* fix(artifacts): name CPU_SUBTYPE_ARM64E_X1 Mach-O slices arm64e
macOS 27 system binaries such as /bin/ls carry a third FAT slice with
cpusubtype 0x8000000c. The SDK defines subtype 12 as
CPU_SUBTYPE_ARM64E_X1, and CPU_SUBTYPE_HAS_ARM64E puts it in the arm64e
family. architectureName knew only subtype 2, so it named the slice
arm64, lipoNameMatches rejected lipo's arm64e.x1 name, and
inspect-artifact /bin/ls failed with integrity_mismatch.
The arm64e.x1 test fixture now uses the real subtype. It fails without
this change.
Refs #1168, #1765
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(artifacts): classify en-GB hdiutil verify refusals as format failures
hdiutil follows AppleLanguages. With en-GB it writes "image not
recognised", and verifyFailureReason matched only the US spelling, so a
file that is not a disk image got the reason unavailable. Both spellings
now give format. Other languages still give unavailable.
Refs #1765
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(process): read the full process table in the lifecycle cleanup check
The check ran ps -axo command= with the Node default maxBuffer of 1 MiB.
On a host with 1905 processes, the output is 1.52 MB, and the test failed
with "stdout maxBuffer length exceeded".
Refs #1765
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(process): bound cleanup observations to sampled descendants
* fix(apple): retain CPU subtype identity during dylib resolution
* test(process): retain OS PIDs for descendant observations
---------
Co-authored-by: David Lee <davidlee@Dav3Book-Pro-M4-Max.local>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* feat(server): frame analysis-target content as untrusted data
Strings, decompiled code, page text, scripts, captures and logs returned
by REA originate from the analysed target, which can embed text aimed at
the agent. Nothing marked that content, so a prompt injection inside a
binary or page could steer the agent toward tools that run code.
- Server instructions now state that target-derived tool output is data,
never instructions, and that any request in it to run, install,
download or send something is a prompt-injection attempt to report.
- Tool contracts declare returnsTargetContent. It is on by default; only
seven tools whose results hold REA session state or caller-supplied
values (binary_session, close_binary, export_evidence_bundle,
record_unknown, update_unknown, inspect_adb_client,
inspect_apktool_client) are audited exceptions, so new tools start
marked.
- Successful results of marked tools carry
_meta.provenance = { untrusted_content: true, source: "analysis_target" }.
content and structuredContent are unchanged, and the response budget
accounts for the fixed marker bytes.
Tests cover the encoding and budget accounting, the audited exception
list against the contract catalog, and an in-memory MCP client that sees
the instructions and the marker on a marked tool but not on
binary_session.
* fix(server): mark failed target-content results as untrusted too
The untrusted provenance marker was attached only on the success branch
of ToolResultDelivery, but failures can carry target-derived text as
well: error messages, captured output, and partial observations such as
the terminal frames a failed capture_process_scenario retains in
partial_observation. A host gating later process or device actions on the
marker would have treated an adversarial error payload as trusted.
Every error path in src/server now carries the failing tool's provenance:
- toErrorToolResult and toRecordedPartialErrorToolResult take the tool's
target-content flag explicitly, so each caller is audited by the type
checker; toCallToolResult, toEvidenceToolResult (recording failure), the
response-budget error that replaces an oversized success, and the
Electron partial-evidence path pass contract.returnsTargetContent.
- Admission failures in runAdmittedToolOperation resolve the flag from the
tool name through the new toolReturnsTargetContent, the same predicate
the contract metadata uses; unknown names stay marked.
- The transport's oversized-error replacement keeps the original's marker
and counts it in its budget check, matching the projected error.
- Errors the MCP SDK builds itself (input validation, thrown handlers) are
marked in the transport for target-content tools only.
Error encoding accounts for the marker bytes exactly as success encoding
does. docs/mcp-contracts.md now states that error results carry the
marker. Tests cover marked and unmarked error results, the
capture_process_scenario partial-observation errors in-process and over
the SDK transport, the recorded partial-evidence path, the budget
replacement, the oversized-error replacement, and SDK validation errors
for a marked and a content-free tool.
* fix(contracts): mark session status, close and client probes as target content
The content-free exception set was audited against successful results
only. Now that failures carry the marker too, every exception was
re-audited against its error paths with one question: can the result
include text produced by an external process, provider or device, or
stored content REA did not author?
Removed from TARGET_CONTENT_FREE_TOOLS:
- close_binary: a failed provider cleanup (GhidraClient cleanup error via
createGhidraDiagnostics) projects the launcher's captured stdout and
stderr, which can contain target-derived log text.
- binary_session: capability status reports runtime unavailability as the
provider error message (ProviderAdapterError and similar), and the
provider candidate and Android availability diagnostics include probe
output such as the apktool stderr excerpt.
- inspect_adb_client: the success result carries the version and
installed path parsed from `adb version` stdout, and a failed probe
keeps a stderr_excerpt.
- inspect_apktool_client: the success result carries the first line of
the launcher's `--version` output, and a failed probe keeps a
stderr_excerpt.
- update_unknown: the new revision carries forward the stored question
and domain, which an imported evidence bundle can supply; list_unknowns
and get_evidence_bundle already return that data marked.
The set now holds only export_evidence_bundle (path, byte and record
counts; failures are filesystem, cancellation, or a fixed integrity
message) and record_unknown (the caller's input plus REA-computed ids
and digests; registry errors project only a reason code).
The conformance test pins the reduced set and asserts each removed tool
is marked. Tests that used binary_session as the unmarked example now use
record_unknown. docs/mcp-contracts.md lists the remaining exceptions and
why the removed tools stay marked.
* fix(contracts): include target-content policy in the catalog identity
createToolCatalog() and CATALOG_IDENTITY.tools projected every
client-visible contract field except returnsTargetContent. Because the
untrusted-content framing left schemas, effects, prompts, CLI names and
the package version unchanged, a same-version MCP process started before
the update digested equal to the new server and binary_session reported
it aligned although it never emits _meta.provenance.
Project the flag as returns_target_content in both the digested tool
catalog and CATALOG_IDENTITY.tools, and export createCatalogDigests so a
boundary test proves that toggling the flag on one tool changes the
tools and combined digests.
URL.search and URL.hash are empty for both a missing component and an explicit empty one, so endpoint and origin checks were accepting URLs the callers meant to reject.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(javascript): bound value evaluation through nested closure aliases
A closure that aliases a binding from its enclosing scope evaluates that
binding once live and once at its own capture point. The capture cannot
order a binding declared in another function, so it fell back to a second
live evaluation. Only primitive results were kept, so every level of
nesting doubled the work: 20 nested closures over a non-constant value
took about 14 s, and a 443 KB generated wrapper reached the 300 s
analysis timeout.
Keep a live binding value when no caller could change it: no alias cycle
closes on or above the binding and the expression depth budget was not
reached. Bindings on a cycle are still re-evaluated, so their values and
reasons are unchanged.
Found while building a source-owned reproduction for #1847; that report's
worker heap exhaustion is a separate, already closed defect.
* fix(javascript): keep the expression depth budget for retained values
Reusing a retained composite value from any depth let a chain of wrappers
(s1 = {x: s0}, s2 = {x: s1}, ...) nest past the expression depth budget:
300 wrappers produced a 299-level value where the budget stops at 86, the
serialized IR grew quadratically, and 3,000 wrappers overflowed the stack
in JSON.stringify.
Record the expression depth at which each caller-independent value was
derived and reuse it only at that depth or shallower. A deeper reader
derives the value again and reaches the budget exactly as before. Closure
aliases read their outer binding at one depth on both paths, so nested
closures stay linear.
inspect_wasm_artifact's configuration-phase error handler assumed every
non-OwnedCommandFailure thrown before the artifact phase was about WABT
tool/directory resolution, and unconditionally appended "Supply WABT 1.0.42
via absolute REA_WABT_BIN_DIRECTORY; REA does not install tools." to the
reason. WabtArtifactProvider.inspect() also allocates its PrivateRuntimeRoot
during the configuration phase, so a PrivateRuntimeRootUnavailableError (the
Windows native-authority capability, unrelated to WABT) got the same false
remediation glued onto its own already-correct message, misdirecting anyone
troubleshooting a correctly-configured WABT install toward re-checking WABT
instead of the real blocker.
Classify PrivateRuntimeRootUnavailableError before the generic
configuration-phase branch, matching the existing convention already used
in JavaScriptAnalysisWorker.ts and SourceMapCommand.ts: surface it as its
own AnalysisCapabilityUnavailableError with no userMessage override, letting
its own message stand.
Verified against a real WABT 1.0.42 Windows release (hash-checked against
GitHub release metadata) and a real compiled .wasm module: before the fix,
`reason` read "REA Windows native controls are unavailable: ENOENT: ...
Supply WABT 1.0.42 ..." even with a fully correct REA_WABT_BIN_DIRECTORY;
after, `reason` reports only the real cause.
* fix(reference): report sloppy scripts, PHP, C# and enclosing Git facts
Importing the source/ subtree of a legacy PHP and C# checkout produced
four wrong or missing facts:
- Babel's unambiguous mode parses a module first and keeps its
strict-mode recovery errors after falling back to a script, so a valid
sloppy-mode `delete messages;` was reported as a parse failure and the
semantic IR marked the file as parser-recovered. Parsing now goes
through one owner that re-parses such a script as a script.
- .php and .cs files had no language; they are now detected as PHP and
C# (detection only, no parsing).
- A root without its own .git was reported as having no version
control even inside a checkout. The adapter now resolves the nearest
enclosing repository, as Git does, and reports `none` only when no
ancestor has one.
- The CLI recorded importer_version as null although the package
version is known.
* fix(reference): keep module strictness and classify C# and PHP as source
Review follow-up. `.mjs` and `.mts` files are modules by extension, but
the script fallback re-parsed them in sloppy mode and erased real
strict-mode errors. The parser boundary now owns the extension rule,
parses explicit modules as modules and applies the fallback only when
the mode is ambiguous; the require-binding analysis uses the same rule.
`.cs` and `.php` received a language but stayed classified as unknown,
so source-to-bundle comparison dropped them. Both are now source
extensions, like the other detected languages.
* fix(reference): honor package module scope and keep unknown-mode strict errors
Review follow-up. A `.js` file under a package.json with `"type":
"module"` is an ES module, but it went through the script fallback that
erases strict-mode errors. The import now reads each in-inventory
package.json and parses `.js` files in a module scope as modules. Other
`.js` and `.ts` sources keep an unknown mode, because bundlers treat them
as modules without that field: they are no longer forced to script, and
errors that only the module parse reports are kept as an entry limitation.
* fix(reference): bound package scopes and keep application parsing unchanged
Review follow-up.
- `.cjs` files parse as scripts, so module-only strict errors no longer
become limitations on CommonJS trees. `.cts` keeps an unknown mode
because TypeScript compiles ES import syntax there.
- An unreadable or linked package.json now bounds the module scope with
an unknown type instead of letting an outer `"type": "module"` apply.
- The application pipeline parses as on main again. Its module-only
strict errors keep a source partial; package-mode handling stays in the
reference importer.
* refactor(reference): keep unmarked module modes unknown instead of resolving package scope
Review follow-up. Resolving Node package scope for `.js` kept pulling in
more of Node's lookup semantics (require bindings, excluded or
directory-named manifests). Only extensions now select a mode: `.mjs`
and `.mts` parse as modules and `.cjs` as scripts. Every other source
keeps an unknown mode, so module-only strict errors stay visible as an
entry limitation instead of being dropped or asserted. This matches the
require-binding analysis, which also treats unmarked paths as unknown.
Decode MZ and Windows signature bytes without ASCII high-bit masking. Move malformed signature cases into the shared Go diagnostics fixtures so provider, compiled CLI, and MCP workflows cover NE, LE, LX, and PE lookalikes; remove the superseded provider-only cases.
Validation: runtime build, check:ci, and focused DOS/Go/target/CLI/MCP tests. Windows-only reference-import cases await CI.
Resolve each caller-selected Electron path once at the application boundary, collect field-specific failures with original causes, and pass canonical paths directly to the provider. Update direct provider consumers and route the real verifier through the shared application workflow.
Validation: public CLI/MCP invalid-path acceptance, canonicalization boundary, provider tests, and cancellation/cleanup checks. Live Electron launch was not verified. This does not address the separate Playwright launch-rejection crash in #2269.
Fixes#2272
Treat missing, partial, unknown, and unresolved call coverage as insufficient for call-set equality. Combine IDA caller/callee coverage without losing either explanation and recheck source coverage when verifying historical comparisons.
Scope generated function-comparison unknowns with optional typed comparison/dimension facts, preserving legacy IDs and revision digests when absent. Keep unrelated dimensions conclusive, explicit dependencies applicable, and ambiguous function identity unknown. A confirmed change for matched functions remains a failure even if calls are incomplete.
Validation: function/reconstruction/residual-unknown and bundle tests, recorded IDA protocol regression, public MCP comparison/verification, and advertised SDK JSON Schema checks. Real IDA provider execution was not verified.
Fixes#2222
Evaluate stdout and stderr together, giving hard errors and non-ok wait statuses precedence over an attempted Starting line. Retain raw streams and typed partial launch facts when Activity Manager refuses or times out; retain successful raw output too.
Validation: parser/provider tests and public MCP app/activity starts through a fake executable at the real process boundary. Real Android device behavior was not verified; device-reported timeout remains distinct from a host deadline.
Fixes#2225
Identify the factory exports parameter through the existing semantic binding owner, including renamed and minified Webpack parameters. Reject nested targets, bare alias rebinding, and shadowed identifiers. Prune the duplicated source happy path after validating the public CLI workflow.
Validation: public analyze-javascript-application CLI acceptance and focused semantic traversal/module-index tests.
Fixes#2171
Classify FAT containers using shared architecture, range, and observed slice-header facts rather than the CAFEBABE signature alone. Preserve typed unsupported and malformed reasons, reject malformed sibling slices consistently, and remove repeated range parsing in consumers.
Validation: valid small and large JVM classes in public JAR/ZIP inspection, four FAT width/byte-order controls, and focused executable/Mach-O boundary tests. The class fixtures were checked with JDK javap; no new JVM analysis capability is advertised.
The Apple selection and Hopper readers retain their existing parsing because they also own oversized-count diagnostics, reserved fields, streaming final-size validation, or source freshness that the shared header result does not represent.
Representative public mixed-ZIP inventory probe, four fresh processes with an already warm filesystem cache: first process elapsed 626.8 -> 548.3 ms, CPU 740 -> 740 ms, sampled family peak RSS 375.5 -> 345.4 MiB. Subsequent three-process medians: elapsed 767.2 -> 454.0 ms, CPU 890 -> 640 ms, peak RSS 368.7 -> 346.2 MiB. Estimated output tokens 2310 -> 2292. Concurrent load was uncontrolled, so these observations do not establish a speed improvement. Whole-agent task tokens and schema/call overhead were not measured.
Fixes#2317
Skip empty home environment values while preserving nonempty caller selections and platform precedence. Setup, doctor, and uninstall now use the same fallback home instead of writing relative paths in the current directory.
Validation: focused home tests and the public setup/doctor/uninstall lifecycle under Node 24.18. Package verification and pack dry run are included in the integrated gate.
Fixes#2325
Share contextBridge argument ownership between direct and wrapped analysis, seed intact isolated-world calls, and preserve dynamic world IDs and conditional metadata as unknown. Keep callee evaluation scoped to each invocation.
Validation: 99 focused source tests and bundled isolated-world acceptance through dedicated and generic CLI routes. Removed duplicate literal source regression in favor of public workflow coverage. Fixes#2174.
Classify coverage once at inventory production and require it in current outputs. Normalize older retained Evidence at the inventory parser, reject inconsistent page statuses, and remove comparison heuristics over limitation prose.
Validation: public CLI-to-MCP plist comparison, changed-behavior and unknowns workflow; boundary checks for legacy Evidence, partial coverage, page disagreement, content and executable changes; generated docs checks. Fixes#2173.
Select authoritative connection fields by runtime class in a shared owner used by keyed archives and NIBArchive projection. Preserve archive identities, endpoint relationships and coverage, and replace duplicate synthetic outlet assertions with a rooted public CLI fixture.
Validation: compiled CLI acceptance against the issue-sourced rooted UIKit fixture and 41 filesystem tests. Native UIKit framework validation was not rerun. Fixes#2177.
Validate unmapped segment file ranges without adding virtual mappings or fallback metadata addresses. Keep mapped-segment bounds and reject build-info sections declared in unmapped segments.
Validation: real unstripped Go 1.26.5 Darwin arm64 and amd64 artifacts passed CLI/MCP oracle, Evidence, source-range and unchanged-byte checks. Retained malformed-input regressions; removed the weaker synthetic acceptance test. Full cross-platform verifier not run. Fixes#2167.
Observe MZ header declarations and signature uncertainty at the domain boundary, then reuse those facts in bounded target resolution, inventory, and Go PE parsing. Preserve DOS relocation ownership and fail closed when the relevant bytes remain unexamined.
Validation: 59 domain tests, 130 target/Go boundary tests, and public CLI inspection of a clang/lld-link PE DLL with a short DOS header. Fixes#2172.
Inspect selected fields in existing captures through shared CLI/MCP Evidence.
Page matching frames without changing dissection or two-pass read coverage;
retain failed descriptor/process cleanup ownership and reject excess admission.
Verified with real TShark 4.0.17 and 4.2.2, public Wireshark recordings,
owned HTTP application traffic, real CLI/MCP workflows and schema checks.
Co-authored-by: Olorinm <59321404+Olorinm@users.noreply.github.com>
Assisted-by: Kogasa (gensokyo-skills:kogasa-surprise-testing)
Collect stage and check references at their owning stage boundary before evaluating required checks. Emit one missing-reference finding per stage and Evidence ID, including stages with no supplied checks, while preserving independent failed checks and cross-stage identity. The 21 readiness tests pass.
Normalize empty XDG config homes once for configuration writers and registration readers. Render the resolved plan for terminal dry runs while preserving structured output. Cover registration placement and a real PTY dry run, including absence of configuration writes.
Classify unresolvable and unusable executable or working-directory selections before launch, retaining field paths, attempted candidates and OS diagnostics. Preserve tagged analysis errors through capture lifecycle handling. Replace duplicate missing-path source assertions with public CLI/MCP regression coverage and retain distinct permission, file-kind and PATH checks.
Reuse the immutable domain parser at the file boundary instead of rebuilding and revalidating an already authenticated bundle. Existing filesystem and integrity regressions pass (27 tests); a 4 MiB parsing probe preserves immutable output and reduces repeated validation work.
Add an explicit attached native IDA MCP compatibility profile while preserving the legacy and headless workflows. Keep recorded input identity, complete source observations and the user's open GUI database. Batch unique callee lookups, conservatively prefilter literal strings, preserve Unicode matching and unavailable caller coverage, and require distinct-function switching in native verification.
Verified with Windows IDA Pro 9.3 and the native MCP plugin built from c9d15d98e8a143fc74f6e14037c48f63889b6d6c with a loopback listener. Production CLI/MCP and an unchanged Windows whoami.exe passed. Windows IDA Free was not used. No personal paths or license registration information are included.
Validation: 200 focused tests, static and documentation checks, architecture guards and discovery passed. Historical same-main warm task measurements improved from 675 to 317 ms; response token estimates fell from 16618 to 11509. Fresh-start, overall CPU and memory improvement were not established. Detailed scope is in experiments/analysis-efficiency/ida-native-windows.md.
Co-authored-by: imissapixel <140405963+imissapixel@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: N0zoM1z0 <161784452+N0zoM1z0@users.noreply.github.com>
Add opt-in EVM MCP summary delivery and retained/portable function and text views. Keep existing complete defaults, provider arguments and original full Evidence identities and records usable. Require retention before selected analysis; preserve inferred candidates, nulls, diagnostics, provenance and explicit unknowns.
Verify old schemas, historical Evidence migration, 64 focused feature checks, 30 affected MCP checks after latest-main rebase, 59 real EVM CLI/MCP cases and six shared-domain artifact workflows. Repeated SDK delivery is smaller; all real model trials and mixed token/time outcomes, limitation failures and resource qualification gaps remain documented without a uniform performance claim.
Refs #2016
Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
* fix(cli): retain correction details across browser commands
Fold the complementary implementation from #2247 into the earlier #2230 proposal, retaining the original contribution history.
Co-authored-by: LouisChen <3361008+chqcjq@users.noreply.github.com>
* fix(browser): retain diagnostics from the selected input branch
* fix(cli): recognize incomplete browser scenarios by their owned fields
---------
Co-authored-by: LouisChen <3361008+chqcjq@users.noreply.github.com>
Add offline SQLite database/WAL inspection through the shared CLI/MCP.
Use compact inventory, selected-table schema, exact ordered columns and
bounded row samples while preserving source identity and lossless values.
Retain completed Evidence through late cancellation and private-root cleanup
failures, bound concurrent inspections, and recheck journal validity after
acquisition. Reuse prepared CI runtime artifacts across supported Node lines.
Co-authored-by: Y0UW1SH <29498139+Y0UW1SH@users.noreply.github.com>
Assisted-by: Kogasa (gensokyo-skills:kogasa-surprise-testing)
Reuse the shared safe issue projection for all six CLI validation boundaries
and cover malformed inputs, credential safety, and parser diagnostics.
Fixesmorluto/rea#2180
Developed with assistance from OpenAI Codex.
* fix(apple): join recognized keys from NIBArchive field maps
inspect-asset-catalog returned no resource-key matches for a decoded
NIBArchive resource whose recognized image key is present in the archive. The
NIBArchive projection groups an object's decoded fields under the node's
nibValues attribute beside the class, label and ID it derives from the same
archive, while the AppKit keyed-archive parser flattens the same fields onto
the node. The catalog join read only the top level, so a recognized NSImageName
on a decoded resource produced no row, and the empty-catalog limitation still
claimed resource keys were reported as unmatched.
Read an archive object's fields from both shapes, and give the namespace one
owner by exporting NIB_VALUES_ATTRIBUTE from the projection that writes it.
The recognized-field set stays in the catalog analysis, which owns that
contract.
Verified with 5 focused catalog tests, check:fast and verify:test-discovery on
macOS arm64, and through the public CLI against the reported NIBArchive
fixture. The keyed-archive control keeps its existing row, and a NIBArchive
with only an unrecognized title still produces none. The ibtool-dependent
native hierarchy lane requires full Xcode and was unavailable here.
* fix(apple): normalize resource keys at the NIBArchive producer
Fold the complementary implementation from #2256 into the earlier #2248 proposal, retaining the original contribution history.
Co-authored-by: Thang Luu <76550342+thangldw@users.noreply.github.com>
* fix(apple): restrict catalog joins to parsed resource nodes
---------
Co-authored-by: 吴佳浩 <wujiahao@supremind.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Co-authored-by: Thang Luu <76550342+thangldw@users.noreply.github.com>
* fix(adb): count UTF-8 bytes in logcat total_bytes
read_adb_logcat assigned the decoded stdout string's UTF-16 code-unit
length to total_bytes, undercounting non-ASCII dumps (a 37-byte CJK line
reported 33). Measure the UTF-8 byte length instead, matching the field's
byte-unit contract and the repo's Buffer.byteLength convention.
Fixes#2175
* fix(adb): count raw captured bytes in logcat total_bytes
Carry the pre-decode stdout byte length through runAdb so malformed
UTF-8 (replaced by U+FFFD on decode) still reports the true captured
byte count instead of the re-encoded replacement size.
* fix(native): retain failed pwntools process and runtime cleanup
Keep failed worker/debugger cleanup owners and their private files until
verified process shutdown and successful directory removal. Retry cleanup
before another inspection and expose shutdown through the shared services,
CLI and MCP server. Serialize acquisition without delaying cancellation of
queued callers, and preserve primary diagnostics when CLI cleanup fails.
The original decoder reproduction left a live owned Node child with its
runtime deleted and no retained cleanup owner; the regression now verifies
retention, admission blocking, retries and eventual process/file cleanup.
Validation: 10 lifecycle/public CLI/MCP tests, 20 service tests, test
discovery 814/814, build:cached and full check including architecture guards.
Actual POSIX process/filesystem boundaries used injected Node workers on
macOS; real Linux pwntools/GDB provider parsing was not exercised.
* test(native): verify retained recorded-crash cleanup lifecycle
* fix(native): finalize diagnostic cleanup before publishing results
* fix(native): retire decoder stages before reading runtime evidence
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Bind existing EVM CLI/MCP metadata and the legacy service factory through one static descriptor, and register the same selected service that owns cleanup. Preserve original names, schemas, inputs/full defaults, configuration, Evidence and injection identity.
Verified exact SDK catalog/fact parity across six real EVM pairs, 59 actual EVM CLI/MCP cases, six real cross-domain workflows, six actual Codex CLI fact/reference/lifecycle trials and relevant local checks. CI run 38283026233 succeeds for exact head d999d9874b. No token reduction from wiring or fully qualified resource claim; final counters remain unknown. Continues #2016.
Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
* fix(adb): preserve existing bugreport archives
* fix(adb): retain bugreports after publication failure
Record acquired archive facts before publishing, preserve the staging archive when the exclusive copy fails, and keep retained evidence out of automatic cleanup. Reuse the same observation through publication and digest failures.
Add a real-filesystem ENOSPC regression covering error projection, existing evidence, subsequent collection and provider shutdown. Verified 96 focused tests, static checks and docs on macOS with simulated ADB; no real-device verification.
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
observeSettlement classified a group as alive_at_deadline when the settlement
budget expired after one empty observation, though no live member was ever
seen. Track observed liveness and return unverifiable when the deadline ends
observation with only empty rounds. The two-consecutive-empty rule for
quiesced is unchanged, and the cleanup_outcome projection for non-quiesced
settlements stays cleaned.
Fixes#2170.
* feat(ghidra): admit RISC-V ELF analysis through a pinned generic lane
Refs #2063.
Recognize `EM_RISCV` (243) as a provider-neutral family that carries its own ELF
facts, and bound the first Ghidra lane to little-endian `ET_EXEC` images that
declare the double-float ABI, pinned to `RISCV:LE:32:default` or
`RISCV:LE:64:default` with the `gcc` compiler specification.
Ghidra 12.1.4 is not reliable without the pin: its `riscv.opinion` pins
`variant="default"` for the 64-bit constraint and leaves the 32-bit constraint
with the comment that the loader "must choose between 'default' and 'AndeStar'
variants". Measured on the installed release tree, a source-owned generic RV32
ELF with no Andes declaration imports as `RISCV:LE:32:AndeStar_v5:gcc`, while
RV64 resolves to `RISCV:LE:64:default:gcc`. The lane therefore states the
language per width and refuses any caller-selected processor outside it instead
of inheriting the header guess.
Refusals name the declaration rather than falling back to a guessed profile:
big-endian images (Ghidra 12.1.4 ships no big-endian RISC-V language),
non-`ET_EXEC` images, `EF_RISCV_RVE`, `EF_RISCV_TSO`, and every float ABI other
than double, which the pinned generic compiler specifications model. The raw
e_flags bits stay in Evidence as declarations, not as instruction semantics.
Family recognition now has one owner for the ELF-metadata families:
`ELF_METADATA_ARCHITECTURES`, `isElfMetadataArchitecture`, and
`HeaderArchitecture` replace ten `Exclude<BinaryArchitecture, "mips">` sites that
would each have needed a second hand-edited member. `ghidraPinnedLanguage` is
that owner for the import commitment: the session readback previously re-derived
only the MIPS pin, so a profile committing any second pinned language failed with
a language-mismatch error before the bridge started.
Hopper and IDA refuse RISC-V by name, matching the MIPS wording. Caller-supplied
raw-binary entry keeps its existing four-family list; this lane is ELF-header
admission only.
Verification on macOS 15.7.7 arm64, Node 24.18.0, Ghidra 12.1.4 PUBLIC and JDK
21.0.12.1, with the host decompiler built from the release sources for
`os/mac_arm_64`: `npm run check` passes; `npm run test:fast` passes 396 test
files; `npm run verify:test-discovery` discovers all 917 test files exactly once.
Source-owned RV32 and RV64 double-float `ET_EXEC` fixtures complete
`rea analyze --provider ghidra` and `rea decompile rea_riscv_entry`, committing
`RISCV:LE:32:default` and `RISCV:LE:64:default` with `language_selection:
provider-pinned-v1`; the soft-float RV64 shape is refused with the named
float-ABI reason, and the full refusal matrix including the RV32 and
width-mismatch shapes is covered by unit tests. No `verify:ghidra:riscv` lane
script is included in this change; the runs above were performed manually against
the installed engine. Fork pull requests receive no CI check runs, so the
real-provider scope above is the evidence available for review.
* fix(ghidra): bound the RISC-V lane by declared flags and document it
Address review feedback on #2151.
`e_flags` bits outside `EF_RISCV_RVC | EF_RISCV_FLOAT_ABI` are now refused with
the offending value named, so a reserved or newer producer declaration cannot
enter the verified generic lane by falling through the RVE, TSO and float-ABI
checks.
Admission reads `e_type` only, so the refusal text no longer calls `ET_EXEC`
static: a fixed-address declaration does not establish static linkage, and
`PT_INTERP` and dynamic dependencies are not inspected. The session records that
boundary as a limitation rather than inferring a load model.
Add `docs/ghidra-riscv.md` in the shape of the MIPS lane guide (admission
boundary, refusal table, why the language is pinned per class, Evidence fields,
and what the fixtures do not establish), and correct
`docs/firmware-analysis.md`, which still told users that MIPS and RISC-V
decompilation had not been added.
Verification after the change: `npm run check` and `npm run docs:check` pass; the
RISC-V module tests pass (17 lane cases plus 9 parser cases); on Ghidra 12.1.4
with JDK 21.0.12.1 the RV32 and RV64 double-float fixtures still analyze and
decompile with `RISCV:LE:32:default` / `RISCV:LE:64:default` and
`language_selection: provider-pinned-v1`, the soft-float fixture is still refused
by name, and the new linkage limitation appears in the emitted Evidence.
* fix(ghidra): align RISC-V pinning with admission
Reuse the complete lane admission boundary when selecting a language, retain unsigned unsupported flag bits in diagnostics, and derive float ABI metadata from the source declaration.
* test(ghidra): verify RISC-V CLI and MCP workflows
Build source-owned RV32/RV64 fixtures with Clang and LLD, establish independent ELF and symbol facts with LLVM readelf, and verify real provider evidence, instruction bytes, calls, decompilation, schema projections, snapshot replay, refusal diagnostics, source immutability and cleanup.
Verified on macOS ARM64 with Ghidra 12.1.4, JDK 21 and LLVM/LLD 23.1.2. Dynamic linkage, big-endian, Andes and RVE/TSO workflows remain unverified; ISA attributes are not inspected.
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Report malformed Hermes and DeepSeek Harness documents as YAML. Carry the
compiled setup/uninstall diagnostics coverage from #2153 onto the existing
validation from #2157, including unchanged files and no backup writes.
Add malformed YAML scenarios for both clients.
Validation: 174 focused tests, npm run check, test discovery, package
verification, and npm pack --dry-run.
Co-authored-by: Jax367 <239124526+Jax367@users.noreply.github.com>
* fix(setup): name invalid client config shape errors clearly
Setup was dumping raw Zod issues for non-object roots or server tables,
and uninstall always claimed the file was not valid JSON. Report the
structural problem instead, and reserve parse failures for real syntax errors.
Fixes#2152
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(setup): preserve client configuration record validation
Keep descriptive shape errors while retaining Zod record validation and its own-property copy. Reject tagged YAML objects before setup or uninstall writes and prevent inherited JSONC properties from fabricating client registrations.
Add production-boundary regressions for tagged YAML roots and server tables, plus inherited Cursor and OpenCode settings. All 91 focused tests, npm run check, verify:test-discovery, verify:package, and npm pack --dry-run passed.
---------
Co-authored-by: Beqa Ksovreli <beqa@Beqas-Macbook-Pro-2.local>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(apple): preflight XML plist nesting before keyed archive decoding
inspect_keyed_archive documented a 128-level nesting limit, but the XML
plist path reached recursive decoding before any depth check: plist and
the prototype-key traversal in parseXmlPropertyList recurse once per
nesting level, so a ~75 KiB, 3000-level XML plist exhausted the call
stack and reported Maximum call stack size exceeded instead of the
bounded-resource limit that shallower over-limit documents (129, 200)
already produced. The CLI classified both as truncated, so no crash or
classification bypass was involved; the diagnostic was the defect.
Reject excessive nesting in parseXmlPropertyList before any recursive
pass, using the projection's own limit (MAX_PLIST_DEPTH) so the error is
identical to the one projectPlistValue raises: callers keep the
limit/truncated classification with no error-handler translation. Depth
counts values the way the projection does - container members nest one
level deeper, dictionary keys name members rather than nesting, and a
self-closing element is the value at its own level - so the guard
rejects exactly the documents the projection rejected, at the same
boundary, and every in-limit document decodes unchanged. The guard sits
at the shared decoder boundary, so Interface Builder, app Info.plist
resolution and native inspect_plist gain the same deterministic limit;
the binary plist path keeps its existing expansion preflight.
Binary plist behavior, ordinary XML decode, __proto__ omission counting,
and archive digest/Evidence provenance are unchanged. OpenStep
autodetection is untouched by this guard, as the existing Interface
Builder preflight also scopes its depth check to XML.
Regression coverage: decoder unit tests at and just past the limit for
dictionary and array nesting with self-closing and opaque markup,
module tests through decodeKeyedArchiveBytes including a valid
NSKeyedArchiver XML envelope at the limit, a production-boundary test
through inspectBundleKeyedArchive, and the compiled public CLI
reporting code truncated / reason limit for a 3000-level plist.
* fix(apple): check XML nesting against parsed element structure
Reuse the iterative XML decoder document for depth preflight and prototype-key lookup. Bound recursive plist wrappers alongside dictionaries and arrays without interpreting markup inside attributes or DOCTYPE declarations as values.
Validated with 104 focused tests including compiled CLI, MCP prototype-key, and Interface Builder coverage; npm run check and verify:test-discovery passed. No real-provider verification was run.
---------
Co-authored-by: Zburgers <Zburgers@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(frida): reject an unreadable script file before starting a session
The shared instrument_with_frida workflow started the Frida session, which
attaches or spawns the target, before the provider read a file script
source. A missing script file then failed loadScript with a bare
invalid_request, and session cleanup resumed the paused spawn, so the
target ran without instrumentation. Check that the selected file is a
readable regular file first and report the path, or access_denied for
EACCES/EPERM, before any device effect.
Fixes#2209
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(frida): let cancellation win over the script file check
An already-aborted request with an unreadable file source returned the
input or access-denied error instead of AnalysisCancelledError. Check the
signal before the script file preflight and again after it, so file
sources keep the existing cancellation semantics.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(frida): keep the pre-start checks out of instrument
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Frida session lookups, script lookups, unreadable script files and the
instrument-with-frida PID and script checks returned invalid_request with
no issues, while the remediation told callers to correct the listed
arguments. Attach the session_id, script_id, path, target or script issue
with the reason so CLI and MCP callers can see what to fix.
Fixes#2213
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(clients): register Qoder CLI during setup
- SupportedClients: qoder definition (~/.qoder/settings.json, format
json, QODER_CONFIG_DIR override honored)
- Qoder CLI documents mcpServers as the user-scope server table with
command/args/env and stdio as the default transport; no personal
skills directory is documented, so setup skips the skill copy
- tests: qoderClientConfiguration.test.ts (path resolution and
QODER_CONFIG_DIR override, upsert preserving unrelated settings,
fresh/empty configs, uninstall)
Verified: tsc --noEmit clean; vitest run
tests/boundary/filesystem/qoderClientConfiguration.test.ts 6/6.
* fix(clients): install Qoder CLI's personal REA skill
Review follow-up (#1572): the Qoder CLI Skills documentation lists
~/.qoder/skills/{skill-name}/SKILL.md with automatic discovery, so the
definition now carries skillPath following Qoder's configuration
location (QODER_CONFIG_DIR honored) and setup installs the bundled
skill copy there. The installation docs no longer claim Qoder has no
personal skills directory, and the bundled REA skill's supported
client list includes qoder.
Verified: tsc --noEmit clean; vitest run
tests/boundary/filesystem/qoderClientConfiguration.test.ts 7/7.
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
