Commit Graph
206 Commits
Author SHA1 Message Date
morluto f8cec67e44 refactor(mcp): bind plain-value application tools under analysis admission (#11591)
* refactor(mcp): bind plain-value application tools under analysis admission

A value binding that declares admitted: true runs under analysis admission. evaluate_reconstruction_coverage and export_environment_bundle move to the shared binding table; registerApplicationTools and its folder are removed.

* refactor(mcp): bind the managed-code workflows as session derivations (#11593)

Each managed workflow's application owner returns its sources and residual unknown; the four MCP tools bind through the shared session-derivation kind, and registerManagedWorkflowTools is removed.
2026-11-16 07:55:43 +08:00
morluto 33557c4715 refactor(mcp): bind the application workflows through an application binding kind on the shared table (#11578)
Thirteen application tools (feature and semantic traces, version, endpoint,
source-to-bundle and export-shape comparisons, API contract, NTFS journal and
Playwright trace inspections, the obligation ledger and the three mobile graph
projections) each had a bespoke registration repeating admission, retention
checks, logging, source retention, unknown recording and view delivery.

The shared table gains an application binding kind: a workflow that returns
the completed Evidence, its sources and its residual unknown, and an optional
view selector. One handler runs it under analysis admission, refuses a view
or summary before running when the complete Evidence cannot be retained,
delivers a failure's partial Evidence, and stages the view against the
completed Evidence before recording (deliverApplicationExecution, moved from
registerApplicationTools/helpers.ts). The comparison unknowns and the ledger's
request resolution move to src/application/ApplicationToolExecutions.ts with
their exact question text, so recorded unknown identities do not change.
2026-11-16 07:28:20 +08:00
morluto 5f718ce30e refactor(kernel): give error-message, byte-hex and SHA-256 hex helpers one owner in a new src/kernel layer (#11542)
* refactor(kernel): give error-to-message text one owner in a new src/kernel layer

* fix(kernel): keep the regex worker's own error text inside its isolated source

* refactor(kernel): exclude copies that run in another realm from the error-message guard by name

The regex worker's source string and the page.evaluate fetch in PlaywrightScenarioRequest cannot import errorMessage. The guard now names them, so the counted baseline does not invite a later slice to rewrite them.

* refactor(kernel): migrate the long tail of inline error-message copies to errorMessage (#11552)

Every provider and adapter folder with six or fewer copies now calls errorMessage from src/kernel. Local helpers whose whole body was the copy (failureMessage, gdbErrorMessage, message, describe, reasonOf, recoveryFailureMessage and four local errorMessage definitions) are removed and their call sites call the owner.

Refs #11548

* refactor(kernel): give byte-to-hex formatting one owner in src/kernel (#11560)

byteToHex returns the two lowercase hexadecimal digits of one byte. All eight inline copies (native, contracts, domain firmware and desktop, pcap) call it, keeping their own 0x, % or \x notation, and a counted guard with an empty baseline keeps new copies out.

Refs #11548

* refactor(kernel): give one-shot SHA-256 hex digests one owner in src/kernel (#11562)

sha256Hex digests one in-memory value. A counted guard records the one-shot createHash("sha256").update(x).digest("hex") chains (115 in 101 files) and the first slice migrates domain, contracts and process (35 copies), removing local wrappers (sha256, sha256Text, hash) and the exported sha256Text. Incremental and streamed hashing is not counted.

Refs #11548

* refactor(kernel): migrate the application layer's inline error-message copies to errorMessage (#11563)

All 69 copies in src/application call errorMessage from src/kernel. Local helpers whose whole body was the copy (errorDetail, causeReason and five local errorMessage definitions) are removed and their call sites call the owner.

Refs #11548

* refactor(kernel): migrate the native provider's inline error-message copies to errorMessage (#11567)

All 49 copies in src/native call errorMessage from src/kernel. The local errorReason helpers in LldbCallTracer and LldbRetainedObservations, whose whole body was the copy, are removed and their call sites call the owner.

Refs #11548

* refactor(kernel): migrate the remaining inline error-message copies to errorMessage (#11569)
2026-11-16 07:05:30 +08:00
morluto 5d3e1dc9cd refactor(mcp): bind analysis settings and session-derived comparison tools through the shared binding table (#11511)
* refactor(mcp): bind analysis settings tools through the shared binding table and count bespoke registrations

* refactor(mcp): bind inspect_analysis_view through the shared binding table (#11514)

* refactor(javascript): bind JavaScript source analysis in composition (#11467)

* refactor(mcp): bind inspect_analysis_view through the shared binding table

* ci(guards): count typed tool registrations and name each registration owner with its reason

* refactor(mcp): bind session-derived comparison tools through a derived binding kind on the shared table (#11547)
2026-11-16 06:18:49 +08:00
morluto 960a387731 refactor(contracts): own CLI command names, catalog identity and schema presentation in contracts (#11425)
* refactor(contracts): move the CLI command name owner into src/contracts

* refactor(contracts): own catalog identity in contracts and server identity in process (#11435)

* refactor(contracts): move catalog identity into contracts and server identity into process

* refactor(contracts): own the advertised MCP schema presentation in contracts

* test(mcp): import the advertised presentation from contracts in toolSchemaValidity

* test(release): expect the catalog identity at its contracts path in the canary

* test(mcp): import the catalog identity from contracts in syntheticContractSurfaces

* refactor(contracts): import CLI command names from contracts in the capability conformance test
2026-11-16 05:31:37 +08:00
morluto 0a9e736aea refactor(mcp): bind web script export, guidance, Cutter discovery and session status through the shared binding table (#11400)
* refactor(mcp): bind web script export through the shared service binding table

* refactor(mcp): give SDK-admitted bindings their contract's parsed input type

* refactor(mcp): type SDK-admitted bindings by the contract's request frame

* refactor(mcp): bind guidance, Cutter discovery and session status as value-returning bindings (#11439)

The shared binding table gains a value binding kind for contracts that return a plain JSON value rather than Evidence. It logs and delivers against the contract's output schema, records no Evidence and runs outside analysis admission, so provider-free guidance stays outside admission even though the shared dependencies carry one.

search_guidance, read_guidance_section, list_cutter_sessions and binary_session move onto it; registerGuidanceTools.ts is deleted and registerSessionStatusTool.ts becomes the sessionStatusTool.ts reader its binding calls.
2026-11-16 04:50:58 +08:00
morluto cb49383269 refactor(browser): take the host in Chromium, Electron and capture adapters (#11130)
* refactor(browser): take the host in Chromium, Electron and capture adapters

chromiumSandboxEnabled and safeElectronEnvironment take a required platform. The scenario browser, native module resolver, Electron provider and capture cleanup, mitmproxy adapter and web script artifacts take an optional platform seam (default HOST_PLATFORM); Electron fuses read the host architecture from HOST_PLATFORM. Their tests inject the platform instead of redefining process.platform. src/browser no longer reads process.platform or process.arch.

* refactor(pcap): take the host platform in the packet tools, native bound and batch launch (#11141)

* refactor(pcap): take the host platform in the packet tools, native bound and batch launch

* test(pcap): cover the selected host in batch launch and capture-host support

* refactor(apktool): take the host platform in launcher selection, version probe and decode (#11164)

* refactor(apktool): take the host platform in launcher selection, version probe and decode

* test(apktool): cover the provider's selected host in launcher selection

* refactor(hopper): give the launcher the provider's host platform

* refactor(native): inject the host into the pwntools decoders and native capability defaults (#11173)

* refactor(native): inject the host into the pwntools decoders and native capability defaults

* fix(native): give composed pwntools services the capability host and drop empty test hooks

* refactor(adb): take the host platform in binary selection and pulled file names (#11176)

* refactor(adb): take the host platform in binary selection and pulled file names

* refactor(cli): pass the resolved host platform through rea mcp add (#11179)

* refactor(cli): pass the resolved host platform through rea mcp add

* refactor(composition): default composition entry points to HOST_PLATFORM and pass it to Hopper and IDA (#11183)

* refactor(kuna): require the host platform in admission and installation, and give it to the session client (#11186)

* refactor(kuna): require the host platform in admission and installation, and give it to the session client

* fix(kuna): pass the host platform from the JavaScript installation callers

verify-real-kuna, both native oracle helpers and the command-lifetime child script call inspectKunaInstallation from dist, where TypeScript cannot see the now-required platform. They passed undefined and got unsupported_host on supported hosts.

* refactor(server): give createServer a host option and pass it to session tools, sigrok and EVM (#11203)

* refactor(server): give createServer a host option and pass it to session tools, sigrok and EVM

createServer read process.platform and process.arch live for its capability services, and registerSessionTools read process.platform for process capture. A host option (default HOST_PLATFORM) now selects both, and composition passes the capability host to the sigrok provider and the EVM provider (execution gate, limiter probe and environment rules), as it does for pwntools.

hostRestrictedAvailabilityMcp selects its host through the option instead of redefining process.platform and process.arch, and moves to the MCP boundary lane since it no longer touches process-global state. evmCleanupRetry passes SIMULATED_LINUX_X64 to the provider instead of 52 property redefinitions; the fixture moves to tests/fixtures/simulatedHost.ts.

* refactor(desktop): require the host platform in build computer and secret folder paths (#11210)

* refactor(desktop): require the host platform in build computer and secret folder paths

buildComputerArchiveDirectory, defaultBuildComputerArchive, resolveBuildComputerEnvironment and buildSecretsDirectory defaulted to process.platform. Every caller but DesktopBuildComputers' secret folder already passed the platform; it now passes HOST_PLATFORM.platform, which it reads for the build environment.

* refactor(filesystem): default caller paths and ChangeTime settling to HOST_PLATFORM (#11212)

appendCallerPath, absoluteCallerPath and the ChangeTime settle seams defaulted to a live process.platform read. Their defaults now come from HOST_PLATFORM: the 59 callers that omit the platform join or settle paths they open on this host, and callers describing another host's layout already pass it.

The webScriptExport case that redefined process.platform to reach the Windows settle budget moves to its own file, which selects Windows by mocking HostPlatform.

* fix(server): report process capture availability for the server's host

sessionAvailabilityPolicy imported platform from node:process, so with a selected host binary_session reported capture_process_scenario from the running host while planning and capture used the selected one. It takes the platform from createServer and registerSessionTools. hostRestrictedAvailabilityMcp checks discovery against a dry-run plan on win32 and linux hosts.

* build(guards): count platform and arch imports from node:process and node:os as ambient host reads

sessionAvailabilityPolicy's named import of platform from node:process escaped the ambient-host-platform guard. The guards now also match named platform/arch imports from node:process or node:os and os.platform()/os.arch(). The one other site, DesktopApplicationCapture's os.arch(), records the capturing host and now reads HOST_PLATFORM.architecture.

* refactor(config): take the host platform in parseConfig and give debugger managers the capability host (#11237)

* refactor(config): take the host platform in parseConfig and give debugger managers the capability host

parseConfig read process.platform for the default Hopper launcher; it takes the platform (default HOST_PLATFORM). GdbSessionManager and RizinDebugSessionManager default their platform to HOST_PLATFORM, and composition passes the capability host that already decides their availability. GDB's /proc descriptor check reads HOST_PLATFORM explicitly: it inspects this host's /proc whatever host the manager models.

* refactor(firmware): give the firmware and Wakaru providers the capability host (#11239)

* refactor(firmware): give the firmware and Wakaru providers the capability host

resolveFirmwareCommand and resolveWakaruCommand read process.platform (and process.arch) for their host gate. They take the platform or host as a parameter; FirmwareProvider and WakaruProvider take a host (default HOST_PLATFORM) for the gate and their environment snapshots, and composition passes the capability host that already decides their availability. The JavaScript analysis source transfer defaults to HOST_PLATFORM.

artifactSelectedCommand passes SIMULATED_LINUX_X64 to both providers instead of redefining process.platform and process.arch, and the Wakaru limiter configuration cases now run on any POSIX host.

* refactor(captures): give tshark the capability host and default Cutter to HOST_PLATFORM (#11240)

* refactor(captures): give tshark the capability host and default Cutter to HOST_PLATFORM

TsharkCaptureProvider and CutterBridgeClient defaulted their platform to process.platform. Both default to HOST_PLATFORM now, and composition passes the capability host to the packet capture service, which already decides its availability.

* refactor(platform): read HOST_PLATFORM explicitly in host-only path checks (#11241)

* refactor(platform): read HOST_PLATFORM explicitly in host-only path checks

Electron's readlink target split, IDA's host path comparison, the inspector's reported runtime path, the server identity check and WABT's executable names all describe paths on the running host. They read HOST_PLATFORM instead of process.platform.

* test(native): select the native UI verifier's host through HostPlatform (#11242)

verify-native-ui.mjs reads its macOS requirement from HOST_PLATFORM, and its action test resolves the script's HostPlatform import to a simulated darwin host instead of redefining process.platform in the child preload. Simulating linux there fails the macOS requirement, so the redirect is effective.

* test(mcp): check that composition forwards the selected host to execution

hostRestrictedAvailabilityMcp ran execution only on darwin, where the runner's own host also refuses. The new case calls the firmware, Wakaru, tshark, sigrok and EVM tools under a darwin and a linux x64 host and requires different outcomes, so dropping a provider's host forwarding fails on any runner (checked by removing the firmware forwarding).

* fix(composition): give the pcap, Apktool and ADB providers the capability host

createServer's selected host decided pcap capture discovery, but the default pcap, Apktool and ADB factories built their providers with HOST_PLATFORM. The factories take the platform (default HOST_PLATFORM) and capabilityServices passes its context platform, so availability and execution follow one host.

* test: import moved resources owners in host-platform fixtures

* refactor(config): drop the unused host parameter from parseConfig
2026-11-16 04:27:07 +08:00
morluto 66ec6a7261 refactor(mcp): bind Android tools through the shared service binding table (#11336)
registerAndroidTools built six handlers by hand: two through the
service's execute and four through package, JNI and symbol inspection
workflows that take the retention dependencies. SERVICE_TOOL_BINDINGS
now has an androidAnalysis entry for all six, registerContractTools
registers them, and registerAndroidTools.ts is deleted.
2026-11-16 04:23:54 +08:00
morluto 4b014fa9ad refactor(mcp): bind web network capture and web runtime tools through the shared service binding table (#11385)
* refactor(mcp): bind web network capture through the shared service binding table

* refactor(mcp): bind web runtime tools through the shared service binding table (#11395)
2026-11-16 03:15:41 +08:00
morluto 7aea9843bc refactor(mcp): register service-backed tools through one contract binding loop (#11264)
* refactor(mcp): register service-backed tools through one contract binding loop

registerContractTools registers every bound contract from the registry through the shared Evidence or retained-summary handler its binding selects, and refuses a binding that names no registered contract. createServer declares the bindings by exact contract name; 41 per-tool register files are deleted.

* refactor(mcp): bind service-backed tools from one shared table and make each binding select one handler

serviceToolBindings owns the binding of every service-backed contract, keyed by its service; createServer and the MCP tests bind from it, so tests exercise the production handler kind, admission and dependencies. A binding can no longer carry both run and summary. The Go real-binary workflow follows the new registration owners.

* test(mcp): assert binding exclusivity with type-level checks that hold under every checker
2026-11-15 23:55:31 +08:00
morluto 5c53bcaed8 feat(mcp)!: retire the full schema profiles and declare schema facts as typed annotations (#10693)
* test(mcp): check advertised schemas against a portable dialect in every profile

* fix(mcp): re-emit advertised patterns in a portable regex grammar

* fix(mcp): present discover_tools input through the selected profile

The hot-cold catalog registered discover_tools directly with the SDK, so
the default compact input profile still advertised its canonical schema,
including page-size defaults clients apply on their own (#5322). Route
every listed input schema, transport discovery included, through one
server-owned presentation, and pass the selected presentation to the
deferred server, which serves recorded tool schemas but not discovery.

* fix(mcp): keep the Evidence union on described borrowed inputs

borrowEvidenceInput installed its advertised projection on one schema
instance. A call site's .describe() rebuilds the schema from its
definition, so match_native_functions advertised left and right as
unconstrained values in every input profile. Install the projection in
the borrowed parser's constructor so derived schemas keep it, for both
borrowed inputs and borrowed arrays.

* fix(mcp): validate advertised dialect by selected profile

* fix(mcp): preserve schema widening through references

* feat(mcp): project compact input schemas into the portable advertised dialect

One projection (advertisedDialectProjection.ts) owns the compact input
advertisement: one type per node, items on every array, string literals
as enums, oneOf widened to anyOf, allOf merged, tuples as element unions,
exclusive bounds made inclusive, and every keyword outside the dialect
dropped. Each rewrite is exact or widening; server validation is
unchanged. The root description and sentences for dropped cross-field
constraints move to the tool description, which Codex and Zed keep.

The compact baseline falls to nullable values only (29), which the
dialect cannot express without narrowing an input. The gateway shape
repairs (typeArrayItemUnions, hasGatewayIncompatibleShape), the second
dependentRequired-to-prose copy and the strict-gateway test denylist are
removed; the dialect checker gates the inline presentation instead.

* feat(mcp): omit output schemas from tools/list by default

No surveyed model API reads an output schema, MCP client validators
repeat REA's own result validation, and complete output schemas were
about 92% of the default tools/list. REA_MCP_OUTPUT_SCHEMA_PROFILE now
resolves to off through one owner (resolveMcpOutputSchemaProfile);
compact and full remain selectable. structuredContent is still
delivered, and the deferred server serves both off and full from the
recorded catalog with the matching recorded identity.

Tests and artifact verifiers that validate results against complete
advertised output contracts select full explicitly.

* docs: describe omitted output schemas as the default in README translations

* fix(mcp): select schemas in verification workflows

* fix(mcp): refuse zero-padded repeat bounds and record host-specific path patterns

* fix(mcp): keep projected schemas permissive

* fix(mcp): keep references valid when presentation widens their targets

* fix(mcp): keep a single union branch's closure scoped to its properties

* fix(mcp): select complete output schemas in the Electron verifier

* fix(mcp): select complete output schemas in standalone verifier probes

* fix(mcp): keep an allOf member's tuple tail with its own positions

* fix(mcp): relax references into negations and conditionals presentation may remove

* refactor(contracts): declare root discriminators as typed schema annotations

* test(mcp): select the borrowed Evidence contract from its family

* fix(mcp): keep astral literals and dynamic references widening in every mode

* fix(mcp): omit astral literals and relax untracked references when presentation widens

* fix(mcp): select complete output schemas in every verifier transport

* fix(mcp): read reference pointers by schema position, not segment spelling

* fix(mcp): protect references into collapsing boolean unions and keep referencing branches out of the collapse

* fix(mcp): keep reduced schemas open when their root admits names by pattern

* test(mcp): keep the reduced-schema pattern case outside the size-limited block

* fix(mcp): omit patterns whose classes name astral characters

* fix(mcp): decide union collapse and pointer protection from canonical branches

* docs(mcp): document non-u astral matching and pin BMP agreement across V8 modes

* fix(mcp): keep prototype-named properties as own properties in the dialect projection

* test(mcp): select full output schemas in packet and signal schema checks

* test(mcp): expect the advertised base64 schema to omit the unbounded pad-bit probe

* fix(mcp): protect references into unevaluated keywords lost evaluation removes

* fix(mcp): validate the doctor probe reply against the canonical output contract (#10655)

* fix(mcp): validate the doctor probe reply against the canonical output contract

* test(mcp): replay doctor probe validation under every output profile

* test(mcp): accept typed-branch nullable unions in the advertised dialect (#10649)

* test(mcp): select full output schemas in HTTP session workflows

* fix(mcp): emit stack-safe classes with non-ASCII members and protect draft-07 reference siblings

* fix(mcp): protect references into literal siblings of relaxed draft-07 references

* feat(mcp): omit output schemas from tools/list by default (#10646)

* feat(mcp): omit output schemas from tools/list by default

No surveyed model API reads an output schema, MCP client validators
repeat REA's own result validation, and complete output schemas were
about 92% of the default tools/list. REA_MCP_OUTPUT_SCHEMA_PROFILE now
resolves to off through one owner (resolveMcpOutputSchemaProfile);
compact and full remain selectable. structuredContent is still
delivered, and the deferred server serves both off and full from the
recorded catalog with the matching recorded identity.

Tests and artifact verifiers that validate results against complete
advertised output contracts select full explicitly.

* fix(mcp): select schemas in verification workflows

* fix(mcp): select complete output schemas in the Electron verifier

* fix(mcp): select complete output schemas in standalone verifier probes

* fix(mcp): select complete output schemas in every verifier transport

* test(mcp): select full output schemas in packet and signal schema checks

* fix(mcp): validate the doctor probe reply against the canonical output contract (#10655)

* fix(mcp): validate the doctor probe reply against the canonical output contract

* test(mcp): replay doctor probe validation under every output profile

* test(mcp): accept typed-branch nullable unions in the advertised dialect (#10649)

* test(mcp): select full output schemas in HTTP session workflows

* fix(mcp): keep the Evidence union on described borrowed inputs (#10597)

* fix(mcp): keep the Evidence union on described borrowed inputs

borrowEvidenceInput installed its advertised projection on one schema
instance. A call site's .describe() rebuilds the schema from its
definition, so match_native_functions advertised left and right as
unconstrained values in every input profile. Install the projection in
the borrowed parser's constructor so derived schemas keep it, for both
borrowed inputs and borrowed arrays.

* test(mcp): select the borrowed Evidence contract from its family

* docs(mcp): admit baseline growth only for restored contract declarations

* fix(mcp): state reduced multi-type properties as typed branches, not type lists

The reduced fallback derived a property's types and emitted them as a type
list, which Gemini, Moonshot and Ark reject. A property whose canonical
schema is anyOf string and number therefore traded a typed union for a
rejected construct when the inline projection fell back. Emit one type, or
anyOf typed branches, so [T, null] becomes the dialect's nullable union.

* feat(mcp)!: retire the full schema profiles and declare complete observations as a typed annotation (#10718)

* refactor(mcp): declare complete observations as a typed annotation applied only on the wire

* test(mcp): compare complete-observation declarations with their owning text

* refactor(schema): couple Zod checks with their JSON Schema constraint keywords (#10725)

* refactor(schema): couple Zod checks with their JSON Schema constraint keywords

* refactor(schema): declare pattern approximations and object output roots as typed annotations (#10754)

* refactor(schema): declare pattern approximations and object output roots as typed annotations

* fix(schema): refuse pattern declarations JSON Schema cannot state

* test(schema): select guided_workflows from its contract family

* feat(mcp)!: retire the full input and output schema profiles (#10841)

* feat(mcp)!: retire the full input and output schema profiles

The compact input presentation is now the only advertised input schema, and
output schemas are either omitted (`off`, the default) or bounded
(`compact`). Catalog identity, the build-recorded digests and the recorded
startup catalog collapse to those presentations: two digest profiles instead
of six, one recorded catalog instead of two, and no input dialect baseline.

Tests and verifiers that validated replies against listed full output
schemas now validate against the canonical output contract, which REA
already enforces server-side; tests that asserted canonical input
restrictions read the canonical input schema instead of a full listing.

BREAKING CHANGE: `REA_MCP_INPUT_SCHEMA_PROFILE=full` and
`REA_MCP_OUTPUT_SCHEMA_PROFILE=full` are refused at startup and by `rea setup`
with a configuration error naming the setting. Remove the setting from the
client registration (or rerun setup with `REA_MCP_INPUT_SCHEMA_PROFILE=compact`
/ `REA_MCP_OUTPUT_SCHEMA_PROFILE=compact`). `REA_MCP_INPUT_SCHEMA_PROFILE`
still accepts `compact`. Server-side validation is unchanged.

* test(mcp): validate replies against canonical output contracts at every listed-schema reader

tools/list omits output schemas by default, so fixtures, verifiers and the
Android analysis runners that read a listed outputSchema now bind the
canonical contract instead. connectLocalToolsMcp and advertisedMcpToolSchemas
hand the SDK a definition carrying the canonical output schema, so the SDK
client and caller-supplied validators check every successful reply.

* test(mcp): resolve Android runners' output contracts from the selected repository

The Android efficiency runners imported the canonical output-schema helper from the harness checkout while --repository selects the revision under test. Import it from the selected repository, as the runners already do for the SDK and Ajv.

* test(mcp): let Android runners compare revisions before the canonical output helper

A selected revision without scripts/lib/canonical-output-schema.mjs could not start the Android runners. Choose the output contracts in one harness helper: the selected revision's canonical contracts when it has them, otherwise its listed schemas under the full output profile it still supports.

* fix(scripts): keep caller launch settings when the real Hopper verifier builds its server environment

The verifier deleted REA_VERIFY_SERVER_COMMAND and REA_VERIFY_SERVER_ARGS_JSON from process.env itself before reading them, so a caller selecting an installed REA command silently ran the checkout's dist/main.js. Copy the environment, as main does, so only the child's environment changes.

* test(mcp): validate whole replies against canonical output contracts wherever a listed definition is held

tools/list omits output schemas by default and the full output profile is gone, so tests that handed the SDK a listed definition no longer checked complete replies. Attach the canonical output schema with withCanonicalOutputSchema at every held REA tool definition. Environment-served tools (not REA contracts), and the deferred-runtime and doctor tests that exercise per-profile listings, keep their listed definitions.

* test(mcp): record the advertised catalog summary without the retired full profiles

Retiring the full input and output profiles leaves only the compact input and compact output listings, so the summary drops the full-profile digests and every pattern only the full listings carried. Regenerated with npm run verify:advertised-schemas -- --update-summary.
2026-11-15 21:51:18 +08:00
morluto 5cfaac0834 refactor(process): require the platform for environment snapshots instead of reading the host (#11028) 2026-11-15 18:33:23 +08:00
morluto 4126698407 feat(mcp): project compact input schemas into the portable advertised dialect (#10632)
* test(mcp): check advertised schemas against a portable dialect in every profile

* fix(mcp): re-emit advertised patterns in a portable regex grammar

* fix(mcp): present discover_tools input through the selected profile

The hot-cold catalog registered discover_tools directly with the SDK, so
the default compact input profile still advertised its canonical schema,
including page-size defaults clients apply on their own (#5322). Route
every listed input schema, transport discovery included, through one
server-owned presentation, and pass the selected presentation to the
deferred server, which serves recorded tool schemas but not discovery.

* fix(mcp): keep the Evidence union on described borrowed inputs

borrowEvidenceInput installed its advertised projection on one schema
instance. A call site's .describe() rebuilds the schema from its
definition, so match_native_functions advertised left and right as
unconstrained values in every input profile. Install the projection in
the borrowed parser's constructor so derived schemas keep it, for both
borrowed inputs and borrowed arrays.

* fix(mcp): validate advertised dialect by selected profile

* fix(mcp): preserve schema widening through references

* feat(mcp): project compact input schemas into the portable advertised dialect

One projection (advertisedDialectProjection.ts) owns the compact input
advertisement: one type per node, items on every array, string literals
as enums, oneOf widened to anyOf, allOf merged, tuples as element unions,
exclusive bounds made inclusive, and every keyword outside the dialect
dropped. Each rewrite is exact or widening; server validation is
unchanged. The root description and sentences for dropped cross-field
constraints move to the tool description, which Codex and Zed keep.

The compact baseline falls to nullable values only (29), which the
dialect cannot express without narrowing an input. The gateway shape
repairs (typeArrayItemUnions, hasGatewayIncompatibleShape), the second
dependentRequired-to-prose copy and the strict-gateway test denylist are
removed; the dialect checker gates the inline presentation instead.

* fix(mcp): refuse zero-padded repeat bounds and record host-specific path patterns

* fix(mcp): keep projected schemas permissive

* fix(mcp): keep references valid when presentation widens their targets

* fix(mcp): keep a single union branch's closure scoped to its properties

* fix(mcp): keep an allOf member's tuple tail with its own positions

* fix(mcp): relax references into negations and conditionals presentation may remove

* test(mcp): select the borrowed Evidence contract from its family

* fix(mcp): keep astral literals and dynamic references widening in every mode

* fix(mcp): omit astral literals and relax untracked references when presentation widens

* fix(mcp): read reference pointers by schema position, not segment spelling

* fix(mcp): protect references into collapsing boolean unions and keep referencing branches out of the collapse

* fix(mcp): keep reduced schemas open when their root admits names by pattern

* test(mcp): keep the reduced-schema pattern case outside the size-limited block

* fix(mcp): omit patterns whose classes name astral characters

* fix(mcp): decide union collapse and pointer protection from canonical branches

* docs(mcp): document non-u astral matching and pin BMP agreement across V8 modes

* fix(mcp): keep prototype-named properties as own properties in the dialect projection

* test(mcp): expect the advertised base64 schema to omit the unbounded pad-bit probe

* fix(mcp): protect references into unevaluated keywords lost evaluation removes

* fix(mcp): emit stack-safe classes with non-ASCII members and protect draft-07 reference siblings

* fix(mcp): protect references into literal siblings of relaxed draft-07 references

* feat(mcp): omit output schemas from tools/list by default (#10646)

* feat(mcp): omit output schemas from tools/list by default

No surveyed model API reads an output schema, MCP client validators
repeat REA's own result validation, and complete output schemas were
about 92% of the default tools/list. REA_MCP_OUTPUT_SCHEMA_PROFILE now
resolves to off through one owner (resolveMcpOutputSchemaProfile);
compact and full remain selectable. structuredContent is still
delivered, and the deferred server serves both off and full from the
recorded catalog with the matching recorded identity.

Tests and artifact verifiers that validate results against complete
advertised output contracts select full explicitly.

* fix(mcp): select schemas in verification workflows

* fix(mcp): select complete output schemas in the Electron verifier

* fix(mcp): select complete output schemas in standalone verifier probes

* fix(mcp): select complete output schemas in every verifier transport

* test(mcp): select full output schemas in packet and signal schema checks

* fix(mcp): validate the doctor probe reply against the canonical output contract (#10655)

* fix(mcp): validate the doctor probe reply against the canonical output contract

* test(mcp): replay doctor probe validation under every output profile

* test(mcp): accept typed-branch nullable unions in the advertised dialect (#10649)

* test(mcp): select full output schemas in HTTP session workflows

* fix(mcp): keep the Evidence union on described borrowed inputs (#10597)

* fix(mcp): keep the Evidence union on described borrowed inputs

borrowEvidenceInput installed its advertised projection on one schema
instance. A call site's .describe() rebuilds the schema from its
definition, so match_native_functions advertised left and right as
unconstrained values in every input profile. Install the projection in
the borrowed parser's constructor so derived schemas keep it, for both
borrowed inputs and borrowed arrays.

* test(mcp): select the borrowed Evidence contract from its family

* docs(mcp): admit baseline growth only for restored contract declarations

* fix(mcp): state reduced multi-type properties as typed branches, not type lists

The reduced fallback derived a property's types and emitted them as a type
list, which Gemini, Moonshot and Ark reject. A property whose canonical
schema is anyOf string and number therefore traded a typed union for a
rejected construct when the inline projection fell back. Emit one type, or
anyOf typed branches, so [T, null] becomes the dialect's nullable union.
2026-11-15 18:26:13 +08:00
7aca4cef64 feat(audit): report what a shipped app exposes to static analysis (#8525)
* feat(audit): report what a shipped app exposes to static analysis

Adds `audit_release_exposure` and `rea audit-release-exposure` to report
source maps, Electron fuse implications, extractable payloads, debug
information, build metadata, and credential markers. Checks carry stable
IDs, observations, statuses, and fixed remediation, while incomplete and
unassessed checks remain visible without implying a security verdict.
Credential findings include only rule IDs and locations. The CLI
supports `--fail_on` severity thresholds and exits 3 when an exposed
check meets the threshold.

* fix(audit): keep incomplete coverage unknown across text, fuses, Go flags and cancellation

Probe files with unrecognized names as text, keep undecided fuse slices
unknown, forward opaque container coverage, read paths from Go build flags,
and compare inventoried digests before accepting owner facts.

Keep typed cleanup failures and attach the partial report when a later owner
is cancelled. Pin the verifier's @electron/fuses version and move the tests
to the named contract family.

* fix(audit): match package members by source spelling and close remaining coverage gaps

Reopen package members by their source spelling and keep caller path
traversal in directory reads. Parse quoted and attached Go build flags, bind a
native image fuse read to its own digest, and keep reader resources and the
audit's partial report when cleanup or cancellation combine failures.

* fix(release-exposure): bind package members by ordinal and follow named map references

Package members were matched to container entries by sourcePath, which is
not unique in a PyInstaller executable, through a queue that only moved when
a member was reopened. Bind by container-relative path and duplicate ordinal,
the numbering inventory uses, so read order cannot change which bytes a
member gets.

A sourceMappingURL naming a file without a map extension was ignored. Local
targets are now examined as maps whatever their name, and a target already
scanned as plain text leaves coverage unknown.

* fix(release-exposure): match data: source map scheme case-insensitively

URL schemes are case-insensitive, so an inline declaration such as
//# sourceMappingURL=DATA:application/json;base64,... was routed to the
local-reference path, which ignores scheme URLs. Its embedded sources were
silently dropped and source_map_sources reported not_observed.

* fix(release-exposure): normalize declared map targets and count unreached ones

A sourceMappingURL that decodes to a decomposed name (cafe%CC%81.json)
never matched the inventoried member, whose logical path is NFC, so the
map was scanned as ordinary text and its sourcesContent went uncounted.
Map-target lookups now use an NFC key on both the record and lookup
side; the declared URL is left as written.

A declared map target that is never reached is now counted as a skipped
map, so the source-map checks read as unknown coverage instead of
not_observed.

* fix(release-exposure): recover indexed leaves after any map format failure

An indexed map whose one leaf has malformed mappings (such as AA) threw a
format failure that only recovered leaves for unsupported sections, so a
valid sibling leaf with sourcesContent was dropped and
javascript.source_map_sources read unknown instead of exposed. Leaves are
now recovered for any format failure, with the envelope parse guarded so
invalid JSON still falls through to the unparsed count.

A nested package that inventory recognized but did not expand was probed
as a plain binary and left the source-map checks at not_observed. It now
marks coverage unknown and records a limitation.

Reader tests assert the javascript.source_map_sources status for the NFC
map-name cases as well.

* fix(release-exposure): count an unexpanded nested package as a readable member

An unencrypted nested package that inventory recognized but did not
expand is still a member its parent package already read. Count it as
readable so packaging.extractable_payload reports exposed, while its
own contents keep coverage unknown.

* fix(release-exposure): bound the indexed leaf fallback by decoder limits

The fallback walk after a map format failure expanded every section and
built a full ancestor path per leaf, so a map refused for nesting depth
could still allocate far past the decoder budgets. Walk only to the
64-level section depth and the shared record budget, keeping leaf maps
without materializing section paths.

* fix(release-exposure): collect UNC roots from Go flag paths

A flag such as -I\\buildhost\sdk\include matched none of the path roots
in GO_FLAG_PATH, so a network path that was the only one in complete
build info left go.build_paths at not_observed. Accept \\host\share
roots alongside the slash-rooted, dot-relative and drive-letter forms.

* fix(release-exposure): keep Electron undecided without positive proof

The desktop owner names Electron only from an ASAR layout, so an asar: false
build has no candidate. A complete scan with no candidate was reported as
not_electron, which outranks unresolved and safe fuse outcomes and hid them
as not_applicable.

Leave the family unknown unless the fuse holder rules Electron out: an app
bundle with no Electron Framework, or a directory with complete coverage
whose native images were all read and hold no sentinel. A directory with no
native image is not proof, so its fuse checks now report unknown.

* fix(release-exposure): decode wide text members before scanning

A member's kind bypassed the NUL guard, so a UTF-16 plist was decoded as
UTF-8, matched nothing and still counted as scanned. A directory holding
only that member reported text.credential_markers as not_observed.

Decode by byte order mark at one boundary instead. UTF-16 is scanned in
either byte order, and inline maps read the same text. UTF-32 and NULs no
mark explains count as skipped, whatever the member's kind. A binary
plist keeps its raw scan. Unnamed members with a mark now reach the
scanner, and the plist decoder shares the same mark detection.

* fix(release-exposure): follow file: URL source map declarations

The scheme guard in noteLocalMapReference returned on every URL scheme, so
a declaration such as file:///release/debug.json was dropped with the remote
ones. A member without a .map name was then scanned as ordinary text and
javascript.source_map_sources reported not_observed.

Convert file: URLs with fileURLToPath and record them in mapTargets through
the same step as relative declarations. A target that is never reached, or
that was already scanned as text, counts as skipped, and so does a malformed
file: URL. A file: URL names one absolute location, so a .map target outside
the inventory stays unassessed rather than being trusted by its name. Remote
schemes stay ignored.

* fix(release-exposure): match backslash-relative Go flag paths

GO_FLAG_PATH accepted dot-relative roots only with a forward slash, so
`-I..\sdk\include` and `-I.\sdk` produced no path and go.build_paths came
back not_observed. Accept either separator after the dots.

The CLI acceptance test now expects unknown for electron.run_as_node on a
directory with no native images, matching the service.

* fix(release-exposure): keep fuses unknown on incomplete inventory

A directory inventory that stops at an acquisition limit never reports
the images it did not reach, so the native image sample has no omission
and safe fuses on an included helper settled to not_observed even though
scan coverage was unknown.

Treat unknown coverage like a truncated sample until the Electron
executable is accounted for. A selected native file is that executable,
so it keeps its decided fuse result.

* fix(release-exposure): decode binary plists before scanning

A binary plist stores each string as ASCII or UTF-16BE. Scanning the raw
bytes as UTF-8 missed any string the encoder kept in UTF-16, such as one
with a non-ASCII character, and still counted the member as fully
scanned, so text.credential_markers read not_observed.

Decode through the plist owner and scan its strings, dictionary keys and
data. A plist that cannot be decoded keeps its raw bytes in the scan and
counts as skipped, so coverage stays unknown.

* fix(release-exposure): keep UTF-8 members that hold a NUL

Any NUL in a member's first 512 bytes made the decoder give up, so a script
with a literal NUL in a string never reached the credential or inline-map
checks. An exposed marker in it read as unknown, and --fail_on high exited 0.

Validity alone cannot tell wide text from UTF-8, since ASCII in UTF-16 is
valid UTF-8. Treat a header that is an eighth or more NUL bytes as wide text
and anything not UTF-8 as undecoded, and scan the rest.

The docs now separate BOM-marked UTF-16, which is scanned, from wide text
with no byte order mark, which is not.

* fix(audit): scan every string a binary plist stores

The plist scan read only the decoded value, which drops objects the root no
longer references and entries a later duplicate key replaced. Those bytes
still ship, but the audit marked the scan complete, so a credential stored
there read as not observed and --fail_on high exited 0.

Read strings, keys and data from the object table instead. A stored object
that cannot be read leaves the member unassessed, and a total byte budget
bounds table entries that overlap to repeat the same text.

* fix(audit): keep packaged native images from ruling out Electron

The applicability check only looked at the "native images" omission. An
Electron executable inside a compressed .NET or PyInstaller member is recorded
under "packaged native images", so a directory whose on-disk hosts had no fuse
sentinel read as not Electron and all four fuse checks as not applicable.

Treat that omission as an unread image too.

* fix(audit): keep readable plist text when one object is unreadable

Throwing on one unreadable stored object discarded every string already
decoded, so a UTF-16BE credential beside a truncated unused object was lost:
the raw bytes that replaced it cannot match NUL-separated text.

Return the decoded text with complete false instead, and keep the raw bytes
in the scan beside it so ASCII in the unread object is still seen. The member
stays unassessed. Text past the budget ends the same way.

* fix(audit): read escaped inline maps and extensionless SEA entries

A data URL may percent-escape its base64 payload, so a map such as one padded
with %3D did not decode and the audit counted no embedded source map. Decode
the escapes before reading the base64.

A directory walk classed a SEA entry by its code path, so a main script whose
code path names no script was never scanned as JavaScript and its inline map
went unseen. Class it by the role the SEA metadata gives it.

* fix(audit): check plist text against the budget before decoding it

decodeBinaryPlistText decoded each stored object before comparing its size
with the remaining budget, and it keeps going after an object is rejected so
later text is still read. A plist under the input limit could repeat a large
range through overlapping table entries and force gigabytes of decoding that
the audit then discarded.

Separate finding a stored payload from decoding it, so a range past the
remaining budget is rejected without allocating its text. Earlier text and
the incomplete result stay as they were.

* fix(audit): count a packaged PE with an unread header as a native image

A package member whose PE signature lies beyond the 4 KiB probe recorded only
a PDB failure. The image was neither read nor listed under packaged native
images, so a directory whose on-disk hosts had no fuse sentinel read as not
Electron and all four fuse checks as not applicable.

Count it as a packaged native image, the omission that keeps the question
open, and keep probing it for text as before.

* fix(audit): keep cleanup-incomplete errors from composed readers

A desktop, fuse, Go or Rust failure became a string on that check, so an
error with unfinished cleanup lost its resources and live bindings while the
audit reported success, and the MCP handler does not close the service after
each request.

Treat unfinished cleanup like cancellation: stop the remaining owners and
return the error with the partial audit. When the owner attached its own
partial, keep the error and replace the partial, so cleanup stays attributed.

* fix(audit): keep safe fuses unknown past a packaged native image

The fuse check counted sampled-out images as possibly unread but not native
images inside a package. A complete directory scan with a safe-fuse image on
disk, no ASAR hint, and an unopened packaged image reported the four Electron
checks as not_observed, although the Electron executable could be that image.

The fuse check, the Go and Rust checks, and the Electron rule-out each decided
this for themselves and had drifted apart. Move the decision to one domain
predicate, nativeImagesMayBeUnread, and use it in all three. A hinted
executable that was read still settles the check.

* fix(audit): settle fuses only on an identified host, and read Bun and HTML maps

A retained helper set electron_executable_included because the rank marks
every .exe or extensionless sibling of resources/app.asar. With more than four
siblings, or an inventory that stopped early, safe fuses on a helper reported
not_observed while the host went unread. Rank stays a sampling hint. The check
now settles only on an executable named as the host, or when inventory was
complete and every executable that could be the host was sampled.

A Bun sourcemap region is named .bunmap, so its role was lost and a complete
audit reported no source maps. Carry the region role from the reader's metadata
into the member view, in the expanded and unexpanded paths, and treat it as a
JavaScript map so an unparsable region stays unresolved, not a debug companion.

An HTML member was scanned as text but its inline scripts were never checked
for source maps. The HTML owner now returns each body, the reader examines it
like a script file, and bodies past the owner's limit count as skipped.

* fix(audit): charge map reads to the count before opening the member

An oversized or unreadable .map returned before readSourceMap counted it, so
the allowance of 32 never ran out. An unexpanded .NET or PyInstaller package
could then decompress every listed map member, about 1 MiB each, up to the
4,096 entry cap.

The count bounds reads, so examineMap now reserves the slot before it opens the
member. A map past the count is skipped unopened, and a linker map past it is
still settled from its bounded prefix. An inline data: map is already in memory
and keeps its charge at the point it is parsed.

* fix(audit): examine scripts HTML loads whatever they are called

isJavaScriptMember knows a script by its suffix or an assigned kind, and the
reader dropped the script references the HTML owner parses. An extensionless
main loaded by <script src="main"> was scanned as ordinary text, so a complete
inventory reported javascript.source_map_sources not_observed with shipped
sources in its inline map.

The reader now resolves each JavaScript script reference against the page and
its base href, and a local target without a JavaScript name is examined as a
script when reached. A target read before its HTML file, never reached, or not
placeable on disk counts as a skipped map, so the check stays unknown.

* fix(audit): charge sibling map reads to the map count

Every generic .map beside a selected file got a full 1 MiB read and a
parse, with no cumulative bound, so the 4,096-entry listing allowed about
4 GiB of reading for one small selection.

Move the slot rule into chargeMapRead, which examineMap and sibling
classification now share. Past the count a sibling reads only the header:
a linker map is still identified, and one that may be a source map stays
a skipped companion.

* fix(audit): report cleanup failures raised by member reads

A close failure during an embedded-package probe, member read or inventory
reaches the shared read filter as an ordinary reader failure. The filter kept
only cancellation and integrity, so the member became a skipped read, inspect
succeeded, and the retained descriptor was reported nowhere. MCP does not close
the service after each request, so no caller saw it.

Propagate any failure that carries cleanup, whatever its reason, so it reaches
releaseExposureFailure and the audit reports its retained resources.

* fix(audit): charge unsuccessful text reads to the byte allowance

An oversized, undecodable, or failed text read left textBytes unchanged, so
every member got the full per-member budget and the cumulative allowance did
not bound what a package decompressed. Count the bytes member reads produce
and charge them whatever the read yields. An oversized file on disk still
costs only its probe.

* fix(audit): bound the paths kept for HTML script targets

A base href is copied into every script target it resolves, so a page under
the text member allowance could retain hundreds of megabytes of paths. Charge
each new target's resolved path to a byte allowance and leave targets past it
unassessed.

* fix(audit): keep examining a page loaded as a script as HTML

A script element naming an .html member promoted that member to a
script, and the text pass then chose one reading. The page's own script
elements were never read, so an inline map in it was lost and a shipped
sourcesContent went unreported.

Run the script reading and the HTML reading independently, and list the
script target allowance among the cases that leave source map coverage
undecided.

* fix(audit): detect credential markers that end in a hyphen

The Google, GitLab, and Slack rules allow a hyphen in the token but ended
on a word boundary, so a token ending in a hyphen at the rule's minimum
length was missed and the audit reported the check as not observed.
Open-ended rules now take the whole token, and the fixed-length Google
rule checks the full token alphabet after its last character.

* fix(audit): confirm FAT identity for probed package members

The package probe treated a decoded FAT table as a universal Mach-O. A
JVM class past the 4 KiB probe shares the FAT magic, so its truncated
prefix decoded as a table, the member counted as an unread native image,
and Electron could not be ruled out of a complete scan.

Pass the member's declared size to the sniff and take identity from
classifyFatContainerEvidence. A member with no usable size, or with
slices past the probe, stays counted as possibly native.

* fix(audit): admit binary plists in package member probes

A package member's probe judged text by its own NUL test, which accepted
byte order marks but not the binary plist signature decodeMemberText also
reads. A PyInstaller executable shipping a binary Localizable.strings was
rejected without a scan or a skipped count, so a UTF-16BE private key
marker inside it could report not_observed.

Name the encodings that explain their own NULs in one predicate, and move
the plist fixture builders to module scope.

* fix(audit): locate Go findings only in images that retain build paths

The exposed decision required non-empty paths, but the locations took
every image whose build info read. An image with no paths was named as a
source of the leak beside the one that had them.

One predicate now decides both.

* fix(audit): keep completed findings when member cleanup fails

A descriptor close failure aborts the member scan, and the scan's state
was dropped with it, so the caller got a typed cleanup failure and no
report. Source maps already counted and credential hits already collected
were lost.

The reader now hands the service what it assembled, with coverage unknown
and a limitation saying so, alongside the failure. The service runs that
scan through its existing stopped-audit path: no later owner runs, and the
report goes on the error as its partial observation.

* fix(audit): keep coverage unknown for nested ASARs in package hosts

A member of a PyInstaller or .NET host reaches the binary probe as a
plain file. An ASAR among them has a NUL-bearing header, so the probe
dropped it with no traversal and no omission, and source map checks could
read not_observed with complete coverage.

Ask the ASAR owner's signature probe, and record a match as a nested
package that was not expanded, as inventory-classified ones already are.

* fix(audit): let native format facts outrank member names

examineMember tried the name-based text and map routes first, so a PE an
owner identified inside an ASAR, named payload.json, was read as text.
Its CodeView path and the packaged-native omission were both lost, and
native.pdb_path read not_applicable.

Route an owner-established native format to the native probe first, and
count an encrypted one as a packaged native image for the same reason.

* fix(audit): report relative paths given to compiler path flags

go.build_paths only kept rooted paths, so retained build settings like
CGO_CFLAGS=-Iprivate-sdk read not_observed even though the setting names
a directory.

Take the relative argument of a C compiler path flag, attached or in the
next field. Scope this to compiler settings, because Go's own tools use
the same letters for non-paths such as -B gobuildid and -T 0x1000.

* test(javascript): expect retained text on SVG inline scripts

The SVG test from main predates this branch's inlineScripts text field,
so each side passed alone and only their merge failed.

* fix(audit): keep an unresolved Electron host's fuses unknown

fuseCheck decided from whatever outcomes the images reported, so a host
with no fuse answer read not_observed as soon as any other image in the
directory reported safe fuses.

Pass the images named as the host into the report and leave a check
unknown when one of them has no answer for it. Other images stay free to
have no fuses, because helpers in a real Electron install have none.

* fix(audit): bound the paths kept for local source map targets

noteMapTarget kept every declared target resolved against its member's
directory, so a short URL under a long directory could retain far more
than the script's own size, with no cumulative limit.

Add the allowance noteScriptTarget already has. A target past it stays
unassessed and counts as a skipped map, and is never read.

* fix(audit): borrow the audit's scope admission for inventory

inspect holds resources.run for the whole audit, and inventoryArtifact
entered the same scope again. With failed cleanup pending from another
audit, the inner run waits for a drain that waits for the outer run.

Add inventoryArtifactInScope, which takes the caller's admission.

* fix(audit): run the exposure verifier's commands under supervision

The verifier launched npm and the CLI with raw execFile and then removed
its temporary tree, so a timed-out child could outlive the tree it was
using.

Use the shared verifier commands and close them before removing the tree.
Their exit error carries stdout on its snapshot, so the audit's exit-code
branch reads it from there.

* fix(audit): read relative -L paths and split flag arguments

The relative-path reader handled C compiler settings only, so a directory
kept through Go's own linker flag, such as -ldflags='-L private-sdk -s -w',
reported go.build_paths as not_observed.

A quoted flag and argument also reached it as one field. The separating
space stayed on the argument, which reported rooted paths twice and kept a
leading space on relative ones.

Give -ldflags its own pattern that matches only -L, so -B gobuildid and
-T 0x1000 stay ignored, and consume the separator in the flag patterns.

* fix(audit): keep embedded leaves beside schema-invalid map fields

The reader recovers independent indexed leaves only after a
SourceMapFormatFailure, but inspectSourceMap let the schema's ZodError
through. A sibling with mappings "!", a section with a bad offset or a
root with a bad field skipped recovery, so a valid leaf with
sourcesContent reported javascript.source_map_sources as unknown.

Normalize schema rejection in SourceMapFormat, which owns these parses,
so every structural failure reaches consumers as a format failure. The
message is unchanged, and the codec and web fetcher already map both
cases to the same result.

* fix(audit): keep packaged executables' own packages unknown

examinePackagedBinary counted a native image inside a package and moved
on, while only on-disk images reached the embedded-package probes. An
ASAR carrying a PyInstaller, Bun, Node SEA or .NET executable with a
source map therefore scanned complete and reported
javascript.source_map_sources as not_observed.

Run the same classifier on packaged images. The probes read carrier
offsets from a file, so the member's bytes stay in a temporary file for
the probe alone. A detected package keeps member coverage unknown, as a
nested ASAR does, and an image past one file's total byte allowance
stays unprobed and unknown. A plain native member leaves coverage intact.

* fix(audit): read an app's sampled images for fuses

A .app target sent only the bundle candidate to the fuse reader, which
resolves to the outer Electron Framework. A sidecar with RunAsNode enabled
was never read, so the check reported not_observed in the app view while
the directory view reported exposed.

Send the bundle candidate first and then the sampled native images. An
image whose digest matches the bundle's read is skipped, so the framework
is not read twice. A sidecar's safe fuses cannot settle the outer framework,
so a bundle read with no answer for a fuse keeps that check unknown.

* fix(audit): read a Node shebang as script identity

The path classifier gives an extensionless executable a resource kind,
so a script such as bin/runner with a `#!/usr/bin/env node` header was
scanned as text without its inline maps being examined or counted. A
complete inventory then reported no shipped sources for it.

The probe holds the first bytes a member reader sees, so establish the
identity there. The script keeps the usual outcomes: a map that is read
counts, and one past an allowance or undecodable stays unknown.

* fix(audit): keep linker strings out of Go build paths

Reading `-X 'main.usage=-L level'` as a library directory made
go.build_paths report `level`, so a plain string could fail an audit
under --fail_on low.

Read the optional package selector once, before the setting's first
flag, and keep the argument of -X attached to it, so a string it sets
is never scanned as another option.

* fix(audit): keep a failed probe directory removal owned

A packaged image's probe copies the member's full bytes into a temporary
directory. When removing it failed, the error became an ordinary
unreadable probe: the audit could still succeed, nothing could retry the
removal, and the client heard nothing about the leftover bytes.

Hold the directory in the reader's resource scope, so close retries it,
and report the unfinished cleanup without replacing a primary failure.

* fix(audit): keep a declared .map from reading as a linker map

A script's sourceMappingURL=assets.map was dropped because the name
already looked like a map. A malformed assets.map then settled as a
linker map, so both source-map checks read not_observed and
native.debug_companions reported an exposure the release does not ship.

Keep a generic .map name as a declared target and read it as a
JavaScript map, in either scan order, so it stays unresolved.

* test(audit): rethrow typed errors from release exposure checks

A failed Result was rewrapped in a plain Error, which dropped its type,
cause and cleanup details from the test output.

* fix(audit): match companions by member identity, not spelling

An ASAR member's location is relative, and a declared map target is
absolute, so a malformed assets.map read before z.js stayed a linker
companion. The examined-map lookup, which is normalized, then returned
early, leaving both source-map checks not_observed.

Index companions by memberLocationKey and keep each reported spelling
under it. Spellings stay distinct because symlinks can make them differ.

* fix(audit): retract only guessed linker maps on a declaration

A script declaring sourceMappingURL=symbols.pdb after the scan reached
symbols.pdb removed the shipped PDB from the companion list, because a
late declaration retracted any companion. native.debug_companions could
then read not_observed.

Keep name-identified companions in their own set, and retract only maps
that were read as non-JSON and guessed to be linker maps.

* fix(audit): record an unfollowed map link as skipped

The directory and ASAR readers do not follow links, so a symlinked
app.js.map dropped out before any map counter moved. Its target was
scanned as plain text, and both source-map checks read not_observed.

Count a link that names a map as skipped wherever entries reach the
audit, and mark it examined so a declaration does not count it twice.

* fix(audit): read relative paths in Go compiler and assembler flags

A binary built with -gcflags=-I private-sdk keeps that setting, but
relative parsing covered only C compiler settings and -ldflags, so the
directory dropped out of go.build_paths.

Give -gcflags and -asmflags their own path options, taken from what
go tool compile and go tool asm document, and strip the package selector
as -ldflags does. The assembler's -D defines a symbol, so it is not one.

* docs(audit): state the packaged image read allowance

Packaged ELF, Mach-O and PE images are read in full and copied to a
temporary file, sharing one max_file_bytes allowance, but the docs and
coverage.allowances still said only a PE gets more than a header read.

Name the allowance and the copies in docs/release-exposure.md,
RELEASE_EXPOSURE_ALLOWANCES and the max_file_bytes description.

* docs(audit): say how go.build_paths treats rooted -X values

A rooted -X value cannot be told from a file path, so the audit reports
it. State that, and why, so the rule is a contract rather than a side
effect of how flags are scanned.

* fix(audit): keep a declared sibling map out of the linker guess

When the selected file declared a sibling map, the sibling classifier
never consulted the declaration. A malformed assets.map then read as a
linker map and raised native.debug_companions.

Leave a declared sibling to the unreached-target accounting, as members
already are, so it stays an unresolved source map.

* fix(audit): keep undecodable Go and Rust metadata unknown

Unsupported meant two things: a container this reader does not handle,
and metadata the image holds in a form the reader cannot decode, such as
a compressed .dep-v0 section. Both read as not_applicable, so an image
that carries metadata reported that the check did not apply.

Mark the second case on the format failure and carry it on
AnalysisUnsupportedTargetError, so the audit records a failed read and
stays unknown. The tools' own errors and messages do not change.

* fix(audit): keep coverage unknown for a PE header past the probe

A packaged member whose PE signature lies beyond the probe prefix stayed
unclassified, so the embedded-package probe never ran and coverage stayed
complete. A map inside such an image then read as not_observed.

Record the member as an unexamined package where it is already marked as
a possible native image.

* fix(audit): read the scripts of a standalone SVG

An SVG can carry a script with an inline source map, but only .html and
.htm members reached the script owner, so such a file left every map
counter at zero and javascript.source_map_sources read not_observed.

Treat .svg as a script-bearing page. htmlArtifactReferences already
reads SVG script elements, so this adds no parsing.

* fix(audit): keep every possible host when no name picks one out

With no named Electron host, every sibling ranks alike and
electron_host_paths stayed empty, so one image's safe fuses settled the
checks even when another possible host reported none.

Keep all sampled candidates as hosts in that case, so a candidate
without fuses leaves the checks unknown.

* fix(audit): read a text-named native member by its bytes

A PE member called payload.json took the text route, so it reached no
native probe, and its CodeView path and packaged-native uncertainty were
lost.

Decide from the bytes the text read already loaded, so real text costs
no extra read and the map and text allowances stay as they were.

* fix(audit): let a map or script read identify a native member

A package lists names without formats, so a PE called payload.js.map
took the map route and never reached a native probe. Its CodeView path
was lost, and with nothing counted as failed, native.pdb_path could read
not_applicable. A PE at a declared script or map target was lost the same
way, because declaration overrode the route.

Decide from the bytes the route already read, in the map route as in the
text route, and keep the listed kind for a promoted script target. Real
text and maps cost no extra read, so those allowances are unchanged.

* fix(audit): name a symlinked sibling debug companion

Selecting an executable skipped every symlink beside it before the name
check, so an app.pdb link left no companion path and no skip count, and
native.debug_companions read not_observed. Selecting the directory
reports the same link.

Check the name first. The map branch already needs a regular file, so no
link is followed.

* fix(audit): keep map coverage unknown for a linked script

A link named app.js to an extensionless payload was skipped as a link,
and its target was read as plain text with no map directives parsed. An
inline map there left no parsed or skipped count, so
javascript.source_map_sources read not_observed. Only links with a map
name were counted.

Count a link under any name that can declare a map, the set an unread
member already uses. It stays one omission when a page or a directive
also names it. The target is still not followed.

* fix(audit): keep a declared JavaScript map beside a selected file unknown

A local sourceMappingURL with a JavaScript map name such as app.js.map
was dropped on the assumption that it would be examined as a member.
When the selection is a file, sibling .js.map files are never read, so
the declared map left javascript.source_map_sources not_observed while
the same target spelled as a file: URL left it unknown. Record every
local target so an unreached one counts as skipped.

* fix(audit): read paths a C compiler forwards to the linker

CGO flags can hand linker options through -Wl,a,b or -Xlinker a. The
compiler flag parser saw -Wl,-L,private-sdk as one field and missed the
relative library directory, so go.build_paths reported not_observed.

Split forwarded options out before reading paths and read them with the
linker's own path options (-L, -T, --library-path, --script). This also
stops -Xlinker -Bsymbolic from being read as the compiler's -B prefix.

* fix(audit): read forwarded linker options by their linker meaning

The forwarded linker flag pattern read -Ttext=0x400000 as a script path
named text=0x400000, and it ignored -I, so -Xlinker -Ilib/ld-custom.so
lost the interpreter path the compiler flag pattern found before the
split.

Exclude the -T section and segment address options (bss, data, text,
text-segment, rodata-segment, ldata-segment) and read -I and
--dynamic-linker as paths.

* fix(audit): split Go flags as Go does and read external linker flags

Go splits a flag value with quoted.Split: a field that starts with a
quote runs to the matching quote, and any other quote is a literal
character. The reader stripped every quote, so CGO_CFLAGS
-I"/home/O'Neil/include" reported /home/ONeil/include, while cc was
given the directory "/home/O'Neil/include", quotes included.

-ldflags kept -extldflags values with the Go linker flags, where only
-L counts, so -extldflags=-Lprivate-sdk reported no path. The values
are now read as the C compiler driver reads them, including -Wl, and
-Xlinker forwarding.

A package member named payload.json or payload.js.map that holds an
ASAR took the text or map route, which checked only native signatures,
so the nested archive was scanned as text and coverage read complete.
The route's own read now recognizes the ASAR signature as well, and the
nested package leaves coverage unknown.

* fix(audit): keep a named script beside a selected HTML file unknown

A script element whose target had a JavaScript name such as app.js was
dropped on the assumption that it would be examined as a member. When
the selection is an HTML file, sibling scripts are never read, so a
sibling holding an inline map left javascript.source_map_sources
not_observed while the same target named app left it unknown. Record
every local script target so an unreached one counts as skipped.

* fix(audit): read a retained -pgo setting as one profile path

`-pgo file` names a single profile, and under -trimpath Go keeps only
its file name, so the flag readers dropped values such as
`private-training.pgo` and split paths with spaces or commas. Report
the value whole, skipping `auto` and `off`.

* fix(audit): read -extldflags paths as the external linker splits them

With -ldflags set to -extldflags '-L "/srv/private sdk"', the outer
flag reader could not match the path behind the inner quote, and the
external reader kept only relative paths, so go.build_paths read
not_observed. Read each -extldflags value once, as go tool link splits
it with quoted.Split, and collect rooted and relative paths from those
same arguments. The rest of the setting is read as before.

* fix(audit): read the Go linker's other path options

go.build_paths read only -L as a Go linker path option, so a retained
-ldflags='-tmpdir=private-obj -s -w' reported no path and --fail_on low
missed the directory.

Read every option `go tool link -h` documents as taking a directory,
path or file: -I, -L, -benchmarkprofile, -capturehostobjs, -cpuprofile,
-fipso, -importcfg, -memprofile, -o, -r and -tmpdir. -extld, -extar and
-libgcc name a program or `none`, so only their rooted values count.

* fix(audit): keep a quoted Go argument whole

goFields already reads -L "/tmp/private - sdk" as two arguments, but
each field was then split again before every whitespace-led dash, so
go.build_paths reported /tmp/private instead of the retained directory.

Split again only a field that begins with a flag, after any package
selector, as a quoted "-L dir -s -w" does. Any other field is one
argument and is kept whole, in -ldflags, -extldflags and CGO flags.

* fix(audit): keep coverage unknown for an unread packaged image

When a listed package member is a native image but its first read
fails, examinePackagedBinary returned before probePackagedImage could
mark its possible embedded package unexamined. Coverage stayed complete
and javascript.source_map_sources could report not_observed for a
payload nobody read.

Mark member coverage unknown wherever a native member goes unread,
including an encrypted one, as an image the probe could not read
already does, and say so in the limitations.

* fix(audit): read rooted Go tool path arguments whole

`-L "/tmp/private,sdk"` and `-L=/tmp/a=b` name one directory, so a `,` or
`=` in a documented path flag's argument is literal. `-trimpath` keeps its
prefix=replacement structure.

* fix(audit): count an unread packaged member once

* fix(audit): keep trimpath rewrites and directory fields apart from selectors

* fix(audit): keep a .map unsettled once a declaration was discarded

* fix(audit): skip env option arguments when reading a Node shebang

* fix(audit): keep companion coverage unresolved for a generic .map link

* fix(audit): count a dropped linker-map guess as a skipped companion

* fix(audit): read env settings after -- in a Node shebang

* fix(audit): route a selected native image named .map by its format

A directly selected or listed native image has no package format, so
examineMember skipped its native-first branch and sent a file such as
release.map through map routing. Its bytes then became a linker-map
guess and marked native.debug_companions exposed. Native format now
outranks the name for every member, packaged or not.

* fix(audit): read forwarded -rpath and -rpath-link as linker paths

A cgo build with CGO_LDFLAGS=-Wl,-rpath,private-runtime kept the
relative directory, but the forwarded linker parser only knew -L, -T,
-I, --library-path, --dynamic-linker and --script, so go.build_paths
reported not_observed. Add -rpath and -rpath-link to the same parser,
in the attached, = and next-field forms. Loader-expanded values such as
$ORIGIN/../lib and @loader_path/Frameworks name no build-host directory
and stay unreported.

* fix(audit): read each entry of a forwarded linker search list

-rpath and -rpath-link take a colon-separated search list. The loader
token filter tested the whole list, so -Wl,-rpath,$ORIGIN:/opt/private-runtime
dropped /opt/private-runtime along with $ORIGIN, and the rooted-path pass
cannot see an entry that follows a colon. Split each search list into its
entries and drop only the loader-expanded ones.

* fix(audit): expand loader tokens only in a linker search list

The runtime loader expands $ORIGIN and @loader_path only in a -rpath or
-rpath-link search list. The token filter also ran on the argument of -T,
-L and -I, so -Wl,-T,$ORIGIN.ld dropped a literal linker script path and
a complete read could report go.build_paths as not observed. Search list
entries keep the filter; every other linker path argument is kept as is.

* fix(audit): count a .map link beside a selected file as a skipped companion

The sibling scan skipped any .map entry that was not a regular file, so a
symlink such as symbols.map -> payload beside a selected executable left
debug companion coverage at not observed. The reader never follows links,
so the linker map such a link may name is unread, not absent. Count a
generic .map sibling link as a skipped companion without following it,
as noteUnfollowedMapLink already does inside a selected directory.

* fix(audit): split a linker search list only for an ELF image

Only the ELF linker reads a -rpath or -rpath-link argument as a
colon-separated list. The Mach-O linker stores each -rpath argument as one
directory, so -Wl,-rpath,@loader_path/lib:extra is a single loader-relative
entry, yet it was split and extra was reported as a build path. Carry the
native image format into the Go flag reader and split search lists only for
ELF; other formats read the whole argument as one entry.

* fix(audit): expand only the image loader's runtime tokens

A runtime search path token is expanded only by the loader of the image's
format. The ELF loader expands $ORIGIN, $LIB and $PLATFORM; dyld expands
@executable_path, @loader_path and @rpath. Each is literal text to the
other loader, so -Wl,-rpath,$ORIGIN:/opt/private-runtime on a Mach-O image
names a literal directory, yet it was dropped and go.build_paths reported
not_observed. Pick the token set from the native image format; the PE
loader reads no search list, so it expands none.

* fix(audit): keep a declared .map link beside a selected script a source map

The sibling scan counted any .map symlink as a skipped debug companion
before checking whether the selected script declared it as its source
map. classifySiblingMap already excludes a declared target that is a
regular file, so a declared symbols.map link added a false unknown to
debug_companions. Apply the declared-target check before either file-type
branch; noteUnreachedTargets still counts the unread map as skipped.

* fix(audit): byte-check members whose kind is only an inventory path role

Inventory gives a packaged member a role from its name, such as resource
for payload.json or source-map for payload.map. The packaged-format guard
treated any kind other than "file" as owner-stated, so a nested ASAR an
ASAR lists under one of those names skipped signature recognition and
cleared coverage. Track whether the package owner stated the kind (Bun
sourcemap region, SEA main script) and only skip the byte check then.

* fix(audit): read Go's linker -r setting as a runtime search list

go tool link -r sets the ELF dynamic linker search path, dir1:dir2:...,
which the linker writes to RUNPATH. It was read as one whole path, so a
release built with -ldflags='-r=$ORIGIN/../lib' reported the
loader-relative entry as a build path and go.build_paths as exposed.
Read -r entry by entry with the same image-loader-aware split and token
filter a forwarded -rpath uses: ELF splits at ':' and drops $ORIGIN
entries; Mach-O keeps the argument as one directory.

* fix(audit): keep a selected package executable's native carrier format

A directly selected PyInstaller executable named release.map recorded the
package format as its member format, even though its ELF or PE carrier was
already recognized. examineMember then missed the native-first branch and
treated the executable as a linker map, reporting a debug companion.
Forward the native carrier format for the selected file, as listed members
already do.

* fix(audit): keep coverage unknown for an unread unclassified member

A packaged member with no format of its own, such as a PyInstaller
payload.bin, recorded only a skipped text member when its read failed
before the header. Member coverage then stayed complete, though the
member could be a native image or carry a nested package. Leave
coverage unknown and say which member could not be read, as an unread
native member already does.

* fix(audit): keep ASAR header admission distinct from max_file_bytes

A directory holding an ASAR whose header exceeds the header admission
was reported as a file-size refusal with max_file_bytes advice, which
cannot fix it. Report the header admission as its own bound instead.

* fix(audit): thread the shared ASAR header admission into release exposure

createReleaseExposureService built ReleaseExposureReader with no admission,
so its inventory and every reopened ASAR reader stayed on the 16 MiB
default and ignored REA_ASAR_MAX_HEADER_BYTES and asar.max_header_bytes.

Resolve the admission in the composition with runtimeAsarHeaderAdmission,
as the Electron and V8 observation providers do, and pass it to both
inventoryArtifactInScope calls, the reopened container readers, and the
nested ASAR signature probe. The refusal now names the setting that can
raise it instead of claiming there is no runtime override.

* fix(audit): keep a nested ASAR probe stopped at the admission undecided

The release exposure reader treated an ASAR signature probe that stopped
at the header admission as "not an ASAR". A packaged member such as
payload.json holding an ASAR whose JSON starts past asar.max_header_bytes
was then read as plain text, its inner maps went unread, and
javascript.source_map_sources could stay not_observed.

A probe that reaches the admission limit, or runs out of a prefix read
after its extents parse, now marks package coverage unknown and records
a limitation, while the member is still examined as text.

* fix(cli): parse audit-release-exposure --max-file-bytes as a decimal token

Main now requires every numeric CLI option to admit decimal tokens and
reject 0x10 and 1e3 (tests/boundary/cli/cliDecimalArguments.test.ts).
Wrap the release exposure maxFileBytes option in cliDecimalField like the
other maxFileBytes options, so the domain field keeps its bounds and default.

* docs(specs): add the release exposure audit specification

Add spec.md, plan.md and tasks.md under docs/specs/artifacts/release-exposure/
as docs/development/documentation.md requires for a new feature, and list
them in docs/specs/README.md. The existing guide at docs/release-exposure.md
keeps describing current behavior and its route is unchanged.

* fix(audit): bind unpacked ASAR member reads to their inventoried digests

A selected ASAR is captured and checked against its inventoried digest, which
binds packed members. An unpacked member is read from its .unpacked companion
on disk instead, and readContainerMember and readEntryBytes never compared
those bytes with the member digest inventory recorded, so a companion
replaced between inventory and the scan was accepted.

readContainerMember now passes the member's inventoried digest for unpacked
entries, and readStreamBounded hashes the full stream, draining past the read
allowance when needed, and fails with "ASAR member changed before
interpretation", which surfaces as an artifact_changed error like the
container check. A regression test replaces a companion after inventory. The
guide, spec, plan and tasks describe the binding.

* fix(audit): let stated JavaScript roles outrank a .map name

isDeclaredMap() routed any member whose name ended in .map to map
handling before the owner's stated role or a page's script declaration
was considered. A Node SEA main script at boot.map, which the SEA reader
marks as javascript, and a .map script an HTML file loads both became
linker-map guesses, so their inline maps were never read and
native.debug_companions was reported exposed.

An owner-stated source-map role or a declared map target still makes a
member a map. Otherwise a stated javascript kind or a script target now
wins, and only then does the .map suffix, or the inventory path role that
repeats it, decide. Regression tests cover both cases, and the guide
notes that a SEA main script is read as a script whatever it is called.

* fix(audit): keep coverage unknown when a packaged text member fails to read

A compressed member named as text, such as payload.json in an unexpanded
PyInstaller host, can fail inflation before any header is returned. The
.bin case already leaves member coverage unknown because an unread header
may hide a nested image or package, but examineTextMember() only counted
the member as skipped text, so otherwise complete coverage stayed
complete and package-dependent checks could settle.

An unread packaged member whose kind its package did not state now goes
through noteUnreadBinary(), which counts it as skipped text, marks
coverage unknown, and records the limitation. A regression test corrupts
the zlib stream of payload.json in a CArchive host and checks the intact
member keeps coverage complete. The guide describes the rule.

* fix(audit): keep the unpacked ASAR digest check inside the read allowance

ea87cf8 made readStreamBounded hash an unpacked ASAR companion to its end
so it could compare against the inventoried digest, which drained the
stream past the read allowance. A 1 GiB unpacked module was read in full
for a 4 KiB header probe and again for a package probe, and a companion
that grew after inventory had no byte limit on that path.

readStreamBounded now stops at the allowance as every other member read
does, and compares the digest only when the member ended within the
allowance. A companion past the allowance stays over_allowance without a
digest verification. A test counts the bytes each scan read pulls from an
8 MiB companion. The guide and spec state the bound.

* fix(audit): read a script target whose bytes are a source map as a map

1d3f89f let a stated javascript kind or an HTML script target outrank a
.map name, so an HTML `<script src="main.map">` that points at a valid
source map with sourcesContent was read only as a script. Its sources
were never counted and javascript.source_map_sources could miss an
exposed finding.

A script role says how a member is loaded, not what its bytes hold. A
.map member or loaded script whose text parses as a source map, a JSON
object with version and mappings, sources, or sections, is now read as
a source map. Anything else is still read as a script, so a SEA main
script named boot.map and a .map script carrying an inline map keep
reaching examineInlineMaps. A regression test covers the map case, and
the guide states the rule.

* fix(audit): retract a linker-map guess when a page loads the map as a script

noteScriptTarget() returned early for a target already read, counting its
source-map coverage as skipped but leaving any linker-map guess the first
read recorded. When a directory or ASAR listed a non-JSON app.map before
the index.html that loads it with <script src="app.map">, app.map stayed in
native.debug_companions and the result depended on enumeration order.
noteMapTarget() already retracted such a guess for a late map declaration.

A late script declaration now retracts the guess too, and still counts the
script as skipped because it was never examined as JavaScript. A test
covers the directory and ASAR order, and the guide states the outcome.

* fix(audit): respect script and native identity for sibling linker maps

classifySiblingMap() guessed a linker map for any sibling .map whose bytes
could not begin a JSON object. A sibling boot.map that a selected index.html
loads with <script src="boot.map">, or an ELF image named release.map beside
a selected executable, was therefore reported in native.debug_companions.

scanSiblings() now skips a sibling that is a script target as well as a map
target, leaving noteUnreachedTargets() to count it, and classifySiblingMap()
records no guess when sniffNativeFormat() recognizes the header. Tests cover
both cases and the guide states the outcome.

* fix(audit): read a script-loaded map from the text that recognized it

scriptHoldsMap() recognizes a .map or script target as a source map from
decodeMemberText(), which honors a UTF-16 byte order mark, but
readScriptAsMap() then passed the raw bytes to readSourceMap(), which reads
UTF-8. A BOM-marked UTF-16 main.map loaded by <script src="main.map"> failed
the second parse and its prefix became a linker-map guess.

readScriptAsMap() now hands the decoded text to readSourceMapText(), which
readSourceMap() also uses for byte input, so both stages read the same text.
A test covers a UTF-16 map with embedded sources.

* fix(release-exposure): honor byte order marks in binary plist data

A binary plist data object holding BOM-marked UTF-16 text was decoded as
UTF-8, so credential markers in it went unmatched while coverage still
read complete. Decode such payloads with the encoding their mark declares,
and leave coverage incomplete for UTF-32, which no reader scans.

* fix(audit): keep an unprobed package member past the text allowance unknown

Once the cumulative text allowance was spent, examineTextMember() counted a
packaged member named like text, such as payload.json in a .NET or
PyInstaller bundle, as skipped text without reading its header. Coverage
stayed complete even though the member could be a native image or a nested
ASAR, while the same bytes named payload.bin kept coverage unknown.

An unexpanded package member whose kind the owner did not state now leaves
coverage unknown with a limitation when the allowance skips it, the same
uncertainty an unread .bin member carries. No extra read is made, so the
allowance still bounds what is opened. A test covers an ASAR with an
embedded source map named payload.json after sixteen full text members.

* fix(release-exposure): require a read host for every Electron layout

A directory with several resources/app.asar layouts counted one named
host as fuse coverage for all of them. Track host coverage per layout so
safe fuses stay unknown when a whole layout's possible hosts were
dropped from the native image sample.

* fix(release-exposure): keep a .map script past the target allowance unassessed

When HTML script references spent the resolved-path allowance, a later
.map script target was dropped without its script role, so its JavaScript
bytes settled as a linker-map guess and debug_companions reported an
exposure. A refused script target now sets mapTargetsDiscarded, the same
flag a refused map target sets, so a later .map stays unassessed instead
of becoming companion evidence.

* fix(release-exposure): read a Node shebang as script in any text member

A selected or listed text file such as runner.txt that starts with a Node
shebang went straight to the text reader, which never gave it a JavaScript
role, so its inline map went unread and source_map_sources read
not_observed. The text reader now takes the role from the bytes it read,
so every route into it sees the shebang, and the probed-text path no longer
sets the role itself.

* fix(release-exposure): exclude Go search lists by position, not by value

compilerPaths dropped every literal path whose text equalled a forwarded
-rpath or -rpath-link list, so CGO_LDFLAGS=-Wl,-L,$ORIGIN -Wl,-rpath,$ORIGIN
lost the literal -L directory and go.build_paths read not_observed. The
literal flags are now read with each search list option and its argument
removed by position, and the search lists are still read entry by entry.

* fix(release-exposure): keep a probed Node launcher's role past the allowance

The probe recognized a Node shebang but handed the text reader the member
without its JavaScript role. Once earlier text spent the shared allowance,
the text reader returned before reading the header again, so an
extensionless launcher added no skipped map and both source-map checks read
not_observed instead of unknown. The probe now passes the role it found.

* fix(release-exposure): decode a map in the encoding its byte order mark names

The map reader, its prefix check and the sibling classifier decoded map
bytes as UTF-8, so a generic symbols.map holding BOM-marked UTF-16 source
map JSON failed to parse and its prefix read as a linker map: a spurious
debug companion and source_map_sources not_observed. Map bytes are now
decoded by their byte order mark at that boundary, and a UTF-32 map, which
this reader does not decode, stays unresolved.

* fix(release-exposure): keep a lone dash inside a quoted Go path argument

splitGoField split a quoted field before every whitespace-led dash, so the
retained -ldflags '-s "-L=/tmp/private - sdk" -w' read the dash inside the
directory as a new flag and goPaths reported /tmp/private. A quoted field
now splits only before a dash that begins a flag, so a dash followed by a
space stays part of the path.

* fix(release-exposure): choose possible Electron hosts within each layout

The named-or-possible host choice ran across every kept image, so a named
a/a.exe removed the unnamed b/launcher.exe of a second ASAR layout from
electron_host_paths, and its unresolved fuses no longer held
electron.run_as_node at unknown. Each layout now chooses its own named or
possible hosts, and the sets are combined.

* fix(release-exposure): read a relative compiler --sysroot as a build path

The compiler path flag list had no --sysroot, and the rooted-path fallback
cannot see a relative operand, so CGO_LDFLAGS=--sysroot=private-sdk left
go.build_paths at not_observed. --sysroot is now a compiler path flag,
attached with = or given as the next field.

* fix(release-exposure): recognize Windows Electron layouts by the owner's path

The desktop owner gives each file a native local_path, such as
C:\ship\App\resources\app.asar on Windows, and desktopPathConventions only
splits on /, so every ASAR basis was skipped: host ranking and
electron_host_paths stayed empty and a named host with no sentinel let a
safe helper read electron.run_as_node as not_observed. Each basis now
carries whether the owner's slash-separated path is an ASAR layout, and the
layout folders read a Windows absolute path with Windows separators on any
host.

* fix(release-exposure): read a Bun module by its JavaScript loader

An extensionless Bun contents member such as $bunfs/root/release carries
loader_name js, but ownerStatedKind kept only the sourcemap region, so the
member was never read as a script and its inline map went unseen. Give a
Bun contents region the javascript role for the js, jsx, ts and tsx
loaders, as SEA main scripts get it.

* fix(release-exposure): read a Node shebang .map as a script

An undeclared `.map` was routed as a source or linker map before its
first bytes were checked, so a Node script named `runner.map` recorded a
linker-map companion and never had its inline map read. When the map
read of an undeclared `.map` starts with a Node shebang, its map slot is
returned and the text reader examines it as JavaScript, so its inline
map is parsed and no companion is reported.

* Detect full-length GitHub refresh tokens

Narrow github.token to gh[opus]_ with 36 characters and give ghr_ its
own 76-character github.refresh_token rule, with a test.

* fix(release-exposure): bind a negative package probe to the inventoried image

An on-disk native image whose embedded-package probe finds nothing clears member coverage, so it now hashes the probed descriptor and compares it with the inventoried digest before accepting absent. A replacement after inventory fails as a changed member instead of reporting complete coverage.

* fix(release-exposure): let an app bundle read settle only its own layout

An app target, an app-bundle fuse read or a sampled selected executable used to mark the Electron executable as included for every ASAR layout, so safe outer fuses reported not_observed even when a nested Electron app's host was never read. Those reads now settle only their own layout; every other discovered layout still needs its own host read, so an omitted nested host keeps fuse checks unknown.

* fix(release-exposure): honor a Node shebang before guessing a sibling linker map

A sibling .map beside the selected file that starts with a Node shebang is a script, as it already is for a member .map, so classifySiblingMap no longer records it as a linker-map debug companion.

* fix(release-exposure): name a bundle's own layout by the owner's path

The settled layout set joined the caller's bundle spelling with Contents, so a selected /tmp/App.app missed the owner's resolved /private/tmp/App.app/Contents layout and safe fuse checks fell to unknown. Each archive basis now carries the owner's relative layout folder, and the bundle's own layout is the one whose relative folder is Contents, looked up by the owner's resolved folder.

* fix(release-exposure): read file-valued linker options as build paths

LINKER_PATH_FLAG listed --script but not the other linker options that read a file, so -extldflags=-Wl,--version-script=private-versions.map left go.build_paths not_observed. --version-script, --dynamic-list and --retain-symbols-file now read their operand as a path, attached or in the next field.

* fix(release-exposure): count a macOS bundle's framework and MacOS hosts in its layout

A .app layout resolves to its Contents folder, but its hosts are Contents/MacOS/* and the Electron framework binary under Contents/Frameworks, so per-layout host ranking and fuse coverage dropped them and one bundle's safe fuses could settle another bundle. Those images now belong to the bundle's layout, with the framework binary ranked as its host.

* fix(release-exposure): keep a bundle's launcher out of its fuse hosts

A macOS bundle's Electron Framework binary carries its fuses. When it
is among a layout's named hosts, it is now the layout's only fuse host;
the MacOS launcher beside it is still sampled but no longer leaves safe
fuse checks undecided when it has no sentinel.

* fix(release-exposure): keep an undecided PE package member out of linker maps

An unexpanded package member named .map whose PE signature lies past the
retained prefix fell through to noteUnreadMap, which read its MZ prefix as
a linker map and reported native.debug_companions exposed with complete
coverage. The map route now records the same undecided-PE uncertainty the
binary route does, so the member stays unexamined and coverage unknown.

* fix(release-exposure): scan the text of an undecided PE map member

A packaged .map member whose MZ header is undecided returned before its
fully read bytes reached the credential scanner, so a PEM key in it read
as unknown instead of exposed. Scan bytes read in full before returning,
while still keeping PE coverage undecided.

* fix(release-exposure): keep an oversized text member's undecided PE header

A packaged text member past the per-member allowance keeps only its prefix,
so an MZ header whose PE signature lies further in returned null and the
text route dropped the native and package uncertainty. Note the undecided
header as the map and binary routes do.

* fix(release-exposure): join Electron layouts by caller path and keep partial discovery unknown

The desktop owner resolves its root, so a target selected through a
symlinked parent got canonical local_paths that never matched the
audit's caller-spelled images, and the named host dropped out of
electron_host_paths. Rebase owner paths under the canonical root onto
the selected path.

When discovery stopped after it found one Electron layout, the basis
list was treated as exhaustive. Report the layouts as incomplete so safe
fuses on the listed host stay unknown.

* fix(release-exposure): keep Electron hosts under a trailing-slash selection

Join the rebased archive path with appendCallerPath so a selected
`/work/app/` yields `/work/app/resources/app.asar`, whose layout folder
matches the inventoried executables' parent.

* fix(release-exposure): keep maps unassessed for unread text past the allowance

A text-named member such as `runner.txt` can start with a Node shebang.
Once the text allowance is spent its header is not read either, so count
every unread member that could hold a script as a skipped map instead of
trusting the filename-derived role.

* fix(release-exposure): keep an unread packaged map member undecided

When a packaged map member yields no header (a failed read or no map
slot), its name is no evidence of its format. Route it through
noteUnreadBinary, as the text route does, so a member that may be a
native image or package keeps coverage unknown.

* fix(release-exposure): keep an undecided sibling PE header out of linker-map guesses

A sibling .map past the map allowance keeps only its probe, so an MZ
header whose PE signature lies further in is undecided. Share the
undecided-PE test with packaged members and count such a sibling as a
skipped companion instead of guessing a linker map.

* fix(release-exposure): keep possible fuse hosts when no ASAR layout names one

Without an ASAR basis no image is ranked, so no image counted as a fuse
host and one image's safe fuses could settle a sentinel-less sibling.
While the desktop family is undecided and no layout names a host, treat
every sampled image as a possible host.

* fix(release-exposure): keep images outside a found ASAR layout as possible fuse hosts

An asar: false build beside a discovered ASAR layout gets no basis, so
its executables were never fuse hosts and the layout's safe fuses could
settle a sentinel-less image in the other build. When no name picks out
a host outside the found layouts, treat every sampled image there as a
possible host, as the no-layout case already does while the family is
undecided.

* fix(release-exposure): keep a dropped host outside a found ASAR layout unread

An asar: false build beside a found ASAR layout names no host, so an
executable outside every layout that the native-image sample dropped may
be that build's unread fuse host. Coverage no longer counts as proven
when such an image was inventoried but not sampled.

* fix(release-exposure): read Go build-path response files named in compiler and linker flags

A CGO setting such as `-Wl,@/tmp/private-link.rsp` or
`-Xlinker @/tmp/private-link.rsp` names a response file the linker reads
options from, and a bare `@FILE` is read by the compiler driver. Report
the file's path, relative or rooted like other path arguments, without
opening it. An option's own value, such as `-install_name
@rpath/libx.dylib`, and dyld runtime tokens are not response files.

* fix(release-exposure): keep a dropped host outside an ASAR layout unread whatever its name

The guard for a sampled-out host beside a found ASAR layout ruled images
out by filename, so an asar: false host named like `b/My.App` was missed
and safe fuses from the sampled layout settled the installation. Every
inventoried native image outside the found layouts that the sample
dropped now keeps the fuse checks unknown.

* fix(release-exposure): keep maps unassessed for an unreadable extensionless launcher

An unclassified file whose initial header probe fails, such as an
extensionless `runner` that became unreadable after inventory, may carry
a Node shebang and a map directive. The probe failure now leaves its
maps unassessed, as an unread `runner.js` or packaged member does,
instead of counting the file as holding no script.

* fix(release-exposure): read a response file named like a dyld token

`-Wl,@rpath/private-link.rsp` names a linker response file, not a dyld
runtime path, so dropping every `@rpath` or `@loader_path` argument hid
it. A dyld token is now skipped only as the value of an option that takes
one, such as `-install_name` or `-dylib_file`.

* fix(release-exposure): keep the partial member scan on ordinary cancellation

A scan cancelled after some members were examined reported its partial
scan only when cleanup also failed, so a clean cancellation dropped the
recovered source-map and credential observations. A cancelled scan now
hands its assembled facts to the interrupted-scan visitor with unknown
coverage and a cancellation limitation, and the caller still receives
the cancellation error carrying that partial audit.

* docs(release-exposure): describe the partial report kept when the member scan is cancelled

Since the member scan keeps what it examined on ordinary cancellation,
the status guide no longer says that cancelling the scan returns no
partial report. It now describes the partial observation and limits the
no-report case to cancellation before the scan can be assembled.

* fix(release-exposure): keep maps unassessed when a text header is cut short

A text-named member read with one or two bytes of the text allowance
left returns only `#` or `#!`, which cannot establish a Node shebang, so
`runner.txt` kept its non-script role and its unexamined inline map was
treated as absent. A partial or missing header that could still start a
shebang line now leaves the member's maps unassessed, as an unread
member's do once the allowance is spent.

* fix(release-exposure): keep dotted executables as possible hosts in ASAR layouts

Inside an ASAR layout, a sibling whose name had a dot and no `.exe`
suffix, such as `my.app`, was neither ranked for the native sample nor
counted as a possible Electron host. With shared libraries filling the
sample, that image could be dropped while the layout still counted its
host as read, so safe fuses on a helper reported `run_as_node` as not
observed. A dotted sibling that is not named like a shared library is
now sampled after plain siblings and counted as a possible host.

* fix(release-exposure): split env -S shebangs on unquoted blanks only

The env shebang reader split the whole line on spaces before handling -S,
so a launcher such as `#!/usr/bin/env -S NODE_OPTIONS='--conditions
development' node` named `development'` as its program and its maps were
never assessed. An -S operand now takes the rest of the line and is split
the way GNU env -S splits it: single and double quotes, backslash escapes,
\_ and \c, and a leading # comment. A string env itself rejects, such as an
unterminated quote, runs no program.

* fix(release-exposure): keep owner facts completed before cancellation

When a Go, Rust, fuse or desktop owner finished its Evidence and then saw
cancellation, completeProviderEvidence returned that Evidence as the
error's partial observation, but the audit only kept the error and
substituted a not-run fact, so a recovered Go build path or Rust package
list reported unknown. Each owner's Evidence-to-fact reading now lives in
one function used for both a successful read and a completed partial, so
the partial passes the same schema and inventory digest checks before its
fact is kept. Later owners still do not run.

* fix(release-exposure): keep env -S programs named by a variable unassessed

`env -S` expands `${NAME}` from the environment, so a launcher such as
`#!/usr/bin/env -S ${NODE_BIN}` may run Node, but the shebang reader kept
the placeholder as the program name and treated the file as plain text,
reporting its inline map as absent. The env splitter now marks fields that
expand a variable, and a program whose name holds one leaves the shebang
undecided: the member's maps count as unassessed, and a `.map` companion
with such a header is not taken as a linker map. Quoted or escaped `${`
stays literal, and a resolved basename such as `${DIR}/bin/deno` still
decides.

* fix(release-exposure): keep the ASAR header admission for selected files

A directly selected app.asar whose header exceeds asar.max_header_bytes
now returns the typed header admission refusal and its recovery, as the
directory scan already does, instead of a generic partial report.

* fix(release-exposure): keep a map member with a variable env -S program unassessed

An undeclared `runner.map` whose `env -S` program is `${NODE_BIN}` may
run Node, so `examineMapMember` now treats an undecided `nodeShebang`
result as an unassessed map and companion, as `classifySiblingMap`
already does, rather than falling through to a linker-map guess.

* fix(release-exposure): read a forwarded -Map linker map path

splitForwardedLinkerFlags hands -Wl,-Map=private-link.map to the linker
side, but LINKER_PATH_FLAG had no Map entry, so the relative map path was
dropped and go.build_paths read not_observed. Add Map to the file-valued
linker options so -Map=FILE and -Map FILE are both reported.

* fix(release-exposure): count unread text in undecided map branches

An oversized .map whose shebang names its program with an unresolved
variable, or a package member with an undecided PE header, returned
before counting its unscanned text, so credential markers could read
not_observed. Count any non-bytes read as skipped text in both branches.

* fix(audit): bound map reads and preserve linker paths

* fix(audit): retain unread package uncertainty

* fix(audit): retain ambiguous linker operands

* fix(audit): preserve compiler path separators

* fix(audit): retain distinct linker map occurrences

* fix(audit): apply map declarations to their exact source path

A declaration recorded only its NFC lookup key, so every member or
sibling sharing that key took the declared JavaScript-map route. On a
normalization-sensitive filesystem an unrelated linker map spelled in
another form lost its companion evidence when the declaring script was
scanned first.

Record the resolved spellings each declaration writes and treat only an
exact match as declared. Another spelling in the bucket is read as an
undeclared map; when no declared spelling is reached, its linker-map
guess stays unresolved coverage rather than companion evidence.

* fix(audit): keep partial inventory from settling fuse-host coverage

When a directory read found ASAR layouts but the inventory stopped
early, a separate asar: false build outside those layouts may never
have been listed. Only sampled-then-dropped images counted as a
possible missed host, so reading every found layout's host reported
safe fuses as not_observed for the whole selection.

In limitNativeSample, treat an incomplete inventory as a possible
missed host outside the found layouts unless the read root is itself
a found layout, where host identity already settles coverage.

* fix(audit): apply script targets to their exact source path

A page's script element was recorded only by its NFC lookup key, so on
a normalization-sensitive filesystem a distinct file spelled in another
normalization form inherited the JavaScript role. An unrelated linker
map such as café.map beside a loaded café.map was never classified
and debug companion coverage could settle as not observed.

Keep the resolved spellings each page wrote under the lookup key, as map
declarations already do, and give the script role only to a member or
sibling whose exact path a page loaded.

* fix(audit): keep partial inventory uncertain beside a root ASAR layout

When the read root was itself a found ASAR layout, an incomplete
inventory was trusted to have listed every possible fuse host. A second
asar: false build under a subdirectory names no layout and sits outside
the root layout, so an inventory that stopped after the root host could
miss its launcher and still report the root's safe fuses as
not_observed. The same launcher, once listed, already kept the result
unknown.

Drop the root-layout exemption in limitNativeSample so any incomplete
inventory beside a found layout keeps fuse-host coverage unknown. A
complete inventory that lists only the named host still settles.

* fix(audit): read the linker's --just-symbols file operand

A Go build that forwarded -Wl,--just-symbols=private.syms through
-extldflags or CGO_LDFLAGS kept the relative symbol file in its build
settings, but the forwarded linker matcher knew only the short -R
spelling. Complete metadata then reported go.build_paths as
not_observed while the equivalent -R build was exposed.

Add --just-symbols to the linker options that read a file. Only -R
turns into -rpath when given a directory, so the $ORIGIN ambiguity
check stays on the short spelling.

* fix(audit): anchor expanded package members to the selected path

An inventory-expanded package member reports its location relative to
the selected path, such as app.asar/boot.map, but its lookup identity
resolved that location against the process working directory. An
on-disk page loading app.asar/boot.map, or an absolute file: map
declaration into the archive, resolved under the selected root, so the
declaration never matched the member it named. A script target with an
inline map then settled as a linker map, and a declared map stayed
skipped.

Resolve every member lookup and declaration base against the selected
path, and keep the reported location as written. Linker-map retraction
compares the same anchored spelling.

* fix(audit): read GCC's -I- include barrier as an option

CGO_CFLAGS=-I- is GCC's include barrier, the same as --include-barrier,
not -I with the directory "-". The compiler path matcher read it as a
build path, so a Go binary retaining it reported go.build_paths exposed.
Skip that exact field before path matching.

* fix(audit): keep packaged-image uncertainty after reading the host

A native image stored inside a package is never opened by a fuse reader,
so reading the Electron host on disk does not settle its fuses. Count the
packaged omission toward a partial fuse sample on its own, so safe host
implications stay unknown instead of not_observed while a packaged image
is unread.

* fix(audit): require every named host in a layout to be sampled

An ASAR layout can name more than one host, such as `a/a.exe` beside
`a/electron.exe`. The sample counted the layout as read when any one
named host was kept, so dropping the other still reported the Electron
executable as included and let safe fuses from the sampled images settle
run_as_node as not_observed. Count a layout as read only when every host
it could have was sampled, reusing the named-host selection the fuse
reducer already applies.

* fix(audit): keep a layout-less bundle read from settling partial inventory

A bundle with no ASAR layout, such as an asar: false build, counted its
framework read as covering the Electron host even when inventory stopped
early or the sample dropped an image. A nested build's host could sit in
the part never listed, so the root framework settled RunAsNode as
not_observed. Apply the same incomplete-inventory guard the layout branch
uses to the bundle read.

* fix(audit): read the linker's --sysroot directory operand

A Go ELF build that forwarded -Wl,--sysroot=private-sdk through
-extldflags or CGO_LDFLAGS kept the relative sysroot in its build
settings, but the forwarded linker matcher did not know the option.
Complete metadata then reported go.build_paths as not_observed, while
the compiler-owned --sysroot was already recognized.

Add --sysroot to the linker options that take a path, so the -Wl comma
form, the -Wl = form and both -Xlinker forms are read.

* test(audit): group Deno boundary regressions

* fix(audit): preserve Go linker file operands

* fix(audit): track script completion per source occurrence

* refactor(audit): give Go path parsing one semantic owner

* refactor(audit): preserve map roles at the parsing boundary

* refactor(audit): consume toolchain operands once

---------

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
2026-11-11 09:57:55 -05:00
morluto cb4e5cfbff feat(analysis): make resource budgets recoverable across CLI and MCP (#9706)
* refactor(analysis): centralize resource budgets and recovery facts

* feat(analysis): expose ASAR and Ghidra runtime admission controls

* feat(artifacts): expose reader admission and preserve limited recognition

* feat(javascript): expose recoverable semantic precision budgets

* feat(web): add recoverable source-map budgets and compact retained views

* feat(web): make historical capture decoding budgets recoverable

* fix(desktop): enforce selected view policies across capture and export

* feat(runtime): make Frida and LLDB retention budgets selectable

* feat(native): expose diagnostic budgets and preserve exhaustion causes

* feat(managed): expose recoverable snapshot and representation budgets

* feat(metadata): make Go and Rust inspection budgets recoverable

* feat(wasm): expose representation and producer admissions

* feat(evm): expose per-call worker resource budgets

* feat(android): make resource and string-pool omissions recoverable

* feat(producers): expose command and recorded-capture budgets

* docs(analysis): record recovery contracts and integrated validation

* fix(docs): link calibration oracle from published verification pages

* fix(docs): restore desktop task anchors and receipt links

* test(analysis): align fixtures with selectable budget contracts

* fix(analysis): remove source-location projector dependency cycle

* test(analysis): refresh budget recovery and verifier regressions

* test(analysis): preserve format and publication recovery expectations

* test(windows): align conformance with owned runtime admission

* fix(desktop): preserve selected alias kinds in exported state

* test(analysis): align provider fixtures with recoverable budgets

* test(desktop): validate emitted alias metadata structurally

* fix(ghidra): classify bounded Windows endpoint refusals

* test(cli): assert selectable artifact admission recovery

* fix(cli): preserve capture provenance in environment builds
2026-11-10 05:11:20 -05:00
8bc01b48b4 feat(native): list ELF and Mach-O initializer and finalizer functions (#9574)
* feat(native): list ELF and Mach-O initializer and finalizer functions

inspect_native_initializers and inspect-native-initializers list the
functions an ELF or Mach-O image runs before and after main: ELF
DT_INIT, DT_FINI, preinit, init and fini arrays and .ctors/.dtors, and
Mach-O __mod_init_func, __mod_term_func and __init_offsets, with slot
addresses, resolved targets, symbols where available and relocation
facts for relocatable objects. Nothing is executed. Slot, symbol,
relocation and note budgets are shared across tables, sentinels are
matched at the image width, and target-derived names are bounded and
marked untrusted.

* fix(native): keep ELF initializer slots unresolved when relocation coverage is partial

A skipped, unreadable or over-budget section relocation table now keeps
every slot it could reach unresolved instead of reading the stored word as
raw or dropping a .ctors slot as a sentinel. Dynamic writes that only
partly overlap a slot, or write it twice, leave it unresolved. Linked
images no longer reapply retained non-SHF_ALLOC link-time records.

The dynamic relocation walk now pauses at the shared reader's step
boundaries and observes cancellation there, resolved targets need only
their first byte mapped, and unreadable symbol names make the report
incomplete.

* fix(native): stop reporting unchecked Mach-O initializer words as resolved

Object files now apply their own section relocations: an UNSIGNED entry
against a defined symbol gives symbol_addend, one against an undefined
symbol gives unresolved_import, and any other entry leaves the slot
unresolved. Linked images without a fixup decoder, such as 32-bit ones,
keep pointer slots unresolved when an LC_DYLD_INFO bind stream or
LC_DYSYMTAB external relocations could replace the stored word.

LC_DYLD_INFO rebases now report dyld_info_rebase instead of
chained_fixup, and unreadable symbol names make the report incomplete.
Section headers keep their declared reloff and nreloc for this.

* docs(native): move initializer guide under docs/guides/native

The documentation guide assigns analyst workflows to docs/guides/<domain>/.
Move the guide there, place its navigation and directory entries in the
Native binaries group, and update links from the docs index, CLI guide,
spec and authored skill. The public /native-initializers route is
unchanged.

The guide and spec also describe per-slot relocation coverage, retained
link-time records, Mach-O object relocations, undecoded bind streams,
the dyld_info_rebase resolution and unreadable names.

* fix(native): keep IFUNC and partly relocated ELF initializer slots unresolved

An absolute relocation against an STT_GNU_IFUNC symbol no longer reports
the resolver address as the target. Section relocation records now use
their real write width, so a record that writes only part of a slot,
starts inside it or repeats it leaves the slot unresolved instead of
reading a raw zero or dropping a .ctors slot as a sentinel.

On 32-bit ARM the Thumb bit is cleared when locating a target's section
and file offset, while value_va keeps the stored callable value.

* fix(native): locate Thumb Mach-O initializer targets and reject arm64_32 with a 64-bit header

On CPU_TYPE_ARM the Thumb bit is cleared when locating and naming a
target, since nlist values do not carry it. An arm64_32 image that
declares a 64-bit header is rejected as malformed rather than read with
8-byte slots. The guide and spec describe IFUNC relocations, partly
relocated section slots, Thumb targets and this rejection.

* fix(native): index initializer address lookups and keep ambiguous mappings unknown

ELF and Mach-O initializer lookups now build the shared file mapping
index once per image, so each slot and target lookup is logarithmic in
the number of load segments rather than a scan of every mapping.

An address that load segments map ambiguously no longer falls back to a
section header offset. Its file offset is reported as unknown and the
report is incomplete; only genuinely unmapped spans use the section
fallback.

* fix(native): report Mach-O bind names over the name budget

A bound import name longer than the 1024-byte name budget was dropped
without a trace, although it is the only identity of an unresolved
initializer target. It now counts toward the over-long name limitation
and makes the report incomplete.

* test(native): count mapping lookups instead of timing the load index regression

The load-mapping regression asserted a wall-clock bound, which is
sensitive to machine load. It now spies on the shared mapping module and
requires the same number of mapping scans and index builds for 8 and 64
slots, so a per-slot scan of every load segment fails deterministically
on a small fixture.

* fix(native): yield while decoding legacy rebase runs

* fix(native): preserve initializer mapping and coverage

* fix(native): index legacy fixup spans

* fix(native): preserve relocation uncertainty and bound bind lookup work

* fix(native): validate ELF slot mappings and yield section relocation scans

---------

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
2026-11-09 08:58:12 -05:00
43f06eb1e8 feat(native): recover ELF function ranges from unwind tables (#9585)
* feat(native): recover ELF function ranges from unwind tables

inspect_elf_unwind_ranges and inspect-elf-unwind-ranges read .eh_frame
and .eh_frame_hdr from one local ELF32 or ELF64 image and list the
function ranges its FDEs name, with CIE facts, LSDA and personality,
the containing section and an exact symbol where one starts at the
address. Nothing is executed and no disassembler is needed. Pointer
encodings follow the DW_EH_PE rules with raw zero as null, discarded
FDEs are counted and skipped, the .eh_frame_hdr search table is
compared with the decoded FDEs, and record, row and symbol budgets are
bounded. verify:elf:unwind-ranges compares the reported starts with
llvm-dwarfdump or readelf on a compiled C++ program.

* fix(elf): keep unresolved unwind facts out of reported ranges

Skip FDEs whose initial location is indirect and do not locate
.eh_frame through an indirect eh_frame_ptr, since pointer slots are
not functions. Stop an unbounded header-located scan at a CIE with an
unrecognized augmentation, report ignored duplicate PT_GNU_EH_FRAME
headers as incomplete coverage, and accept a null personality stored
under an indirect encoding.

Keep only symbols at decoded FDE starts, cache name observations per
string-table offset, and decode names only for the returned page under
a fixed 16 MiB text allowance, so repeated or long names no longer
grow memory with the symbol count.

* fix(scripts): run the ELF unwind verifier under owned process trees

Launch probes, compiler drivers, the stripper, the frame dump and the
inspection through createVerifierCommands, as verify-binary-hardening
does, so a timeout or interrupt stops the whole tree. The temporary
directory is removed only after the owned commands have closed.

* docs(elf): move ELF unwind ranges guide under docs/guides/native

The documentation guide assigns analyst workflows to docs/guides/<domain>/.
Move the guide there, place its navigation and directory entries in the
Native binaries group, and update links from the docs index, CLI guide,
spec and authored skill. The public /elf-unwind-ranges route is unchanged.

* fix(elf): give unwind text limits their own recovery and skip indirect tables

Throw a dedicated failure when a page's symbol names exceed the 16 MiB
text allowance, and map it to a resource error that tells the caller
to request a smaller limit rather than raise max_ranges.

Read a CIE augmentation with one bounded NUL search and keep it as a
byte view, decoded only for returned CIEs, so a long augmentation is
never built up character by character.

Leave an .eh_frame_hdr search table with an indirect encoding unread:
the header is incomplete, the comparison is unavailable, and the table
no longer bounds a header-located .eh_frame scan.

* fix(elf): resolve header count and table forms and name IFUNC resolvers

Leave a non-zero .eh_frame_hdr FDE count under an indirect encoding
unresolved, since it is the address of a slot that is not followed, and
mark the header incomplete without reading the table.

Treat a DW_EH_PE_omit table encoding as an omitted search table before
testing the indirect bit, so a header that declares a count without a
table stays present with complete coverage, and accept that form in
output validation.

Match STT_GNU_IFUNC symbols at exact FDE starts after function symbols,
with a limitation that the name is the resolver, not the implementation
it selects at run time.

* fix(elf): require allocated unwind pointer bases

* fix(elf): map unwind header relative base

* fix(elf): validate unwind load mappings

* test(elf): cover mapped nonallocated frame agreement

---------

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
2026-11-09 21:42:53 +08:00
f21fda0fc0 feat(elf): report a Linux kernel image's version banner and embedded build config (#9573)
* feat(elf): report a Linux kernel image's version banner and embedded build config

inspect_kernel_build_config and inspect-kernel-build-config read one
local vmlinux, raw arm64 Image, gzip bzImage or zImage, or configs.ko
and report its Linux version banner and the IKCONFIG .config embedded
between IKCFG_ST and IKCFG_ED, without executing it. A config is present
only when a gzip member ends exactly at IKCFG_ED, passes its CRC-32 and
ISIZE checks and parses as .config text; framing that fails is reported
malformed, never absent. bzImage gzip payloads are inflated within a
budget and other codecs are named as unsupported, using codec detection
now shared with the kernel module parser. The config search, inflate
sizes and option counts are bounded, scans honor cancellation, and
target-derived text is marked untrusted.

* fix(elf): reject a gzip member whose header checksum fails

When FHCRC is set, compare the low 16 bits of the header CRC-32 before
inflating, so a damaged header no longer yields a present config or an
inflated bzImage payload.

* fix(elf): detect the bzImage codec inside the declared payload

Codec detection now reads the same payload_offset/payload_length slice that
is inflated, so bytes past a short payload cannot name its codec.

* fix(elf): report a framed config that does not decode as malformed

A damaged or cut-off deflate stream between IKCFG_ST and IKCFG_ED was
skipped and could end as an absent config with complete coverage. It is
now malformed with config_gzip_invalid, and a later valid config still wins.

* fix(elf): scan config lines in place and accept long string values

Config text is read one line at a time instead of split into an array, and
a quoted value is validated before it is decoded once. The 4096-character
string cap is removed: Kconfig has no such limit and max_config_bytes
already bounds every line, so a long CONFIG_EXTRA_FIRMWARE list is no
longer reported as malformed.

* fix(elf): let a live cancellation reach the kernel config parser

The parser ran synchronously, so an abort raised during parsing could not
change the signal before it returned. Inflation now runs on the zlib thread
pool and each scan yields to the event loop once per MiB, so the existing
cancellation checks observe a real abort.

* refactor(elf): carry the refused kernel config budget as a typed fact

Budget failures are now KernelBuildConfigBudgetFailure with a knob field,
and the provider builds its recovery advice from that field instead of
searching the diagnostic message for an option name.

* docs(elf): move the kernel build config guide under docs/guides/native

The documentation guide assigns analyst workflows to docs/guides/<domain>/.
Move the guide there, place its navigation and directory entries in the
Native binaries group, and update links from the docs index, CLI guide,
spec and authored skill. The public /kernel-build-config route is unchanged.

* fix(elf): yield while checking one long config line

A config made of one long quoted value never reached the per-line yield, so
the control-byte check, value validation and SHA-256 ran in one synchronous
block. Each now works one MiB window per event-loop turn, and the line
patterns match only a bounded name prefix instead of the whole value.

* fix(elf): return paged config entries that do not pin the decoded text

Entry names and values were V8 slices of the whole decoded config, so a
one-entry page kept the full text alive. Entries in the returned page are
now standalone copies, so retained memory follows the page size.

* fix(elf): yield while kernel marker searches find nothing

IKCFG_ST, IKCFG_ED and banner searches used one Buffer.indexOf over the
rest of the image, so a large image with no marker never yielded. They now
search overlapping one MiB windows and yield after each window without a
match.

* fix(elf): preserve long kernel config names

* fix(elf): frame kernel configs by gzip member boundaries

---------

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
2026-11-09 21:41:56 +08:00
Matt Van HornandMatt Van Horn 645735fc0f feat(windows): inspect Shell Link (.lnk) shortcut files (#7859)
* feat(windows): inspect Shell Link (.lnk) shortcut files

Add inspect_windows_shortcut / inspect-windows-shortcut to read one
local Shell Link without resolving or executing the target. The report
keeps header flags and FILETIMEs, IDList items, LinkInfo volume and
network names, StringData, ExtraData including Tracker droid facts
labelled derived, and derived observations such as argument length
and a minimized ShowCommand.

* fix(windows): tighten Shell Link string reads and coverage

- Decode counted StringData over its full CountCharacters payload.
- Locate LinkInfo string terminators before charging max_string_bytes
  and before decoding; an unterminated string is a coverage issue.
- Mark an IDList without a TerminalID and an out-of-range
  LinkInfoHeaderSize as partial coverage.
- Report arguments_present and argument_length as null when
  HasArguments is set but the field could not be read.
- Derive target_extension from the joined base path and suffix.
- Keep ISO dates for FILETIME and UUIDv1 timestamps before 1970.
- Retain the completed file snapshot on parse failures.
- Reject an empty CLI path before normalization.
- Select the shortcut contract from its family in focused tests.
- Fixture StringData no longer writes a NUL into the counted payload.

* style: format docs index and Shell Link boundary type

Apply oxfmt to the two files format:check reports so Static checks passes.

* fix(windows): reject network share paths for Shell Link reads

inspect_windows_shortcut is advertised as offline, but the Windows path
schema admits UNC paths, and lstat on one can contact an SMB server and
offer the caller's credentials.

Reject UNC, extended UNC, and other host-prefixed device paths in the
provider before any filesystem call. Extended drive paths such as
\\?\C:\x stay allowed. A mapped network drive cannot be told from a
local drive by its path, and the docs say so.

* fix(windows): decode Unicode Shell Link file-entry names

File-entry items with bit 0x04 of the type byte, such as 0x36, store a
UTF-16LE name, but the parser always read it as Latin-1. The read stopped
at the first zero byte, so report.exe became r and an IDList-only
shortcut reported the wrong target_extension. Read those names as
UTF-16LE.

* fix(windows): read volume labels by exact Shell Link discriminator

VolumeLabelOffset 0x14 selects VolumeLabelOffsetUnicode and is not itself a
label offset, and that Unicode field exists for no other value. The parser
treated any offset of 0x14 or more as having both, so a Unicode-only label such
as "OS" also produced a false ANSI label "O", and an ANSI label stored past 0x14
had its own bytes read as a Unicode offset.

Branch on the exact value in parseVolume, the one place that interprets it, and
report a VolumeID too short to hold the selected field. The fixture wrote both
encodings for a Unicode volume, which hid the false label, so it now writes one.

* fix(windows): limit network share rejection to Windows hosts

POSIX hosts read repeated leading slashes as one local root, so an
accepted selection such as ///tmp/shortcut.lnk is a local file there.
Apply the host-prefixed path check only on Windows, and take the
platform through the provider constructor so the Windows path forms
stay covered on every test host.

* fix(windows): compare shortcut icons only against full paths

A stored IDList filename or relative path cannot show that an icon at a
full path is elsewhere, so the comparison reported a false difference.
Report null unless the icon and a recorded target are both full paths,
and assert the missing TerminalID issue by code and offset instead of
its message.

* fix(windows): keep unread ExtraData blocks unknown

A walk that stopped before or inside ExtraData still reported the blocks
that HasExpString and HasExpIcon expect as absent, though they were never
checked for. Report a block as absent only after the TerminalBlock shows the
section was read in full. Without one the block stays unknown and coverage
stays partial.

* fix(native): ignore LNK network provider type unless ValidNetType is set

CommonNetworkRelativeLink NetworkProviderType is only meaningful when the
ValidNetType flag (0x2) is set. Return null otherwise.

* fix(native): accept any ExtraData TerminalSize below 4 as the terminal

MS-SHLLINK requires TerminalSize to be less than 4, not exactly 0. Treat
1, 2 and 3 as the terminal block instead of a malformed block.

* fix(native): derive shortcut target facts from the stored environment target

Include the EnvironmentVariableDataBlock literal target among the target
candidates for target_extension and the icon comparison. Variables are not
expanded and the target is not opened.

* fix(windows-shortcut): keep unexpanded variable paths unresolved in icon comparison

A stored environment target such as C:\Users\%USERNAME%\app.exe could be
compared as different from C:\Users\Alice\app.exe. Variable-bearing paths
are never expanded, so the icon comparison now returns null unless a fixed
target matches the icon exactly.

* test: drop redundant ok assertion before typed error guard

The testing guide asks for the typed error to be thrown before asserting a
successful Result value, and says a preceding .ok assertion adds no coverage.

* fix(windows): point out-of-range LinkInfo offsets at their field

A LinkInfo, VolumeID or CommonNetworkRelativeLink offset that points
past its structure was reported at base + offset. With an offset such
as 0xffffffff that position exceeds uint32, so the report failed its
schema and the service discarded the whole partial result.

Report such issues at the offending offset field instead, which keeps
every issue offset and coverage.failing_offset inside the file. Cover
each LinkInfo offset field in the parser tests and the
LocalBasePathOffset case through the stable filesystem service.

* fix(windows): require RFC 4122 variant before deriving Tracker UUID facts

A version nibble of 1 alone does not identify a version-1 UUID. Also require the clock_seq_hi_and_reserved top bits to be 10, so other GUID layouts keep their raw value and derive no MAC or timestamp.

* fix(windows): ignore DeviceNameOffset fields unless ValidDevice is set

MS-SHLLINK 2.3.2 says DeviceNameOffset and DeviceNameOffsetUnicode are
ignored when the ValidDevice flag is clear. parseNetwork read both
regardless, so garbage in those fields became device-name facts or
malformed_structure issues. Read the flags once and only read the device
name fields when ValidDevice is set.

* fix(windows): keep unsupported Tracker headers as raw data

Check TrackerDataBlock Length (0x58) and Version (0) before decoding.
A block with any other header is retained as raw bytes with a
malformed_structure issue, so it no longer yields MAC and timestamp
facts under complete coverage.

* fix(windows): compare shortcut icon paths by normalized location

Collapse "." and ".." segments and repeated separators in the icon
comparison as text only, never opening a path. ".." stops at the drive
or UNC share root, and \\?\ and \\.\ paths stay verbatim.

Join LocalBasePath with CommonPathSuffix as the target and use the bare
base only when the suffix is empty, as the network share branch does.

Read the selected shortcut with lstat and open only, so a rejected
symlink's destination is never resolved.

* fix(windows): keep the UNC share in the root after repeated separators

normalizePath took the host and the next split part as the UNC root, so
\\server\\share\..\app.exe kept an empty share and let .. remove the
real share. Drop empty parts before picking the root so paths on
different shares stay different.

* fix(cli): parse windows shortcut limits as decimals

The four inspect-windows-shortcut limit options now go through
cliDecimalField like the other inspection commands, so values such as
0x10 are rejected and cliDecimalArguments.test.ts passes.

* fix(native): reject linked parents of windows shortcuts

A parent directory symlink or junction pointing at a UNC share was
followed by the first lstat, since the network check reads only the
path text and the symlink rejection covers only the final component.
On Windows, lstat each ancestor from the drive root down and reject
the path at the first link, before anything follows it.

* docs(native): move Windows shortcut guide and add its feature spec

Move the Shell Link guide to docs/guides/native/windows-shortcut.md and
list it under Native binaries in the site navigation, keeping the
/windows-shortcut public route. Update README, translated README, CLI
page and authored skill links to the new physical path.

Add docs/specs/native/windows-shortcut/ with spec.md, plan.md and
tasks.md, as required by docs/development/documentation.md.

* fix(native): admit the windows shortcut parent walk

On Windows the ancestor link walk ran before the provider's resource
admission, so close() could finish while the walk kept calling lstat.
Run the walk inside the same admission as the read, and check the
caller's signal before each ancestor lstat.

* fix(native): walk the effective parents of windows shortcuts

The ancestor link walk checked every prefix of the path as written, so
C:\alias\..\real\a.lnk was refused when C:\alias is a directory link,
although Windows removes alias\.. before opening the file. Walk the
ancestors left after removing . and .. components, keeping the
caller's spelling, never climbing above an absolute root, and leaving
\\?\ paths as written since Windows does not normalize them.

* test(native): assert the linked parent as a structured fact

The parent link regression matched the phrase "junction, which" in the
refusal message, so a wording edit would fail it. Report the linked
ancestor as observed.linked_ancestor on the input issue and assert that
fact instead of the explanatory text.

* test(native): bound the shortcut close gate and release it in finally

Race the parent walk lstat acknowledgement against inspection settling
with awaitAcknowledgement, then release the gate and join the inspect
and close work in finally so a failure reports its cause instead of
waiting for the Vitest timeout.

* fix(native): walk Windows shortcut parents lazily

windowsAncestors built and kept every parent prefix before the first
lstat, so a path with thousands of components allocated memory
quadratic in its length. It now keeps only the normalized components
and yields each parent from a running prefix as the walk reaches it.

---------

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
2026-11-09 20:19:44 +08:00
吴佳浩Albenandmorluto 25885b05c0 feat(v8): inspect V8 code cache containers and retained byte patterns (#8993)
* feat(v8): inspect V8 code cache containers and their retained byte patterns

Bytenode-compiled `.jsc` files — the V8 code cache container that `vm.Script`,
`vm.compileFunction` and bytenode 2.x write, and that Node single-executable applications
and Electron apps ship — had no format-native reader. `inspect_binary_layout` shows the
surrounding ELF/PE and `inspect_artifact` inventories the file, but neither decodes the V8
structures, so an agent could not tell which V8 line produced the file, could not learn the
container is metadata rather than source, and had nothing to point at a version-pinned
decompiler. Issue #8543.

`inspect_v8_code_cache` joins the format-native family with `inspect_python_pyc`,
`inspect_erlang_beam` and `inspect_kernel_module`: one explicit local file, bounded parsing,
no execution, no installation, no child process.

**The container is version-dependent, so the reader reports roles with confidence instead
of assuming an ABI.** All eight header words are reported with byte offset, observed value,
the role this reader believes the word has, and whether that role is `confirmed`,
`inferred`, or `unconfirmed`. Only three facts gate admission: the header size this reader
pins, the V8 code cache magic high half `0xc0de` (the low half is
`ExternalReferenceTable::kSize` and varies per build, so it is a build fingerprint rather
than a constant), and a declared payload length that reaches exactly the end of the file.

I derived the field semantics from the producer rather than assuming them, and two findings
changed the design:

- **The word V8 calls the source hash holds the source length on the measured producers.**
  A 54-code-unit source stored `54`, and sources with the same code-unit length but
  different content stored the same value, including a source whose UTF-8 length was 154.
  The reader reports it as `source_length` and the docs say that this name is what the
  measurement supports, not a claim about what another build stores there.
- **`vm.compileFunction` sets a high bit in that word.** The reader masks it out and reports
  the producer family as `script` or `compile_function`, so the source length stays correct
  for both producers.

Embedded strings are reported as **byte-pattern observations with absolute byte ranges**,
labelled as such in the field name and in the limitations. A reported sequence is not proof
V8 retained it as a string, and a missing literal is not proof it is absent.

**Fail-closed boundaries, all covered by tests:** an unknown magic is a typed
`unsupported_target` failure because the file is not a V8 code cache; a short header or a
payload-length disagreement is a typed `invalid_request` failure. Neither yields a partial
parse and neither records Evidence.

Non-goals are explicit in the tool description and the docs: the payload is serialized V8
heap data, not source; recovering logic needs a version-pinned View8-style decompiler that
this tool neither provides nor depends on.

Fixtures are generated from the running V8 instead of committing a version-pinned binary, so
the tests exercise the real producer and stay valid across Node upgrades.

- Semantic owner and earliest changed stage: format-native V8 code cache container decoding.
- CLI and MCP/tool-catalog contract: one new `artifact-provider` tool and one new
  `inspect-v8-code-cache` CLI command, registered in `TOOL_CONTRACTS`, the `artifact`
  language family, `TOOL_EFFECTS`, the capability declarations and services, and the
  capability inventory. `docs:check` passes, so the generated catalog agrees.
- Provider, bridge, target-format, or platform compatibility: no provider runs and no
  external tool is required, so the tool is available on every supported Node host.
- Evidence, artifact, provenance, or reconstruction contract: complete Evidence with the
  payload byte range as its location; no artifact format, reader, or reconstruction change.
- Generated metadata, package, or installation impact: catalog and document facts
  regenerated by the existing checks; no new dependency.

Expected effects:

- [x] Target content is returned as bounded structured facts.

- `src/v8/V8CodeCacheParser.test.ts` (13 cases): real-producer header facts, per-field role
  and confidence, source length for both producers, an observation's absolute offset
  verified against the file bytes, the limitation text, foreign magic, payload-length
  mismatch, short header, missing payload bytes, a different build constant still admitted,
  and the observation budget.
- `tests/boundary/mcp/v8CodeCacheMcp.test.ts` (4 cases): advertisement and annotations,
  header facts over MCP, relative-path rejection, unsupported magic, malformed container.
- `tests/acceptance/v8CodeCacheCli.test.ts` (4 cases): the public CLI, relative-path
  resolution, byte-range observations, and the typed failure.

- `npx vitest run src/contracts src/v8 src/application/v8CodeCache src/domain/v8` — 242 passed
- `npx vitest run tests/boundary/mcp/v8CodeCacheMcp.test.ts tests/acceptance/v8CodeCacheCli.test.ts tests/boundary/filesystem/productCatalog.test.ts tests/acceptance/tools/localToolContracts.test.ts` — 20 passed, 4 skipped
- `npx tsc --noEmit -p .` — exit 0
- `npm run build` — exit 0
- `npm run lint` — 0 errors
- `npm run knip` — exit 0
- `npm run docs:check` — exit 0
- `npx oxfmt --check src tests docs` — exit 0

Gap: the header table is measured on node v24.22.1 / V8 13.6.233.17-node.58 only. No other
V8 build is available here, so a different build's field order is untested; that is exactly
why only three facts gate admission and the remaining roles carry a confidence level rather
than being asserted.

Additive. No existing tool, schema, or behaviour changes. The reader never executes the
selected file, never launches a process, and never accesses the network; its audited effects
are `mutatesSession: true` only, matching the other static format readers.

Human owner: @XiaoBinGan

- [x] I reviewed the final diff and relevant existing code, understand the change and its edge cases, and take responsibility for the complete PR, including AI-assisted work.
- [x] I checked the evidence above, confirmed it supports the current change, and recorded meaningful validation gaps.
- [x] I declare this PR ready for initial maintainer review.

- [x] This PR addresses a concrete problem or an agreed enhancement.
- [x] Material scope and authority questions are resolved, and any required maintainer scope decision covers the final diff.
- [x] The PR has one focused outcome and the title follows `type(scope): outcome`.
- [x] Related issue: #8543.
- [x] Tests cover changed observable behavior and meaningful failure paths.
- [x] Owning docs, contracts, and generated metadata are updated where needed.
- [x] I checked the final diff for secrets, unrelated cleanup, and unsupported claims.

* chore: retrigger title and readiness checks after the title was corrected

* fix(v8): bound the payload scan and use the repository path conventions

Review feedback from @greptile-apps on four points.

**Bound the synchronous payload scan.** The scan is one synchronous pass over the
payload, so the 512 MiB ceiling allowed a single request to hold the process for
several seconds. The ceiling is now 64 MiB, matching the pinned CPython pyc reader,
and the default is 32 MiB. The scan stays synchronous, like the BEAM and ELF note
readers in this family, but the budget that bounds it is no longer large enough to
matter. The new `max_file_bytes` ceiling and default are documented.

**Use the absolute-path schema.** `path` now uses `absoluteLocalPathStringSchema`
instead of `localPathStringSchema`, matching `inspect_erlang_beam`. A relative name
previously passed validation and was resolved against the server's working
directory, so `module.jsc` could select a different file than the caller meant. The
rejection is now at the schema boundary.

**State the extraction step.** The handoff in `docs/v8-code-cache.md` now says a
cache embedded in a single-executable application or an Electron app must be
extracted first, because this reader requires the magic at byte zero and returns
`unsupported_target` for the surrounding ELF/PE executable.

**Load only the owning contract group in tests.** Both tests now import
`V8_CODE_CACHE_TOOL_CONTRACTS` directly instead of resolving the tool through the
full catalog, so a focused test no longer constructs every other tool's schema.

21 cases pass across the parser, MCP boundary and CLI acceptance files; `tsc` 0;
`build` 0; `knip` 0; `docs:check` 0; `oxfmt --check` 0.

* fix(v8): select the tool contract by name in the focused tests

Reverts the catalog-position selection I added for a review suggestion. This
repository has a lint rule (`rea(no-positional-tool-contract)`) that forbids
selecting a tool contract by catalog order, and `toolContract("inspect_v8_code_cache")`
is how every neighbouring focused test selects one. The positional form I tried
introduced two lint errors, so the named form stays.

The review concern was that resolving through the full catalog constructs every
other tool's schema. That cost is real but it is the repository's chosen trade:
catalog ordering must never select an operation or schema, and `toolContract` is
the named accessor that enforces it.

* test(v8): cover the new tool in the absolute-path boundary test

Main added `tests/boundary/mcp/absoluteLocalPathTools.test.ts` after this branch
was opened. It enumerates every tool whose MCP `tools/list` schema must reject a
relative path, and `inspect_v8_code_cache` uses `absoluteLocalPathStringSchema`,
so the tool belongs in that list. Adding it keeps the enumerated field set complete
instead of leaving one absolute-path tool untested.

`tests/boundary/mcp` + v8 + contracts: 256 passing. `tsc` 0; `build` 0; `knip` 0;
`docs:check` 0; `npm run lint` 0 errors.

* perf(v8): read the payload by index instead of allocating per byte

`for (const [index, byte] of payload.entries())` allocates a two-element array for
every byte of the cache payload. On a 1 MiB payload that allocation dominated the
scan.

Measured with a real cache built from this parser, alternated so the two forms run
against the same input:

| payload | entries() | indexed reads |
| --- | --- | --- |
| 1 MiB | 15.5ms | 4.4ms |

That is 3.49x on the scan, and the gap widens on smaller payloads where the
allocation cost is a larger share (7.3x at 64 KiB in a synthetic run of the same
shape). The scan is the whole cost of `inspect_v8_code_cache` once the file is read,
so this is the tool's own hot loop.

Output is unchanged: the parser returns one retained string for the 64 KiB fixture,
and `src/v8` plus the CLI and MCP boundary files stay at 21 passing. `tsc` 0;
`build` 0; `npm run lint` 0 errors.

* fix(v8): admit only decimal tokens on the numeric CLI options

Main added `tests/boundary/cli/cliDecimalArguments.test.ts` after this branch was
opened. It enumerates every numeric CLI option and requires `0x10` and `1e3` to be
rejected, because plain `Number` coercion would read them as 16 and 1000. This
command's three options forwarded the domain fields unchanged, so the test failed on
them:

    inspect-v8-code-cache options.maxFileBytes: Parser.parse accepts 1e3
    inspect-v8-code-cache options.maxStrings: Parser.parse accepts 0x10
    inspect-v8-code-cache options.maxStrings: Parser.parse accepts 1e3
    inspect-v8-code-cache options.minStringBytes: Parser.parse accepts 0x10

`cliDecimalField` publishes the input view that parses the decimal token first and
then lets the domain field own bounds, defaults and help, which is what the other
numeric commands use.

`tests/boundary/cli/cliDecimalArguments.test.ts` now passes. `tsc` 0; `build` 0;
`npm run lint` 0 errors; `knip` 0; `docs:check` 0.

* fix(v8): recognize module caches, and advise from remaining headroom

Two review findings, each measured on Node 24.22.1 before changing anything.

**A module cache was reported as a script.** `vm.SourceTextModule.createCachedData()`
sets bit 30 of the source-length word, which the source-length mask `0x1fffffff` leaves
out, and the producer check only tested the function bit at bit 29:

| producer | source word |
| --- | --- |
| `vm.Script` | `0x00000013` for a 19-unit source |
| `vm.compileFunction` | `0x20000009` for a 9-unit source |
| `vm.SourceTextModule` | `0x40000013` for a 19-unit source |

So a real module cache was labelled `script`, and `vm.Script` rejects that container. The
producer enum gains `module` and the module bit is tested first; the source length is
unchanged because both flags sit outside the mask.

**The budget advice recommended a retry that could not work.** Both limits have
ceilings — `max_strings` at 65536 and `min_string_bytes` at 256 — so an admitted file
with more than 65536 runs of 256 bytes reached the branch while every suggested retry
was a rejected input. `observationBudgetRemediation` now reads the remaining headroom
and only recommends a change that is still admissible, falling back to selecting a
smaller artifact when neither can move. The parser message states the fact and leaves
recovery to the provider, which is the only place that can see both budgets.

`MAX_INSPECT_V8_CODE_CACHE_MIN_STRING_BYTES` names the `min_string_bytes` ceiling that
the schema previously spelled inline, so the advice and the schema cannot drift.

`docs/v8-code-cache.md` describes both flag bits. `src/v8` plus the CLI and MCP boundary
files: 28 passing; 7 new cases fail on the base commit. `tsc` 0; `build` 0;
`npm run lint` 0 errors; `knip` 0; `docs:check` 0.

* refactor(v8): decide the budget recovery as data, and test the decision

The remediation tests matched fragments of the advice sentence and banned old phrases.
`docs/testing.md` asks for error codes, recorded limits and recovery behaviour instead of
frozen prose, so a harmless reword could fail those tests while wrong advice passed them.

`observationBudgetRecovery` now returns the decision — which retries remain within the
schema's admitted range, and whether a smaller artifact is the only option — and
`observationBudgetRemediation` renders one sentence from it. The tests assert the
decision, so the wording is free to change and an incorrect choice cannot pass.

This also fixes a bug in the previous shape: it took the `min_string_bytes` branch to mean
"lower it", but that branch is reachable with the value already at the schema minimum,
where lowering is not a legal retry. The decision now carries `canLowerMinStringBytes`
and a `requiresSmallerArtifact` flag for that case.

5 cases, all failing on the base commit, covering both headroom states, either ceiling
alone, and the run-length floor. `src/v8` plus the CLI and MCP boundary files: 28
passing. `tsc` 0; `build` 0; `npm run lint` 0 errors; `knip` 0; `docs:check` 0.

* ci: rerun after the wakaru release download recovered

The only failing job on this branch was `javascript_recovery / real-javascript-recovery`,
which failed on `curl: (22) The requested URL returned error: 504` while downloading
`wakaru-linux-x64.tar.gz`. The release is reachable again (HTTP 200), so this empty
commit re-triggers the lane.

* docs(v8): place the guide in the tree and record the feature

The guide sat at `docs/v8-code-cache.md`, outside the `docs/guides/` tree every other
guide uses, and neither index linked it, so it could not be reached from
`docs/README.md` or the VitePress sidebar.

Moved to `docs/guides/javascript/`, linked from both indexes, and recorded the feature
under `docs/specs/v8-code-cache-inspection/` with `spec.md`, `plan.md` and `tasks.md`.
The spec keeps only the premises that were measured on Node 24.22.1 — the three producer
words and the retained-string layout — rather than restating the implementation.

`npm run docs:check` 0; `npm run format:check` 0; `tsc` 0; `build` 0; `lint` 0 errors;
`knip` 0. V8 reader, CLI and MCP files: 28 passing.

* docs(v8): use the documented spec path and test the advice against its struct

Two review findings.

**The records were at `docs/specs/v8-code-cache-inspection/`.** `AGENTS.md:73` asks for
`docs/specs/<domain>/<feature-slug>/`, so they move to
`docs/specs/v8/code-cache-inspection/`.

**The verification section claimed more than the tests do.** It said the producer cases
were built from real `vm.Script`, `vm.compileFunction` and `vm.SourceTextModule` output,
and that the budget advice was compared with its owning formatter. In fact only
`vm.Script` and `vm.compileFunction` are real producers — the module case sets the
measured bit on a `vm.Script` cache — and only `observationBudgetRecovery` was asserted.

Corrected the claim, and closed the second gap rather than only describing it:
`observationBudgetRemediation` now has four cases holding the sentence to the struct it
renders from. The sentence must not advise a knob the struct closed, and must mention a
smaller artifact exactly when the struct requires one. Verified by making the sentence
advise `max_strings` after the ceiling was reached — the test fails. Wording stays free;
wrong advice does not.

`npm run docs:check` 0; `npm run format:check` 0; `tsc` 0; `build` 0; `lint` 0 errors;
`knip` 0. V8 reader, CLI, MCP, absolute-path and decimal-argument files: 34 passing.

* fix(v8): restore the absolute-path contract entry after the rebase

The rebase onto `main` conflicted in `tests/boundary/mcp/absoluteLocalPathTools.test.ts`:
`main` gained `inspect_api_contract`, `inspect_devicetree`, `inspect_playwright_trace` and
`inspect_python_pyc` in both the contract list and the path-field map while this branch
added `inspect_v8_code_cache`. Both sides are additive, so both are kept.

Without the resolution the branch carried conflict markers, which made `mergeable_state`
`dirty` and every check on it meaningless. Re-verified after the rebase: `tsc` 0, `build` 0,
`format:check` 0, `lint` 0 errors, `knip` 0, `docs:check` 0, and the V8 reader plus its CLI,
MCP, absolute-path and decimal-argument boundary files at 34 passing.

* test(v8): pin the producer from the runtime's own header word

An independent reviewer ran these files on Node 22.23.4, 24.22.1 and 26.12.1. Only 22.23.4
failed, on `separates the compileFunction producer bit from the source length`, expecting
`compile_function` and reading `script`.

The cause is the fixture, not the parser: `vm.compileFunction(...).cachedData` carries
`0x20000009` in the source-length word on Node 24.22.1, with bit 29 set, while Node 22.23.4
does not set that bit at all. Whether a given V8 build marks the word is a build detail, so
asserting one build's spelling made the check version-dependent.

The case now reads the word the runtime produced and asserts the parser maps it correctly,
and keeps the part that must hold everywhere: the source length is the low bits, so it
equals both `word & 0x1fffffff` and the source's length. The flag meanings themselves stay
pinned by the synthetic `withHeaderWord` cases above, which do not depend on any build.

Reverting the parser's `compile_function` arm fails this case and
`keeps the function and module flags distinct`. `src/v8` plus the CLI, MCP and absolute-path
boundary files: 24 and 34 passing. `tsc` 0; `build` 0; `npm run format:check` 0;
`npm run lint` 0 errors; `knip` 0.

* fix(v8): report an unflagged producer word as unknown, not as a script

`vm.compileFunction` marks bit 29 of the source-length word on Node 24.22.1 (`0x20000004`
for a 4-unit source) and does not mark it on Node 22.23.4 (`0x00000004` for the same call).
Whether a V8 build marks a function cache is therefore a build detail, and an unflagged
word is either a `vm.Script` cache or an unmarked function cache. The container carries
nothing that separates the two.

The reader reported `script` for that case, which is a confident claim the bytes do not
support. My previous commit made the test accept whichever value the runtime produced,
which recorded the ambiguity instead of resolving it in the contract. `unknown` is now a
producer value, the unflagged word maps to it, and the observed word stays in
`header.fields` for a caller that trusts a specific build fingerprint. A limitation line
says the same in the report itself.

`header.producer` widens to `["script", "compile_function", "module", "unknown"]`; `script`
is no longer produced, because no unflagged word can be shown to be one.

The spec's producer rule and its verification notes are updated to match. Reverting the
mapping to `script` fails 2 of the 24 `src/v8` cases. `src/v8` with the CLI, MCP and
absolute-path boundary files: 33 passing. `tsc` 0; `build` 0; `docs:check` 0;
`npm run format:check` 0; `npm run lint` 0 errors. `knip` reports one unused export
(`builtinDataEffectMethod`) on `main` as well, so it is not from this change.

* fix(artifacts): retain completed identity across native parser failures

* docs(v8): preserve unknown unflagged cache provenance

---------

Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
2026-11-09 20:17:25 +08:00
Rishikesh Gowda K K 891371e738 fix(objdump): reflect observed launch failure in discovery availability (#9051)
* fix(objdump): reflect observed launch failure in discovery availability

Distinguish a statically registered standalone command from an executable
whose launch has failed in the active server session. Discovery leaves
unprobed commands optimistically available, reports an observed launch
failure as provider_unavailable with the command remediation, keeps the
tool callable for retries, and clears the failure after a subsequent
successful launch.

Fixes #8826.

* test(reverse): verify overlapping command launch dispatch ordering

* test(reverse): unblock gated test workers before awaiting service closure

* test(reverse): bound lifecycle acknowledgement and join launched workers

* fix(reverse): distinguish host capacity exhaustion from command launch failures

* fix(reverse): preserve host capacity across Linux process ownership preparation

* fix(reverse): exclude process ownership admission failures from command launch state

* fix(reverse): preserve Darwin process ownership preparation error across command launch

* fix(reverse): disambiguate cleanup declaration in provider observation

* fix(reverse): preserve command deadline timeout during Darwin ownership preparation

* fix(reverse): isolate Darwin baseline ps errors and provide platform-aware admission guidance

* fix(reverse): rely on typed baseline admission failures and use owning repair formatter in tests

* test(reverse): use owning formatters for process ownership remediation assertions

* fix(reverse): restrict Linux ownership prerequisite remediation to missing ps and preserve system error codes

* fix(reverse): retain typed process ownership prerequisite with preserved error codes

* fix(reverse): remove unvalidated error label fallbacks and keep admission tags out of system_code

* fix(reverse): keep Windows native runner failures out of command-launch state

* fix(process): preserve current_process_unreported refusal in Linux provider readiness

* fix(process): distinguish no-code ps failure from missing ps and cover rizin launch failure

* fix(windows): classify process_spawn host-control rejections as RuntimeControlUnavailableError
2026-11-09 17:39:34 +08:00
morlutoandCursor Agent 88cf498036 fix!: preserve results when source publication fails (#8444)
* fix: retain address context cleanup failures

* fix!: preserve address context input errors

BREAKING CHANGE: Strict clients that cached the previous full inspect_address_context output schema must refresh tools/list and accept the typed error field on unavailable facets.

* test: use session contract for address context

* fix!: preserve address context resource recovery

BREAKING CHANGE: Clients validating unavailable context facets against a cached strict full output schema must refresh tools/list to accept resource_constraint errors.

* fix: expose authenticated records refused by the evidence ledger (#8426)

* fix: preserve completed evidence on retention refusal

* fix: preserve context when unretained evidence exceeds delivery limits

* fix(mcp): recover oversized context from session evidence

* fix(evidence): preserve completed atomic write results

* fix: retain completed results when source publication fails

* test: cover bounded recovery with unknown retention

* fix(evidence): preserve recovery identity and finalization facts

* fix(evidence): clarify retention of omitted observations

* test(evidence): compare omission guidance with its owner

* fix(evidence): preserve completed composed workflows

* fix(evidence): clarify unretained recovery export

* test(evidence): cover mixed recovery guidance

* fix(evidence): preserve caller scope in secondary failures

* refactor(evidence): remove superseded cleanup projection

* test(evidence): align recovery checks with owners

* fix: preserve cleanup on partial recovery limits

* test: use owners for diagnostic wording

* fix: admit completed analysis before evidence construction

* fix: preserve selected paths when reusing evidence

* fix(cli): preserve caller path in unretained evidence

* fix(evidence): preserve recovery metadata and admission bounds

* refactor: consolidate evidence closure imports

* style(test): format CLI help default annotations (#9280)

* test(mcp): keep recovery checks focused on contract facts

* fix(evidence): preserve feature trace replay revision

* test: align evidence recovery checks with ownership

* test: align partial recovery and cleanup diagnostics

* fix(evidence): preserve nested partials during cleanup

* fix(evidence): respect outer recovery ownership

* fix(capture): discard rejected nested observations

* test(frida): assert inline evidence recovery

---------

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
2026-11-09 02:41:54 +08:00
Edgar ArakelyanandClaude Opus 5.5 fc4c381dd7 feat(desktop): keep an agentenv computer up and drive it through REA's MCP tools (#9041)
* feat(desktop): keep an agentenv computer up and drive it through REA's MCP tools

- A Python helper keeps one agentenv computer up per build with
  run_bundle(keep=True) and reports its address, MCP URL, control token
  and sandbox id; it tears the computer down when stdin closes.
- AgentEnvComputer owns the helper, waits for readiness, holds one MCP
  session to the computer's computer_* tools, records through the
  computer-recording routes, and removes the sandbox's containers by
  label when agentenv cannot confirm teardown. The token never leaves it.
- start_build_computer starts a computer from a request (archive package,
  apt repository or .deb, opened on demand with REA in /opt/rea) or from
  a bundle folder; act_on_build_computer forwards screenshot, click,
  drag, scroll, type, key and wait and returns the driver's images
  unchanged; close_build_computer tears it down.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(desktop): remove a build computer's container by name when agentenv cannot confirm teardown

agentenv's local provider runs an ordinary computer as agent-<sandbox id>
without the agentenv.sandbox label, so the label-only fallback found
nothing and reported a running computer as removed. The fallback now
lists containers by exact name and by label. The real test holds its
recording open long enough for a first frame, checks the container by
name, and runs rea inside the computer; the guide and tool descriptions
say the app opens from the Applications menu and that a screenshot after
opening a menu should come from a wait.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(desktop): own failed build-computer starts, keep ids past oversized screenshots, bound recordings

- A start is owned from the helper's spawn. One that fails and whose
  teardown is unconfirmed stays held under a computer_id close can retry;
  closeAll refuses new starts, cancels and joins starts in progress, then
  closes every held computer.
- When the first screenshot is too large, the error carries REA's result
  with the computer_id, and start_build_computer takes screenshot region,
  format and quality.
- Recordings are refused before reading when the manifest's total exceeds
  the allowance, then fetched one segment at a time, each read bounded and
  checked against its size and sha256.
- The helper keeps the tail of its output (a new tail retention mode in
  ProviderProcessSupervisor), so a build's own error is not crowded out.
- A helper that reported no computer exits on its own before SIGTERM, so
  no phantom owner is kept.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(desktop): bound a build computer's action log

Every action kept its full arguments for the computer's whole life, so a long
session grew REA's memory without limit. The log now keeps the latest 256
actions within 256 KiB, cuts each string to 1 KiB and each list to 64 items
(naming what it cut), and keeps the total and the last applied index of each
kind of action, so rules over the whole session hold after entries drop.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-06 11:01:11 +08:00
gensokyo-dev 0beedcd712 perf(mcp): add opt-in hot/cold tool discovery (#9046)
Add per-connection discovery that loads canonical named tools on demand while preserving the default full catalog. Bound discovery replies before changing visibility, preserve family exclusions and Evidence workflows, and reset visibility on HTTP reconnect.

Verified with focused protocol tests and real Codex CLI artifact investigations; token costs remain task-dependent, so the profile is opt-in.
2026-11-05 20:27:31 +00:00
560674e2de feat(native): report binary hardening for ELF, PE and Mach-O (#8524)
* feat(native): report binary hardening for ELF, PE and Mach-O

Documents inspect_binary_hardening and its CLI equivalent, including
per-property states, supporting evidence, read budgets, and one report
per Mach-O slice. Explains that missing protection imports remain
unknown, signature presence does not establish validity, and hardened
runtime requires inspect_signature. Adds hardening links throughout the
documentation. The supplied diff also removes the advanced CodeQL
workflow and configuration, along with snapshot-binding mismatch
guidance.

* fix(native): validate declared payloads and dynamic tables before hardening rows

Run the shared range checks for ELF, read imports from the dynamic tables
rather than section metadata, and keep an unresolved symbol count apart from
an absent table. Check every Mach-O command before building rows and retain
only the offsets the rows need.

Make the verifier's readelf output wide, check the emitted format before
parsing Mach-O, and place the sources before the libraries in the link.

* fix(native): share Mach-O segment admission with hardening reads

A segment with a section count its cmdsize cannot hold passed the hardening
reader and could enable restrict_segment. Move the size, section count and
file range checks to a shared admission that readMachoImage and the hardening
reader both use, and pin the ELF load-size case with a regression test.

* fix(hardening): reject Mach-O section payloads past the slice and empty ELF interpreters

Segment admission never checked each section's payload range, so a
file-backed section past the slice still produced rows. Decode sections with
the section header owner and mark the slice malformed when a payload leaves
it; zero-fill and empty sections own no file bytes and are exempt.

admitInterp accepted a NUL first byte in a longer extent, which counted an
empty interpreter as PIE evidence. It now requires a nonempty path.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KuWHGvzJkrhCqjtEHUwqyH

* fix(native): read CFG instrumentation from decoded GuardFlags

Use the IMAGE_GUARD_CF_INSTRUMENTED flag that guardFlagRecord already
decodes instead of masking the raw GuardFlags word a second time.

* fix(native): reject ELF table spans a second PT_LOAD remaps

The hardening reader mapped DT_STRTAB, DT_SYMTAB and hash tables through
the one PT_LOAD containing the table's first address. A second PT_LOAD
starting inside the table could map the rest of it from other file bytes
and the reader still reported parsed import evidence.

Cap each mapping at the next PT_LOAD that starts inside it, reject a
requested range that crosses that boundary, and stop hash reads at the
same limit, matching the conflicting-mapping refusal in readElfSymbols.

* fix(native): dedupe ELF note tables only on shared boundaries

The hardening reader dropped any note table whose bytes another table
already covered. A covering SHT_NOTE can hold a non-GNU note whose
descriptor is itself a complete GNU property note, while PT_NOTE names
that inner note directly. Walking only the outer stream missed the GNU
marker and reported CET disabled.

Drop an alias only when it starts and ends on boundaries of admitted
tables, which are note boundaries of the walked streams. Any other
covered alias leaves the feature rows unknown, as a partial overlap
already does.

* fix(native): admit every Mach-O symbol record after a canary match

The Mach-O stack protector scan returned as soon as it found an
undefined ___stack_chk_fail, so a later undefined symbol whose n_strx
lay outside the string table was never checked. The same damaged table
parsed or failed as malformed depending only on record order.

Keep the match and finish the bounded record walk. Records after the
match are only checked against the string table bounds, so their names
are not read or charged to the scan budget. The ELF reader already
admits its whole symbol table before matching names.

* fix(native): accept ELF load aliases with the same file translation

The hardening reader treated every PT_LOAD that starts inside a table,
or that also contains the table's address, as a conflicting mapping.
A load with the same address-to-file translation names the same file
bytes, so tables it overlaps still have one interpretation, and the
canonical ElfLoadMap already merges such spans.

Treat equivalent loads as one interpretation. A load with a different
translation, or the zero-filled tail of an equivalent load, still ends
the usable range, so a remapped table is still rejected.

* fix(native): keep same-span ELF note tables with different alignments unknown

A section and a program header that share endpoints were treated as one
note stream even when their effective note alignments differed. Walking the
same bytes at 4 and 8 can frame different notes, so dropping the second table
could hide a GNU property word and report CET or BTI as disabled.

Require matching effective alignment at both endpoints, and no alignment
change inside the covered span, before treating a table as an alias.
Otherwise mark the feature rows unknown. The fixture's .note.gnu.property
section now declares the class's note alignment, as toolchains emit it.

* refactor(native): read LC_SYMTAB through the shared fixed-word decoder

Move the Mach-O fixed uint32 command layouts out of MachoCommandNames
into domain/apple/machoLoadCommands so there is one owner of those
offsets. MachoCommandNames and the hardening reader now both decode
through it; the hardening reader keeps its own bounds and import policy.

* fix(native): skip Mach-O common definitions in canary imports

A common definition sets N_UNDF|N_EXT like an import but stores its
allocation size in n_value. An object holding only
`int __stack_chk_fail;` therefore reported stack_protector: enabled.

Require a zero n_value before an undefined external counts as an
import, for both nlist widths.

* refactor(native): read linkedit data commands through the shared fixed-word decoder

readMachoLinkeditDataCommand reproduced the dataoff/datasize offsets that
FIXED_WORD_FIELDS already owns. Take the command and decode through
decodeMachoFixedWordFields, as readMachoSymtabCommand does, keeping the
truncation and endianness behavior.

* fix(native): skip empty SHT_NOTE sections in ELF hardening

A zero-sized SHT_NOTE owns no bytes, so the section directory marks it
not file-backed and the hardening reader rejected the whole image as
malformed, while readElfImage accepted the same bytes. Skip empty note
sections before the file-backing check; they have nothing to walk.

* fix(scripts): run hardening verifier commands through the shared owner

The binary hardening verifier started its probes and compiler drivers with
raw spawnSync, which bypassed the src/process/ supervision that AGENTS.md
requires. Every launch now goes through createVerifierCommands, and the
temporary root is removed only after the owner closes cleanly.

* fix(scripts): clean up hardening verifier commands on Ctrl-C

SIGINT or SIGTERM ended Node before withVerifierCleanup ran, so detached
compiler trees and the temporary root could outlive the verifier. The
verifier now turns those signals into an abort that every owned command
receives, so cleanup closes the owner and removes the root before exit.

* fix(scripts): fail the hardening verifier when a signal lands outside a command

Only the owned commands read the interruption signal, so a SIGINT or
SIGTERM during the final cleanup or between commands was recorded and
then dropped, and the verifier exited 0. After cleanup settles, the
verifier now throws the interruption reason. A failure the operation or
its cleanup already raised still leaves first and is not replaced.

* fix(native): fail Mach-O slices with a malformed encryption command

The hardening reader turned every malformed status from the shared
encryption reader into an unknown `encrypted` row, so a short
LC_ENCRYPTION_INFO_64 or an encrypted range past the slice still left the
slice parsed with a full report.

Only two conflicting encryption commands stay an unknown row now. Any other
encryption defect fails the slice like other structural command failures.

* fix(native): ignore STB_LOCAL undefined ELF symbols as imports

A local undefined __stack_chk_fail or __memcpy_chk is never resolved
against another object, so it does not establish an external import.
Skip STB_LOCAL bindings when reading protector signs, and give the
fixture's positive symbols STB_GLOBAL binding.

---------

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-06 01:07:00 +08:00
morluto 6259ac41a4 fix(mcp): guide target recovery for scoped tools (#8659)
* fix(mcp): guide target recovery for scoped tools

* test(mcp): cover unscoped target recovery

* fix(mcp): warn before scoped session restart

* refactor(mcp): own scoped target recovery guidance

* fix(mcp): include unknown history in restart guidance
2026-11-03 09:49:49 +08:00
Kaoru0822andClaude Opus 5.5 144843ccc2 fix(debugger): let REA exit when a debugger could not be stopped (#8652)
* fix(debugger): let REA exit when a debugger could not be stopped

`rea debug-with-gdb` and `rea debug-with-rizin` printed cleanup_incomplete
and set exit status 1, but the retained child and its pipes kept the event
loop alive, so the process never exited. A stdio MCP server whose client
disconnected hung the same way.

closeOwnedDebuggerSessions is the owner both managers finish through. It now
unreferences retained sessions, as supervised providers already do after
incomplete cleanup, restores the references on retry, and names the retained
process IDs. The one-shot commands report those IDs, since no later command
can retry. Debugger ownership, signals and detach behavior are unchanged.

GDB cleanup requests also no longer hold the host until their 120 s command
deadline, because cleanup already bounds its own wait.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(debugger): name retained debuggers when an MCP server shuts down

An MCP server that exits with a retained debugger logged only "REA provider
cleanup failed" and "could not close cleanly", so nothing named the process
left for the caller to check. The PIDs were on OwnedDebuggerCleanupError, but
each shutdown reporter read only the outer aggregate's message, and the debug
failure_cause description dropped the field too.

The failure type now lives in its own dependency-free module with one reader
that walks composed failures. The shutdown report, the server's onclose log,
the debug description and the one-shot CLI all use it, and the shutdown report
writes the same remediation the CLI returns.

The fake debuggers now model each request exactly, with real reply shapes, and
a request they do not model fails the test.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* test(debugger): compare retained-debugger guidance with its formatter

The new checks froze the wording "Check process ... and its target".
docs/testing.md asks guidance checks to compare with the owning formatter,
alongside independent checks of the facts. They now do, and each still proves
the process ID is named, because a formatter comparison alone would pass if the
formatter dropped it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-02 05:16:34 -08:00
gensokyo-dev e18594b247 refactor(composition): unify capability bindings and cleanup ownership (#8441)
Bind typed service creation, observed availability and retryable cleanup in one composition owner. Reuse selected CLI/MCP factories and canonical contract groups without copying provider operation inventories.

Closes #8429. Refs #740.

Validated paired real application workflows, actual local Codex CLI Terra/high use, complete Evidence and Unknown checkpoint parity, compiler-backed import boundaries, SDK schemas and final real mixed routing. Automated review findings were fixed at their producing owners. Existing exact-base CodeQL alerts are documented on the PR under the owner-authorized unrelated-check exception; no global alerts or security rules were changed.

Assisted-by: Suika (gensokyo-skills:suika-scatter-gather-planning)
2026-11-01 13:30:21 +08:00
gensokyo-dev f72638f6b1 feat(investigation)!: retain records across engine cleanup (#8361)
Separate engine cleanup, investigation reset and terminal disposal in the shared application owner. Keep stable Evidence IDs and full Unknown revision history through engine close and target changes; serialize reset with all record writers and cleanup owners, optionally acknowledging a complete checkpoint before clearing. Bound retained representation without eviction.

Validation: 43 focused files / 533 distinct tests, source/type/lint/schema/documentation checks, nine genuine source cases across eight families, real Ghidra session identity and cleanup, and nine actual local Codex CLI terra/high runs with independent full-result and checkpoint audits. See docs/verification/investigation-lifecycle.md for measurements, partial attempts and verification limits.

Fixes #8347.

BREAKING CHANGE: close_binary retains investigation records. Use reset_investigation to clear them. Cross-process recovery requires an explicitly acknowledged Evidence bundle checkpoint.

Assisted-by: Suika (gensokyo-skills:suika-scatter-gather-planning)
2026-10-31 21:34:45 +00:00
morlutoandCursor Agent 1c14f66356 refactor(contracts): register CLI and MCP tools by family id (#8051)
* refactor(contracts): register CLI and MCP tools by family id

Walk the published tool-family list once for each adapter so a new
family id must supply both a CLI registrar and an MCP registrar.

* fix(cli): keep family registration off the contract tables

Walk family ids from the startup-safe list, and register the web script
commands with the application family that owns their MCP contracts.

---------

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
2026-10-30 04:25:48 +08:00
gensokyo-dev f60e42a9ae fix(mcp)!: bound validation work before tool dispatch (#7905)
Reuse the pinned SDK's combined array-element and object-member admission
before canonical validation, deferred runtime loading and tool execution.
Share a configurable finite allowance across eager/deferred HTTP and stdio,
and project refusals to bounded, actionable transport resource errors.

Preserve canonical CLI/application contracts, caller-selected input values,
ordinary validation semantics and complete Evidence access. Exercise actual
transport boundaries, real artifact/TShark workflows, packaged installation,
continued service and downstream identity/capacity regressions.

Fixes #7882

BREAKING CHANGE: MCP tool arguments now default to 1,024 combined array
elements and object members per call. Select a larger positive finite
REA_MCP_MAX_INPUT_ELEMENTS allowance and restart for larger inline inputs,
or use retained/file selectors where supported. Inputs are never truncated.

Assisted-by: Reimu (gensokyo-skills:reimu-incident-triage)
Assisted-by: Kogasa (gensokyo-skills:kogasa-surprise-testing)
Assisted-by: Yukari (gensokyo-skills:yukari-boundary-analysis)
2026-10-29 05:56:38 +00:00
morluto 3536ea2f49 fix(discovery): report debugger start host restrictions (#7965) 2026-10-29 09:20:33 +08:00
morluto 7cc33fcf7e fix(server): preserve cleanup failure diagnostics 2026-10-27 11:52:46 +08:00
morluto 3515b26221 refactor(ownership)!: remove redundant aliases
Use canonical owners for literal matches, tool families, MCP schema presentation and IDA protocol selection; remove unused web-bundle reexports. Remove the duplicate CLI command name and keep the canonical compare workflow.

BREAKING CHANGE: rea compare-bundles has been removed. Use rea compare instead. The MCP compare_bundles tool is unchanged.
2026-10-27 06:27:10 +08:00
b675073b73 feat(mcp): scope advertised tools by family (#7228)
Scope MCP discovery, setup, doctor, status and recovery guidance by selected tool families using one contract inventory and shared catalog identity. Preserve default discovery, canonical validation and complete Evidence.

Fixes #3246. Builds on #3026. Real libcrypto workflow verifies all 5,367 exports and identical complete Evidence. Static/source/build/docs/Windows/package/Greptile checks pass; ten assertions reproduced identically on clean main and feature across Node 24/22, with cancelled coverage explicitly unverified. Merge uses the user-authorized bypass for baseline failures and unchecked human-readiness declarations, without attesting human review.

Co-authored-by: 0xLLVEL <fakhrezan1@gmail.com>
Co-authored-by: Codebuff <noreply@codebuff.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-10-26 19:50:12 +08:00
gensokyo-devandAkshad Jogi 9085222ca1 feat(mcp): support compact output schema advertisements (#7218)
Add opt-in bounded output-schema advertisements while preserving the full default, canonical server validation and complete structured content. Verify real stdio discovery under a 1 MiB receive cap, all 5,367 libcrypto export coordinates, complete Evidence parity, whole-task serialization estimates and resource bounds.

Fixes #2512. Integrates #7054 by Akshad Jogi with original authorship retained.

Co-authored-by: Akshad Jogi <akshadjogi@gmail.com>
2026-10-26 17:59:51 +08:00
fc43e285ea feat(browser): retain compact scenario summaries and selected views (#7120)
Retain complete browser scenario Evidence while delivering compact summaries and selected step, event, transaction and text views through shared CLI/MCP workflows. Preserve initial CLI failures across cleanup retries and validate with real Chromium/TodoMVC.

Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-25 19:03:06 +08:00
morluto f3ed1657f8 feat(mcp): support runtime analysis limit recovery (#7049)
* feat(mcp): support runtime analysis limit recovery

* refactor(config): remove unused limit re-exports

* fix(android): keep fixture release metadata dependency-free

* fix(settings): preserve provider budgets and recovery facts

* fix(tests): resolve lint plugins from their repository owner

* refactor(mcp): verify guidance delivery without frozen wording

* fix(android): align availability with runtime timeout

* test(recovery): assert limit facts instead of frozen wording

* fix(mcp): share runtime settings with the default JADX provider

* test(mcp): bound acknowledgements and release held work on failure

* fix(catalog): declare runtime settings tools as MCP-only
2026-10-25 09:45:26 +08:00
morluto faf219e90b refactor(application): use current Evidence and CLI contracts
Require the immutable Evidence read APIs already implemented by production owners. Remove the detached-only compatibility fallbacks and forwarding helpers, and bind consumers directly to their owner. Preserve editable detached exports and existing snapshot semantics.

Use the shared canonical CLI error classifier for process commands instead of reinterpreting their projection. Validation: 172 focused application and process tests, build, static checks and documentation checks passed using local fixtures.
2026-10-25 02:44:21 +08:00
morluto 92f32981da refactor(mcp): remove the duplicate result delivery class
Construct the canonical ToolResultDelivery through the native continuation policy factory. Move type and lifecycle spy consumers to the core owner while preserving explicit response budgets and the deferred runtime composition.

Validation: runtime build and 335 focused tests passed; 41 Linux-only firmware cases were skipped on macOS. Tests use authored local fixtures, not real external providers.
2026-10-25 02:28:09 +08:00
b4ba96a9a5 feat(firmware): inspect flattened devicetree blobs (#7009)
* feat(firmware): inspect flattened devicetree blobs

Adds inspect-devicetree and inspect_devicetree for FDT v16/v17 blobs
without target execution or external dependencies. Results include
validated headers, reservations, a node/property tree with bounded raw
previews, and derived standard values and selected phandle references.
Parsing enforces file, tree, and decoded representation budgets and
rejects malformed input without partial success. Summary delivery
supports retained Evidence pagination, with tests covering parsing,
service validation, CLI delivery, and MCP access after the input is
removed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016y53PdpyMAp9RSW65fatHe

* fix(firmware): reject conflicting devicetree phandles and bound page summaries

A node whose phandle and linux,phandle disagree no longer resolves under
either value; the conflict is reported as a limitation. Finalization
failures after a completed read retain the file-snapshot-hash partial
observation, matching the Go provider. Devicetree view pages carry
fixed-size context and counts instead of repeating summary strings, so a
large bootargs no longer pushes every page over the MCP response budget.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016y53PdpyMAp9RSW65fatHe

---------

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
2026-10-24 22:45:22 +08:00
dba8a31978 feat(elf): inspect Linux kernel module metadata (#7010)
* feat(elf): inspect Linux kernel module metadata

Adds inspect-kernel-module and inspect_kernel_module for ELF32/ELF64
ET_REL Linux kernel modules, read statically without executing them.
Reports ordered .modinfo entries with duplicates preserved, __versions
symbol CRCs, the presence of .gnu.linkonce.this_module and __ksymtab,
.init.text and .exit.text sizes, and an appended module signature
trailer with its PKCS#7 byte range and SHA-256. gzip, xz and zstd
containers are reported unsupported with external decompression as the
next probe. Malformed headers and exhausted entry budgets return no
partial success.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016y53PdpyMAp9RSW65fatHe

* fix(elf): bound kernel module ksymtab listing and tighten contracts

Charge listed __ksymtab* sections against max_entries, omit section names
longer than 256 bytes with ksymtab_name_exceeds_limit, and aggregate the
non-file-backed limitation so repeated huge names cannot inflate Evidence.
Use fixedTupleSchema for the signature pad bytes with projection parity
tests, and assert the late-cancellation reason and retained observation.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016y53PdpyMAp9RSW65fatHe

* fix: resolve merge integration errors

---------

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
2026-10-24 22:35:20 +08:00
3d84215bb2 feat(pe): inspect TLS callbacks and load configuration (#7008)
* feat(pe): inspect TLS callbacks and load configuration

Adds inspect-pe-load-config and inspect_pe_load_config for PE32 and
PE32+ images, with artifact hashes and file ranges in the returned
evidence. Reports callback addresses and mappings, plus size-covered
security-cookie, SafeSEH, CFG, EH-continuation and XFG fields;
flag-derived mitigation claims are explicitly labeled inferred. Missing
directories remain null, and malformed storage or exhausted read budgets
return errors without partial success. Adds documentation and tests for
service validation, CLI/MCP access, malformed inputs and stable-file
handling.

x86 images that declare the legacy 64-byte load-config directory are
read to the structure's own Size when those bytes are file-backed, as
the Windows loader does.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016y53PdpyMAp9RSW65fatHe

* fix(pe): retain snapshot facts and bound load-config section lookup

Keep the completed stable-read path, SHA-256 and byte count as a
file-snapshot-hash partial observation when descriptor close fails or is
cancelled. Report the checked Size span as structure_location and add it
to Evidence locations beside the declared directory range. Resolve
callback sections through a sorted segment index built once per image,
keeping the ambiguity failure. Pin every load-config field to offsets
transcribed from the PE/COFF spec and llvm COFF.h at its exact Size
boundary for PE32 and PE32+.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016y53PdpyMAp9RSW65fatHe

---------

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-24 22:06:41 +08:00
781760a195 feat(apple): inspect IPS crash reports (#6511)
* feat(apple): inspect IPS crash reports

Add inspect_apple_ips for caller-selected Apple .ips files. Parse the
first-line metadata object and, when bug_type is the string 309, the
crash body: process identity, exception and termination facts, threads,
usedImages, lossless integer registers, and documented classification
rules. Other bug types including 288 stackshots return no crash result.
Opened Mach-O and dSYM UUID joins remain unobserved.

Related to #4297

* fix(apple): keep IPS register literals exact

Use the arm64e fixture integer that matches 0x8527000193ed4944 instead
of an IEEE-754-rounded decimal, and count the lossless JSON example
length from the source text.

* fix(apple): budget IPS parsing and keep typed presence

Parse the IPS metadata and crash body through a projection of the
documented members. Members the report does not interpret are
syntax-checked without being materialized, and thread, frame, image and
the new max_registers budgets run before each row is parsed, so a large
selected file fails with a typed resource error instead of first
expanding every row. Fixed allowances bound JSON nesting, unlisted key
names and termination reasons.

Present values whose JSON type differs from the documented one are no
longer read as absence: they are listed in coverage.type_mismatches and
coverage becomes partial, including wrong-type or missing threads and
usedImages collections. osVersion keeps the producer dictionary or the
string form. Frame roles follow physical frame depth, so the top frame
of a non-faulting thread is current_instruction and inlined rows share
the following frame's role. SIGKILL classes compare the canonical hex
once, so 0xdeadd0cc no longer matches dead10cc. Invalid UTF-8 is refused
instead of being replaced, and JSON strings are decoded by copying
unescaped spans.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ra7LM2NyQ7DumdssxbUq5g

* fix(apple): bound integer text expansion and reconcile coverage

* refactor(apple): retain image metadata at its owning row

* refactor(apple): validate row capacity through reconciled coverage

* test(apple): verify resource recovery at the public error boundary

* docs(apple): record complete IPS task and corpus measurements

* docs(apple): record integrated checks and stdio corpus evidence

* docs(apple): record PE integration and baseline cleanup controls

* fix(native): bound IPS string decoding and preserve invalid identities

* refactor(native): use exact optional IPS source evidence

* docs(native): verify IPS escape capacity and complete-task costs

* test(native): verify IPS through production stdio MCP

* style(tests): format IPS advertised schema bindings

* fix(native): preserve unresolved IPS references and recorded frame semantics

* docs(native): record IPS reference review evidence

* fix(apple-ips): enforce JSON and report consistency

* docs(apple-ips): record final consistency verification

---------

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
2026-10-24 16:02:59 +08:00
gensokyo-dev ead9fdcda9 feat(erlang): query selected BEAM table facts (#6943)
* feat(erlang): query selected BEAM table facts

* docs(perf): record BEAM selected-view measurements
2026-10-24 12:00:18 +08:00
4ef916da03 feat(pe): inspect import and delay-load tables (#6509)
* feat(pe): inspect import and delay-load tables

Add inspect_pe_imports / inspect-pe-imports to read PE32/PE32+ import
and delay-load directories without executing the image. The report keeps
DLL names, named and ordinal functions, ILT/IAT RVAs, and a pefile
import hash of the standard import table. A missing directory is
observed absence, not proof the image never imported APIs.

* fix(pe): default PE import fixture options

Allow peImportImage() without an argument so CLI and filesystem tests
can use the KERNEL32 sample without repeating an empty options object.

* test(pe): keep import path cases on the dedicated MCP contract

inspect_pe_imports already rejects relative MCP paths in its own
boundary test. Leave the shared absolute-path matrix to tools whose
handler issue projection matches that suite's expected shape.

* fix(native): index virtual-to-file mapping lookups

mappedFileOffset scans every mapping twice per span, so a parser that
translates one span per table entry costs sections x entries. Add
fileMappingLookup, which builds an O(n log n) interval index once and
answers each span in O(log n) with the same accept, reject and
translation results. A differential test checks it against
mappedFileOffset over randomized overlapping mappings and zero fills.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ra7LM2NyQ7DumdssxbUq5g

* fix(pe): report partial import evidence and exact pefile hashes

- Withhold imphash with an explicit reason whenever the standard import
  table is partial (unreadable names or thunks, missing tables), or when
  pefile 2024.8.26 would hash a different row sequence: invalid or
  truncated names, ordinal 0, descriptors without functions, tables its
  bounds or repeat/spread heuristics discard, and its 8192-thunk import
  symbol cap. Delay-load issues do not withhold it.
- Hash ordinal imports with pefile's complete ordlookup tables and its
  ord<decimal> fallback instead of declining unmapped ordinals.
- Report a nonterminal descriptor whose FirstThunk is zero, and a
  delay-load descriptor whose IAT or INT RVA is zero, as malformed_entry
  partial coverage instead of a complete empty inventory.
- Directory evidence lists the file spans actually consumed for the
  descriptor array and terminator; discontiguous mappings produce
  separate spans and separate Evidence ranges.
- Carry exhausted limits as PeImportLimitFailure.limit and derive
  recovery from it; the fixed 16 MiB retained-text ceiling reports a
  fixed allowance instead of max_file_bytes advice.
- Translate RVAs through the indexed mapping lookup and yield with a
  cancellation check every 1024 table entries.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ra7LM2NyQ7DumdssxbUq5g

* fix(pe): read only own ordinal tables for DLL names

DLL names such as constructor and __proto__ pass pefile's name rule but
resolved inherited object members, throwing before the ord<n> fallback.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ra7LM2NyQ7DumdssxbUq5g

* fix(native): preserve PE import facts and prove hash compatibility

* test(native): verify PE import workflows and measured costs

* refactor(native): narrow PE facts across the container union

* docs(native): record integrated PE import validation

* fix(native): bound every PE import RVA span

* docs(native): measure RVA guard completed workflows

* fix(pe): reconcile import coverage with retained rows

* fix(pe): retain partial imports at descriptor mapping gaps

---------

Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
2026-10-24 11:44:12 +08:00
gensokyo-dev e2463372ea feat(python): query selected pyc code-object facts (#6928)
* feat(python): query retained pyc code-object facts

* test(python): bind complete pyc selection benchmark tasks

* test(python): include server instructions in token accounting

* docs(perf): record pyc selected-facts task measurements

* test(python): keep oracle dependencies separate from selected targets

* docs(perf): record guarded oracle requalification

* test(python): narrow complete Evidence through its owning contract
2026-10-24 10:23:06 +08:00
morluto 53601d0bdf merge: consolidate upstream identity and representation owners 2026-10-24 05:13:41 +08:00
gensokyo-dev 57b1f82367 perf(native): query retained export table facts (#6907)
* perf(native): query retained export table facts

* docs(perf): record native export task comparisons

* fix(native): preserve omitted export projection fields

* docs(perf): refresh final export view qualification

* test(native): verify sparse export view digests
2026-10-24 05:00:05 +08:00
morluto ad629b2e31 fix(pcap): share capture host admission with discovery 2026-10-24 02:36:50 +08:00