Check the ARM/Thumb success marker with the runner-provided grep -F. The real Ghidra probe succeeded, but the previous rg dependency made the lane exit 127.
Validated with the actual workflow shell, producer-failure and missing/incorrect marker controls, 17 focused tests, and hosted Ghidra 12.1.4 / JDK 21 compilation and ARM/Thumb context checks.
Inject interrupts through imported projection and Evidence resolution boundaries so lexical service calls cannot bypass the test. Preserve exact normalized partial results, authenticated Evidence provenance and signal exit codes.
Assisted-by: Reimu (gensokyo-skills:reimu-incident-triage)
Require parsed seed facts in the pure trace projection and leave raw request defaults at the shared workflow boundary. Remove the redundant seed parse, unused raw-seed alias and duplicate seed-match binding.
Move omitted-versus-explicit defaults and full Evidence identity checks to compiled CLI/MCP acceptance. Reuse the owned MCP fixture and request files in the tests. Matching, native handoff, empty-literal, coverage and immutability checks remain covered; public schemas and successful outputs are unchanged.
Return expected source and semantic failures from the runtime input owner as tagged Results. Bind source schema validation to concrete schemas, preserve nested active-capture paths, and let unexpected interpretation and derivation failures reach the adapter boundary.
Remove broad protocol-error wrappers, duplicate active identity validation, and redundant JSON result parsing. Migrate parsed-fact consumers and retain CLI/MCP contracts. Relevant source and public-boundary tests pass; six injected implementation-fault controls fail on the prior revision. Live Electron behavior was not exercised.
Route CLI and MCP coverage requests through the shared strict workflow boundary before authenticating source Evidence. Preserve canonical advertised schemas, use integrity errors for malformed source bundles, and use the workflow operation for input diagnostics.
Remove the unused throwing domain facade and keep semantic fixture assertions on the parsed Result APIs. Validate compiled CLI and stdio MCP parity, strict request rejection, domain authority/time semantics, generated schemas, and unexpected failure propagation. No external providers execute in these checks.
Separate source corruption and selected-role errors from unexpected derivation failures. Normalize coordinates before body membership and conflict checks, reuse parsed requests, and account for outer observation arrays before validation copies.
Authenticate inventory requests once before family projection, remove obsolete schema aliases and raw facades, and preserve unresolved occurrence references from partial pages without fabricating content identities.
* fix(reconstruction): dedupe residual unknown ids with uniqueSorted
Current main calls uniqueBy, which is not defined, so typecheck fails.
Residual unknown ids are strings, and the adjacent lists already use
uniqueSorted.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* feat(config): make the bridge request ceiling configurable
REA_BRIDGE_MAX_REQUEST_BYTES replaces the hard-coded 16 MiB Hopper and
Cutter request payload ceiling. The default stays 16777216 bytes. Values
outside 7464 to 268435456 bytes are configuration_invalid.
Part of #6176.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(cutter): keep cleanup ownership when a sent command's reply is oversized
An authenticated command may already have changed the Cutter document
when its reply exceeds the frame ceiling. Return the existing unknown
execution, with the ceiling diagnostic in the message, so receipt
cleanup still waits. A frame that overflows before the command is sent
remains a resource constraint.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* style: format the bridge request ceiling sources
oxfmt --check failed on the five files that set and test
REA_BRIDGE_MAX_REQUEST_BYTES. Formatting only; an oversized Cutter
reply still stays an unknown execution.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(cutter): exclude the LF from the response frame budget
REA_BRIDGE_MAX_REQUEST_BYTES counts payload bytes. The Cutter response
reader and authenticated encoder were reserving one byte for the
terminating LF, so an exact-limit response was rejected. A reply that
exceeds the ceiling after the command was sent still stays an unknown
execution.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(bridge): measure escaped JSON before allocating the request
A command of U+0001 characters can expand sixfold. JSON.stringify of
that expansion can throw past Node's string ceiling before the payload
check. Count the escaped size within REA_BRIDGE_MAX_REQUEST_BYTES
first and refuse it as a resource constraint.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(config): derive the bridge request maximum from JSON expansion
A 256 MiB ceiling still lets a shorter U+0001 command expand past Node's
536870888-character string limit. The maximum is now
floor((536870888 - 296) / 7) = 76695798, the longest such command whose
authenticated frame fits in one string. Requests are still measured
before JSON.stringify.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(bridge): size the Cutter frame deadline and send the measured Hopper payload
The authenticated Cutter command is encoded after the challenge, so a
one-second read rejected an in-budget frame. That read now allows one
second per mebibyte of the ceiling and never waits longer than 25
seconds. Hello stays at one second. Hopper writes the payload it already
measured instead of serializing the same request again.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(config): bound the bridge request maximum by the serialized string
The ceiling already measures escaped JSON, so dividing Node's string limit by the seven-byte expansion rejected encoded payloads that still fit. The maximum is 536870887, one byte under that string ceiling, leaving room for the framing LF.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(hopper): serialize a bridge request when it reaches the wire
The queue kept each encoded frame, plus the caller's params, until Hopper finished the active request. A raised ceiling made a few waiting calls enough to exhaust the heap. The FIFO now keeps the params, and the frame is encoded only for the request being written.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(cutter): hash the bridge proof prefix and payload separately
An exact-limit Cutter frame plus the HMAC prefix exceeds Node's single-string ceiling, so the client threw before sending an admitted command. The prefix and payload are now separate HMAC updates, including the response proof.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(cutter): keep recovery fields when a status path exceeds the frame
A long current_file made the status reply miss the ceiling, and the fallback then dropped session_id and the completion receipts. The status now omits the path and version and still returns the session, connection proof, and up to 64 receipts. That required recovery frame is smaller than the 7464 floor.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(cutter): report an unknown path when status identity does not fit
A bounded status kept the new document generation and omitted the path, so the client filled in the descriptor's older path. The observed path is sent when it fits. When it does not, current_file is null and is not replaced with that stale path.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(cutter): count structured replies by their encoded JSON size
The response guard charged 16 bytes per primitive and two bytes per list slot, so a dense array under the configured ceiling was replaced with output_truncated. Numbers, booleans, null, and container punctuation now use their JSON lengths.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(cutter): stream authenticated responses in bounded chunks
A fitting reply was signed from retained raw bytes, escaped bytes, and a
joined frame, so a ceiling near 512 MiB could exhaust Cutter before send.
Measure and sign JSON pieces, then write the same frame in 64 KiB chunks.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(bridge): keep Cutter admission from stalling or hiding sessions
Read an unauthenticated frame off the accept loop. A command reply that
does not fit keeps completion and generation and reports an oversized path
as null. Discovery reports a ceiling mismatch instead of an absent session.
The maximum HMAC proof is opt-in, and both bridges reject non-ASCII digits.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(cutter): bound frame scans and client response copies
Search only newly read bytes for the framing LF. Unescape each socket
chunk into one response string, check the proof, and then parse it.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(cutter): keep completion, admission, and dense replies bounded
Report completed_connection only through connections that have
terminated. Cap connection threads and keep the listener up if one
fails to start. Batch scalar JSON and fail discovery when any ceiling
mismatches.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(cutter): hash requests in parts and scan escapes once
Update the client-proof prefix and request separately. Stop the response
decoder at the first quote or backslash instead of rescanning the tail.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(cutter): confirm live ceiling mismatches and omitted output
A stale descriptor from a previous ceiling is not a discovery mismatch
until its hello proves a live authenticated bridge. A response discarded
for the frame ceiling after dispatch reports output_truncated.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* test(cutter): close the unauthenticated ceiling-mismatch fixture
The listener that fails the hello proof is destroyed after its challenge
so discovery cleanup does not wait on that socket.
Co-authored-by: morluto <morluto@users.noreply.github.com>
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: morluto <morluto@users.noreply.github.com>
The registry fixture sent response headers before reading the upstream body,
then always wrote a 500 on error. A body failure hit ERR_HTTP_HEADERS_SENT,
which hid the real error and crashed the verifier with its helper still running.
The fixture now reads the body first, destroys a response that already started,
and the verifier rethrows the first fixture error.
Refs #6038.
🤖 Generated with Claude Code (Sonnet 5.5) 🦊
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
fix(managed): retain types after member limits
Admit type identity and a bounded recovery envelope before member decoding so callers can discover smaller selections after capacity exhaustion.
Fixes#5146
Co-authored-by: eco-null <eco-null@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(native): preserve whitespace in otool load-command strings
otool prints dylib install names and rpaths as `<key> %s (offset %u)`
with exactly one separator space around the recorded string. The parser
trimmed the value and stripped the offset suffix with `\s+`, so a valid
install name such as " leading.dylib " became "leading.dylib" in both
the dependency inventory and the raw load-command fields.
Parse these offset-suffixed string fields by their exact format and
strip only the single-space offset suffix.
Fixes#3384
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(native): verify exact Mach-O install names through CLI and MCP
---------
Co-authored-by: Bhaaraadwaj <bharadwajdaya@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
`start_xvfb` treated `--display-number 0` as omitted because it used the
option's truthiness, so an explicit `:0` was replaced by a random display
and its collision check ran against that random display instead.
Select the random display only when the option is absent, and add a
fixture-backed regression test covering explicit zero, an occupied
explicit zero, explicit one, and omission.
Fixes#5189
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(scripts): remove the ADB verifier's pushed probe file from the device
The verifier pushed `/data/local/tmp/rea-verify-<timestamp>.txt` for its
transfer roundtrip but never deleted it, so every successful run left a
probe behind while reporting that all checks passed.
After a successful no-overwrite push, the verifier now removes that exact
destination with the provider-reported adb client and confirms its absence
through `list_adb_directory`. Removal runs even when the pull fails, and a
removal failure is reported as a failed check instead of a pass.
`pull_adb_file` still retains device originals.
Fixes#5190
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(scripts): cover ADB probe cleanup after pull failure and leftover probes
Add verifier cases where the pull returns an error or throws after a
successful push, and where `adb shell rm` exits 0 but the listing still
shows the probe. Pull failures must still remove the probe, and a leftover
probe must fail the run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(setup): keep client and skill setup when the Hopper install fails
A failed Hopper package install returned from runSetup before any client
registration or skill install ran, so one optional provider's package
conflict (zlib-ng-compat on Arch and CachyOS) cost every other approved
change. Record the failure instead, apply the remaining plan without a
Hopper launcher, and report needs_human with the install's code and
remediation.
Fixes#5113
* fix(setup): keep the Hopper failure when a client write or the skill install fails too
* fix(setup): keep client and skill setup when the Hopper install fails
A failed Hopper package install returned from runSetup before any client
registration or skill install ran, so one optional provider's package
conflict (zlib-ng-compat on Arch and CachyOS) cost every other approved
change. Record the failure instead, apply the remaining plan without a
Hopper launcher, and report needs_human with the install's code and
remediation.
Fixes#5113
* fix(setup): keep the Hopper failure when a client write or the skill install fails too
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(apple): classify keyed-archive objects by declared superclasses
Interface Builder keyed archives record each object's superclass chain
in `$classes`, but the decoder classified objects by class-name
substrings alone. UINavigationController, UITabBarController and
UISearchController contain "control" but not "viewcontroller", so they
were reported as controls.
Classify keyed-archive objects by UIKit/AppKit view-controller, control
and view bases in their declared chain first, falling back to the name
heuristic only when the chain names none. The decoder now reads
`$classes` through one helper shared with NSArray detection.
Fixes#5218
* test(apple): cover each declared keyed-archive base with neutral class names
Neutral names classify only through the archived superclass chain, so each
UIKit and AppKit base entry and the control-before-view ordering now has a
case that fails without it.
* perf(apple): parse each keyed-archive class chain once
Objects sharing one class descriptor reparsed and scanned its whole
`$classes` array per object, so 20,000 objects sharing a 20,000-name chain
took about 16 seconds to decode. Cache each descriptor's chain as a set by
archive UID; the same archive decodes in about 3.5 seconds, matching main.
---------
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Pin admitted script identities and metadata before selected source reads. After the final CDP read, count unique additional retained IDs as dropped script inventory so incomplete captures stay unknown without extending source reads or inventing budget failures. Cover producer and comparison behavior through registered MCP tools.
Co-authored-by: gongfpp <22683417+gongfpp@users.noreply.github.com>
* feat(firmware): attribute hardware MMIO peripheral memory regions in native disassembly
- introduce hardwareMemoryMap domain model supporting standard ARM Cortex-M, RISC-V, MIPS, x86, and ARM64 physical memory region classifications (Code, SRAM, Peripheral MMIO, System Control Space, External Device)
- distinguish RAM/ROM and address-space segments from MMIO (isolate ARM External RAM / Device and MIPS kseg0/kseg1 uncached mirrors and boot ROM as non-MMIO)
- add hardwarePeripheralAttribution to map analyzed instruction references to hardware peripheral blocks and summarize peripheral register interaction profiles with numerical offset ordering
- expand ReaGhidraBridge.java decoder_context to capture ARM TMode execution register alongside MIPS ISA_MODE for native instruction inspection
- add focused unit and boundary tests across hardwareMemoryMap.test.ts, hardwarePeripheralAttribution.test.ts, GhidraFunctionValues.test.ts, and ghidraCompactReferencesMcp.test.ts verifying region bounds, MMIO isolation, numerical sorting, 32-bit/64-bit overflow rejection, ARM TMode MCP preservation, and bridge contract bindings
- make fake Ghidra installation host platform-aware on Windows in GhidraLanguageOverride.test.ts
* test(ghidra): inspect ARM and Thumb context in the Java bridge
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(hopper): keep macOS and Linux launcher paths POSIX
* fix(hopper): keep Linux and macOS bundle paths POSIX
Linux package archives and macOS app-bundle launcher paths describe the target host, not the host running the test. Build them with path.posix so exercising these installers from Windows produces the same paths as the target platform.
* fix(electron): follow Windows symlink targets with native separators
readlink returns backslash-separated targets on Windows. Splitting only on '/' kept a whole relative target as one component, so nested links were not followed and a relative target could pass the framework-root check without being walked. Split on both separators on Windows and add a regression for an escaping target.
* test(windows): align host-specific path and lifecycle expectations
Compare generated skill and provider-command paths with native joins, bind ADB device-path validation to an explicit platform, and prepare simulated POSIX Ghidra install roots on Windows. Skip POSIX-only Hopper lease, permission-bit, and symlink-parent cases, assert Windows' unsupported Hopper state instead, and cover the Windows Ghidra native-source admission step before snapshot replay. Launch Node child seams with SystemRoot available and expect Windows close-failure resource retention.
* test(windows): close review coverage gaps
---------
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(native): admit universal Mach-O files without a host slice
FAT Mach-O admission selected only the host's slice and rejected a valid
container whose slices are all foreign, so list-architectures returned
target_unavailable instead of the container's inventory. A thin foreign
image was already admitted.
Prefer the host's slice and otherwise select the first declared supported
slice. Remove the Ghidra installation probe's re-parse under other host
architectures, which existed only to work around the rejection.
Fixes#5404
* fix(native): select the first valid foreign FAT slice
Without a host slice, FAT selection took the first declared slice even when
it was too short for its Mach-O header, rejecting a container whose later
foreign slice is valid. The Ghidra decompiler probe then reported such a
foreign-only decompiler as unknown rather than incompatible. Select the first
foreign slice that can hold its header; a malformed host slice still fails.
* fix(native): select foreign FAT slices from every declaration
Foreign-slice fallback chose among architecture families after keeping only
each family's first declaration, so an undersized first arm64 subtype hid a
valid later arm64 subtype. Select from all declarations, and report the
deduplicated families separately. The host-slice rule is unchanged. Move the
FAT selection tests into their own block.
* fix(native): admit only unambiguous foreign FAT families
With no host slice, a family declared more than once, such as arm64 and
arm64e, has no provider-neutral choice: IDA analyzes the recorded slice while
Hopper prefers its own subtype. Select only families declared exactly once and
keep ambiguous foreign-only containers rejected, as on main.
* fix(ghidra): decide FAT decompiler compatibility from validated families
Slice selection now refuses a foreign-only FAT table that declares one family
more than once, such as arm64 and arm64e, and the decompiler probe reported that
refusal as unknown, so an x64 host accepted an arm64-only decompiler. When
selection fails, read the table's validated CPU families and report the
decompiler incompatible when they exclude the host. Selection and host
exclusion now have separate owners in binaryTarget.ts.
Authenticate source bundles through their shared owner and retain workspace lookup facts for evaluation. Preserve source integrity failures and precise caller constraints while allowing unexpected generated-output failures to propagate. Carry the typed closure result through the service and CLI without redundant JSON validation.
Validation: 242 focused tests, source checks, docs conformance, test discovery and duplication gate passed. Six public SDK regressions fail on the baseline. No analysis provider execution is claimed for the coverage workflow.
* fix(native): admit thin arm64_32 Mach-O targets for static inspection
Target admission mapped Mach-O CPU types only to x86, x86_64, arm and
arm64, so a thin arm64_32 object was rejected as "unsupported Mach-O
architecture" before list-architectures, inspect-macho or
inspect-signature could run, although the same slice inside a FAT file
was inventoried.
Recognize CPU_TYPE_ARM64_32 as its own Mach-O-only architecture with a
dedicated thin target variant, so ELF, PE and raw-binary paths cannot
carry it. Add its slice policy so 32-bit dispatch metadata refuses it
like other 32-bit Mach-O images instead of reporting a CPU mismatch.
Ghidra, Hopper and IDA have no verified arm64_32 lane and decline it
explicitly; Hopper's declined families now share one predicate.
Fixes#5713
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* ci(apple): run the thin arm64_32 CLI regression on macOS
The regression compiles an arm64_32 object with Apple clang, so it skips
on the Ubuntu jobs. List it in the macOS Apple artifact job and its
change scope so the job runs it and changes to it select that job.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(mcp): report Linux-only integrations as unsupported_host off their hosts
Session discovery folded the host check into each optional integration's
enabled flag, so on macOS or Windows the Linux-only packet, signal,
firmware, pwntools, EVM and Wakaru tools reported provider_missing.
Their execution guards refuse those hosts whatever is configured, so
installing a provider there could never make them available.
The server now records, per integration, when the host is outside its
verified host and no service is injected. Discovery checks that before
configuration and reports unsupported_host with the existing
remediation. Injected services and supported hosts behave as before.
Fixes#5730
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* refactor(mcp): share provider host checks with session discovery
Discovery restated each Linux-only integration's host rule in
createServer while its provider kept a separate execution check, so the
two could drift apart again. Each provider now exports its host
predicate, uses it in its execution guard, and composition hands the
same predicate to createServer for both the host restriction and the
enabled flag.
Observed host restrictions now also apply when an embedder supplies its
own availability policy, as provider load failures already do, because
no policy lets the default providers pass their host checks.
The MCP regression moves to tests/process-global and emulates darwin
arm64, linux arm64 and linux x64 through the real createServer, so CI on
Linux x64 exercises the unsupported-host branch. It also covers a
configured policy and the matching execution refusals.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(mcp): keep caller host restrictions beside observed ones
The observed host restriction replaced a configured policy's whole
hostUnsupported map. For an injected service the observed value is
false, so a caller's own restriction was lost and the tool was reported
available. Combine both per integration: either the caller or the
observed host can mark it unsupported.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(pcap): expect unsupported_host for packet tools off Linux
The host-independent PATH case asserted provider_missing on every host.
Off Linux the packet tools are unavailable because of the host, so
discovery now reports unsupported_host there; Linux still reports
provider_missing for an unusable PATH.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(reconstruction): dedupe residual unknown ids with uniqueSorted
Merged obligation candidates called uniqueBy, which is not defined, so
the project did not typecheck. String ids already use uniqueSorted.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* docs(limits): describe how to design admission recovery
Add a short contributor guide for named defaults, settings, recovery
text, and the agent recovery ladder, and mirror that ladder into the
authored skill.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(hopper): pass distribution readers into install-host tests
systemLinuxHopperInstallHost now requires the caller's distribution
reader. The Linux install tests still passed only the environment, so
typecheck failed.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* docs(limits): name the parser doctor shares with startup
Co-authored-by: morluto <morluto@users.noreply.github.com>
* test: align pcap, storage, and regex fixtures with current contracts
Source and portable CI failed on fixtures that still described the
previous pcap client identity, legacy storage fingerprints, and the
old regex deadline sentence.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* docs: link repository experiments outside the built site
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: morluto <morluto@users.noreply.github.com>
Prevent polling readers from observing truncated readiness JSON in compiler fixtures. Retain deterministic cancellation controls and strengthen the real Swift compiler case to run native identity inspection, verify parent/group facts, and check stable identity across reads.
Validated all 32 Darwin tests with the real host toolchain. A delayed write reproduces the old JSON race; delayed atomic publication passes. No CI timeout or retry changes.
morluto