* feat(ghidra): configurable import language and analysis seeds
REA_GHIDRA_LANGUAGE_ID and REA_GHIDRA_COMPILER_SPEC_ID set the import
language when the header cannot identify the compiler (for example,
Borland C++ PE images).
REA_GHIDRA_SEED_FILE seeds functions, code and labels before
auto-analysis, for code reached only through computed jumps or data.
Both are validated before launch, recorded in the analysis profile and
verified in the bridge handshake.
* fix(ghidra): parse language XML and preserve seed ordering
* test(ghidra): use explicit platform paths for seed launch checks
* fix(javascript): retain cleanup failure reason in error causes
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Validate Windows destination representability at the writer before registering or creating any entry. Preserve literal logical names in inventory and POSIX output. Refuse NTFS alternate stream syntax with typed path evidence and roll back prior extracted files.
Return producer path diagnostics directly, removing the presentation layer English-prefix classifier. Verify stream leaf/parent syntax, input stream cleanup, extraction rollback, inventory fidelity, POSIX materialization, and error projection: 63 passed, one platform-specific skip. Native Windows execution remains unverified here.
Fixes#1574. Adapted from and supersedes #1575.
* fix(test): preserve Ghidra interior-label controls on Mach-O
* test: preserve selector semantics and accept signaled fixture exits
* test(javascript): exercise deep returns through the public CLI
Keep failed reader, root descriptor, and output-tree cleanup with the caller. Drain admitted work before close, retry the same retained owners, and block new resource work while cleanup is incomplete. Preserve primary failures and complete inventory observations.
Thread ownership through JavaScript reconstruction, the CLI's one-shot scope, MCP server shutdown, verifier scripts, and test fixtures. Remove redundant ownership flags and the unused validated test wrapper.
Validate with 850 passing focused tests across 90 suites (two skipped), targeted classification and cleanup retries, static and unused-code checks, and real macOS CLI/MCP directory, ZIP, and DMG workflows. Native Windows and deep-analysis providers were not verified.
BREAKING CHANGE: Direct artifact inventory/extraction and JavaScript analysis APIs require an explicit caller-owned ArtifactResourceScope. CLI and MCP input contracts are unchanged.
* feat(ghidra): admit set_address_name and set_addresses_names
Ghidra sessions rejected the existing naming tools with "operation is not declared by this provider", so agents had to use annotate_native_function, which returns a complete function dossier per rename and cannot name data addresses.
The Java bridge now names a local function entry (namespace-aware, like annotate_native_function) or creates/renames the primary label at any other mapped address, verifying readback. set_addresses_names applies each address in its own transaction and reports per-address booleans. Both are mutating capabilities: excluded from read-only (Windows P0) handshakes and from snapshot replay.
* fix(ghidra): normalize qualified symbol names and reject empty leaves
Share namespace-aware leaf normalization across functions and primary labels. Invalid qualified leaves now return false for naming operations without marking the provider unavailable; function annotations retain typed invalid-input errors.
Extend real CLI/MCP namespace verification with namespaced globals, readback replay, failed-name rollback, mixed batches, and capability checks.
Validation: 75 focused tests and npm run check pass. Ghidra 12.1.4/JDK 21 namespace verification passes on macOS arm64 with unchanged source bytes and complete cleanup. The full host lane passes this workflow before the existing entry-alias fixture fails because the compiler places its interior label at the function entry.
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Make inventory reader construction synchronous and DMG attachment lazy, so the inventory registers its reader before native operations start. Share initialization, reject work after close, wait for an in-flight attach before cleanup, and preserve retryable close state.
Cover failed-attach inventory cleanup, concurrent initialization, close-before-use, and close-during-attach alongside existing ownership, cancellation, and integrity cases. Session-level retention after inventory cleanup failure remains a separate concern.
Reject Mach-O and ZIP entries whose metadata no longer matches the reader inventory, preventing a sibling adapter key from selecting bytes under another entry. Reuse the existing ASAR metadata snapshot and equality rules and read Mach-O ranges from owned records.
Move the real system-lipo test to the provider boundary suite, release its reader, and derive slice expectations from the host FAT table.
Validation: 68 focused artifact, ASAR and MCP tests plus ZIP cancellation passed. Real macOS lipo was exercised against /bin/ls; Ghidra, IDA and other deep providers were not run. Static, docs, architecture and test discovery checks passed.
Give Ghidra and Hopper an 8 MiB aggregate retained-output budget while continuing to drain their process pipes. Expose total observed bytes at the supervisor boundary and retained bytes plus explicit truncation in provider diagnostics.
Redact known transport tokens even when the retained prefix ends inside one, preserving adjacent caller path/query evidence. Existing partial-launch and cleanup ownership remain intact.
Validation: 285 tests across 34 focused suites passed; the final cutoff-fixture refinement passed all 31 Ghidra/Hopper diagnostic tests. check:fast and architecture guards passed. Provider fixtures use real child processes; real Ghidra/Hopper applications were not verified.
* fix(electron): resolve pages served from inside app.asar
Packaged Electron applications load their renderer pages from file URLs
inside app.asar. The host filesystem cannot stat those paths, so
list_electron_targets excluded every such page as unsupported_url and
inspect_electron_page could not select it.
Resolve these URLs the way Electron's file protocol does: the first .asar
path component that is a regular file is the archive, and the remainder
must name a regular file in its header. Directories, links, and missing
members stay rejected. Parsed headers are cached by archive identity.
* fix(artifacts): extract and explain rewritten ASAR unpacked companions
Packaging tools commonly sign or strip .asar.unpacked native binaries after
writing the archive header, so their declared hashes no longer match.
analyze_javascript_application already accepts integrity_policy, but
extract_artifact did not, and inspect_artifact and extract_artifact still
told callers to "get a fresh copy", which cannot help for a shipped app.
- Accept integrity_policy on extract_artifact (MCP and CLI). The default
still fails closed; record-and-continue writes the observed bytes and
returns the extracted files' integrity_contradictions records.
- Name the unpacked entry and the post-packaging cause in integrity
failures, and point every operation that accepts integrity_policy at
record-and-continue.
- Report an active unpacked entry whose companion file is absent as
unavailable with its logical path instead of a format failure.
- Share one integrity_policy schema across the inventory, extraction, and
JavaScript application inputs.
* fix(electron): bound ASAR headers before allocation
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(ghidra): clear wrong no-return flags before serving a session
Auto-analysis can flag ordinary functions as no-return (stream readers,
__cxa_guard_acquire, ...). Their callers then end at the call and the
decompiler truncates them. Run a postScript before the bridge that clears the
flag on functions containing a real RET and re-disassembles the call sites that
were terminators.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* fix(ghidra): refit in place and scope no-return call-site repair
Keep function types by refitting the owning functions with fixupFunctionBody
instead of removing and recreating them. Only redo call sites whose callee was
actually cleared (including thunks to it), skip sites followed by a function
start or x86 padding so real end calls do not run on, and always log the result.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
* fix(ghidra): constrain no-return repair to admitted return evidence
---------
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
Every inventory edge is created from one occurrence and copied that
occurrence's artifact_id and logical_path as child_artifact_id and
logical_path. Edges now keep the occurrence_id reference alone. Edge IDs
still hash the same child and path, read from the occurrence, so they are
unchanged; validation and artifact comparison read the occurrence too.
inspect_artifact output: Transmission.app 1.10 MB to 0.97 MB, Visual
Studio Code.app 15.37 MB to 13.48 MB.
BREAKING CHANGE: Artifact inventory edges no longer include
child_artifact_id or logical_path; read them from the occurrence named by
occurrence_id. The graph commitment changes, so inventory Evidence
recorded by earlier versions no longer validates.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
inventory_artifact, inspect_artifact, and extract_artifact copied each
occurrence or extracted file path into Evidence locations, although the
result already names every one. inspect_artifact did so twice, in its own
Evidence and in the nested inventory Evidence. Like other whole-target
Evidence, these now report no locations.
Visual Studio Code.app: inspect_artifact 15.37 MB to 13.07 MB,
extract_artifact 1.40 MB to 1.06 MB.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
ZIP member names are case-sensitive. Inventory was folding them with a
filesystem case rule, so shrunk APKs never became evidence for the Android
application graph. Logical registration now keeps both spellings and records
an English Unicode case-folding limitation. Extraction fails only when the
destination directory already holds the other spelling.
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
inspect_artifact nested the complete inventory Evidence and then restated
every artifact node and occurrence as an observation and every edge as a
derived relationship. No caller reads those copies; comparison and tests
use the nested inventory Evidence. On Transmission.app the result shrinks
from 2.46 MB to 1.10 MB.
Observations keep the root manifest and integrity findings.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
inspect_macho, inspect_signature, inspect_plist, list_architectures and
demangle_swift already return their command provenance inside the
normalized result. The provider also copied the same array into
raw_result, so every response carried it twice. raw_result now stays
null, as docs/mcp-contracts.md describes for results without a distinct
provider representation.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Reject omitted segment bounds, procedure classification and body facts, unresolved-call observations, dossier facets, and Ghidra parameter flow metadata. Providers emit explicit nulls for unsupported facets. Verify those unknowns in the real Hopper workflow.
BREAKING CHANGE: Provider analysis outputs must explicitly supply metadata previously repaired with defaults; unsupported nullable facets must be reported as null.
list_architectures ran lipo on every admitted target. For a PE DLL, ELF
binary or plist, lipo's refusal surfaced as execution_failure with
"Retry once; if it continues, run rea doctor", which neither can fix.
Refuse non-Mach-O targets as capability_unavailable before running
lipo, as inspect_macho already does.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Deliver complete Evidence directly through MCP with required nullable analysis profiles and immutable per-server delivery settings. Bind selected environments at construction and require typed cleanup outcomes with retained retry ownership.
Remove test-only routing, duplicate session and snapshot metadata, producer-field repairs, and ambiguous provider arguments. Separate artifact byte identity from occurrence classification and retain typed source-map and partial native facts across CLI, MCP, and verifier consumers.
BREAKING CHANGE: obsolete Evidence envelopes, session aliases, incomplete snapshots, and missing normalized producer facts are no longer accepted.
* perf(hopper): resolve string search records once per query
search_strings called _string_inventory(document) once per match, and each
call sorts every string address and rebuilds the record dict, so a query cost
O(matches x strings log strings). Resolve the inventory once per search.
Warm search, 50% matches, fake document, best of 7: 1000 strings 186 -> 0.5 ms,
4000 strings 3161 -> 2.1 ms, 8000 strings 13121 -> 4.9 ms; output byte-identical.
* perf(hopper): skip record resolution for empty string searches
Resolving the full string inventory for a query with no matches added a sort
of every string address that the previous code never performed. Return early
when nothing matches. Zero-match search on a warm document, best of 7:
64000 strings 35.9 ms before this change, 4.0 ms after (main: 3.9 ms).
Host and tool-configuration refusals from the pwntools, EVMole, firmware,
mitmproxy, Wakaru, LLDB and xcrun adapters carried no userMessage, so CLI
and MCP callers saw the generic "This analysis is unavailable for the
current target. Choose another analysis or target." as both message and
remediation. Choosing another target cannot satisfy a Linux-only profile
or an unset REA_*_COMMAND. The actual requirement appeared only in
details.reason.
Pass each refusal's reason as its userMessage, as the adapters' other
configuration refusals (EVMole limits, JADX, pwndbg) already do.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
When removing Ghidra's private runtime directory failed, close_binary
reported only a generic `provider-client` resource, and a later close,
after the permission was restored, returned success while the directory
and its contents remained. The client dropped its reference to the root
before removing it, and PrivateRuntimeRoot cached the rejected removal.
Keep each runtime or socket root until its removal succeeds, report a
failure as a ProviderCleanupError naming the leftover paths, let
PrivateRuntimeRoot retry a failed removal, and pass such typed cleanup
failures through closeAnalysisClient instead of replacing them.
Fixes#1190
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Target admission accepted a thin Mach-O after reading only its magic and
CPU type, so a 12-byte prefix opened as a valid arm64 executable and the
first provider query failed as a generic execution failure. ELF headers
were likewise accepted after 20 bytes, and a FAT file whose selected slice
lies beyond its end was admitted.
Require the complete fixed header (Mach-O 28 or 32 bytes, ELF 52 or 64),
Mach-O load commands that fit in the file, and a selected FAT slice within
the file, checking commitments against the file size rather than the
probed prefix. Each refusal names the failed constraint.
Fixes#1189
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Newer lipo names arm64e ABI variants `arm64e.<variant>`: `arm64e.v1` for
pointer-auth ABI version 1 on freshly built universal binaries, and
`arm64e.x1` on recent macOS system binaries such as /bin/ls. The
structural FAT parser names the same slice `arm64e`, so the exact name
comparison failed and inspect_artifact reported a valid, unmodified
universal binary as an integrity mismatch ("Artifact is invalid or has
changed").
Accept `arm64e.<variant>` for an `arm64e` FAT slice. Offset, size, CPU
type, subtype (including the pointer-auth version bits when lipo prints
them) and alignment are still compared field by field. The slice keeps
lipo's name in its logical path.
Fixes#1168
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two follow-ups to #1161 from its post-merge review:
- The new format preflight ran before scanArtifactInventory, which was
the first cancellation check. An already-cancelled extraction of a
DMG or PKG therefore reported unsupported_target instead of
cancellation. Check the signal before the preflight.
- The unsupported-format and encrypted-entry refusals reported the
realpath. When the caller selected the target through a symlink,
junction, or other alias, the diagnostic named a different path.
Report the caller's input path; I/O still uses the canonical path.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
extract_artifact on a format without an extraction reader, such as a
DMG, first ran the full inventory scan (mounting the image) and then
failed with an `unavailable` artifact failure: "Check artifact support
and required tools". The target format, not the host, is unsupported.
Check the format before any inventory or output work and raise
AnalysisUnsupportedTargetError, which the provider now keeps, as it does
for the other target-kind checks in this PR.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(artifacts): recheck failed DMG detaches before reporting cleanup failure
NativeDmgArtifactReader.close() detached each device once and, when a
detach failed, asked `hdiutil info` once whether the device was still
attached. On APFS images, detaching the synthesized container ejects the
image disk too. `hdiutil info` can still list that ejected device
briefly, and a just-read volume can report busy. Either way a real
cleanup succeeded but close() reported "DMG detach or mount-root cleanup
failed" and kept the mount root.
macOS CI hit this intermittently in verify:macos-bundle: `hdiutil
detach /dev/disk23` failed with "No such file or directory" (already
ejected) and the device was still counted as attached. The #1103
review also raised the missing recheck.
After the first pass, recheck the remaining devices up to three times
with 250/500/1000 ms settle delays. Drop any that hdiutil no longer
lists, retry the detach of any still listed, and fail only for devices
that stay attached. The host seam gains an optional delay so tests run
without real waits.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(artifacts): share DMG detach fixtures to stay under the duplication gate
The new busy-device and recheck cases copied the fake hdiutil host and
the APFS close sequence from neighbouring tests, which pushed jscpd to
its 2.5% threshold in Static checks. Share one single-device host
between the busy and retained-attachment cases and one APFS close
helper between the ejected-device cases.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>