 会先命中同目录的 rea.ps1 垫片,两者行为(尤其退出码)并不相同。
$reaCmd = (Get-Command 'rea.cmd' -ErrorAction SilentlyContinue).Source
if (-not $reaCmd) { $reaCmd = (Get-Command rea -ErrorAction SilentlyContinue).Source }
# 2.x:优先用 bundle 内置的 CLI 和它自己的 Node。PATH 上的 rea 可能是另一个,
# 甚至完全不在 PATH —— 那会让整段自检被静默跳过。这正是 1.x 口径的残留。
$reaCmd = $null
$reaArgs = @()
if ($bundleReaEntry -and (Test-Path $bundleReaEntry)) {
$bundleNode = $null
if ($bundleState -and $bundleState.paths.node) {
$candidateNode = [string]$bundleState.paths.node
if (Test-Path $candidateNode) { $bundleNode = $candidateNode }
}
if (-not $bundleNode) { $bundleNode = (Get-Command node -ErrorAction SilentlyContinue).Source }
if ($bundleNode) {
$reaCmd = $bundleNode; $reaArgs = @($bundleReaEntry)
Info '使用一体包内置 REA(node <bundle 入口>),不依赖 PATH 上的 rea'
}
}
if (-not $reaCmd) {
# 优先取 .cmd:发布出去的那一行是 `cmd.exe /c rea mcp`(走 cmd 的 PATHEXT,命中 rea.cmd),
# 而 PowerShell 的 Get-Command rea 会先命中同目录的 rea.ps1 垫片,两者行为(尤其退出码)并不相同。
$reaCmd = (Get-Command 'rea.cmd' -ErrorAction SilentlyContinue).Source
if (-not $reaCmd) { $reaCmd = (Get-Command rea -ErrorAction SilentlyContinue).Source }
}
if (-not $reaCmd) {
if ($reaEntry -and (Test-Path $reaEntry) -and (Get-Command node -ErrorAction SilentlyContinue)) {
$reaCmd = 'node'; $reaArgs = @($reaEntry)
@@ -247,6 +320,8 @@ if (-not (Test-Path $manifestPath)) {
$state = Get-Content $bundleJson -Raw -Encoding UTF8 | ConvertFrom-Json
Ok "已供给:$($state.provisionedAt)"
$drift = 0
$compared = 0
$unrecorded = 0
foreach ($name in $manifest.pins.PSObject.Properties.Name) {
if ($name -eq 'nodeEngines') { continue }
$pinned = [string]$manifest.pins.$name
@@ -254,18 +329,46 @@ if (-not (Test-Path $manifestPath)) {
if ($state.versions -and ($state.versions.PSObject.Properties.Name -contains $name)) {
$installed = [string]$state.versions.$name
}
if (-not $installed) { continue }
if (-not $installed) {
# 被 --reuse 采纳的组件没有 version 字段(bundle.json 里是 null)。静默
# continue 会让这一项从未被比对,却在下面打印"全部一致"——那是假绿。
$unrecorded++
Warn " $name 未记录版本(reuse 采纳的组件?)—— 无法与 pin $pinned 对账"
continue
}
$compared++
# Node 的 pin 是引擎范围,不是版本号;其它组件比前缀即可
$same = $installed.Equals($pinned) -or $installed.StartsWith(($pinned -split '\.')[0] + '.')
if ($same) { Info " $name = $installed(pin $pinned)" }
else { Warn "$name 盘上为 $installed,manifest 已钉 $pinned —— 重跑 setup 对齐"; $drift++ }
}
if ($drift -eq 0) { Ok '所有组件与 manifest 的 pin 一致' }
if ($drift -eq 0 -and $unrecorded -eq 0) { Ok '所有组件与 manifest 的 pin 一致' }
elseif ($drift -eq 0) { Warn "已比对的 $compared 项与 pin 一致,但 $unrecorded 项没有记录版本(见上)—— 不是全绿" }
if ($state.unverifiedChecksums -and @($state.unverifiedChecksums).Count -gt 0) {
Warn ("TOFU(无厂商校验和,取首次实测值): " + (@($state.unverifiedChecksums) -join ', '))
}
$mirrored = Join-Path $bundleRoot 'bin\rea-shim.mjs'
if (Test-Path $mirrored) { Ok 'MCP 行指向的 shim 已就位' }
if (Test-Path $mirrored) {
Ok 'MCP 行指向的 shim 已就位'
# The row launches this COPY, so a plugin upgrade that does not re-run
# setup leaves the old launcher in place — visible only as "the new
# version behaves like the old one". Compare it with the checkout the
# mirror record names.
$mirrorRecord = Join-Path $bundleRoot 'mirror.json'
if (Test-Path $mirrorRecord) {
try {
$record = Get-Content $mirrorRecord -Raw -Encoding UTF8 | ConvertFrom-Json
$pluginShim = if ($record.pluginRoot) { Join-Path $record.pluginRoot 'bin\rea-shim.mjs' } else { $null }
if ($pluginShim -and (Test-Path $pluginShim)) {
$same = (Get-FileHash $mirrored -Algorithm SHA256).Hash -eq (Get-FileHash $pluginShim -Algorithm SHA256).Hash
if ($same) { Ok '镜像 shim 与插件一致(行跑的就是这个版本)' }
else { Warn "镜像 shim 落后于插件($($record.pluginRoot))—— 行仍在跑旧副本;重跑 node scripts/setup.mjs 让它换成新版本" }
} else {
Warn "mirror.json 指向的插件目录已不存在($($record.pluginRoot))—— 无法比对镜像 shim 是否最新"
}
} catch { Warn "mirror.json 解析失败: $($_.Exception.Message)" }
}
}
else { Bad "缺少 MCP 行指向的 shim: $mirrored —— 重跑 setup --reuse" }
# 镜像副本的 setup 用它自己所在的位置推 plugin root,所以 manifest 必须
# 一起镜像。旧版镜像缺这个文件:auto-provision 会以 ENOENT 崩掉且不可见,
+30 -6
View File
@@ -31,11 +31,11 @@ Two upstream rules shape the runtime wiring:
## 2. Verified facts this design rests on
Measured on Windows 11 x64, REA 6.4.0, Ghidra 12.1.4, Temurin JDK 25.0.4:
Measured on Windows 11 x64, REA 6.5.0, Ghidra 12.1.4, Temurin JDK 25.0.4:
| Fact | Evidence |
| --- | --- |
| A row can launch REA with **no PATH `rea`** | `node <pkg>/scripts/rea.mjs mcp` answered the MCP `initialize` handshake (`serverInfo.name = rea`, version 6.4.0) and `rea doctor --json` reported 170 tools, `healthy: true` |
| A row can launch REA with **no PATH `rea`** | `node <pkg>/scripts/rea.mjs mcp` answered the MCP `initialize` handshake (`serverInfo.name = rea`, version 6.5.0) and `rea doctor --json` reported 207 tools. `healthy` is `false` on a bundle-only install: 6.5.0 also checks the host's own setup (Hopper, IDA registration, `rea setup` skill/registrations), which provisioning deliberately never touches — the row does not require `healthy`, only a CLI that runs and a catalog it can read |
| `!!js` expressions are supported in this profile layer | the existing profile patch already used one; the loader evaluates them |
| A row-level `env:` block reaches the MCP child | `binary_session` reported `registrationEnvironment` = `GHIDRA_INSTALL_DIR`, `JAVA_HOME` |
| No fixed home/layout is promised | the disk layout differs per install; nothing may assume one |
@@ -54,8 +54,9 @@ Three responsibilities, kept strictly apart:
checksums, unpacks, runs the readiness probe, and writes `bundle.json`.
Idempotent and re-runnable; `--reuse` adopts pre-existing installs.
3. **Patch (wiring, published).** `cordis.patch.yml` on `main`, carrying
**no machine path**: the row points at `<plugin>/bin/rea-shim.mjs` and the
shim is responsible for the rest.
**no machine path**: the row computes `<DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs`
at load time from `DSH_HOME` alone, and that mirrored shim is responsible for the
rest. (The checkout's own copy is what a human runs; the row never points at it.)
The published plugin therefore stays portable, while the machine-specific truth
lives in one small JSON file under the DSH home.
@@ -82,7 +83,10 @@ through `failOnStartupError: false` without breaking the harness.
```
dsh starts
└─ MCP row: node <bundle root>/bin/rea-shim.mjs (setup mirrors it there)
├─ bundle.json present & components verify → exec REA (normal path)
├─ mirrored copy differs from the installed plugin → re-exec that copy, once
├─ a working `rea` on PATH is at least as new as the bundle
│ → exec it (the 1.x contract)
├─ bundle.json present & components verify → exec the provisioned REA
├─ missing, auto-provision allowed → start setup detached, print
│ remediation, exit non-zero
└─ missing, auto-provision refused → print exact command, exit
@@ -128,6 +132,26 @@ Two notes on stages that exist because their absence was a bug:
harness's own dialect, `args[0]` became `{"[object Object]": "..."}` and the
row could not launch at all. Any future expression with a ternary needs the
quotes; `audit-manifest.mjs` rejects the bare form.
- **Launch mode (2.0.7).** 2.x exists to supply a runtime a machine lacks, so a
machine that already has one should not be provisioned. Before touching bundle
state the shim asks whether `rea` is on PATH and answers `--version`; if it
does, and its version is not older than `bundle.json`'s `versions.reaAgents`,
the row execs that install exactly as the 1.x row did. The probe is asynchronous
and bounded (8 s: a `.cmd` shim needs ~2.4 s), runs with
`REA_BUNDLE_DELEGATED=1` and `REA_BUNDLE_AUTO_PROVISION=0` so a candidate that
wraps this shim cannot recurse or start an 800 MB run, and kills its whole tree
on timeout. `REA_BUNDLE_LAUNCH=auto|system|bundle` overrides the choice; an
unorderable pair (a prerelease on either side) keeps the bundle.
- **Self-healing mirror (2.0.8).** The row launches a COPY, and a market upgrade
runs no setup, so without this the copy keeps executing the code it was made
from — invisibly, because it still works. The copy therefore compares its own
hash with `<mirror.json.pluginRoot>/bin/rea-shim.mjs` and re-execs that file
when they differ: one hop, guarded by `REA_BUNDLE_SHIM_REEXEC`, and only from a
mirror (a checkout ships `cordis.patch.yml`, so it never hops into another
copy). `mirror.json` carries `shimSha256`, `pluginVersion` and `dshHome` —
the last one because the harness strips `DSH_*` from an MCP child, so the
duplicate-registration check would otherwise read `~/.dsh` on a machine with a
custom home. `--shim-status` reports `mirror.stale` instead of hopping.
- **3 (fetch).** Route order is deliberate and environment-dependent: direct
first, then the system proxy, and on Windows the .NET route ahead of both
because it follows the GitHub release-asset redirect that the Node routes
@@ -161,7 +185,7 @@ Windows and is what this bundle targets to complete.
| Component | Needed for | Provisioned as |
| --- | --- | --- |
| Node runtime | running REA at all | reuse a compatible existing node, else portable `node.exe` |
| `rea-agents` | the MCP server itself | vendored copy under `vendor/rea-agents` |
| `rea-agents` | the MCP server itself | vendored copy under `vendor/node_modules/rea-agents` |
| Temurin JDK 21/25 | Ghidra headless + JADX bridge | portable JDK zip (full JDK, not a JRE) |
| Ghidra 12.1.x | native analysis | official release zip, extracted |
| apktool | Android resource decode | `apktool.bat` + jar in one directory |
-18
View File
@@ -1,18 +0,0 @@
# macOS / Linux variant of the row.
#
# `npm i -g rea-agents` installs a `rea` executable on PATH, so no shell wrapper
# is needed here (on Windows the shim is a .cmd/.ps1 file, which is why the
# default cordis.patch.yml goes through cmd.exe).
#
# Copy the block below into cordis.patch.yml, replacing the Windows row.
- insert:
- id: mcp-rea
name: '@deepseek-ai/dsh-mcp-client'
config:
serverName: rea
transport: stdio
command: 'rea'
args:
- 'mcp'
failOnStartupError: false
-42
View File
@@ -1,42 +0,0 @@
# Windows variant that also pins the analysis-engine environment for the MCP child.
#
# Use this when GHIDRA_INSTALL_DIR is NOT set as an OS-level environment variable
# the harness can inherit — for example after setting it up just now, without
# restarting the harness.
#
# The value below is a placeholder in angle brackets, not a real directory:
# replace it with your own, or use the `!!js` form to read the variable the
# harness already has. REA reads only GHIDRA_INSTALL_DIR; `java` comes from PATH
# or JAVA_HOME.
#
# Copy the block into cordis.patch.yml, or — better — keep it out of the plugin
# entirely and override the row from your profile's own cordis.patch.yml:
#
# - id: mcp-rea
# config:
# env:
# GHIDRA_INSTALL_DIR: '<your Ghidra directory>'
#
# A profile-level override is machine-local and is never published.
- insert:
- id: mcp-rea
name: '@deepseek-ai/dsh-mcp-client'
config:
serverName: rea
transport: stdio
command: 'cmd.exe'
args:
- '/c'
- 'rea'
- 'mcp'
env:
# Replace with your own Ghidra 12.1.x extraction directory.
# Leaving the placeholder in place is harmless: REA then reports the
# native engine as unavailable, exactly as when the variable is unset.
GHIDRA_INSTALL_DIR: '<your Ghidra 12.1.x extraction directory>'
# Or read it from the environment the harness already has — the
# entry-list dialect evaluates `!!js` expressions at activation:
# GHIDRA_INSTALL_DIR: !!js process.env.GHIDRA_INSTALL_DIR
# JAVA_HOME: '<your JDK 21 or newer home>' # only when java is not on PATH
failOnStartupError: false
+43
View File
@@ -0,0 +1,43 @@
/**
* Package entry — the smallest thing that makes this bundle a *loadable* package.
*
* Everything this plugin DOES lives in `cordis.patch.yml`: it inserts one row of
* `@deepseek-ai/dsh-mcp-client` that launches `bin/rea-shim.mjs`, and that row is
* what the harness loads. This file registers nothing, starts nothing and
* provisions nothing.
*
* It exists because a package that declares no `main`/`exports` is judged by its
* consumers as "has no entry file", whether or not the loader ever needs one:
*
* - The market's update path (`hasLoadableEntry` in dshmarket) asks that
* question on every update and, when the answer is false, treats the new
* build as broken and rolls back. A bundle with no entry of its own is
* *supposed* to survive the test through the packages it mounts — but the
* package this one mounts (`@deepseek-ai/dsh-mcp-client`) is part of the
* harness build, not of `node_modules`, so every location that test probes
* comes back empty. Measured on 2026-10-18: updating an installed, working
* 2.0.1 to 2.0.4 failed three times with "updated build has no loadable
* entry", and the rollback was then reported as unverifiable for the same
* reason.
* - `npm pack` metadata and `import 'dsh-plugin-rea'` both resolve `main`;
* absent, both fall back to `index.js`, which this repository never shipped.
*
* So this is a real module with a real Cordis plugin shape rather than a stub
* for a checker: if the harness ever does load this package as a plugin, it
* loads cleanly and does nothing.
*/
/** Cordis plugin name. Also the package name, which is what the loader expects. */
export const name = 'dsh-plugin-rea'
/**
* Intentionally empty.
*
* The row that does the work is inserted by the bundle patch, not by this
* function: registering the MCP client from here would create a second
* `serverName: rea` entry, which the harness refuses as a duplicate (the same
* collision `bin/rea-shim.mjs` warns about for upstream's `rea setup --client
* dsh`). Doing nothing is the correct behaviour, and the function present keeps
* this a valid Cordis plugin instead of an arbitrary module.
*/
export function apply() {}
+48 -41
View File
@@ -12,12 +12,14 @@
" binaries - workspace-relative entries inside the unpacked directory that are",
" directly spawnable; the first that exists becomes the component's",
" recorded path so another component can depend on it.",
" executable - a placed file that must carry the POSIX exec bit (a wrapper REA",
" spawns without a shell); setup chmods exactly these to 0755.",
"Nothing here is redistributed inside the npm package; setup downloads it."
],
"manifestVersion": 1,
"generatedFor": "dsh-plugin-rea one-plugin bundle",
"pins": {
"reaAgents": "6.4.0",
"reaAgents": "6.5.0",
"ghidra": "12.1.4",
"jdk": "21",
"apktool": "3.0.3",
@@ -203,16 +205,16 @@
"win32-x64": [
{
"id": "wrapper",
"version": "master",
"url": "https://raw.githubusercontent.com/iBotPeaches/Apktool/master/scripts/windows/apktool.bat",
"version": "3.0.3",
"url": "https://raw.githubusercontent.com/iBotPeaches/Apktool/v3.0.3/scripts/windows/apktool.bat",
"format": "file",
"size": 2348,
"sha256": "9440ec9d98efdef14136f4cefd79ac200d72baa016b5991d128234a3a3ef3b43",
"sha256Source": "measured on 2026-10-13 (raw.githubusercontent.com publishes no checksum for the wrapper); a single 2,348-byte file, so the archive-free case needs no TOFU",
"sha256VerifiedOn": "2026-10-13",
"size": 2386,
"sha256": "d81a91bd694459a61a88e4a1a8184efc21130390ff7cb5c172f631551ad9842a",
"sha256Source": "measured on 2026-10-16 from the immutable v3.0.3 tag (raw.githubusercontent.com publishes no checksum for the wrapper). The tag, not master: pinning master meant any upstream commit to the wrapper broke every new install with a hash mismatch, and four retries could not fix a download whose content had legitimately changed.",
"sha256VerifiedOn": "2026-10-16",
"notes": [
"The wrapper pushd's to its own directory and resolves apktool.jar first, then globs apktool_*.jar, so the renamed jar next to it is found.",
"Its final line is `for %%i in (%cmdcmdline%) do if /i \"%%~i\"==\"/c\" pause & exit /b`: measured here it returns exit 0 in 0.2 s, but it appends 'Press any key to continue . . .' to stdout. That trailing text is upstream's, not a provisioning defect; tools that parse apktool's stdout should tolerate it.",
"The previous entry measured master's copy (2,348 bytes) and recorded that it returns exit 0 in 0.2 s while appending 'Press any key to continue . . .' to stdout. This is the tag's copy (2,386 bytes) and that runtime behaviour has NOT been re-measured; treat the trailing-text claim as unverified for this file.",
"REA is pointed at this .bat (REA_APKTOOL_COMMAND), not at the jar, exactly as the wrapper expects."
],
"moveTo": "apktool",
@@ -222,13 +224,14 @@
"darwin-x64": [
{
"id": "wrapper",
"version": "master",
"url": "https://raw.githubusercontent.com/iBotPeaches/Apktool/master/scripts/osx/apktool",
"version": "3.0.3",
"url": "https://raw.githubusercontent.com/iBotPeaches/Apktool/v3.0.3/scripts/osx/apktool",
"format": "file",
"size": 2672,
"sha256": "d0c86f0475ed73bfffa6ac39faa69de7aa2dd40ff5e7e6540f82002be3a38f0b",
"sha256Source": "measured on 2026-10-13 from the osx wrapper (the correct one for macOS; the linux script is a different file)",
"sha256VerifiedOn": "2026-10-13",
"size": 2709,
"sha256": "3b9a95c41680c4a758b0b8d6e91e122b2ea3f1a189685a915613ca1ccba00fd9",
"sha256Source": "measured on 2026-10-16 from the v3.0.3 tag (the correct file for macOS; the linux script is a different file). The tag is immutable, so this pin cannot drift under a new install the way master's did.",
"sha256VerifiedOn": "2026-10-16",
"executable": true,
"moveTo": "apktool",
"saveAs": "apktool"
}
@@ -236,13 +239,14 @@
"darwin-arm64": [
{
"id": "wrapper",
"version": "master",
"url": "https://raw.githubusercontent.com/iBotPeaches/Apktool/master/scripts/osx/apktool",
"version": "3.0.3",
"url": "https://raw.githubusercontent.com/iBotPeaches/Apktool/v3.0.3/scripts/osx/apktool",
"format": "file",
"size": 2672,
"sha256": "d0c86f0475ed73bfffa6ac39faa69de7aa2dd40ff5e7e6540f82002be3a38f0b",
"sha256Source": "measured on 2026-10-13 from the osx wrapper",
"sha256VerifiedOn": "2026-10-13",
"size": 2709,
"sha256": "3b9a95c41680c4a758b0b8d6e91e122b2ea3f1a189685a915613ca1ccba00fd9",
"sha256Source": "measured on 2026-10-16 from the v3.0.3 tag; the same osx wrapper serves arm64.",
"sha256VerifiedOn": "2026-10-16",
"executable": true,
"moveTo": "apktool",
"saveAs": "apktool"
}
@@ -250,13 +254,14 @@
"linux-x64": [
{
"id": "wrapper",
"version": "master",
"url": "https://raw.githubusercontent.com/iBotPeaches/Apktool/master/scripts/linux/apktool",
"version": "3.0.3",
"url": "https://raw.githubusercontent.com/iBotPeaches/Apktool/v3.0.3/scripts/linux/apktool",
"format": "file",
"size": 2648,
"sha256": "a851371023c43ca9b9a30218f87f9fd536304283de7b85447183ae811dfd033c",
"sha256Source": "measured on 2026-10-13 from the linux wrapper",
"sha256VerifiedOn": "2026-10-13",
"size": 2685,
"sha256": "493a9c98df968a25c373ca9df871ca77730c82954be4959b74e90adaf91a0e77",
"sha256Source": "measured on 2026-10-16 from the v3.0.3 tag.",
"sha256VerifiedOn": "2026-10-16",
"executable": true,
"moveTo": "apktool",
"saveAs": "apktool"
}
@@ -264,13 +269,14 @@
"linux-arm64": [
{
"id": "wrapper",
"version": "master",
"url": "https://raw.githubusercontent.com/iBotPeaches/Apktool/master/scripts/linux/apktool",
"version": "3.0.3",
"url": "https://raw.githubusercontent.com/iBotPeaches/Apktool/v3.0.3/scripts/linux/apktool",
"format": "file",
"size": 2648,
"sha256": "a851371023c43ca9b9a30218f87f9fd536304283de7b85447183ae811dfd033c",
"sha256Source": "measured on 2026-10-13 from the linux wrapper",
"sha256VerifiedOn": "2026-10-13",
"size": 2685,
"sha256": "493a9c98df968a25c373ca9df871ca77730c82954be4959b74e90adaf91a0e77",
"sha256Source": "measured on 2026-10-16 from the v3.0.3 tag; the same linux wrapper serves arm64. It is only correct where the platform-tools archive is too — see the arch caveat on androidPlatformTools.",
"sha256VerifiedOn": "2026-10-16",
"executable": true,
"moveTo": "apktool",
"saveAs": "apktool"
}
@@ -337,9 +343,9 @@
"format": "zip",
"size": 16110554,
"sha256": "ee39ad5967e95c2a07f04dbcbde96b1a0c916ba376096db5d2f498b7727a5d1d",
"sha256Source": "Same darwin archive as x64 (measured 2026-10-13); it serves Intel and Apple silicon.",
"sha256Source": "Same darwin archive as x64 (measured 2026-10-13); Google publishes one darwin zip for both architectures.",
"sha256VerifiedOn": "2026-10-13",
"note": "The same darwin archive serves Intel and Apple silicon."
"note": "The darwin archive contains x86_64 binaries. On Apple silicon they run only through Rosetta 2; without it adb cannot start and this optional component is reported unavailable."
},
"linux-x64": {
"version": "latest",
@@ -349,7 +355,7 @@
"sha256": "d230f13842f60f782a8645f9c813f8f845bf36089ea7289f28c48f17979313f1",
"sha256Source": "Google publishes no checksum for this rolling asset; measured on 2026-10-13 by downloading the archive and hashing it locally (9,054,187 bytes)",
"sha256VerifiedOn": "2026-10-13",
"note": "The same linux archive serves x86-64 and arm64."
"note": "The linux archive contains x86_64 binaries. Google publishes no arm64 platform-tools and this bundle does not emulate one, so on linux-arm64 the entry below is offered but its probe is expected to fail — reported as an optional component being unavailable, never as a broken install."
},
"linux-arm64": {
"version": "latest",
@@ -357,8 +363,9 @@
"format": "zip",
"size": 9054187,
"sha256": "d230f13842f60f782a8645f9c813f8f845bf36089ea7289f28c48f17979313f1",
"sha256Source": "Same linux archive as x64 (measured 2026-10-13).",
"sha256VerifiedOn": "2026-10-13"
"sha256Source": "Google publishes no arm64 platform-tools; this is the same x86_64 linux archive as x64 (measured 2026-10-13), recorded so the mismatch is explicit instead of implied.",
"sha256VerifiedOn": "2026-10-13",
"note": "Kept so an arm64 host reports a precise failure instead of \"no source\". adb here is x86_64: it runs under qemu/binfmt where the host provides it, and otherwise the probe fails and the report names it."
}
}
},
@@ -370,14 +377,14 @@
"unpack": { "moveTo": "rea-agents" },
"probe": {
"argv": ["{node}", "{vendor}/node_modules/rea-agents/scripts/rea.mjs", "--version"],
"expect": { "regex": "^6\\.4\\.0" }
"expect": { "regex": "^6\\.5\\.0(\\s|$)" }
},
"sources": {
"npm": {
"version": "6.4.0",
"version": "6.5.0",
"registry": "https://registry.npmjs.org/rea-agents",
"integrity": "npm's own dist.integrity is the authoritative check for this tarball: the registry tarball ships NO node_modules (verified: 1161 entries, 0 under node_modules), so setup resolves the dependency closure with `npm install --ignore-scripts` instead of unpacking a tarball.",
"note": "`just unpacking the published tarball cannot work` is a verified fact, not a caution: the CLI would fail with ERR_MODULE_NOT_FOUND on zod."
"integrity": "npm's own dist.integrity is the authoritative check for this tarball: the registry tarball ships NO node_modules (verified: 1650 entries in 6.5.0, 0 under node_modules), so setup resolves the dependency closure with `npm install --ignore-scripts` instead of unpacking a tarball.",
"note": "`just unpacking the published tarball cannot work` is a verified fact, not a caution: the CLI would fail with ERR_MODULE_NOT_FOUND on zod. 6.5.0 is the same engines/bin contract as 6.4.0 (^22.19.0 || ^24.11.0 || >=26.0.0, scripts/rea.mjs) and adds 37 MCP tools (170 -> 207 by rea doctor --json on 2026-10-16; the upstream contract files count 206) plus an optional frida dependency; --ignore-scripts means its install script never runs."
}
}
}
+8 -5
View File
@@ -1,8 +1,9 @@
{
"name": "dsh-plugin-rea",
"version": "2.0.2",
"version": "2.0.9",
"description": "One-plugin bundle: provisions Node/REA/JDK/Ghidra for you, then registers the upstream REA MCP server as mcp__rea__* tools. · 一体包:自动准备 Node/REA/JDK/Ghidra,并把上游 REA 的 MCP 服务接入 DeepSeek Harness。",
"type": "module",
"main": "./index.mjs",
"license": "MIT",
"author": "czcdr",
"homepage": "https://github.com/czcdr/dsh-plugin-rea",
@@ -33,6 +34,7 @@
},
"files": [
"cordis.patch.yml",
"index.mjs",
"bin/",
"scripts/",
"manifest/",
@@ -43,8 +45,7 @@
"CHANGELOG.md",
"check-updates.ps1",
"screenshots.json",
"assets/",
"examples/"
"assets/"
],
"dsh": {
"bundle": {
@@ -57,10 +58,12 @@
"bundle:status": "node ./bin/rea-shim.mjs --shim-status",
"bundle:dry-run": "node ./scripts/setup.mjs --reuse --dry-run",
"bundle:audit": "node ./scripts/audit-manifest.mjs",
"bundle:test": "node --test ./scripts/setup.test.mjs"
"bundle:test": "node --test ./scripts/setup.test.mjs",
"bundle:host-compat": "node ./scripts/host-compat.mjs"
},
"engines": {
"node": "^22.19.0 || ^24.11.0 || >=26.0.0"
"node": "^22.19.0 || ^24.11.0 || >=26.0.0",
"dsh": ">=0.1.7-rc.2 <0.3.0-0"
},
"publishConfig": {
"registry": "https://registry.npmjs.org/"
+98
View File
@@ -430,6 +430,104 @@ if (manifest) {
}
}
/* ------------------------- 11. setup and the manifest describe the same set */
// A component added to the manifest but not to setup.mjs's INSTALL_ORDER was
// silently never provisioned, and nothing here said so. The reverse — a name in
// INSTALL_ORDER with no manifest entry — is skipped by the loop's
// `if (!component) continue`, so it is equally silent.
const installOrderMatch = setupText.match(/const INSTALL_ORDER = \[([^\]]*)\]/);
if (!installOrderMatch) {
add("error", "scripts/setup.mjs", "INSTALL_ORDER was not found, so manifest/setup agreement cannot be checked");
} else {
const order = [...installOrderMatch[1].matchAll(/"([^"]+)"/g)].map((match) => match[1]);
const manifestNames = Object.keys(manifest.components ?? {});
for (const name of manifestNames) {
if (!order.includes(name)) {
add("error", "manifest.components", `"${name}" is not in scripts/setup.mjs INSTALL_ORDER, so setup would silently never provision it`);
}
}
for (const name of order) {
if (!manifestNames.includes(name)) {
add("error", "scripts/setup.mjs", `INSTALL_ORDER names "${name}", which the manifest does not describe`);
}
}
}
/* ---------------------------- 12. the Node engines rule has one wording too */
// REA's engines are restated in the manifest pin, package.json and the node
// probe regex. Only the pin and package.json can be compared as text, but a
// drift between those two would advertise one range and enforce another.
const declaredEngines = manifest.pins?.nodeEngines;
let packageEngines = null;
try {
packageEngines = JSON.parse(readFileSync(path.join(HERE, "..", "package.json"), "utf8")).engines?.node ?? null;
} catch (error) {
add("error", "package.json", `could not be read: ${String(error && error.message)}`);
}
if (!declaredEngines) {
add("error", "manifest.pins.nodeEngines", "is missing, so nothing states REA's engines for the reuse path");
} else if (declaredEngines !== packageEngines) {
add(
"error",
"manifest.pins.nodeEngines",
`is ${JSON.stringify(declaredEngines)} but package.json engines.node is ${JSON.stringify(packageEngines)}`,
);
}
/* ------------------- 13. a pinned version's probe must not accept a prefix */
// The concern is a probe that accepts a *newer* release as the pin, so test the
// expression against the versions that would mean that. The audit is not a
// regex linter: `^6\.4\.0` legitimately rejects 6.4.1 and 6.5.0 — the failure
// mode is only real when the next patch or minor still matches.
const reaAgents = manifest.components?.reaAgents;
const pinnedRea = reaAgents?.sources?.npm?.version;
const reaProbe = reaAgents?.probe?.expect?.regex;
if (pinnedRea && reaProbe) {
let probe = null;
try {
probe = new RegExp(reaProbe, "m");
} catch (error) {
add("error", "reaAgents.probe.expect.regex", `is not a valid regular expression: ${String(error && error.message)}`);
}
if (probe) {
if (!probe.test(pinnedRea)) {
add("error", "reaAgents.probe.expect.regex", `does not match the pinned version ${pinnedRea}`);
}
const [major, minor, patch] = pinnedRea.split(".").map((part) => Number(part));
for (const [label, next] of [
["the next patch", `${major}.${minor}.${patch + 1}`],
["the next minor", `${major}.${minor + 1}.0`],
]) {
if (Number.isFinite(next.split(".")[2]) && probe.test(next)) {
add("error", "reaAgents.probe.expect.regex", `accepts ${next} (${label}), so a newer release passes as the pin`);
}
}
}
}
/* --------------------- 14. a POSIX wrapper must be marked executable */
// copyFileSync carries the archive's 0644 and nothing else changes a mode, while
// REA spawns REA_APKTOOL_COMMAND without a shell: on Linux/macOS an unmarked
// wrapper fails with EACCES even though the component probe (`java -jar`) passes.
// Only the manifest can say which files need the bit, so require it here.
const apktoolWrappers = manifest.components?.apktool?.entries?.byPlatform ?? {};
for (const [platform, list] of Object.entries(apktoolWrappers)) {
if (platform.startsWith("win32")) continue;
for (const entry of list) {
if (entry.executable !== true) {
add(
"error",
`apktool.entries.byPlatform.${platform}`,
`the ${platform} wrapper is not marked "executable": true, so it lands 0644 and cannot be spawned without a shell`,
);
}
}
}
/* ------------------------------------------------------------- summary */
const pinned = fetchEntries.filter((entry) => entry.entry.sha256 && entry.entry.sha256 !== "unverified").length;
+224
View File
@@ -0,0 +1,224 @@
#!/usr/bin/env node
/**
* host-compat — does the range this package declares cover the DSH releases it
* claims to support?
*
* `package.json` carries `engines.dsh`, and that single string is what the
* plugin market reads to say "compatible" instead of "unknown" (the market's
* `deriveHostCompatibility` treats `engines.dsh` as a host declaration and
* checks it against the running host version). A declaration is therefore load
* bearing twice over: it decides whether the market HIDES this plugin from users
* on a matching host, and it decides whether users on a host we never tested are
* told it works.
*
* The two ways it goes wrong are silent. A typo makes the range unparseable, and
* the market then reports `unknown` — a plugin that looks undeclared rather than
* mis-declared. A floor that drifted (say `>=0.2.0`) silently drops hosts the
* README still promises. Neither shows up as a test failure unless something
* compares the STRING to the release list, which is what this file does.
*
* `HOST_RELEASES` is the dsh release line this package is built against, newest
* first, as published on github.com/deepseek-ai/deepseek-harness. It is a
* literal list rather than a GitHub query on purpose: a check that needs the
* network fails on a plane and passes on a total outage, which is the wrong
* shape for a gate. When a release is added upstream, add it here too — that
* edit is the reminder to re-check the row against the new host.
*
* SEMANTICS: this file evaluates the declaration the way `includePrerelease:
* true` does, and that choice is not stylistic. Both consumers of `engines.dsh`
* in this ecosystem pass that option — dshmarket's discovery
* (`satisfiesRange(host, declared, { includePrerelease: true })` in
* `deriveHostCompatibility`) and the host's own peer gate — so a plain semver
* reading, where `0.2.0-rc.2` does NOT satisfy `>=0.1.7-rc.2 <0.3.0-0` (a
* prerelease must be named by a comparator on its own tuple), would disagree
* with the verdict users actually see. The tests pin the option's effect.
*
* Usage: `node scripts/host-compat.mjs` (exit 0 = every listed release is
* covered by the declared range; exit 1 = at least one is not).
*/
import { readFileSync } from 'node:fs'
import path from 'node:path'
import process from 'node:process'
import { fileURLToPath } from 'node:url'
const HERE = path.dirname(fileURLToPath(import.meta.url))
const REPO_ROOT = path.join(HERE, '..')
/**
* Every dsh release this bundle is checked against, newest first.
*
* Taken from the release list of github.com/deepseek-ai/deepseek-harness. The
* floor is the oldest release the plugin has been adapted to; everything above
* it up to the declared ceiling must keep working, so a release that is missing
* here is a release nobody checked.
*/
export const HOST_RELEASES = [
'0.2.1-rc.2',
'0.2.1-rc.1',
'0.2.1-alpha.5',
'0.2.1-alpha.4',
'0.2.1-alpha.3',
'0.2.1-alpha.2',
'0.2.1-alpha.1',
'0.2.0-rc.2',
'0.2.0-rc.1',
'0.1.7-rc.2',
]
/** Hosts that must fall OUTSIDE the declaration, for the other direction. */
export const OUTSIDE_HOSTS = [
'0.3.0',
// Prereleases of the NEXT line are the case a plain `<0.3.0` ceiling lets
// through: with `includePrerelease` they sort below `0.3.0` itself. The
// declared ceiling is `<0.3.0-0`, semver's idiom for "nothing from 0.3.0",
// and these entries are what proves it does what it says.
'0.3.0-rc.1',
'0.3.0-alpha.1',
'0.1.7-rc.1',
'0.1.7-alpha.2',
]
const SEMVER = /^(\d+)\.(\d+)\.(\d+)(?:-([0-9A-Za-z.-]+))?$/
/**
* Split a version into comparable parts.
*
* Build metadata (`+sha`) is dropped, as semver says it must be for precedence
* and as node-semver does for comparison: a host that reports
* `0.2.0-rc.2+build.7` is running 0.2.0-rc.2. Everything else is deliberately
* strict — a leading `v` or a zero-padded identifier is rejected rather than
* guessed at, because this gate exists to catch a declaration nobody checked and
* a guess would put the guess in charge.
* @param {string} version - a plain semver string, no range operators.
* @returns {{ parts: number[], pre: (string|number)[] } | null} null when unparseable.
*/
export function parseVersion(version) {
const cleaned = String(version ?? '').trim().replace(/\+.*$/, '')
const match = SEMVER.exec(cleaned)
if (match === null) return null
const raw = match[4] === undefined ? [] : match[4].split('.')
// semver forbids a zero-padded numeric prerelease identifier (`0.2.1-01`);
// accepting one would make this gate read a typo as a version.
if (raw.some(part => /^\d+$/.test(part) && part.length > 1 && part.startsWith('0'))) return null
const pre = raw.map(part => (/^\d+$/.test(part) ? Number(part) : part))
return { parts: [Number(match[1]), Number(match[2]), Number(match[3])], pre }
}
/**
* Compare two versions the way semver does, prerelease included.
*
* A prerelease sorts BELOW its release (`0.2.0-rc.2 < 0.2.0`), and among
* prereleases an identifier that is numeric sorts below one that is not, which
* is why `0.2.1-alpha.5 < 0.2.1-rc.1` — the whole reason this file cannot lean
* on a numeric comparison.
* @returns -1, 0 or 1; NaN-free because unparseable input throws.
*/
export function compareVersions(a, b) {
const left = parseVersion(a)
const right = parseVersion(b)
if (left === null) throw new Error(`not a version: ${a}`)
if (right === null) throw new Error(`not a version: ${b}`)
for (let index = 0; index < 3; index += 1) {
if (left.parts[index] !== right.parts[index]) return left.parts[index] < right.parts[index] ? -1 : 1
}
if (left.pre.length === 0 && right.pre.length === 0) return 0
if (left.pre.length === 0) return 1
if (right.pre.length === 0) return -1
for (let index = 0; index < Math.max(left.pre.length, right.pre.length); index += 1) {
const l = left.pre[index]
const r = right.pre[index]
if (l === undefined) return -1
if (r === undefined) return 1
if (l === r) continue
const lNum = typeof l === 'number'
const rNum = typeof r === 'number'
if (lNum && rNum) return l < r ? -1 : 1
// Numeric identifiers always have lower precedence than alphanumeric ones.
if (lNum !== rNum) return lNum ? -1 : 1
return String(l) < String(r) ? -1 : 1
}
return 0
}
/**
* Evaluate one `>=floor <ceiling` declaration (the only shape this package
* ships, and the shape the market's `engines.dsh` is compared with).
*
* Deliberately narrow: a wider parser here could disagree with the market's
* semver implementation on a range the manifest would then be trusted for. If a
* future declaration needs `||` or `^`, this function has to grow with it.
* @param {string} version - host version to test.
* @param {string} range - the declared range.
* @returns {boolean} whether the host is inside the declaration.
*/
export function satisfiesDeclaredRange(version, range) {
const match = /^>=\s*(\S+)\s+<\s*(\S+)$/.exec(range.trim())
if (match === null) throw new Error(`unsupported host range: ${range}`)
return compareVersions(version, match[1]) >= 0 && compareVersions(version, match[2]) < 0
}
/**
* The host range this repository declares, `engines.dsh`.
* @param {string} [root] - repository root, for tests.
* @returns {string | null} the declared range, or null when absent.
*/
export function declaredHostRange(root = REPO_ROOT) {
try {
const pkg = JSON.parse(readFileSync(path.join(root, 'package.json'), 'utf8'))
const value = pkg?.engines?.dsh
return typeof value === 'string' ? value : null
} catch {
return null
}
}
/**
* Every release covered by the declaration, and every version that must not be.
* @param {string} range - declared host range.
* @returns {{ covered: string[], uncovered: string[], leaked: string[] }} findings.
*/
export function checkHostCoverage(range) {
const covered = []
const uncovered = []
for (const version of HOST_RELEASES) {
if (satisfiesDeclaredRange(version, range)) covered.push(version)
else uncovered.push(version)
}
const leaked = OUTSIDE_HOSTS.filter(version => satisfiesDeclaredRange(version, range))
return { covered, uncovered, leaked }
}
function main() {
const range = declaredHostRange()
if (range === null) {
console.error('FAIL: package.json declares no engines.dsh — the market would report this plugin as "unknown"')
process.exit(1)
}
let result
try {
result = checkHostCoverage(range)
} catch (error) {
console.error(`FAIL: ${error.message}`)
process.exit(1)
}
console.log(`engines.dsh = ${range}`)
for (const version of HOST_RELEASES) {
const inside = satisfiesDeclaredRange(version, range)
console.log(` ${inside ? 'ok ' : 'MISS'} ${version}`)
}
for (const version of OUTSIDE_HOSTS) {
const inside = satisfiesDeclaredRange(version, range)
console.log(` ${inside ? 'LEAK' : 'ok '} ${version} (must stay outside)`)
}
if (result.uncovered.length > 0) {
console.error(`FAIL: the declaration does not cover ${result.uncovered.join(', ')}`)
process.exit(1)
}
if (result.leaked.length > 0) {
console.error(`FAIL: the declaration also admits ${result.leaked.join(', ')}, which is not a tested host`)
process.exit(1)
}
console.log(`host range ok: ${HOST_RELEASES.length} releases covered, ${OUTSIDE_HOSTS.length} correctly outside`)
}
if (process.argv[1] !== undefined && path.resolve(process.argv[1]) === fileURLToPath(import.meta.url)) main()
+42 -10
View File
@@ -1,4 +1,4 @@
# rea-bundle-bootstrap.ps1 — get from "no Node at all" to a provisioned bundle.
# rea-bundle-bootstrap.ps1 — get from "no Node at all" to a provisioned bundle.
#
# Why this exists: setup.mjs is written in Node, so a machine with no Node
# cannot run it. This script uses only PowerShell + Expand-Archive (no
@@ -30,6 +30,11 @@ function Write-Step([string]$Message) { Write-Host "[bootstrap] $Message" }
# --- where things go -------------------------------------------------------
# Same precedence as setup.mjs, the shim and check-updates.ps1: explicit override,
# then DSH_HOME, then ~/.dsh. An override this script alone ignored would put
# ~800 MB into <DSH_HOME>/plugins/rea-bundle while every reader looked in
# REA_BUNDLE_ROOT, saw an empty bundle, and started a second run.
if (-not $Root -or $Root.Trim() -eq '') { $Root = $env:REA_BUNDLE_ROOT }
if (-not $Root -or $Root.Trim() -eq '') {
$home2 = $env:DSH_HOME
if (-not $home2 -or $home2.Trim() -eq '') { $home2 = Join-Path $HOME '.dsh' }
@@ -39,8 +44,12 @@ $Root = [System.IO.Path]::GetFullPath($Root)
$stack = Join-Path $Root 'stack'
$cache = Join-Path $Root 'cache'
$logs = Join-Path $Root 'logs'
foreach ($dir in @($Root, $stack, $cache, $logs)) {
if (-not (Test-Path $dir)) { New-Item -ItemType Directory -Path $dir -Force | Out-Null }
# -DryRun promises that nothing was changed, so it must not create these either:
# it only reads them, and Test-Path / Get-ChildItem handle a missing path.
if (-not $DryRun) {
foreach ($dir in @($Root, $stack, $cache, $logs)) {
if (-not (Test-Path $dir)) { New-Item -ItemType Directory -Path $dir -Force | Out-Null }
}
}
$pluginRoot = Split-Path -Parent $PSScriptRoot
@@ -84,6 +93,32 @@ $nodeFolder = [System.IO.Path]::GetFileNameWithoutExtension($nodeZipName)
if ($nodeSha256 -eq 'unverified') { throw "the manifest pins no checksum for $hostKey; refusing to download unverified bytes" }
Write-Step "Node source: $hostKey v$nodeVersion"
$script:NodeEngines = [string]$manifest.pins.nodeEngines
if (-not $script:NodeEngines) { throw "the manifest pins no nodeEngines, so Node compatibility cannot be judged" }
# REA's engines come from the manifest pin, which the rest of provisioning already
# treats as the single source of truth. This used to be a second copy of the same
# numbers in a comment, and two copies drift: the audit now asserts that the pin
# keeps the shape this parser understands.
function Test-EnginesMatch([string]$Clean, [string]$Range) {
$versions = @()
foreach ($part in $Clean.Split('.')) {
$n = 0
if (-not [int]::TryParse($part, [ref]$n)) { return $false }
$versions += $n
}
if ($versions.Count -lt 2) { return $false }
foreach ($clause in ($Range -split '\|\|')) {
$c = $clause.Trim()
if ($c -match '^\^(\d+)\.(\d+)\.(\d+)$') {
if ($versions[0] -eq [int]$Matches[1] -and $versions[1] -ge [int]$Matches[2]) { return $true }
} elseif ($c -match '^>=(\d+)\.(\d+)\.(\d+)$') {
if ($versions[0] -gt [int]$Matches[1] -or ($versions[0] -eq [int]$Matches[1] -and $versions[1] -ge [int]$Matches[2])) { return $true }
}
}
return $false
}
function Test-NodeUsable([string]$Exe) {
if (-not $Exe -or -not (Test-Path $Exe)) { return $null }
try {
@@ -91,13 +126,10 @@ function Test-NodeUsable([string]$Exe) {
} catch { return $null }
if (-not $version) { return $null }
$clean = $version.TrimStart('v').Trim()
$parts = $clean.Split('.')
if ($parts.Count -lt 2) { return $null }
$major = [int]$parts[0]
$minor = [int]$parts[1]
# REA's engines: ^22.19.0 || ^24.11.0 || >=26.0.0
$ok = ($major -eq 22 -and $minor -ge 19) -or ($major -eq 24 -and $minor -ge 11) -or ($major -ge 26)
if (-not $ok) { return $null }
# A version that is not three numbers (`vX.Y`) used to reach `[int]` and throw
# under $ErrorActionPreference='Stop', aborting the whole bootstrap instead of
# falling through to the portable download — the one path this script exists for.
if (-not (Test-EnginesMatch $clean $script:NodeEngines)) { return $null }
return [pscustomobject]@{ Path = (Resolve-Path $Exe).Path; Version = $clean }
}
+399 -90
View File
@@ -26,7 +26,7 @@
*/
import { spawnSync } from "node:child_process";
import { createHash } from "node:crypto";
import { appendFileSync, closeSync, copyFileSync, createWriteStream, existsSync, mkdirSync, openSync, readFileSync, readSync, rmSync, statSync, writeFileSync } from "node:fs";
import { appendFileSync, chmodSync, closeSync, copyFileSync, createWriteStream, existsSync, mkdirSync, openSync, readFileSync, readdirSync, readSync, renameSync, rmSync, statSync, writeFileSync } from "node:fs";
import { request as httpRequest } from "node:http";
import { request as httpsRequest } from "node:https";
import { homedir } from "node:os";
@@ -69,6 +69,22 @@ function defaultBundleRoot() {
return path.join(dshHome(), "plugins", "rea-bundle");
}
/**
* A numeric environment override, or the default.
*
* `Number("abc")` is NaN, and NaN does not stay local: `REA_BUNDLE_FETCH_RETRIES=abc`
* produced "gave up after NaN attempt(s)", and a NaN socket timeout behaves as 0
* (immediate expiry) rather than as "no override". Anything that is not a positive
* finite number is therefore ignored, and the effective values are logged below so
* an override is never invisible.
*/
function envNumber(name, fallback) {
const raw = process.env[name];
if (raw === undefined || String(raw).trim() === "") return fallback;
const value = Number(raw);
return Number.isFinite(value) && value > 0 ? value : fallback;
}
function parseArgs(argv) {
const options = {
root: defaultBundleRoot(),
@@ -82,9 +98,9 @@ function parseArgs(argv) {
// be declared dead quickly when the socket never opens, but a slow-but-alive
// transfer must not be killed for taking long overall (measured here:
// 0.35–0.54 MiB/s, where 570 MB legitimately takes ~30 minutes).
connectTimeoutMs: Number(process.env.REA_BUNDLE_CONNECT_TIMEOUT_MS ?? 15000),
idleTimeoutMs: Number(process.env.REA_BUNDLE_IDLE_TIMEOUT_MS ?? 120000),
retries: Number(process.env.REA_BUNDLE_FETCH_RETRIES ?? 4),
connectTimeoutMs: envNumber("REA_BUNDLE_CONNECT_TIMEOUT_MS", 15000),
idleTimeoutMs: envNumber("REA_BUNDLE_IDLE_TIMEOUT_MS", 120000),
retries: envNumber("REA_BUNDLE_FETCH_RETRIES", 4),
};
for (let i = 0; i < argv.length; i += 1) {
const arg = argv[i];
@@ -307,11 +323,20 @@ function openGet(url, { proxy, headers, connectTimeoutMs }) {
},
(res) => done(res, "direct"),
);
let connected = false;
req.on("socket", (socket) => {
socket.setTimeout(connectTimeout);
socket.once("connect", () => socket.setTimeout(idleTimeout));
socket.once("connect", () => {
connected = true;
socket.setTimeout(idleTimeout);
});
});
req.on("timeout", () => req.destroy(new Error(`no response within ${idleTimeout}ms`)));
// The socket carries the connect deadline until it connects and the idle
// deadline afterwards. Reporting the idle number for a connection that never
// happened sent the reader after the wrong setting.
req.on("timeout", () =>
req.destroy(new Error(connected ? `no data within ${idleTimeout}ms` : `no connection within ${connectTimeout}ms`)),
);
req.on("error", reject);
req.end();
return;
@@ -387,11 +412,27 @@ async function download(url, destination, expected, label) {
rmSync(temporary, { force: true });
have = 0;
}
if (have > 0 && have === expected.size) {
log(` ${label}: partial file already complete`);
} else {
log(` downloading ${label}${have > 0 ? ` (resuming at ${(have / 1048576).toFixed(1)} MiB)` : ""}`);
// A `.part` already holding every expected byte is finished work, not a resume
// point: a crash between the last write and the rename leaves exactly this
// shape. Resuming it sends `Range: bytes=<size>-`, which servers answer with
// 416, so four attempts later a complete download was reported failed (and on
// Windows the "resume" simply deleted it and fetched 570 MB again). Verify it
// and finish the rename instead.
if (have > 0 && expected.size && have === expected.size) {
const digest = sha256File(temporary);
const matches = !expected.sha256 || expected.sha256 === "unverified" || digest === expected.sha256;
if (matches) {
log(` ${label}: partial file was already complete; verified and kept`);
rmSync(destination, { force: true });
renameSync(temporary, destination);
return { sha256: digest, size: have, tofu: !expected.sha256 || expected.sha256 === "unverified" };
}
log(` ${label}: partial file is complete but the checksum does not match; starting over`);
rmSync(temporary, { force: true });
have = 0;
}
if (have > 0) log(` downloading ${label} (resuming at ${(have / 1048576).toFixed(1)} MiB)`);
else log(` downloading ${label}`);
for (let attempt = 1; attempt <= attempts; attempt += 1) {
try {
@@ -412,29 +453,29 @@ async function download(url, destination, expected, label) {
let via = null;
let nodeFailure = null;
const tryPowerShellFirst = IS_WINDOWS;
const first = tryPowerShellFirst
? async () => {
rmSync(temporary, { force: true });
const written = downloadViaPowerShell(url, temporary);
log(` ${label}: via powershell (${(written / 1048576).toFixed(1)} MiB)`);
return "powershell";
}
: async () => {
const response = await openViaBestRoute(url, { connectTimeoutMs, have });
await writeBody(response.stream, response.headers, temporary, have, response.status, label);
return response.via;
};
const second = tryPowerShellFirst
? async () => {
const response = await openViaBestRoute(url, { connectTimeoutMs, have });
await writeBody(response.stream, response.headers, temporary, have, response.status, label);
return response.via;
}
: async () => {
rmSync(temporary, { force: true });
const written = downloadViaPowerShell(url, temporary);
return `powershell (${(written / 1048576).toFixed(1)} MiB)`;
};
const viaPowerShell = async () => {
// Resume through the .NET route as well: it is the Windows default, it
// honours the system proxy, and the 570 MB Ghidra archive is the transfer
// that matters. Deleting the partial file first — what this used to do —
// made every retry a full re-download, the opposite of the resumable
// behaviour this module promises. downloadViaPowerShell appends only when
// the server answers 206.
const written = downloadViaPowerShell(url, temporary, have);
log(` ${label}: via powershell (${(written / 1048576).toFixed(1)} MiB)`);
return "powershell";
};
const viaNode = async () => {
// Both routes write into the same .part. Trusting the `have` captured
// before this attempt appended a full 200 response onto the bytes the
// failed route had already written, producing a file that could only fail
// the size gate on the way out. Read the file's real length instead.
have = existsSync(temporary) ? statSync(temporary).size : 0;
const response = await openViaBestRoute(url, { connectTimeoutMs, have });
await writeBody(response.stream, response.headers, temporary, have, response.status, label);
return response.via;
};
const first = tryPowerShellFirst ? viaPowerShell : viaNode;
const second = tryPowerShellFirst ? viaNode : viaPowerShell;
try {
via = await first();
@@ -469,7 +510,6 @@ async function download(url, destination, expected, label) {
throw new Error(`sha256 mismatch: got ${digest}, expected ${expected.sha256}`);
}
rmSync(destination, { force: true });
const { renameSync } = await import("node:fs");
renameSync(temporary, destination);
return { sha256: digest, size: produced, tofu: !expected.sha256 || expected.sha256 === "unverified" };
} catch (error) {
@@ -487,30 +527,46 @@ async function download(url, destination, expected, label) {
}
/**
* Windows fallback route: let PowerShell/.NET perform the transfer.
* The PowerShell/.NET route: on Windows this is the **primary** route (see
* `tryPowerShellFirst` in download()), and everywhere else it exists only as the
* last-resort fallback.
*
* Measured here, this is not redundant with the Node routes. The local proxy
* Measured here, it is not redundant with the Node routes. The local proxy
* rewrites GitHub's release-asset redirect back to itself (an infinite 301
* loop), while a direct Node connection to github.com times out or is reset.
* The .NET stack honours the system proxy, follows the redirect chain to
* release-assets.githubusercontent.com correctly, and reports the full asset
* length. It is slower (~0.13 MiB/s vs ~0.5 MiB/s), so it is a fallback, and
* length. It is slower (~0.13 MiB/s vs ~0.5 MiB/s), which is why the routes are
* ordered per platform rather than universally, and
* the result is verified by the same size + checksum gate as every other route.
*/
function downloadViaPowerShell(url, destination) {
function downloadViaPowerShell(url, destination, have = 0) {
const resume = Number.isFinite(Number(have)) && Number(have) > 0;
const script = [
"$ErrorActionPreference = 'Stop'",
"$ProgressPreference = 'SilentlyContinue'",
`$url = '${url.replace(/'/g, "''")}'`,
`$out = '${destination.replace(/'/g, "''")}'`,
`$have = ${resume ? Math.trunc(Number(have)) : 0}`,
"$req = [System.Net.HttpWebRequest]::Create($url)",
"$req.UserAgent = 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) dsh-plugin-rea-bundle'",
"$req.AllowAutoRedirect = $true",
// The response stage. The body-copy stage needs its own deadline: with only
// Timeout set, a stalled socket inside CopyTo hung forever (spawnSync runs
// with no total timeout by design, because a slow-but-alive 570 MB transfer
// is legitimate).
"$req.Timeout = 60000",
"$req.ReadWriteTimeout = 120000",
// Resume instead of starting over. A server that ignores the range answers
// 200, and the file is then rewritten from the start rather than appended to.
"if ($have -gt 0) { $req.AddRange($have) }",
"$resp = $req.GetResponse()",
"$code = [int]$resp.StatusCode",
"$stream = $resp.GetResponseStream()",
"$file = [System.IO.File]::Create($out)",
"if ($have -gt 0 -and $code -eq 206) { $mode = [System.IO.FileMode]::Append } else { $mode = [System.IO.FileMode]::Create }",
"$file = New-Object System.IO.FileStream($out, $mode, [System.IO.FileAccess]::Write, [System.IO.FileShare]::None)",
"$stream.CopyTo($file)",
"$file.Flush()",
"$file.Close()",
"$stream.Close()",
"$resp.Close()",
@@ -609,24 +665,72 @@ async function writeBody(stream, headers, temporary, have, status, label) {
return seen;
}
/** Extract with the platform's own tools. Returns true on success. */
function extract(archive, destination, format) {
/** The provisioned JDK's own `jar`, when one is resolved — a guaranteed unzipper. */
function jdkJarCommand(context) {
const jdk = context?.resolved?.jdk?.path;
if (typeof jdk !== "string" || jdk === "") return null;
const home = /[\\/]bin[\\/][^\\/]+$/.test(jdk) ? path.dirname(path.dirname(jdk)) : jdk;
const candidate = path.join(home, "bin", `jar${EXE}`);
return existsSync(candidate) ? candidate : null;
}
/**
* Extract with whatever the host actually has.
*
* Ghidra ships only as a zip on every platform, so "no unzip" was a certain
* failure on a minimal Linux image rather than an occasional one. The fallbacks
* are `python3 -m zipfile` and — since a JDK is provisioned before Ghidra — that
* JDK's own `jar`. Tarballs fall back to python3, whose tarfile reads xz without
* the xz binary. A host with none of them is told which tools were tried instead
* of getting a bare "extraction failed".
*/
function extract(archive, destination, format, context) {
mkdirSync(destination, { recursive: true });
const stdio = options.json ? "ignore" : "inherit";
const attempts = [];
if (format === "zip") {
if (IS_WINDOWS) {
const result = spawnSync("powershell", ["-NoProfile", "-NonInteractive", "-Command",
`Expand-Archive -LiteralPath '${archive.replace(/'/g, "''")}' -DestinationPath '${destination.replace(/'/g, "''")}' -Force`],
{ stdio: options.json ? "ignore" : "inherit" });
return result.status === 0;
// bsdtar ships with Windows 10+ and unpacks a zip far faster than
// Expand-Archive's single-threaded .NET implementation — the 543 MB Ghidra
// archive is the case that matters. A failing tar still falls through to
// Expand-Archive.
attempts.push({ command: "tar", args: ["-xf", archive, "-C", destination] });
attempts.push({
command: "powershell",
args: [
"-NoProfile",
"-NonInteractive",
"-Command",
`Expand-Archive -LiteralPath '${archive.replace(/'/g, "''")}' -DestinationPath '${destination.replace(/'/g, "''")}' -Force`,
],
});
} else {
attempts.push({ command: "unzip", args: ["-q", "-o", archive, "-d", destination] });
attempts.push({ command: "python3", args: ["-m", "zipfile", "-e", archive, destination] });
const jar = jdkJarCommand(context);
// `jar xf` unpacks into the working directory, hence cwd.
if (jar) attempts.push({ command: jar, args: ["xf", archive], cwd: destination });
}
const result = spawnSync("unzip", ["-q", "-o", archive, "-d", destination], { stdio: options.json ? "ignore" : "inherit" });
return result.status === 0;
} else if (format === "tar.gz" || format === "tar.xz" || format === "tar") {
attempts.push({ command: "tar", args: ["-xf", archive, "-C", destination] });
attempts.push({
command: "python3",
args: ["-c", "import sys,tarfile;tarfile.open(sys.argv[1]).extractall(sys.argv[2])", archive, destination],
});
} else {
throw new Error(`unsupported archive format: ${format}`);
}
if (format === "tar.gz" || format === "tar.xz") {
const result = spawnSync("tar", ["-xf", archive, "-C", destination], { stdio: options.json ? "ignore" : "inherit" });
return result.status === 0;
let anyToolRan = false;
for (const attempt of attempts) {
const result = spawnSync(attempt.command, attempt.args, { stdio, cwd: attempt.cwd });
if (result.status === 0) return true;
if (!(result.error && result.error.code === "ENOENT")) anyToolRan = true;
}
throw new Error(`unsupported archive format: ${format}`);
if (!anyToolRan) {
log(` no ${format === "zip" ? "zip" : "tar"} extractor available (tried ${attempts.map((attempt) => path.basename(attempt.command)).join(", ")}); install unzip or python3`);
}
return false;
}
/* --------------------------------------------------------------- host probing */
@@ -704,7 +808,28 @@ function substituteProbe(value, context) {
function runProbe(probe, context) {
if (probe.file) {
const file = substituteProbe(probe.file, context);
return { ok: existsSync(file), detail: file };
if (!existsSync(file)) return { ok: false, detail: `${file} is missing` };
// `expect` used to be ignored for file probes, which made the Ghidra probe's
// `expect.contains: "application.version="` dead configuration: any file at
// that path passed, including a directory or a stray placeholder. Read the
// file — these are small metadata files — and honour the contract. The bound
// exists so a probe can never be pointed at a 543 MB tree by accident.
const expect = probe.expect ?? {};
if (expect.contains || expect.regex) {
let text = "";
try {
text = readFileSync(file, "utf8").slice(0, 65536);
} catch (error) {
return { ok: false, detail: `${file} could not be read: ${String(error && error.message)}` };
}
if (expect.contains && !text.includes(expect.contains)) {
return { ok: false, detail: `${file} does not contain "${expect.contains}"` };
}
if (expect.regex && !new RegExp(expect.regex, "m").test(text)) {
return { ok: false, detail: `${file} does not match /${expect.regex}/` };
}
}
return { ok: true, detail: file };
}
if (!probe.argv) return { ok: true, detail: "no probe" };
const argv = probe.argv.map((value) => substituteProbe(value, context));
@@ -829,6 +954,21 @@ function reuseExisting(name, root) {
/* ------------------------------------------------------------------- manifest */
/**
* JSON from disk, tolerating the UTF-8 BOM a Windows editor writes.
*
* `JSON.parse` rejects a leading BOM, and every caller in this file treats a
* parse failure as "no record". A BOM-ed mirror.json therefore meant "this is
* not a mirrored run", so setup went on to re-mirror the previous manifest;
* a BOM-ed bundle.json meant "nothing is provisioned". `bin/rea-shim.mjs`
* carries the same helper — the two must agree, or the same file means two
* different things depending on which one read it (a hand-edited record with
* PowerShell's BOM is exactly the case that produced this).
*/
function readJson(file) {
return JSON.parse(readFileSync(file, "utf8").replace(/^\uFEFF/, ""));
}
/**
* Where this run's plugin files actually live.
*
@@ -848,7 +988,7 @@ function sourceRoot() {
const self = path.resolve(PLUGIN_ROOT);
if (!samePath(self, options.root)) return self;
try {
const record = JSON.parse(readFileSync(path.join(options.root, "mirror.json"), "utf8"));
const record = readJson(path.join(options.root, "mirror.json"));
const recorded = record?.pluginRoot;
if (typeof recorded === "string" && recorded.trim() !== "") {
const candidate = path.resolve(recorded);
@@ -862,10 +1002,29 @@ function sourceRoot() {
function loadManifest() {
const file = path.join(sourceRoot(), "manifest", "bundle.manifest.json");
const parsed = JSON.parse(readFileSync(file, "utf8"));
const parsed = readJson(file);
return parsed;
}
/**
* Read a bundle.json, tolerating one that is not valid JSON.
*
* A half-written file (disk full, a kill mid-write) used to throw straight out of
* provisioning — the run died before it could replace the file it could not
* parse, and the shim then failed on it forever. Treating it as absent lets the
* normal path read what is usable and rewrite the record at the end.
*/
function readBundleJson(file) {
if (!existsSync(file)) return null;
try {
const parsed = readJson(file);
return parsed && typeof parsed === "object" && !Array.isArray(parsed) ? parsed : null;
} catch (error) {
log(` ${path.basename(file)} is not readable JSON (${String(error && error.message)}); ignoring it for this run`);
return null;
}
}
const INSTALL_ORDER = ["node", "jdk", "ghidra", "apktool", "jadxHeadlessMcp", "androidPlatformTools", "reaAgents"];
/* -------------------------------------------------------------------- stages */
@@ -878,8 +1037,26 @@ async function provisionComponent(name, component, context) {
// makes the recorded-state check below redundant (and its message wrong).
const reused = reuseExisting(name, root);
if (reused) {
step(name, "skip", `reusing existing ${name} from ${reused.how}: ${reused.path}`);
return { path: reused.path, reused: true };
// A candidate is a claim until the component's own probe agrees. Adoption
// tested only that javac / analyzeHeadless / node exists, so a JDK 11 (Ghidra
// 12.1 needs 21+) or a Ghidra 11 install was recorded in bundle.json and the
// failure surfaced much later, inside REA. Run the same probe a provisioned
// copy has to pass, with the candidate planted so the probe's own tokens
// ({jvm}, {ghidra}) resolve to it instead of to a vendor path it never had.
const planted = context.resolved[name];
context.resolved[name] = { path: reused.path };
let probe;
try {
probe = runProbe(component.probe ?? {}, context);
} finally {
if (planted === undefined) delete context.resolved[name];
else context.resolved[name] = planted;
}
if (probe.ok) {
step(name, "skip", `reusing existing ${name} from ${reused.how}: ${reused.path}`);
return { path: reused.path, reused: true };
}
step(name, "plan", `not reusing ${reused.how} ${name} at ${reused.path}: ${probe.detail}`);
}
// 2) already present in the bundle and still verifiable?
@@ -889,8 +1066,8 @@ async function provisionComponent(name, component, context) {
// deleted afterwards. Only a recorded path that exists AND passes the probe is
// trusted; anything else falls through to provisioning again.
if (existsSync(path.join(root, "bundle.json"))) {
const current = JSON.parse(readFileSync(path.join(root, "bundle.json"), "utf8"));
const recorded = current.components?.[name];
const current = readBundleJson(path.join(root, "bundle.json"));
const recorded = current?.components?.[name];
if (recorded && typeof recorded.path === "string" && existsSync(recorded.path)) {
const probe = runProbe(component.probe ?? {}, context);
if (probe.ok) {
@@ -955,8 +1132,19 @@ async function provisionComponent(name, component, context) {
}
mkdirSync(targetDirectory, { recursive: true });
const target = path.join(targetDirectory, entry.saveAs ?? archiveName);
const { copyFileSync } = await import("node:fs");
copyFileSync(archivePath, target);
// A wrapper script has to be executable. copyFileSync lands the 0644 that
// the archive carries and nothing else in this file changes a mode, while
// REA starts REA_APKTOOL_COMMAND through execFile/no shell — so on POSIX a
// 0644 wrapper fails with EACCES. The component's own probe misses it
// because it runs `java -jar` and never executes the wrapper.
if (entry.executable === true) {
try {
chmodSync(target, 0o755);
} catch {
/* a filesystem that cannot carry the bit must not fail the component */
}
}
placed.push({ target, sha256: fetched.sha256, tofu: fetched.tofu });
}
if (options.dryRun) return { planned: true };
@@ -1015,7 +1203,6 @@ async function provisionComponent(name, component, context) {
const targetDirectory = path.join(vendor, component.unpack?.moveTo ?? name);
mkdirSync(targetDirectory, { recursive: true });
const target = path.join(targetDirectory, source.saveAs ?? archiveName);
const { copyFileSync } = await import("node:fs");
copyFileSync(archivePath, target);
const probe = runProbe(component.probe ?? {}, context);
step(name, probe.ok ? "ok" : "FAIL", `${target}${probe.ok ? "" : " — " + probe.detail}`);
@@ -1026,7 +1213,7 @@ async function provisionComponent(name, component, context) {
rmSync(target, { recursive: true, force: true });
const staging = target + ".staging";
rmSync(staging, { recursive: true, force: true });
const unpacked = extract(archivePath, staging, source.format);
const unpacked = extract(archivePath, staging, source.format, context);
if (!unpacked) {
step(name, "FAIL", `extraction failed for ${archiveName}`);
return null;
@@ -1036,12 +1223,10 @@ async function provisionComponent(name, component, context) {
for (let level = 0; level < strip; level += 1) {
// Descend into the archive's single root directory. If the level is not a
// single directory (some archives are already flat), stop rather than guess.
const { readdirSync } = await import("node:fs");
const entries = readdirSync(effective).filter((entry) => !entry.startsWith("."));
if (entries.length !== 1 || !statSync(path.join(effective, entries[0])).isDirectory()) break;
effective = path.join(effective, entries[0]);
}
const { renameSync } = await import("node:fs");
rmSync(target, { recursive: true, force: true });
renameSync(effective, target);
rmSync(staging, { recursive: true, force: true });
@@ -1094,6 +1279,15 @@ async function provisionReaAgents(component, context) {
return { planned: true, version };
}
// `--offline is a promise, not a hint` — and this is the one step that cannot be
// served from the cache: it is an npm install, not a download. It used to run
// anyway, so a machine with every archive pre-cached still went to the registry
// (or failed there) after the flag had promised it would not.
if (options.offline) {
step("reaAgents", "FAIL", `--offline and the vendored REA at ${vendor} is not usable; npm install needs the registry (drop --offline or pre-install it)`);
return null;
}
/*
* A plain pack+unpack cannot work here: the published tarball carries no
* node_modules, so the CLI would fail with ERR_MODULE_NOT_FOUND on its own
@@ -1149,6 +1343,10 @@ async function provisionReaAgents(component, context) {
});
if (install.status !== 0) {
step("reaAgents", "FAIL", `npm install failed (${install.status}): ${`${install.stderr ?? ""}${install.stdout ?? ""}`.trim().split("\n").slice(-3).join(" | ").slice(0, 300)}`);
// Every other component can fall back to the proxy route this script resolves
// itself; npm cannot, and this step is the one that fails most often on a
// network where the registry needs a mirror.
log(" npm reads HTTPS_PROXY/HTTP_PROXY and NPM_CONFIG_REGISTRY from the environment (this script's own proxy routing does not apply here)");
return null;
}
@@ -1174,6 +1372,9 @@ log(` root ${options.root}`);
log(` cache ${options.cache}`);
log(` host ${hostKey()} node ${process.versions.node}`);
log(` mode ${options.dryRun ? "dry-run " : ""}${options.reuse ? "reuse " : ""}${options.offline ? "offline" : "online"}`.trimEnd());
// Printed so an environment override is visible: a typo used to reach the socket
// layer as NaN (and as "gave up after NaN attempt(s)") with nothing to see.
log(` limits connect=${options.connectTimeoutMs}ms idle=${options.idleTimeoutMs}ms retries=${options.retries}`);
log("");
const manifest = loadManifest();
@@ -1185,17 +1386,72 @@ const context = {
};
const results = {};
/**
* What a previous run recorded.
*
* `--only` names a subset of components, so this run's `results` describes only
* that subset — and writing bundle.json straight from `results` erased every
* other component: the JDK and Ghidra disappeared from `paths`, the shim stopped
* exporting JAVA_HOME/GHIDRA_INSTALL_DIR, so `setup.mjs --only androidPlatformTools`
* silently disabled native analysis on a fully provisioned machine, and the next
* full run had to unpack 750 MB again. Components this run does not touch are
* carried over instead.
*/
const previous = readBundleJson(path.join(options.root, "bundle.json"));
/** One untouched component's record, kept only while its recorded path exists. */
function carriedOver(name) {
const record = previous?.components?.[name];
if (!record || typeof record !== "object") return null;
if (typeof record.path === "string" && record.path !== "") {
if (existsSync(record.path)) return record;
return { ...record, path: null, unavailable: true, reason: "the recorded path is gone" };
}
return record;
}
for (const name of INSTALL_ORDER) {
const component = manifest.components[name];
if (!component) continue;
if (options.only && !options.only.includes(name)) continue;
const missingRequirement = (component.requires ?? []).find((requirement) => !context.resolved[requirement]);
if (missingRequirement && !options.only) {
step(name, "FAIL", `skipped: required component "${missingRequirement}" is unavailable`);
results[name] = null;
continue;
if (missingRequirement) {
// Under --only the requirement is not being provisioned this run, but the
// previous run's record still answers it — and that is the same answer a full
// run would use. Skipping the check entirely made `--only apktool` resolve
// {jvm} to the vendor/jdk that only a full run creates and report FAIL, on a
// machine whose JDK was recorded right next to it.
let carried = options.only ? carriedOver(missingRequirement) : null;
if (!carried && options.only) {
// A requirement can also be answered by the machine instead of the record.
// `node` always can: a Node is already running this script, and
// provisionReaAgents falls back to exactly that interpreter — without this,
// `--only reaAgents` on a fresh root failed on a "missing" node the run is
// itself made of. The others are answered when the caller passed --reuse.
if (missingRequirement === "node") {
carried = { path: process.execPath, reused: true, how: "the interpreter running setup" };
} else {
const reused = reuseExisting(missingRequirement, options.root);
if (reused) carried = { path: reused.path, reused: true, how: reused.how };
}
}
if (!carried || typeof carried.path !== "string" || !existsSync(carried.path)) {
step(name, "FAIL", `skipped: required component "${missingRequirement}" is unavailable`);
results[name] = null;
continue;
}
context.resolved[missingRequirement] = carried;
}
let resolvedValue = null;
try {
resolvedValue = await provisionComponent(name, component, context);
} catch (error) {
// An unexpected throw used to abort the run before it could write a summary
// or record why its lock is still there. Report it as this component's
// failure instead, so the exit code and the log stay truthful.
step(name, "FAIL", `unexpected error: ${String(error?.message ?? error)}`);
resolvedValue = null;
}
const resolvedValue = await provisionComponent(name, component, context);
results[name] = resolvedValue;
// A component satisfies another's `requires` when it has a real path — or when
// this run only *planned* it. In dry-run every component returns
@@ -1208,27 +1464,36 @@ for (const name of INSTALL_ORDER) {
}
}
/* Components this run does not touch keep the record the previous run wrote. */
const merged = { ...results };
for (const name of INSTALL_ORDER) {
if (Object.prototype.hasOwnProperty.call(merged, name)) continue;
if (!manifest.components[name]) continue;
const carried = carriedOver(name);
if (carried) merged[name] = carried;
}
/* browser is adopted, never downloaded */
const browser = reuseExisting("browser", options.root);
if (browser) step("browser", "skip", `adopting ${browser.path}`);
/* --------------------------------------------------------------- wire + probe */
const nodePath = typeof results.node?.path === "string" && existsSync(results.node.path)
? results.node.path
const nodePath = typeof merged.node?.path === "string" && existsSync(merged.node.path)
? merged.node.path
: process.execPath;
const paths = {
node: nodePath,
reaEntry: path.join(options.root, "vendor", "node_modules", "rea-agents", "scripts", "rea.mjs"),
ghidra: results.ghidra?.path ?? null,
jdk: results.jdk?.path ?? null,
adb: results.androidPlatformTools?.path ? path.join(results.androidPlatformTools.path, `adb${EXE}`) : null,
apktoolCommand: results.apktool?.path
? path.join(results.apktool.path, IS_WINDOWS ? "apktool.bat" : "apktool")
ghidra: merged.ghidra?.path ?? null,
jdk: merged.jdk?.path ?? null,
adb: merged.androidPlatformTools?.path ? path.join(merged.androidPlatformTools.path, `adb${EXE}`) : null,
apktoolCommand: merged.apktool?.path
? path.join(merged.apktool.path, IS_WINDOWS ? "apktool.bat" : "apktool")
: null,
jadxJar: results.jadxHeadlessMcp?.path
? path.join(results.jadxHeadlessMcp.path, results.jadxHeadlessMcp.jarName ?? "")
jadxJar: merged.jadxHeadlessMcp?.path
? path.join(merged.jadxHeadlessMcp.path, merged.jadxHeadlessMcp.jarName ?? "")
: null,
browser: browser?.path ?? null,
ilspycmd: process.env.REA_ILSPY_CMD_PATH ?? null,
@@ -1242,17 +1507,17 @@ if (!options.dryRun) {
plugin: { name: "dsh-plugin-rea", root: sourceRoot() },
paths,
versions: {
reaAgents: results.reaAgents?.version ?? null,
ghidra: results.ghidra?.version ?? null,
jdk: results.jdk?.version ?? null,
apktool: results.apktool?.version ?? null,
jadxHeadlessMcp: results.jadxHeadlessMcp?.version ?? null,
node: results.node?.version ?? process.versions.node,
reaAgents: merged.reaAgents?.version ?? null,
ghidra: merged.ghidra?.version ?? null,
jdk: merged.jdk?.version ?? null,
apktool: merged.apktool?.version ?? null,
jadxHeadlessMcp: merged.jadxHeadlessMcp?.version ?? null,
node: merged.node?.version ?? process.versions.node,
},
components: Object.fromEntries(
Object.entries(results).map(([name, value]) => [name, value ?? { path: null, unavailable: true }]),
Object.entries(merged).map(([name, value]) => [name, value ?? { path: null, unavailable: true }]),
),
unverifiedChecksums: Object.entries(results)
unverifiedChecksums: Object.entries(merged)
.filter(([, value]) => value?.tofu)
.map(([name]) => name),
};
@@ -1303,7 +1568,11 @@ if (!options.dryRun) {
const shim = spawnSync(process.execPath, [mirroredShim, "--version"], {
encoding: "utf8",
timeout: 120000,
env: { ...process.env, REA_BUNDLE_ROOT: options.root },
// Pin the bundle path: this step exists to confirm the shim relays the REA
// THIS run provisioned. Under the default `auto` mode a machine that also
// has a newer global REA answers with its own version, and the step would
// report success for something this run never verified.
env: { ...process.env, REA_BUNDLE_ROOT: options.root, REA_BUNDLE_LAUNCH: "bundle" },
});
const output = `${shim.stdout ?? ""}${shim.stderr ?? ""}`.trim();
const ok = shim.status === 0 && /\d+\.\d+\.\d+/.test(output);
@@ -1382,6 +1651,31 @@ function mirrorLauncher(context) {
*/
function writeMirrorRecord(context, source) {
const record = { schema: 1, pluginRoot: source, mirroredAt: new Date().toISOString() };
// Provenance for the self-healing check: the mirrored shim compares itself
// with `<pluginRoot>/bin/rea-shim.mjs` and re-execs that one when the plugin
// has moved on. These two fields are what `--shim-status` shows when the
// comparison cannot be made (missing checkout, unreadable file).
try {
record.shimSha256 = createHash("sha256")
.update(readFileSync(path.join(context.root, "bin", "rea-shim.mjs")))
.digest("hex");
} catch {
/* the copy is missing or unreadable; the shim reports the comparison itself */
}
try {
record.pluginVersion = readJson(path.join(source, "package.json")).version ?? null;
} catch {
record.pluginVersion = null;
}
// The MCP child the harness starts has every DSH_* variable stripped, so the
// mirrored shim cannot re-derive a CUSTOM home from its environment — it would
// read `~/.dsh` and miss a duplicate `serverName: rea` living in the real one.
// Record the home setup ran under; the shim prefers the environment, then this.
try {
record.dshHome = dshHome();
} catch {
record.dshHome = null;
}
try {
mkdirSync(context.root, { recursive: true });
writeFileSync(path.join(context.root, "mirror.json"), JSON.stringify(record, null, 2) + "\n");
@@ -1417,6 +1711,25 @@ const failures = state.steps.filter((entry) => entry.status === "FAIL");
const failed = failures.filter((entry) => !optionalFailure(entry));
const optionalUnavailable = failures.filter(optionalFailure);
/*
* Lock bookkeeping is not presentation, and it used to sit inside the
* non-`--json` branch: a successful `--json` run therefore left the lock behind,
* and the shim then reported "provisioning started in the background" for six
* hours on a fully provisioned machine. A failing run keeps the lock on purpose —
* so a restart loop cannot re-download forever — but marks it failed, which lets
* the shim say what actually happened instead of claiming a run is in flight.
*/
const lockFile = path.join(options.root, "setup.lock");
if (failed.length === 0) {
rmSync(lockFile, { force: true });
} else {
try {
writeFileSync(lockFile, JSON.stringify({ state: "failed", pid: process.pid, at: Date.now(), exit: 1, log: logFilePath() }) + "\n");
} catch {
/* the lock is advisory: a read-only root must not break the report */
}
}
if (options.json) {
process.stdout.write(
JSON.stringify(
@@ -1436,10 +1749,6 @@ if (options.json) {
}
log("");
if (failed.length === 0) {
// Release the provisioning lock the shim set, so a later first run on a new
// machine (or after a manual wipe) can start again. A failed run keeps the
// lock, but the shim treats a lock older than six hours as stale.
rmSync(path.join(options.root, "setup.lock"), { force: true });
log("restart DSH (or reload the profile) so the MCP row reconnects.");
// Point at the mirrored copy, not at this checkout's: the mirror resolves its
// bundle from its own location, so this command is right for the root this run
+1479 -6
View File
File diff suppressed because it is too large. Load diff
+55 -3
View File
@@ -12,11 +12,36 @@ function Check($name, [bool]$ok, $detail) {
}
Write-Host "`n=== 1. JS syntax ===" -ForegroundColor Cyan
foreach ($f in @('bin\rea-shim.mjs','scripts\setup.mjs','scripts\audit-manifest.mjs','scripts\setup.test.mjs')) {
foreach ($f in @('index.mjs','bin\rea-shim.mjs','scripts\setup.mjs','scripts\audit-manifest.mjs','scripts\setup.test.mjs','scripts\host-compat.mjs')) {
$out = & node --check $f 2>&1
Check $f ($LASTEXITCODE -eq 0) ($out -join ' ')
}
Write-Host "`n=== 1b. both JSON readers tolerate a BOM ===" -ForegroundColor Cyan
# bin/rea-shim.mjs and scripts/setup.mjs read the same mirror.json and bundle.json.
# A BOM strip on one side only makes one file mean "there is a checkout" to the
# shim and "there is none" to setup, and the mirrored setup then re-mirrors the
# previous manifest. A bare JSON.parse(readFileSync(...)) is that divergence.
$jsBom = Join-Path $env:TEMP 'rea-verify-bom.cjs'
@'
const fs = require("node:fs");
const blind = [];
for (const file of ["bin/rea-shim.mjs", "scripts/setup.mjs"]) {
const text = fs.readFileSync(file, "utf8");
for (const m of text.matchAll(/JSON\.parse\(\s*readFileSync\(/g)) {
// readJson() itself is the strip; anything else must not parse a file raw.
if (!/replace\(/.test(text.slice(m.index, m.index + 200))) {
blind.push(file + "@" + text.slice(0, m.index).split("\n").length);
}
}
if (!text.includes("replace(/^\\uFEFF/")) blind.push(file + ": no U+FEFF strip");
}
console.log("bom-blind reads: " + JSON.stringify(blind));
if (blind.length) process.exit(1);
'@ | Set-Content $jsBom -Encoding utf8
& node $jsBom 2>&1 | ForEach-Object { Write-Host " $_" -ForegroundColor Gray }
Check 'no BOM-blind JSON read' ($LASTEXITCODE -eq 0) 'see above'
Write-Host "`n=== 2. PowerShell parse + BOM ===" -ForegroundColor Cyan
# This script checks itself too: it is edited as often as the others, and an editor
# that drops the BOM would otherwise go unnoticed until someone adds CJK text here.
@@ -73,10 +98,37 @@ Write-Host "`n=== 6. package payload lists what the bundle needs ===" -Foregroun
# -Encoding UTF8 is required: the package description contains CJK text, and
# PowerShell 5.1 reads a BOM-less file as ANSI, which breaks ConvertFrom-Json.
$pkg = Get-Content (Join-Path $root 'package.json') -Raw -Encoding UTF8 | ConvertFrom-Json
foreach ($entry in @('bin/', 'scripts/', 'manifest/', 'cordis.patch.yml')) {
foreach ($entry in @('index.mjs', 'bin/', 'scripts/', 'manifest/', 'cordis.patch.yml')) {
Check "files[] includes $entry" (@($pkg.files) -contains $entry) ''
}
foreach ($script in @('bundle:setup', 'bundle:status', 'bundle:audit', 'bundle:test')) {
# The market's update path asks whether package.json's main/exports points at a
# file that exists, and rolls the update back when it does not (dshmarket's
# hasLoadableEntry). This package declared neither until 2.0.5, so every update of
# a healthy profile failed with "updated build has no loadable entry". Asserting
# the pointer resolves keeps that from returning silently.
Check 'package.json declares main' (-not [string]::IsNullOrWhiteSpace($pkg.main)) ''
Check 'main resolves to a shipped file' ($pkg.main -and (Test-Path (Join-Path $root $pkg.main))) "main=$($pkg.main)"
# The market decides "compatible" vs "unknown" from engines.dsh, and hides the
# plugin from hosts whose version does not match. An absent or unparseable
# declaration therefore reads as "nothing to say" rather than as a defect, which
# is why the coverage check runs here as well as in CI.
Check 'package.json declares engines.dsh' (-not [string]::IsNullOrWhiteSpace($pkg.engines.dsh)) ''
$compat = & node scripts\host-compat.mjs 2>&1
Check 'engines.dsh covers every released DSH host' ($LASTEXITCODE -eq 0) (($compat | Select-Object -Last 2) -join ' ')
# Both READMEs quote the range verbatim, and a stale quote is invisible: JSON,
# host-compat and the audit all agree while the sentence a human reads names a
# different ceiling. That is exactly how `<0.3.0` survived in the docs after the
# declaration moved to `<0.3.0-0`, so the comparison lives here now.
foreach ($doc in @('README.md', 'README.zh-CN.md')) {
$text = Get-Content (Join-Path $root $doc) -Raw -Encoding UTF8
Check "$doc quotes the declared engines.dsh" $text.Contains([string]$pkg.engines.dsh) "expected $($pkg.engines.dsh)"
# The pin example is copy-pasteable, so it must name a ref that will exist: the
# version this tree publishes. It has drifted twice (v2.0.0 long after 2.0.5,
# then v2.0.8 while the tree was 2.0.9), and both times every other signal was
# green because nothing compared the two.
Check "$doc pins the current version" $text.Contains("dsh-plugin-rea#v$($pkg.version)") "expected dsh-plugin-rea#v$($pkg.version)"
}
foreach ($script in @('bundle:setup', 'bundle:status', 'bundle:audit', 'bundle:test', 'bundle:host-compat')) {
Check "scripts.$script exists" ($pkg.scripts.PSObject.Properties.Name -contains $script) ''
}