Move plist projection into the graph owner and identify generated scalar
wrappers by producer identity through every dictionary/UID consumer. Remove
the intermediate JSON clone that erased those facts. Literal wrapper-shaped
dictionaries remain dictionaries; scalar $top values fail as malformed.
Keep numeric projection after graph interpretation, preserve unknown counts,
and retain binary prototype-key partial facts with an explicit limitation.
Adapt @Jax367's relevant scalar proposal and public regression from PR #4317.
Add scalar-root, literal-root and binary prototype controls. Correct only
the captured Foundation data node expectation and its manifest digest;
original producer bytes and source are unchanged. Unrelated PR changes excluded.
Validation: 175 focused checks across 23 files pass, one skipped; build,
check:fast and docs:check pass. Public CLI and stdio MCP cover both encodings.
Baseline scalar controls fail; 24 tiny completed CLI cost comparisons retain
ordinary full replies and Evidence IDs byte for byte. No statistical speed
claim, heavy profiling, target execution or real deep-provider verification.
Fixes#4295Fixes#4785
Use the matched Contents/Library/LaunchAgents or LaunchDaemons suffix to
select a plist's domain. Remove the independent scan of every ancestor.
Keep regular-file, application-root, sidecar and suffix admission rules.
Adapt @bnkaiteam's proposal and ZIP boundary regression from PR #3504,
adding public CLI/stdio MCP controls for nested bundles, case variants,
misleading ancestors and excluded paths.
Validation: 53 focused checks, build, check:fast and docs:check pass.
Two new controls fail before the fix. 12 completed actual built CLI cost
controls preserve the ordinary reply and projection ID byte for byte.
No heavy profiling, target or real deep-provider execution; npm release
and whole-agent costs remain unverified.
Fixes#3353
Use the owning DEX/JVM and Ark/JavaScript component lists for runtime
summaries. Remove the independent filename checks and Harmony path copy.
Preserve native format checks and framework path conventions.
Adapt @eco-null's proposal and regressions from PR #3627. Cover all eight
supported source extensions and unrelated content, plus the Android
inverse mismatch discovered during consumer review: .dex/.class files
had no bytecode components but produced runtime hints.
Validation: 44 focused checks, build, check:fast and docs:check pass.
Six Harmony assertions and one Android built CLI assertion fail before
this fix. 36 actual built CLI cost controls preserve four ordinary
complete replies and projection IDs byte for byte. No heavy profiling,
target or real deep-provider execution; npm release unverified.
Fixes#3350Fixes#4782
Select file and slice identity once before Android, Apple and Harmony
classification. Keep directory/link facts in source inventory and Apple
anatomy rather than deriving code roles, runtime hints or bridges from them.
Preserve occurrence order, multiplicity and existing unavailable facts.
Adapt @flyl1u's proposal and public CLI/stdio MCP controls from PR #3830,
extending the controls and structural fix to Apple and slice/link semantics.
Validation: 80 focused checks, build, check:fast and docs:check pass.
Four package controls fail before the fix. 72 tiny completed source cost
tasks preserve all regular-file control replies and projection IDs byte
for byte. No heavy profiling, target or real deep-provider execution;
full agent/public-task costs and published npm package unverified.
Fixes#3349
Build separate directory and authenticated-file indexes before applying
anatomy name preferences. Share those facts across application roots,
nested bundles, layout, metadata, executable candidates and signing paths.
Remove mixed child-name interpretation and downstream membership repairs.
A leaf Contents/Versions resource no longer changes a shallow layout, and
leaf files/symlinks cannot manufacture application roots. Real explicit
and implied directories retain their existing semantics and spelling.
Adapt the directory-identity proposal and source regressions from PR #3505
by @bnkaiteam, extending it to metadata file selection and root admission.
Validation: 57 focused checks, build, check:fast and docs:check pass.
30 tiny completed source projection cost controls preserve ordinary deep
and versioned replies byte for byte. No real-provider/target execution,
heavy profiling or published npm verification.
Fixes#3359Fixes#4779
Preserve unread file and slice facts before Apple attribution. Report partial
component coverage while keeping manifest count completeness independent.
Directories and symlinks retain their existing representation.
Consolidate occurrence identity interpretation and component schemas across
Apple, Android and Harmony; remove superseded producers. Android/Harmony
completed source replies and projection IDs remain byte-identical in controls.
Apple's additive unavailable_components field changes its projection IDs.
Adapt the unavailable-component proposal and real ZIP regressions from
PR #4692 by @rodricksz4h5, adding directory, CLI, MCP and non-encrypted controls.
Validation: 77 focused checks; check:fast; docs:check (5 README checks).
18 tiny completed source cost controls; no real-provider/target execution
or heavy profiling. Published npm package and large-input costs unverified.
Fixes#4688
Adapt the correction and authored overlap control proposed by @berelels in #4156 to the bounded prepared-image owner. Retain each fixup's application facts, then read every memory window after all writes. Carry the producer's load segment into comparison and advance only the MZ evidence revision.
Validation: 129 focused parser/comparison/MCP checks, 38 shared MCP harness checks and 41 selected provider checks passed, with 15 checks overlapping those runs. Fourteen lifecycle checks also passed, plus static/build/docs checks. Two baseline controls failed before the correction. Twelve completed normal source comparisons retained byte-identical replies; no real Ghidra or maximal-relocation performance qualification is claimed.
Fixes#3981
FlutterBuildProvider.execute checks the caller signal after its awaited
snapshot-root removal. When the signal was aborted during a successful
first removal, that check replaced the completed AnalysisExecution with a
bare AnalysisCancelledError, so identify-flutter-build and
inspect-dart-aot reported a typed cancellation without
details.partial_observation and the shared CLI/MCP Evidence assembly
never saw the facts.
The post-cleanup check now keeps the caller's abort reason and retains a
completed result as the cancellation's partial observation. The Evidence
composition from executeProviderOperation moves into
createAnalysisExecutionEvidence beside the provider contract, so the
success path and the retained cancellation use one composer and carry
identical Evidence. Failed results keep their existing classification.
Fixes#4717
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Attached IDA binds the open database to the target only through the
recorded whole-input SHA-256, path and image base. Every slice of a
universal Mach-O shares that digest, so an attached database of another
slice passed the identity check and its results were attributed to the
host-selected architecture. Attached admission now reports the target
format as unsupported, with lipo -thin and headless recovery, before any
upstream contact. Headless mode keeps thinning the selected slice.
Fixes#3871
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(evidence): verify staged output identity after its final write
On macOS FAT32 and exFAT volumes an empty file reports a placeholder file
ID, and the first write assigns its cluster-derived ID. The staged JSON
writer sampled the descriptor identity right after exclusive creation, so
publication and cleanup both rejected REA's own staged file as replaced:
every export failed with cleanup_incomplete and retained a .rea-write-*
directory.
The staged output owner now observes the descriptor identity immediately
before closing it, after the last write, and uses that identity for the
publication and cleanup pathname checks. A failed observation leaves the
identity unknown, so cleanup still refuses pathname removal.
Fixes#4613
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(evidence): cover failed final staged identity reads
Reject the staged descriptor's final stat and verify that the writer
still closes the descriptor, refuses publication, and retains the
unverified staged content with incomplete cleanup.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The one-shot Flutter CLI commands passed their command signal to the
inspection but not to finishAnalysisCommand, whose post-close cancellation
guard only runs when a signal is supplied. When the provider's first
snapshot-root removal failed, close() retried it after the inspection had
returned; SIGINT during that retry set exit status 130 while the command
still printed ordinary successful Evidence.
Pass the command signal to the finisher so an abort observed after the
awaited cleanup retry yields typed cancellation that retains the completed
Evidence as partial_observation. A failed retry still reports
cleanup_incomplete ahead of the later signal.
Fixes#4640
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The MCP verify_managed_native_boundaries handler resolved native
observations through an inspect_macho/analyze_function allowlist before
calling the shared verification service, and reported any other authentic
operation as evidence_integrity_mismatch. The domain already owns which
native operations it can read symbols from and counts the rest as
unsupported, which is what the CLI returns. The MCP resolver now only
authenticates native observations and leaves operation support to the
shared workflow, so both adapters return the same partial verification.
Fixes#4639
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
collectResponseConsumers copied the linked request's endpoint onto a
response consumer whenever the receiver binding had exactly one
initializer (added in 0323ce0a2, #3822). A single initializer is a source
fact, not proof that the write reaches the read: an incoming parameter
that is overwritten only conditionally, or only after the read, still
lets the consumer read a caller-supplied response, yet it received the
request's exact endpoint.
The sole write now also has to cover the read, using the same rule the
call resolver applies to incoming values: either it is the binding's
stable unconditional declaration initializer, or the shared read-position
owner (semanticBindingAtRead) reports a definite preceding initializer.
Otherwise the endpoint stays null.
Fixes#4606
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
scanRecoveryFile raised a plain AnalysisOutputError when a selected file
exceeded the remaining operation-wide input allowance, and
snapshotRecoveryInput folded that, read failures and UTF-8 decode
failures into one invalid_value request issue. A 64 MiB + 1 input was
therefore reported as a retryable invalid argument whose only record of
the 67108864-byte budget was message text that also called it an
invalid analysis output, and malformed UTF-8 received the same reason.
The scanner now marks its byte-bound refusals with the bound it
enforced. snapshotRecoveryInput, which owns the meaning of that bound
as the remaining total input allowance, reports it as a non-retryable
file-size AnalysisResourceConstraintError with maximum_input_bytes and
remaining_input_bytes, and decodes UTF-8 separately as an invalid_format
issue. Other scanner consumers still receive an AnalysisOutputError, and
the limit stays enforced.
Fixes#4629
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
V8 Inspector capture counted an execution-context created or destroyed
event whose context ID was missing or not a safe integer as retained,
but recorded nothing about it. The capture counters claimed the event
was kept while the context list silently omitted it, and no unknown
identified the loss.
The producer that parses the event and owns the capture counters now
records an invalid-context count, mirroring the existing invalid-script
accounting. The projection reports a non-zero count as an explicit
unknown, so the retained count and the reported omission agree. No
context is invented from a malformed ID, and all-valid captures are
unchanged.
Fixes#4685
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(native): retain Apple dispatch metadata descriptor cleanup ownership
inspect_native_dispatch_metadata opened the target with raw
openRegularFile and closed it in a finally block, so a rejected close
replaced integrity, capacity or cancellation failures and left no owner
to retry. The read now uses the shared owned regular-file operation in a
client resource scope; the physical failure stays the cleanup primary,
client close retries the exact descriptor, and a confirmed release
clears the original error's cleanup uncertainty. Decoding runs after the
descriptor is released.
Fixes#4743
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(verify): give the Apple dispatch verifier an owned file scope
Pass an ArtifactResourceScope to inspectAppleDispatchMetadata and close
it after inspection so a failed descriptor close is retained and retried
rather than reaching an undefined scope.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Keep initialized image facts and full source identity separate at the snapshot owner. Stream overlay bytes into original and modified digests, retain format-bounded image bytes, and consume prepared relocation/mapping facts. Preserve completed comparisons, exact cancellation and cleanup retry through descriptor finalization. Keep explicit source refusals as typed primary failures.
Validation: 119 focused parser/comparison/MCP checks, 40 selected provider checks and 14 lifecycle checks passed; static checks, runtime build, discovery and docs checks passed. Twelve small completed source comparisons retained byte-identical full replies. Real Ghidra, large input and full public CLI/MCP performance remain unverified. The separate overlap timing issue #3981 remains open.
Fixes#4770
Reject contradictory headers, overlapping virtual sections, DOS header reuse and image intervals outside the 64-bit address domain before metadata interpretation. Preserve empty and adjacent ranges, source section order, and raw padding semantics. Advance the metadata profile revision and retain provider snapshot refusal facts.
Validation: 37 parser checks and 35 selected provider checks passed; check:fast and build:cached passed. Six tiny authored-fixture summary/detail tasks returned byte-identical 791-token replies. No real-engine or Windows loader qualification.
Fixes#4763
Consume the named inventory observation once and compare typed facts through one final cancellation check. Retain measured or completed load-image facts and exact reasons, preserve source partials and cleanup, and remove duplicate interpretation and format-specific abort guards.
Validation: 131 contract/projection checks, 40 selected provider checks, and 24 selected private-file/socket lifecycle checks pass; typecheck, lint and builds pass. Includes the original confirmed descriptor-close boundary and late result consumption. Tiny authored fixtures and simulated installation/engine facts; no real Ghidra, investigated target or heavy profiling.
Equivalent completed DOS/raw/NativeAOT tasks retain byte-identical full replies and one source read. First processes and two alternating warm repeats; no statistically established performance improvement.
Fixes#3977
Attach each caller’s measured observations before source failure and cleanup projection. Preserve those facts through admission, peer and draining cancellation, cached results, cleanup fences and later lease resolution.
Validation: selected NativeAOT provider controls and typecheck/lint passed in the load-image audit tree. Related consumer validation follows in the next commit.
Refs #3977, #4753
Replace the private promise race with a named snapshot owner using SharedStartupAttempt. Keep peers independent, stop and join final departure, preserve source failures and completed facts through draining waits, and fence replacement behind retained cleanup. Close joins derivation before releasing the underlying client.
Validation: 31 selected simulated-provider lifecycle and NativeAOT checks, 80 shared-contract/schema/projection checks, typecheck, lint and builds. Completed tiny summary/detail/close comparison preserves full replies and one snapshot read. No real Ghidra, investigated target or heavy profiler execution.
Refs #4753
Parse only the exact regular packed Python bridge member without extracting into the caller-owned private runtime root. Preserve required Ghidra admission, typed callers, native lease ownership and cleanup on success or failure.
Validation: 263 focused real tar/Python/ownership and CI selection checks, 42 official WABT packaged CLI/MCP cases, full check:ci/check:fast/actionlint, real-package alternating measurements and exact-head hosted CI passed. The required Windows lane exercises the retained native lease without skipping missing authority.
Fixes#4134.
Co-authored-by: abdooo2235 <abdalrhmantheiv15@gmail.com>
Use the shared artifact resource scope for snapshot reads. Preserve completed facts, exact cancellation reasons and descriptor cleanup leases through session and public error projection; keep parent cleanup pending until all retained resources release.
Validation: 22 selected snapshot lifecycle checks and 85 contract, projection and bounded-reader checks; typecheck, lint and runtime builds. Uses small authored Node/socket fixtures, not real Ghidra or investigated target execution.
Refs #4752, #3977
Validate the consumed optional-header extent before interpreting mapping fields. Require complete header-backed read intervals and prevent failed header reads from falling through to section mappings. Advance NativeAOT profile interpretation identity to v4.
Refs #4750
Carry caller and runtime deadline cancellation through helper cleanup, preserve command-owner reasons and physical failures, and retain collected scenario reports with typed terminal errors.
BREAKING CHANGE: failed or cancelled native UI scenarios now return typed errors with the collected report in partial_observation instead of successful reports containing failed or cancelled steps.
Refs #4227
Join admitted helper work before runtime removal, retain one preparation outcome and root, and bind originating and retry errors to exact runtime release leases with live resources.
Refs #4227
Stage only the current portable Turbo task archives, metadata and available manifests for same-run artifact consumers. Share task identity/hash validation with cross-run keys and fail publication on incomplete preparation, while retaining existing cache authority and consumer build/prepack verification.
Validation: 252 focused tests, full check:ci, docs:check, check:fast, actionlint, real Turbo output equivalence and invalidation, restored CLI/MCP scenarios and full Linux package verification passed. Fault injection detected historical-entry inclusion. Exact-head hosted CI and available review passed; measurement limits are recorded in the PR.
Refs #4744.
Own journal, capture, result and PID descriptors through confirmed release. Preserve parsed facts and exact cancellation reasons while retaining unresolved children and gating runtime removal on their completion.
Refs #4742
* fix(electron): retain fuse descriptor ownership through failed final close
Static Electron fuse inspection admitted a file descriptor and closed it
in a bare finally: a rejecting final close() replaced the in-flight read
error, was classified by errno (e.g. EACCES) as a read permission failure,
and silently dropped ownership of the still-live descriptor.
Route the read through withOwnedRegularFile so a failed close keeps its
exact owner, preserves the primary error in the outcome, and surfaces as
RegularFileCleanupFailure. The provider now maps that to a primary error
with its own classification plus a ProviderCleanupError reporting
incomplete cleanup, instead of an access-denied read failure.
Regression tests fault close() (EACCES) with and without a primary read
error (EIO): both fail on the pre-fix revision and pass with the fix.
Fixes#4658
* fix(electron): keep cleanup report and partial evidence through cancel and successful reads
Address automated review on #4668:
- Handle RegularFileCleanupFailure before the cancellation check in
failureFor: a cancelled operation whose final close rejects now reports a
typed cancellation composed with the incomplete-cleanup error, instead of a
bare AnalysisCancelledError that silently dropped the descriptor owner.
- Attach the completed fuse image as an explicitly partial Evidence
observation when only the final close fails, instead of discarding the
successful read. The abort-after-completion path keeps the same evidence on
the composed cancellation error.
- Add provider-level regression tests: combined read+close faults, abort
synchronized with the faulted close boundary, and partial-observation
retention. Abort-during-close and partial-evidence tests fail pre-fix.
29/29 electron tests green; npm run check:fast clean.
---------
Co-authored-by: nosmile99 <nosmile99@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
rea-agents 6.5.0 passed an absolute caller path to Windows native
snapshot admission verbatim, so a spelling with forward slashes or
repeated separators reached the native component walker, which rejects
empty components as "Ambiguous Windows path component is unsupported"
(Win32 50). Current main already resolves every Windows caller path with
win32.resolve (fc791bb60), which collapses those spellings.
Pin that boundary: absoluteCallerPath normalizes four separator spellings
of one absolute Windows path, and on Windows parseBinaryTarget returns the
same canonical sourcePath for each of them. The Windows test fails when
the 6.5.0 verbatim-absolute-path rule is restored.
Fixes#4707
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Reuse stable-file hashing and exact cleanup leases for target verification, join admitted call workflows before tracer close, and block replacement work while descriptor cleanup remains unresolved. Preserve physical failures, exact cancellation reasons and completed call facts; remove the duplicate reader and AbortError-name heuristic.
Validate real inert-file cleanup controls and controlled tracers through provider and MCP boundaries. No LLDB execution, real OS close-failure or Windows qualification.
Refs #4740
Reuse test-owned stdio MCP connections within two stateless source-analysis matrices while retaining every original CLI/MCP scenario, schema/Evidence checks and test-finished cleanup. Add cross-input identity controls.
Validation: 23 focused public-workflow tests and complete check:ci passed; stale-response fault detected. Exact-head hosted CI and available review passed. Pinned workload measurements and their limits are recorded in the PR.
Refs #4737.
Bind signature inspection to its named domain result and retain completed facts in typed cancellation. Check cancellation after descriptor and root finalization, preserve exact caller reasons, and keep physical failures primary. Use explicit cancellation checkpoints instead of signal-state guesses.
Validate root and descriptor cleanup, retries, late root handoff, opaque reasons, provider projections, and existing real codesign/CLI controls. No real OS close-failure or Windows qualification.
Refs #4735
Retain roots and descriptors through shared cleanup leases and bind composed errors to their actual pending owners. Successful retries shrink live attribution and reveal the original typed failure.
Validate mixed root and descriptor failures, partial/final release, existing target binding, and real codesign and CLI controls. No real OS close-failure or Windows qualification.
Refs #4734
Use shared owned file operations for plist admission and source probes. Join admitted work during close, retain exact descriptor leases through retries, and preserve physical failures, cancellation reasons and completed converter output.
Validate with controlled cleanup faults and native plist CLI workflows on inert data. Real OS close-failure incidence and Windows behavior remain unverified.
Refs #4723
Own decoded node columns, attributes and parent validity once for web, Electron and SVG interpretation. Reject dangling, self and cyclic parent facts; preserve source positions and independently unknown fields. Remove fallback conversions and duplicated string/attribute interpretation.
Validation: baseline public CLI controls failed for four malformed cases with valid control passing; final 103 focused checks, 5 final built CLI controls, 1 late-detach cancellation control (20 skipped), check:fast and docs:check passed. Fake CDP producers and generated local files; no real browser/Electron or Windows qualification.
Refs #4709
* fix(native): report unidentified text plists as unknown
Require explicit XML identification in file output instead of treating all
text as XML. Preserve decoded plist values and source-format provenance
when plutil falls back to XML conversion.
Add provider-boundary regressions and real macOS CLI controls for OpenStep,
data-bearing OpenStep, XML, UTF-16 XML and binary property lists.
Fixes#4294
* fix(native): keep XML plists without a declaration labeled xml
`file -b` names XML only when a document declaration is present, so the
narrowed `\bXML\b` check reported a declaration-less XML plist as unknown and
broke the real-plutil `decodes xml plists that plutil cannot express as JSON`
CLI test. When `file` reports only text, read the source head once and label it
XML only when its first non-whitespace character (after a UTF-8 BOM) is `<`;
OpenStep and other text stay unknown.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(native): distinguish OpenStep data from XML source markup
Recognize XML markup instead of treating every leading angle bracket as XML.
Keep data-only OpenStep sources unknown, honor explicit JSON identification,
and decode BOM-marked UTF-16 prefixes before detecting declarationless XML.
Add source-format regressions and macOS CLI cases for data-only sources.
---------
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Fix permissions count issue for `pkg=null` in ADB analysis and add test cases
* fix(adb): handle null package permission counts
Recognize pkg=null as unknown permission data and keep the provider projection aligned with the parser. Add regression coverage for the parser and provider.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Use one typed CDP resource-size owner in web and Electron capture. Supplied invalid sizes retain null plus completeness diagnostics; omitted sizes remain unknown without inventing malformed metadata. Preserve other resource facts and valid zero/positive values.
Validation: public CLI regression failed against baseline clamping; 36 focused controls passed after the fix, check:fast and docs:check passed. Fake CDP producer with real CLI/transport and tiny local files; no real Electron/browser or heavy profiling qualification.
Refs #4708
morluto