* fix(electron): reject non-executable paths and contain launch rejections
An executable_path that runs but never prints the inspector line made
Playwright's electron.launch() leave three waits unawaited. They rejected
unhandled and Node ended the CLI or the MCP server (#2269).
Check the resolved executable_path is a regular file, and on non-Windows
hosts that it passes X_OK, before any launch. Wrap only the launch call
in a scoped unhandledRejection listener that absorbs rejections whose
stack is in playwright-core/lib and re-throws anything else.
Tested on Linux with Node 22.23.1, 24.18.0 and 26.11.1.
Refs #2269🤖 Generated with Claude Code (Sonnet 5.5, verified by Opus 5.5) 🦊
* fix(electron): absorb only this launch's orphaned rejections
The listener around the launch dropped every rejection with a
playwright-core stack, so an unrelated Playwright failure during the
launch window would vanish. Hold those rejections instead. When the
launch settles, drop only the ones whose message is part of the launch's
own error (the orphans of a failed launch). Raise every other one again
through the unhandledRejection path, so the process still fails. A launch
that succeeds raises all of them. Rejections without a Playwright stack
still fail at once.
One setImmediate turn before the decision lets rejections queued by the
settling tick arrive first. It is a single turn, not a timer. A shared
set of absorbed errors stops overlapping launches from raising each
other's orphans. The listener is removed on every path.
Add child-process tests that drive the built provider through its launch
seam, since Vitest's worker has its own unhandledRejection listener.
🤖 Generated with Claude Code (Sonnet 5.5, verified by Opus 5.5) 🦊
* test(electron): distinguish launch success from capture ownership
* test(electron): assert cleanup ownership failures around overlapping launches
* fix(comparison): recheck historical calls-only changes (#2223)
* fix(comparison): preserve incomplete function call evidence
Keep missing, unknown, partial and unresolved call relationships unknown in function comparisons and reconstruction claims. Aggregate IDA caller and callee coverage without overwriting uncertainty.
Validate with 144 focused tests, typecheck/lint, docs checks and test discovery. Real providers were not verified; the full headless IDA session suite requires unavailable Windows native controls.
* fix(comparison): recheck historical calls-only changes
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(android): make DEX descriptor pattern portable
* fix(ci): isolate Linux Hopper process fixture inventory
* fix(browser): avoid replaying observed launch rejections
* style: format comparison and IDA sources
* test(electron): allow cold macOS ownership setup
* test(electron): allow cold macOS setup in the early-exit path test
The test that runs an executable which exits at once builds the Swift
ownership helper twice on macOS: in the CLI process, and again in the
MCP server process on its first tool call that prepares. A cold Swift
cache can spend up to 120 seconds on a build. The test had only the
30 second default.
Give the test 180 seconds, as e1579ec31 did for
electronLaunchRejections.test.ts and 65a809a0d did for
webSourceLocationCli.test.ts. cli.run has no deadline of its own, so
it needs no timeoutMs.
🤖 Generated with Claude Code (Sonnet 5.5, verified by Opus 5.5) 🦊
* test(ci): isolate report fixtures and declare PowerPC Java bounds
---------
Co-authored-by: Claude Opus 5.5 <opus@fox.invalid>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: Lucien <164857476+dakjdakd@users.noreply.github.com>
* fix(cli): refuse an integrity policy the binary analyze lane cannot honor
`analyze` declares --integrity-policy for its JavaScript lane and forwards it
only there. On a binary target the flag was accepted and then dropped:
inspect_artifact-style integrity handling does not exist on this path, and
binary_overview is a strictObject with no input at all, so the request could
never take effect and the run proceeded as if the caller had not asked.
The same branch already refuses source_paths on a binary target with a typed
AnalysisInputError, and the sibling `inspect` command refuses the flag outright,
so the refusal is the established handling here. The policy option carries a
default, so the check keys off a value that differs from the owner schema's own
default rather than the flag's presence; otherwise every binary analyze would be
refused. The default is read from the schema instead of being restated.
Verified with 5 integrity policy CLI cases, check:fast and
verify:test-discovery (1023 files) on macOS arm64, and by running the built CLI
for the no-flag, explicit-default and record-and-continue cases. The regression
case failed with the command reaching the binary lane before the change.
* fix(cli): reject explicit default integrity policies for binaries
* fix(cli): keep an omitted integrity policy absent on the binary lane
`javascriptApplicationOptions.shape.integrityPolicy` fills an omission with its
default, and `.optional()` keeps reporting that parsed value, so
`options.integrityPolicy !== undefined` was true for every `analyze <target>`
call: ordinary binary analysis was rejected before the target was even read.
List `z.undefined()` first in a union so an omitted option stays absent while
the owner still validates a supplied value, and reapply the owner's metadata so
help and `--schema` keep describing the option.
Validation: tests/boundary/cli/javascriptIntegrityPolicyCli.test.ts reports
1 failed | 5 passed without this change (the case "still routes a binary
analyze target without an integrity policy" returned invalid_request for
integrity_policy instead of reaching the target) and 6 passed with it;
`npm run check:fast` passes on macOS 26.5.1 arm64 with Node.js 24.18.0.
* fix(cli): omit integrity defaults from binary option schema
* test(ci): isolate report fixtures and declare PowerPC Java bounds
---------
Co-authored-by: 吴佳浩 <wujiahao@supremind.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(javascript): exclude TypeScript this annotations from runtime parameters
* test(javascript): assert exact typed argument bindings
* test(apple): keep dyld slice regression portable
* fix(process): report no remaining time after a provider deadline fires
Node timers can fire while performance.now() is still a fraction of a
millisecond short of the monotonic deadline. ProviderStartupDeadline then
reported interruption "timeout" and an aborted signal while remainingMs()
returned ceil(0.x) = 1, so polling loops guarded by remainingMs() > 0
(Hopper and Ghidra startup) could run past an elapsed deadline, and the
real-timer test "aborts a bounded wait when its real monotonic deadline
expires" failed intermittently in CI (expected 1 to be 0).
Once the deadline's own timer has classified a timeout, report zero
remaining time. A deterministic regression test fires the timer with the
monotonic clock 0.4 ms short of the deadline.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(ci): restore descriptor and fixture validation
* test(ci): retain verification coverage within lint limits
* fix(test): share admitted Ghidra profile installation fixtures
* fix(ci): isolate timing report fixtures from workflow state
---------
Co-authored-by: hhang <169027046+xu91102@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Remove OS-based assumptions about long and non-ASCII string recognition. Require the dedicated fragmented fixture, independently verify full source bytes, and reject gaps, overlaps, incomplete reads and corrupted decoding. Report unrecognized native strings explicitly and retain the actual procedure count.
Verified through 14 verifier regressions and the complete macOS arm64 Hopper 6.6.0-demo CLI/MCP lane with cleanup.
Declare observed Ghidra and Hopper facets, preserve unavailable pseudocode, and attach dossier call-set coverage. Version the analysis profiles to invalidate older snapshots and remove the retired include_assembly fixture branch.
Add MCP comparison and snapshot-replay regressions plus real-provider checks for empty and unavailable facets. Fixes#2980.
* fix(gemini): compare consolidated allowlists through one normalizer
geminiServerPolicyBlock consolidates allowlists by normalized intersection but
tested the consolidated list against the raw server key. Because the
intersection keeps the spelling of whichever policy came first, the same pair of
policies blocked the server in one arrangement and allowed it in the other, and
the excluded test had the mirror-image asymmetry.
Normalize the comparison key once and use the same normalizer for the exclusion
test, the intersection and the membership test. The messages keep the
caller-facing server key, so lookup and display stay separate.
Verified with 88 client profile, lifecycle and registration tests, check:fast
and verify:test-discovery on macOS arm64. The regression test drives setup twice
with the same policies in both arrangements; before the change one arrangement
failed and the other passed.
* test(gemini): cover normalized exclusions and padded allowlist entries
The regression case added with the fix drove the allowlist intersection in
both spellings but left the other two comparisons the change touches
uncovered: the exclusion membership test and the trimming applied to padded
names. Reverting either behaviour kept the suite green, so neither was
pinned. Greptile raised exactly this on the PR.
An exclusion is now exercised as " REA " against the selected server, and a
padded allowlist entry is exercised as the only policy, so a padded entry has
to permit the same server its exact spelling permits rather than fall to the
membership test.
Verified with `npm run test:focused -- tests/boundary/cli/clientProfiles.test.ts`:
14 passed with the change; reverting only ClientConfigurationDocument.ts to
main fails both cases, the arrangement case with `rea+REA: expected 1 to be +0`
and the new one on both the excluded and the padded-allowlist assertions.
* fix(process): report no remaining time after a provider deadline fires
Node timers can fire while performance.now() is still a fraction of a
millisecond short of the monotonic deadline. ProviderStartupDeadline then
reported interruption "timeout" and an aborted signal while remainingMs()
returned ceil(0.x) = 1, so polling loops guarded by remainingMs() > 0
(Hopper and Ghidra startup) could run past an elapsed deadline, and the
real-timer test "aborts a bounded wait when its real monotonic deadline
expires" failed intermittently in CI (expected 1 to be 0).
Once the deadline's own timer has classified a timeout, report zero
remaining time. A deterministic regression test fires the timer with the
monotonic clock 0.4 ms short of the deadline.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: 吴佳浩 <wujiahao@supremind.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Captured-bundle chunk edges and source-map original module edges only
accepted quoted string literals as import()/require() specifiers. Bundlers
such as rolldown emit plain constants as expression-free template literals
(import(`./chunk.js`)), so statically known dependencies were dropped while
completeness stayed complete. Read both sites through the shared exact-literal
reader (stringValue); template literals with substitutions stay unrecognized.
On OWASP Juice Shop v20.3.0 this recovers 8 lazy chunk edges (previously 0).
Fixes#2841Fixes#2599
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <17176810+morluto@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(javascript): retain bridge conditions on IPC joins
* fix(javascript): retain IPC condition ownership
* style(javascript): format bridge path regression
* fix(process): report no remaining time after a provider deadline fires
Node timers can fire while performance.now() is still a fraction of a
millisecond short of the monotonic deadline. ProviderStartupDeadline then
reported interruption "timeout" and an aborted signal while remainingMs()
returned ceil(0.x) = 1, so polling loops guarded by remainingMs() > 0
(Hopper and Ghidra startup) could run past an elapsed deadline, and the
real-timer test "aborts a bounded wait when its real monotonic deadline
expires" failed intermittently in CI (expected 1 to be 0).
Once the deadline's own timer has classified a timeout, report zero
remaining time. A deterministic regression test fires the timer with the
monotonic clock 0.4 ms short of the deadline.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(javascript): preserve Electron join condition coverage
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Codex <codex@users.noreply.github.com>
* fix(javascript): preserve Electron runtime entrypoint paths
* fix(javascript): resolve runtime entrypoints before modules
* fix(javascript): preserve Electron runtime path semantics
* fix(javascript): resolve Electron renderers exactly
* fix(javascript): retain runtime package aliases
* fix(javascript): retain missing renderer URL paths
* fix(javascript): resolve Electron URL and utility entries correctly
* fix(javascript): reject suffix-only worker URLs
* fix(javascript): respect Electron URL and utility path semantics
* fix(javascript): separate Node runtime entrypoints from source imports
* fix(comparison): recheck historical calls-only changes (#2223)
* fix(comparison): preserve incomplete function call evidence
Keep missing, unknown, partial and unresolved call relationships unknown in function comparisons and reconstruction claims. Aggregate IDA caller and callee coverage without overwriting uncertainty.
Validate with 144 focused tests, typecheck/lint, docs checks and test discovery. Real providers were not verified; the full headless IDA session suite requires unavailable Windows native controls.
* fix(comparison): recheck historical calls-only changes
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
(cherry picked from commit 688defca0f)
* fix(ci): isolate Linux Hopper process fixture inventory
(cherry picked from commit 8cac51e12184d56a9ba80a4427b3e6082ddac1ec)
* fix(android): make DEX descriptor pattern portable
(cherry picked from commit 1335bcddc569a1cfb34b6efedcb6f2e3aaaa3357)
* fix: repair the static-check failures on the default branch (#2895)
* fix: repair the static-check failures on the default branch
The default branch fails its own required `Static checks` job three ways, so
every pull request reports that job as failed regardless of its own change.
- `npm run format:check` reports `src/domain/functionComparison.test.ts` and
`src/ida/IdaOperationRunner.ts`.
- `npm run lint` reports a duplicate `import { join } from "node:path"` in
`tests/boundary/mcp/ghidraInventoryMcp.test.ts` (lines 4 and 17).
- `npm run typecheck` reports `format: "devin"` missing from the client format
union in `src/application/SupportedClients.ts`, and an unchecked index in
`tests/boundary/cli/clientProfileLifecycle.test.ts`, whose pair array is
inferred as `string[][]`.
Apply the pinned formatter, drop the duplicate import, list the `"devin"` format
that its own definition already uses, and mark the pair array `as const`.
Validation: `npm run check:fast` and `npm run format:check` pass.
* fix(setup): keep Devin on the JSON configuration dialect
---------
Co-authored-by: 吴佳浩 <wujiahao@supremind.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
(cherry picked from commit df3d96da80)
---------
Co-authored-by: morluto <morluto@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: Lucien <164857476+dakjdakd@users.noreply.github.com>
Co-authored-by: 吴佳浩Alben <35291372+XiaoBinGan@users.noreply.github.com>
Co-authored-by: 吴佳浩 <wujiahao@supremind.com>
* fix(artifacts): preserve Mach-O load-command string identity
* fix(process): report no remaining time after a provider deadline fires
Node timers can fire while performance.now() is still a fraction of a
millisecond short of the monotonic deadline. ProviderStartupDeadline then
reported interruption "timeout" and an aborted signal while remainingMs()
returned ceil(0.x) = 1, so polling loops guarded by remainingMs() > 0
(Hopper and Ghidra startup) could run past an elapsed deadline, and the
real-timer test "aborts a bounded wait when its real monotonic deadline
expires" failed intermittently in CI (expected 1 to be 0).
Once the deadline's own timer has classified a timeout, report zero
remaining time. A deterministic regression test fires the timer with the
monotonic clock 0.4 ms short of the deadline.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit dfdaa0ca74cc84a45dcb186422a0f51aa61bc427)
* fix(ci): isolate Linux Hopper process fixture inventory
* fix(comparison): recheck historical calls-only changes (#2223)
* fix(comparison): preserve incomplete function call evidence
Keep missing, unknown, partial and unresolved call relationships unknown in function comparisons and reconstruction claims. Aggregate IDA caller and callee coverage without overwriting uncertainty.
Validate with 144 focused tests, typecheck/lint, docs checks and test discovery. Real providers were not verified; the full headless IDA session suite requires unavailable Windows native controls.
* fix(comparison): recheck historical calls-only changes
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* style: format identity comparison and IDA runner
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Lucien <164857476+dakjdakd@users.noreply.github.com>
* fix(managed): pair members by declared identity before name-free remaps
The first matching round keyed methods on signature and normalized CIL,
and fields on signature, without the declared type or name. Identical
members of different types fell into one ambiguous group, so a file
compared with itself reported unknowns, and a lone match across types
was reported as an unchanged exact pair while the member's own
counterpart went unpaired.
Pair by declared type, name, and exact CIL/signature first, then by
declared type, name, and raw signature. The name-free CIL/signature,
field-signature, and structural rounds now run last, as remaps, and
skip a pair across declared identities while either member's own type
and name remain unpaired on the other side.
Nested types that share a bare name still share one declared identity,
so their identical members stay ambiguous in a self-comparison.
* fix(managed): keep members with kept identities out of remap rounds
The cross-identity check applied only to one-to-one groups. A member
whose own declared type and name were still unpaired on the other side
could fall into a larger remap group and be reported ambiguous, and a
blocked candidate could make an otherwise unique remap ambiguous.
A remap round now drops such members before grouping, along with
members whose declaring type is unresolved. Same-identity structural
pairs get their own round before the remaps. Each member's declared
identity is computed once.
Ambiguity limitations now name the evidence a group shares, which
stays true whether the round keyed on names or not.
* fix(managed): keep names in unresolved identities and hold ambiguous ones
A member of an unresolved declaring type had no declared identity, so
the declared-identity rounds keyed it by CIL or signature alone. Two
unresolved members with different names paired as an exact match, and
a file compared with itself reported unresolved members ambiguous.
The declared identity now always includes the name, with a null type
for an unresolved one. Remap rounds still leave such members out.
A remap round now also leaves out a member whose own identity sits in
an ambiguous group on the other side, not only one left one-sided.
The undecoded-counterpart check uses the same declared identity.
* fix(managed): keep remap groups with unresolved owners ambiguous
Remap rounds dropped members of an unresolved declaring type before
grouping. Such a member can be another member's own counterpart, so a
resolved member could remap to a copy in another type while its
unresolved counterpart was reported added.
Unresolved members now join remap grouping, and a group that contains
one stays ambiguous instead of pairing.
The undecoded-counterpart set reuses each member's computed identity.
* fix(managed): keep ownerless members out of declared rounds when shared
A member whose declaring type did not resolve has a declared identity
of a null type and its name. The declared rounds paired two such
same-named members even when the other side also held a resolved
member with the same CIL and signature, which could be the owner. The
old name-free first round reported that group as ambiguous.
An ownerless member now enters the declared rounds only when no other
member on either side shares its exact CIL/signature key, and never
the declared structural round. Otherwise it reaches the remap rounds,
which keep any group with an unresolved member ambiguous.
* fix(managed): keep competitors of an ownerless member out of the declared round
When an ownerless member shares its exact CIL/signature key with another
member on the same side, its owner may be that member. The resolved
members holding the key now stay out of the declared round too, so they
cannot pair and consume the candidate the ownerless member may own, and
the remap round keeps the whole group ambiguous as main does. A lone
ownerless member still pairs with its lone counterpart.
* test(managed): split the ownerless comparison tests under the function length limit
The ownerless describe block reached 181 lines, past oxlint's
max-lines-per-function limit of 150. Move the structural-competitor
case into its own block.
* fix(managed): contest signatures and remap only open candidates
* fix(managed): retain blocked raw-signature choices as unknown
* fix(managed): keep member names in signature contention
* fix(managed): scope residual signature contention by name once
ownerlessContested applies competitionIdentityKey itself, but
unresolvedSignatureAmbiguities passed it keys that were already
name-scoped, so the contested set never matched the groups and the
same-signature, different-body case reported additions and removals
instead of one unknown group. Pass the raw signature key.
* fix(managed): constrain matchObserved to declared members
matchObserved passes its items to matchByKeys, which needs Declared, so
its looser { declaring_type } bound failed typecheck (TS2344, TS2322).
* fix(managed): retain held ownership contention for residual matches
* fix(test): bind workspace fixture in Hopper launch cases
* fix(ci): prepare worker tests and preserve Hopper cleanup ordering
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* feat(ghidra): admit 32-bit big-endian PowerPC ELF
Recognize EM_PPC as a provider-neutral `powerpc` family that keeps its ELF
class, byte order, type and flags, and admit ELF32 big-endian ET_EXEC through
a bounded Ghidra lane. ELF64, little-endian, relocatable/shared objects and
EF_PPC_EMB are refused with a specific reason; Hopper and IDA do not gain
PowerPC support, and Windows Ghidra P0 is unchanged.
Ghidra's ELF loader selects PowerPC:BE:32:e500:default for a plain EM_PPC
header, which misreads 750-family code. The lane binds
PowerPC:BE:32:default/default in the profile (language_selection
"admitted-lane-v1"), imports with it and checks the reported language. A
language set through REA_GHIDRA_LANGUAGE_ID still replaces the lane default.
Add a source-owned fixture, fixed-encoding probe, GNU readelf cross-check and
the optional `verify:ghidra:ppc` real-provider lane, plus docs/ghidra-ppc.md.
Refs #1741
* style(ghidra): format shared PowerPC profile integration
* fix(ghidra): preserve caller-selected PowerPC languages
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Verify the native helper's sealed identifier and pinned Apple signing team in BridgeLauncher, then carry that provenance to HopperClient. Only a verified native helper's complete not-found exit confirms no document handoff, allowing runtime and target-lease cleanup and retry; wrappers retain unknown document state.
Share publisher identity with setup through the domain contract. Correct fixture teardown ordering and parameterized MCP coverage. Validation: 54 focused tests passed across boundary, CLI, MCP acceptance, diagnostics and installer coverage (one Linux-only case skipped); check:fast passed. Real codesign verification accepted the installed official Hopper helper; Spotlight failures and bridge recovery used process fixtures.
* fix(hopper): recover from a Spotlight not-found launch
Hopper's native launcher locates the application through a fixed Spotlight
query before any Apple event. With indexing disabled it exits 3 with a
not-found report, but REA treated the external document as unconfirmed,
retained the runtime and failed every later open_binary until the server
restarted. Recognize the complete report as a pre-handoff failure, confirm
no document exists, and tell the caller to enable Spotlight indexing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(hopper): require empty stdout for native launch signatures
* fix(hopper): retain ownership after unverified Spotlight reports
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* perf(ci): start source feedback early and diagnose stalled runs
* fix(ci): verify workflow identities before main comparison limits
* fix(ci): preserve diagnostics across passing checks and install failures
* fix(ci): preserve ambiguous workflow run ordering
* fix(ci): preserve failure-report candidate and request budgets
* fix(ci): retain candidate errors while checking later main evidence
* fix(ci): retain partial reports and require complete health scans
* test(ci): prune stale fixtures and duplicated assumptions
* fix(setup): configure a comment-only Hermes config.yaml
comments, or only a `---` marker, also parses to a null YAML document.
Setup still rejected it as malformed ("expected record, received null")
and reported Hermes as invalid.
Treat a null Hermes document as holding no settings, as #1983 does for
an empty file. The serializer already edits a comment-only document in
place. A bare `---` document holds a null scalar that takes no keys, so
replace it with a map and keep the comments written after the marker.
* fix(setup): keep rejecting an explicit null Hermes root
Only a document with no root, or a bare --- marker whose null scalar has
no source text, holds no settings. An explicit null or ~ is a value, so
setup and uninstall reject it as before, the same as any other non-map
root.
* test(setup): split the Hermes empty-configuration tests under the function length limit
The Hermes YAML registration block reached 159 lines, past oxlint's
max-lines-per-function limit of 150. Move the empty-document cases and
the multiple-document rejection into their own block.
* fix(setup): type the Hermes edit document as a plain Document
parseDocument with options returns a Document.Parsed, whose contents
accept only parsed nodes, so assigning the new map for a bare --- root
failed typecheck (TS2322). Annotate the edited document as Document, as
the code did before it moved into HermesConfigurationYaml.ts.
* fix(ci): isolate Linux Hopper process fixture inventory
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
boundedViewMessages returns uniqueSorted output, so the 100-message fixture
came back in lexicographic order ("message 59", "message 6", "message 60")
while the assertion compared against generation order. Pad the index to a
fixed width so both orders coincide and the test checks what it means to:
that no count-based cutoff applies below the byte budget.
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Document the pinned JADX release download and digest verification for npm users and repository checkouts. Keep caller-selected local paths; identify the fixture helper as checkout-only. Include shell and PowerShell examples, with the POSIX path set only after successful download and verification.
Validation: actual pinned JAR/APK fetch verification; real curl download and SHA-256 checks using sha256sum and the shasum fallback; controlled corrupt-file and failed-download cases; POSIX syntax; docs:check (5 conformance tests), typecheck/lint, formatting and diff checks. Provider execution and native macOS/Windows command runs were not repeated.
Co-authored-by: Seelly <93262000+Seelly@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
* docs(i18n): add Greek translations
Give Greek-speaking readers localized onboarding guidance and begin website coverage while keeping incomplete pages on the explicit English fallback.
- Add the reviewed Greek README and place Greek last in every language selector.
- Register Greek website localization and include 206 translated messages.
- Keep incomplete routes nonindexable and replace fixed translation counts with inventory-based coverage.
Refs #2984
* docs(website): improve Greek homepage wording
* fix(ci): restore inherited validation failures
* perf(ci): share trusted caches and fail fast on PR test failures
* fix(ci): bound static cache and hash vitest config for knip
- Bound the rolling .turbo/cache snapshot with 7d/500MB eviction so each
trusted main save cannot accumulate history toward the GitHub limit.
- Hash vitest.config.ts in knip inputs so removing the CiTestSequencer
import invalidates the cached passing check.
- Cover both in cache-policy and task-hash regressions.
push_adb_file claimed to be the only device-mutating ADB tool while
install, uninstall, start-app, start-activity and stop-app are also
device-mutating. Drop the exclusivity claim from the contract
description and the push limitations, keeping the device-mutation fact.
Fixes#2602.
Co-authored-by: me <me@medz.org>
* test(signals): verify native decoder discovery coverage
Compare catalog and help pages against independent native sigrok output,
including full coverage counts, empty matches and boundary offsets.
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
* style(test): format the JVM class fixture
* fix(comparison): recheck historical calls-only changes (#2223)
* fix(comparison): preserve incomplete function call evidence
Keep missing, unknown, partial and unresolved call relationships unknown in function comparisons and reconstruction claims. Aggregate IDA caller and callee coverage without overwriting uncertainty.
Validate with 144 focused tests, typecheck/lint, docs checks and test discovery. Real providers were not verified; the full headless IDA session suite requires unavailable Windows native controls.
* fix(comparison): recheck historical calls-only changes
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(ci): isolate Linux Hopper process fixture inventory
---------
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: Lucien <164857476+dakjdakd@users.noreply.github.com>
* fix(artifacts): describe ZIP non-entry coverage accurately
* fix(process): report no remaining time after a provider deadline fires
Node timers can fire while performance.now() is still a fraction of a
millisecond short of the monotonic deadline. ProviderStartupDeadline then
reported interruption "timeout" and an aborted signal while remainingMs()
returned ceil(0.x) = 1, so polling loops guarded by remainingMs() > 0
(Hopper and Ghidra startup) could run past an elapsed deadline, and the
real-timer test "aborts a bounded wait when its real monotonic deadline
expires" failed intermittently in CI (expected 1 to be 0).
Once the deadline's own timer has classified a timeout, report zero
remaining time. A deterministic regression test fires the timer with the
monotonic clock 0.4 ms short of the deadline.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* style: format comparison and IDA sources
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(cli): preserve import-map option descriptions
* fix(process): report no remaining time after a provider deadline fires
Node timers can fire while performance.now() is still a fraction of a
millisecond short of the monotonic deadline. ProviderStartupDeadline then
reported interruption "timeout" and an aborted signal while remainingMs()
returned ceil(0.x) = 1, so polling loops guarded by remainingMs() > 0
(Hopper and Ghidra startup) could run past an elapsed deadline, and the
real-timer test "aborts a bounded wait when its real monotonic deadline
expires" failed intermittently in CI (expected 1 to be 0).
Once the deadline's own timer has classified a timeout, report zero
remaining time. A deterministic regression test fires the timer with the
monotonic clock 0.4 ms short of the deadline.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(process): keep terminal deadline expiry sticky
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* docs: identify official REA installation sources
Assisted-by: OpenAI GPT-6 <noreply@openai.com>
* docs: show official sources before setup shortcut
Move the unchanged official-source notice before the first setup command in all 19 README editions.
Assisted-by: OpenAI GPT-6 <noreply@openai.com>
---------
Co-authored-by: Aurantium-47 <280805072+Aurantium-47@users.noreply.github.com>
* fix(wasm): preserve original glue source locations
* fix(wasm): retain BOM-prefixed hashbang glue references
* test(wasm): parse the artifact contract before reading glue evidence
* test(wasm): give glue fixtures and assertions a single owner
* fix(comparison): recheck historical calls-only changes (#2223)
* fix(comparison): preserve incomplete function call evidence
Keep missing, unknown, partial and unresolved call relationships unknown in function comparisons and reconstruction claims. Aggregate IDA caller and callee coverage without overwriting uncertainty.
Validate with 144 focused tests, typecheck/lint, docs checks and test discovery. Real providers were not verified; the full headless IDA session suite requires unavailable Windows native controls.
* fix(comparison): recheck historical calls-only changes
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
(cherry picked from commit 688defca0f)
* fix(ci): isolate Linux Hopper process fixture inventory
(cherry picked from commit 8cac51e12184d56a9ba80a4427b3e6082ddac1ec)
* fix(android): make DEX descriptor pattern portable
(cherry picked from commit 1335bcddc569a1cfb34b6efedcb6f2e3aaaa3357)
* fix: repair the static-check failures on the default branch (#2895)
* fix: repair the static-check failures on the default branch
The default branch fails its own required `Static checks` job three ways, so
every pull request reports that job as failed regardless of its own change.
- `npm run format:check` reports `src/domain/functionComparison.test.ts` and
`src/ida/IdaOperationRunner.ts`.
- `npm run lint` reports a duplicate `import { join } from "node:path"` in
`tests/boundary/mcp/ghidraInventoryMcp.test.ts` (lines 4 and 17).
- `npm run typecheck` reports `format: "devin"` missing from the client format
union in `src/application/SupportedClients.ts`, and an unchecked index in
`tests/boundary/cli/clientProfileLifecycle.test.ts`, whose pair array is
inferred as `string[][]`.
Apply the pinned formatter, drop the duplicate import, list the `"devin"` format
that its own definition already uses, and mark the pair array `as const`.
Validation: `npm run check:fast` and `npm run format:check` pass.
* fix(setup): keep Devin on the JSON configuration dialect
---------
Co-authored-by: 吴佳浩 <wujiahao@supremind.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
(cherry picked from commit df3d96da80)
---------
Co-authored-by: morluto <morluto@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: Lucien <164857476+dakjdakd@users.noreply.github.com>
Co-authored-by: 吴佳浩Alben <35291372+XiaoBinGan@users.noreply.github.com>
Co-authored-by: 吴佳浩 <wujiahao@supremind.com>
* fix(verify-package): read Hopper installer applicability from its owner
packageSetupSupport derived hopperSetupSupported from the general doctor
host check plus a non-Windows condition, so a Debian-family host REA
supports but the Hopper package never tested reported an automated
installer the packaged setup never offers. verifySetupApply then expected
needs_confirmation for the explicit --install-hopper request and failed
at verify-package-setup.mjs:403.
Applicability now comes from readLinuxDistribution() in the built
application, so the verifier follows the predicate the package ships, and
it stays unknown when no distribution can be read instead of collapsing
into a known-unsupported host. Generic agent configuration and idempotence
assertions stay independent of installer applicability, and the expected
status is still derived from the host rather than from the request being
asserted.
Verified on Kali 2026.3, a Debian-family host with an installer-untested
distribution and the simulated Hopper package environment: before the
change npm run verify:package fails at verify-package-setup.mjs:403 with
the packaged setup correctly returning needs_human; after it the same run
passes every stage. Both applicability branches are covered by
tests/evaluation/package/packageSetupSupport.test.ts, which runs in the
source lane without a build. No real Hopper installation was exercised,
so no engine compatibility claim follows.
Refs #2939
* fix(ci): preserve Hopper doctor readiness checks
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(analysis-view): bound limitations and unknowns per view
Views of a JavaScript application or binary layout copied every parent
limitation into both limitations and unknowns, so a parent with tens of
thousands of limitations produced views as large as the complete record
and exceeded the MCP response budget even for a one-item page. Each view
now carries at most 64 distinct messages per field plus one line naming
the omitted count and the parent Evidence that retains the full list.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(analysis-view): assert exact bounded message notices
Source inspection and targeted formatting only. Tests, builds, lint, typechecks and provider workloads were not run at the user's explicit request.
* fix(analysis-view): preserve coverage facts in bounded messages
Reserve JavaScript completeness and unknown-count facts before bulk limitations. Direct omission recovery to the source parent without assuming session retention. Preserve exact threshold assertions and add unrun exhausted-page/item and unretained-inline MCP regressions. Source inspection and targeted formatting only; tests/checks/builds/provider workloads/CI reruns skipped by explicit user request.
* fix(analysis-view): bound repeated message serialization bytes
* fix(analysis-view): let byte budget bound messages
* test(analysis-view): exercise byte-based message omissions
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* fix(application): accept alternative document selection in binary_overview (#2460)
Update binary_overview capability negotiation so that either current_document or list_documents satisfies the document identity requirement, matching resolveAnalysisDocument runtime semantics. When a provider declares neither document capability, resolveAnalysisDocument reports AnalysisCapabilityUnavailableError for current_document rather than leaking the internal list_documents fallback. Preserve list_segments as an independently mandatory requirement.
* fix(application): retain document fallback provider errors
Return the fallback producer error unchanged, retaining operation identity, retry advice, diagnostics and cleanup instead of fabricating missing capabilities.
Source inspection and git diff --check; tests/builds/typechecks/CI reruns skipped at explicit user request.
* test(application): retain document fallback error expectations
Composition and MCP controls now expect the original list_documents producer error and reason.
Source inspection and git diff --check only. Tests, builds and typechecks skipped by explicit user request. GitHub review writes paused globally.
* fix(ida): assert attached-mode routing failure
---------
Co-authored-by: yashas1624 <yashas1624@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
The semantic member-mutation collector revisits the same nodes while it closes
alias chains, and each visit re-derived results that cannot change within the
phase:
- resolveSemanticBindingState re-walked the scope chain for the same identifier
node. Definition collection completes before this phase, and addBinding is
reachable only from it, so a scope's bindings no longer change.
- directInitializerPosition, directMutationPosition and
referenceInitializerPosition are pure functions of a node and the analysis
parent map, but were recomputed for every mutation and initializer.
- markValue re-encoded the same capture tuple with JSON.stringify and built a
fresh long key for every Set/Map lookup, so each lookup re-hashed the whole
key. The capture key is now cached per binding, origin and owner, and the
lifetime key is derived from it.
On pdfjs-dist 5.3.93 wasm/openjpeg_nowasm_fallback.js (462316 bytes, sha256
53c228428a8861035a34738966af263fc9d4cee6df25a0ad942af947469a0bd9) the semantic
phase takes 177.8 s instead of 272.9 s in-process, with a byte-identical
semantic IR. With --max-heap-mb 3072 and the default heap the base revision
exhausts the worker; with the default 300 s per-source deadline raised, the
completed CLI run takes 240.6 s instead of 396.7 s and its output is identical
apart from volatile identifiers.
Refs #2498
Co-authored-by: gongfpp <22683417+gongfpp@users.noreply.github.com>
* fix(package-verifier): reject overview responses without a procedure count
The analyze and inspect checks compared procedure_count with `< 1`, and an absent
or null normalized_result makes that comparison false, so a binary_overview
response without its required result passed the platform smoke check. Require an
integer procedure count before the positive-count comparison.
* test(package-verifier): cover malformed overview receipts
---------
Co-authored-by: REA Maintenance <76467478+morluto@users.noreply.github.com>
When every registered deep provider rejected a target, select() returned
an intentionally unbound success for any concrete environment selector,
including one that names no registered provider. A misconfigured
REA_ANALYSIS_PROVIDER therefore opened the target unbound, and the later
deep-operation diagnostic blamed target support instead of the unknown
configuration. The same identifier requested explicitly already failed
with unknown_provider.
Take the unbound shortcut only for a registered environment preference.
An unknown configured identifier now reaches concrete selection and fails
with unknown_provider, matching the explicit request. Registered
preferences and auto still leave artifact-only targets unbound.
Fixes#2592
Cassie_Fox
吴佳浩Alben
hhang
morluto
Shivek
Yann Roberto
Shubham Ahuja
Avocado
xin zhuang
Jeremy Levartovsky
Ariel Fikru Avicenna
Bruno Costa Candia
morconsultoriadigital-afk
Ysopwnable
BeqaKs
wseelly
Pirro Zani
Quentin Limo
Nianze Wu
Akash-M-D
Aurantium
CindyGest
LouisChen
yashas s
gongfpp
Devulapalli Naga Sri Vaishnavi
CodeAlex
Aaditya