* 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): validate advertised dialect by selected profile
* fix(mcp): preserve schema widening through references
* fix(mcp): refuse zero-padded repeat bounds and record host-specific path patterns
* fix(mcp): keep references valid when presentation widens their targets
* fix(mcp): relax references into negations and conditionals presentation may remove
* 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): 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
* 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
* refactor(domain): one owner for the non-array record predicate
Thirty-four source files each declared isRecord as a non-null object that is
not an array, and six more spelled the same predicate as isObject,
isSchemaRecord or isStorageRecord. They now import isRecord from
src/domain/isRecord.ts, whose overloads keep JsonValue narrowing for the
desktop and environment readers. DshCordisPatch's predicate also excludes
Cordis expressions, so it is renamed isPatchRecord.
architecture-guards refuses a new local copy under any name. The seven files
that still hold one are listed and must leave the list when their copy goes.
* fix(guards): count record predicate copies precisely
* test(hopper): retry after confirmed application exit
* fix: reject stale record predicate exceptions
* refactor(errors): require a reason for every analysis cancellation
Add cancelledBy, the one constructor for AnalysisCancelledError. Its
reason argument is required: an aborted signal, an earlier cancellation,
a reason carried from an inner failure, or an OwnerCancellation when REA
closed the work's owner itself.
Closed adb, apktool, firmware, sqlite, gdb, web-bundle analyzer and
binary-session record owners now report OwnerCancellation.ownerClosed
instead of an absent reason. Firmware's predecessor wait returns its
cancellation as a Result, removing an unreachable reason-less fallback.
An architecture guard rejects new direct constructions, with a counted
per-file baseline that can only shrink.
* refactor(errors): share one aborted-signal predicate with cancelledBy
* refactor(process): return a departing caller's reason from a shared startup wait
* refactor(providers): build decompiler-provider cancellations with a required reason (#10587)
Migrate every AnalysisCancelledError construction under src/ghidra,
src/ida, src/cutter, src/jeb and src/kuna to cancelledBy and lower the
guard baseline from 671 to 572. Closed Kuna target commands report
OwnerCancellation.ownerClosed instead of an absent reason, and Kuna and
JEB close abort their lifetimes with that typed reason.
* refactor(process): build infrastructure cancellations with a required reason
Migrate every AnalysisCancelledError construction in shared process,
provider, artifact, domain, server and CLI code and in the smaller
providers to cancelledBy, lowering the guard baseline from 671 to 566.
Option bags that carried a cause are split so the cancellation branch
cannot receive one.
* fix(errors): read the reason of any signal-shaped cancellation source
* test(ci): cover the counted-baseline comparison behind the architecture guard
* fix(cli): record the caller's --join-path spelling in Apple IPS Evidence
* refactor(cli): project CLI-only options for release exposure and IL2CPP from contracts (#10635)
* refactor(cli): project CLI-only options for release exposure and IL2CPP from contracts
The contract command description gains CLI-only options whose effect the handler owns, plus their dependentRequired entries. audit-release-exposure keeps --fail_on (exit status 3) and inspect-il2cpp-metadata keeps --ghidra-seeds/--seed-base; their hand-written options and input spreads are deleted.
* refactor(cli): project GPU, signal and capture inspectors from their contracts (#10643)
The contract command description gains JSON options decoded when the command runs, admitted with the delivery by the contract input, and bare flags that spell a nullable limit's null. inspect-gpu-code-objects, inspect-signal-decoders, decode-signal-recording, inspect-packet-fields and inspect-packet-capture are derived from their contracts; their hand-written options, superRefine blocks and CLI-only option schemas are deleted.
* feat(setup): give five mcp-add clients one client-table row
Cline VS Code extension, Kiro, Mastra Code, Zed and Amp move from
registration-only catalog entries to rows in the client table. rea mcp add
projects those rows, so doctor, update and uninstall now inspect, refresh
and remove what it wrote, and setup offers each client when detected.
Refs #7181
* refactor(setup): keep FreeBuff's comment on its declaration
The client helpers added for the folded mcp-add clients were inserted
between FreeBuff's profile-directory comment and its declaration. Move
them above it, and drop an import the folded locations no longer use.
* test(setup): refresh an older mcp-add registration through setup
Address review: seed each folded client with an older REA-owned package
runner launcher, then check that doctor reports it stale, update selects
it, setup plans and applies the update, and doctor then reads the
current launcher.
* fix(setup): register Amp in settings.jsonc when that is its settings file (#10717)
Amp reads ~/.config/amp/settings.json or settings.jsonc. REA always
used settings.json, creating a second file beside an existing
settings.jsonc. Use settings.jsonc when it is the only file, and refuse
when both exist because Amp does not document which one it reads.
* feat(setup): manage Cline in VS Code Insiders, VSCodium, Cursor and Windsurf (#10729)
* feat(setup): manage Cline in VS Code Insiders, VSCodium, Cursor and Windsurf
The Cline extension keeps its MCP settings in the global storage of
whichever editor runs it. REA resolved only VS Code's, so a Cline in
another editor was never configured, inspected or cleaned. Add one
client-table row per editor, resolved through the shared Code-OSS user
data rule, and detected where that editor's Cline settings exist.
* test(setup): cover VSCODE_PORTABLE for Cline's editor hosts
Address review: VSCODE_PORTABLE moves only VS Code's own storage, so the other editors keep their settings path; and the older-release note now says which hosts mcp add could register before.
* fix(setup): find a Linux Flatpak Zed's configuration (#10721)
* fix(setup): find a Linux Flatpak Zed's configuration
Zed's config_dir honors FLATPAK_XDG_CONFIG_HOME first, and a Flathub Zed
reads ~/.var/app/dev.zed.Zed/config/zed inside its sandbox. REA always
wrote ~/.config/zed, which that Zed never reads. Honor the override, use
the sandbox folder when it is the only one, and refuse when both a
native and a Flatpak configuration exist.
* fix(setup): treat an XDG_CONFIG_HOME naming the Flatpak folder as one Zed folder
* fix(setup): treat linked native and Flatpak Zed folders as one configuration
* test(setup): expect doctor to report the Amp registration mcp add wrote
* refactor(native): select ELF sections through the keyed owner
* test(native): pin returned exports and versions for repeated version sections
* refactor(nativeaot): read neither of repeated frozen object regions (#10790)
* refactor(nativeaot): read neither of repeated frozen object regions through the keyed owner
* refactor(parsers): retire every first-match section lookup (#10793)
* fix(nativeaot): report an ambiguous frozen object region as partial (#10812)
* refactor(native): select program headers and load commands through the keyed owner (#10827)
* refactor(native): select the GOT base and dependency segments through the keyed owner (#10847)
* refactor(native): select the GOT base and dependency segments through the keyed owner
resolveGotBase selects the SHT_DYNAMIC section and the allocated .got.plt/.got sections with uniqueKeyed, so a repeat leaves the GOT base unknown instead of taking the first match; elfDependencies reads PT_LOAD segments through the same keyed program table as PT_DYNAMIC. A new keyed-first-match guard recognizes a find/findLast or filter-then-[0] whose predicate compares a key field, whatever the table is called; existing matches are baselined. The unwind fixture can declare PT_DYNAMIC headers and repeated .dynamic/.got.plt sections.
* test(native): pin the PT_DYNAMIC path and destructured keyed lookups
The keyed-first-match guard moves to scripts/verify/keyed-first-match.mjs with focused tests, and now recognizes a destructured key in any position and a destructured filter predicate. The single PT_DYNAMIC unwind test declares .dynamic as SHT_PROGBITS so only the program header supplies DT_PLTGOT, and asserts the resolved ranges.
* fix(macho): refuse repeated chained-fixup and dyld-info tables (#10850)
parseMachoLayout kept the last LC_DYLD_CHAINED_FIXUPS range and merged the rebase/bind streams of every LC_DYLD_INFO(_ONLY) command. A repeat now refuses the slice as malformed. NativeMachoInspection's chained import-name reader already declined repeated LC_DYLD_CHAINED_FIXUPS; that rule is now covered by a test.
* fix(verify): count a filter index only on the filter's own result
* ci(windows): validate host-absolute path schemas in the Windows lane
Windows hosts advertise their own absolute-path pattern (#7151). The lane
compiled it through advertisedDialect.test.ts but never validated a Windows
path against it. Run absoluteLocalPathSchema.test.ts there, select the lane
when the pattern's owner changes, and guard both schema checks in the lane.
Refs #10517
* ci(windows): run live MCP schema validity in the Windows lane
toolSchemaValidity.test.ts compared the live advertisement with the
POSIX-generated .cache/mcp-tool-catalog.json, so it could not run where
the Windows lane skips artifact generation. Move the generated-catalog
cases into generatedToolCatalog.test.ts, share the reference walkers
through the advertised-schema fixture, and add the live-server file to
the curated Windows lane and its scope.
* refactor(errors): read cleanup completion from dependencies instead of pushing it to dependents
AnalysisError.cleanupIncomplete becomes a getter over the error's own unresolved state and its bound dependencies. The WeakRef dependents graph and resolveCleanup's push propagation are removed, and subclasses mark their initial state through markCleanupIncomplete instead of overriding a public mutable field.
* fix(errors): settle bound dependencies before binding more
A composite whose dependencies had released but whose status was never read
stayed incomplete when a later dependency was bound.
* refactor(errors): let errors read a lease group's release instead of resolving them (#10759)
AnalysisCleanupLeaseGroup no longer calls resolveCleanup() on its error.
The error holds the group as a release claim and reads cleanupReleased when
asked, so completion follows the lease owner without a push.
* chore(verify): count resolveCleanup and bindCleanupDependencies calls so they only shrink (#10763)
* refactor(errors): let errors read a lease group's release instead of resolving them
AnalysisCleanupLeaseGroup no longer calls resolveCleanup() on its error.
The error holds the group as a release claim and reads cleanupReleased when
asked, so completion follows the lease owner without a push.
* chore(verify): count resolveCleanup and bindCleanupDependencies calls so they only shrink
* refactor(process): let worker cleanup errors read their root epoch's release
OwnedWorkerRuntime's root epoch is now an AnalysisCleanupReleaseClaim. Errors bound to it read its release when asked instead of the epoch pushing resolveCleanup() through weak references, and a runtime without a retained root binds an already-released owner. Removes the OwnedWorkerCleanupFailure structural type and four resolveCleanup() pushes; the counted baseline shrinks accordingly.
* refactor(pcap): let pcap cleanup errors read command and snapshot release (#10792)
Retained pcap commands, retained snapshot failures and snapshot completion descriptors now own their release and errors read it: commands and snapshot failures implement AnalysisCleanupReleaseClaim, and a failed completion-descriptor close is retried through an AnalysisCleanupLease bound to its error. Removes four resolveCleanup() pushes; the counted baseline shrinks accordingly.
* refactor(gdb): parse MI result records with one typed record parser
GdbSessionManager recognized ^done with a prefix regex at seven sites, kept a hand-written span scanner and a second character scanner for -list-thread-groups, and decoded C-string fields with a dynamic RegExp. src/gdb/GdbMiRecord.ts now parses a result record into its token, class and typed results (C-string bytes, tuples, value lists and result lists), and returns a typed malformed-mi-record failure with a reason and offset. The done checks, the thread-group PID inventory and the C-string field reader use it; the stream dispatcher reads only the record header through the same module. An architecture guard keeps MI result-record regexes in the owner.
* perf(gdb): encode each unescaped C-string run once
The MI C-string decoder allocated a Buffer per unescaped character, pushed every byte into a number array, and copied the rest of the line for each hex or octal escape. It now validates unescaped runs in place, encodes each run once, reads escapes with sticky patterns, and joins the chunks once. On a 12.1 MiB groups reply: 887/832 ms elapsed and 818/775 ms CPU before, 76/76 ms elapsed and 131/128 ms CPU after; peak RSS 152-170 MiB both.
* test(gdb): refuse a malformed ^done for a startup setting through a real session
A fake GDB answers -gdb-set auto-load off with ^done,msg="unterminated. The session must not start: the typed MI record parser refuses the record, where main's ^done prefix match accepted it (the case fails against main's GdbSessionManager).
* feat(gdb): record the GDB version from -gdb-version in session Evidence (#10865)
* feat(gdb): record the GDB version from -gdb-version in session Evidence
Startup now asks -gdb-version before the auto-load setting and reads its first console stream record through a new MI stream-record parser (parseGdbMiStreamRecord). Each session's Evidence names that version; an unrecognised banner, malformed record or refused query leaves it null (unknown), as before.
Fake GDB fixtures answer -gdb-version from one owner in gdbTestHelpers: a stdin preamble for scripted fakes and answerFakeGdbStartup for in-process fakes, so scenarios model only their own commands.
* fix(gdb): leave a lossily decoded version unknown and frame fake startup replies by line
* test(gdb): reject unmodeled commands in the lossy version banner fixture
* test(gdb): refuse commands the scripted fake GDBs do not model (#10947)
* test(gdb): refuse commands the scripted fake GDBs do not model
* test(gdb): match the migrated fakes' special commands exactly
The case checks that a one-shot CLI exits after a retained-cleanup
failure and that the operating system then releases the Hopper target
lease to a retry. Since #10412, `cli.run` serves plain commands on the
file's shared CLI process. That process keeps the lease, as the
cleanup-ownership rules require while cleanup is incomplete, and never
exits between commands, so the retry was refused as "already open" and
the final lease check failed on macOS.
Run both attempts with `freshProcess: true`, which the fixture provides
for process-exit facts. The case is macOS-only because target leases
are taken only on the macOS launcher route, and no CI lane ran it, so
add it to the Apple lane and its change scope.
* feat(ida)!: require an explicit protocol for attached IDA registrations
An attached registration (the default mode) silently selected the legacy
ida-pro-mcp 1.4 GUI proxy when `protocol` was omitted. The registration
schema now resolves the upstream protocol once: attached registrations
must name `"legacy"` or `"native"`, headless registrations resolve to the
database supervisor, and consumers read the resolved protocol instead of
inferring it from `mode`.
BREAKING CHANGE: an attached IDA registration without `protocol` is refused
with a configuration error naming the choice. Add `"protocol": "legacy"` to
an ida-pro-mcp 1.4 proxy registration (or `"protocol": "native"` for the
ida-free-mcp plugin). Headless registrations are unchanged.
* test(ida): name the protocol in every attached registration fixture
The setup budget fixture still wrote an attached URL registration without a
protocol, and the IDA registration rejection fixtures relied on the missing
protocol as one of several failures. Each now names its protocol, so setup
reaches its timeout checks and every rejection case fails only for its own
constraint.
* refactor(hopper): read shutdown replies through the bridge reply contract
* test(hopper): retain the document through off-contract primary and fallback shutdown replies
* refactor(hopper): read the bridge health reply through the reply contract table (#10943)
* refactor(ida): decide IDA admission once for doctor and provider readiness
* refactor(ida): make both admission projections handle each refusal kind explicitly
* fix(ida): give doctor the runtime's repair for an invalid IDA registration
Move the fake bridge's successful replies (handshake, inventory, naming, file offset, shutdown) into fakeGhidraReplies.mjs and check each against the production parsers, so contract drift fails one focused test instead of surfacing as an unreadable reply. Fixes three replies that had drifted from the inventory contract (search_procedures, search_strings, resolve_containing_procedure) and the checkpoint naming reply.
* refactor(python): read pyc and PYZ marshal values through one shared reader
* refactor(python): read pyc 64-bit values and header fields through shared readers (#10834)
* refactor(limits): declare the Node timer bound and duration schema once
Fifteen modules re-declared 2_147_483_647 as their own timer maximum and
eleven built the same 1..2^31-1 millisecond schema. They now use
MAX_NODE_TIMER_DELAY_MS and timerDurationMsSchema from
src/domain/analysisResourceLimits.ts, and an architecture guard rejects the
literal outside that owner and five non-timer int32 bounds. The generated
MCP tool catalog is byte-identical.
* refactor(limits): keep the JVM processor-count bound separate from the timer bound
REA_JADX_ACTIVE_PROCESSOR_COUNT is a Java int count, not a timer delay, so it keeps its own named bound and a documented exception in the signed 32-bit guard. Remove the timer-bound and zod imports the duration-schema migration left unused.
* refactor(limits): keep the TShark packet-count bound separate from the timer bound
packet_limit is passed to TShark's -c as a C int packet count, not a timer delay, so it keeps its own named bound and a documented exception in the signed 32-bit guard.
* fix(process): run macOS ps from /bin/ps, not REA's PATH
Owned process cleanup and process sampling ran "ps" through REA's PATH, so
an unusable PATH turned a confirmable cleanup into an incomplete one
(spawn ps ENOENT). macOS ships ps at /bin/ps on the sealed system volume.
Linux keeps the documented procps-on-REA's-PATH contract.
* test(process): stop the ps fixture only after its descendant spawned
* fix(process): run Linux ps from fixed system locations before REA's PATH
Linux process ownership now runs the first executable of /bin/ps, /usr/bin/ps and /run/current-system/sw/bin/ps, searching REA's PATH only when none exists. Admission refusals name the ps that ran in their reason and ps_command/ps_source diagnostics; cleanup failure reasons keep the executed command. Doctor and remediation text follow the same selection. The macOS cleanup test now releases its fixture group even when the descendant wait fails.
* fix(application): keep which ps ran in provider availability
ownedProcessPrerequisiteAvailability copied only the admission and system codes from a Linux ps refusal, so Ghidra, JADX and Apktool status lost ps_command and ps_source. Copy the refusal's prerequisite facts from one key list.
* fix(pcap): report a refused tool launch as an unavailable tool
When a capture read's tool launch was refused with EPERM, pcap reported it
as unreadable tool output (AnalysisProtocolError) instead of an unavailable
tool. Pcap also never told "not executable" apart from "missing".
Add classifyLaunchFailure in src/process as the one owner of which spawn
errnos mean missing (ENOENT, ENOTDIR) and which mean present but not
executable (EACCES, EPERM, ENOEXEC, EISDIR). Pcap, apktool and gdb used three
different errno lists; all three now use it.
* refactor(process): classify owned command outcomes through one owner
Add ProviderOutcome, read once from the supervisor's facts, and classifyProviderOutcome, the one mapping from an outcome to a typed error. Migrate pcap and the shared artifact producer command failure (TShark, sigrok). Add the rea/provider-failure-classifier lint rule with a shrink-only baseline of per-site (cause: unknown) => AnalysisError functions.
* refactor(process): classify Java and adb launch refusals with the shared owner (#10598)
The Ghidra and JADX Java probes and adb's version probe now take
missing-versus-not-executable from classifyLaunchFailure instead of their
own errno checks. ELOOP and ENAMETOOLONG join the owner's missing set, so
adb keeps its existing classification and both Java probes now treat a
non-directory path component as a missing runtime.
* fix(pcap): preserve command launch attribution
* test(pcap): expect an unknown failing stage when a limited capture tool exits unsuccessfully
* refactor(process): classify owned command outcomes through one owner
Add ProviderOutcome, read once from the supervisor's facts, and classifyProviderOutcome, the one mapping from an outcome to a typed error. Migrate pcap and the shared artifact producer command failure (TShark, sigrok). Add the rea/provider-failure-classifier lint rule with a shrink-only baseline of per-site (cause: unknown) => AnalysisError functions.
* test(lint): cover baselined files in the provider-failure classifier rule
* fix(apktool): report the decoder JVM's heap exhaustion as memory, not a target refusal
The supervisor now watches for a managed runtime's out-of-memory report while output streams (JVM java.lang.OutOfMemoryError, V8 'JavaScript heap out of memory'), so a retention budget cannot hide it. ProviderOutcome gains a runtime-exhausted kind and classifyProviderOutcome maps it once. Apktool decode opts in and moves to the shared classifier; Ghidra's single-purpose output token becomes the JVM runtime option and its area matching reuses the shared table.
* fix(process): keep the runtime's out-of-memory report on the memory error it supports
* fix(pcap): attribute a limited capture's exit through the prlimit wrapper
* fix(process): keep a shortened out-of-memory line marked as shortened
* fix(process): identify the refused executable for every launcher shape
Name the executable on errnos Node throws synchronously from spawn
(ENOEXEC, ENOTDIR, ...) and read the Windows native launcher's
executable-launch constraint, so pcap attributes those refusals to the
selected tool or limiter instead of an unknown stage.
* test(pcap): prove synchronous launch naming with ENOTDIR on every POSIX host
* fix(apktool): advise the JVM limit that governs the reported memory area
* fix(ghidra): read the supervisor's parsed OOM area instead of reparsing lines
* fix(sqlite,capture): read V8 heap exhaustion from streamed output, not retained stderr (#10724)
The SQLite worker and the Node capture decoders now declare runtime: "v8", so the supervisor observes V8's heap report as output streams. Both classifiers read the runtime-exhausted outcome instead of matching retained stderr, so a report beyond the retention budget, or one whose process then overflowed its output budget, is still reported as memory exhaustion.
* refactor(evm): classify EVMole worker outcomes through the shared provider classifier (#10716)
* refactor(evm): classify EVMole worker outcomes through the shared provider classifier
The EVMole worker's cancellation, deadline, output budget and CPU/file-size signals now go through classifyProviderOutcome; only the prlimit launch refusal and the reserved EFBIG exit remain in its dialect. The worker is opted into V8 exhaustion scanning so a heap report names its fixed 256 MiB heap. providerOutcomeOf now reads a launch refusal only from a spawn errno, so a private-storage or carrier read errno is no longer reported as a limiter launch failure.
* fix(evm): keep a stream failure's exit status and update Linux worker expectations
* refactor(process): read launch refusals through the launcher identity owner
* refactor(wasm): classify WABT command outcomes through the shared provider classifier (#10727)
* refactor(wasm): classify WABT command outcomes through the shared provider classifier
WABT's cancellation, command deadline, output budget, and inherited-limit signals now go through classifyProviderOutcome; its dialect keeps only the configuration-probe and launch-refusal unavailability and the unclassified-exit output error. A tool ended by SIGXCPU/SIGXFSZ or the memory limit is now a resource constraint instead of an output error, and a refused launch in the producer phase is reported as unavailable. The shared timeout context gains the analysis setting that selected the deadline, and the output-limit context may name its exceeded budget.
* fix(wasm): keep setup guidance when a WABT version probe's output cannot be collected
* refactor(providers): classify raw provider failures through one owner
* refactor(process): build classified cancellations with cancelledBy and the carried caller reason
* chore(lint): count runOwned's close-failure mapper from main in the classifier baseline
* fix(process): carry the runtime's exhaustion report on every memory error that claims it
SQLite and historical-capture memory errors recognised a streamed V8 heap report but dropped it, so a report outside retained output reached callers without its source line or omission flags. One projection, runtimeExhaustionReport, now owns that shape for the shared classifier and both consumers. Also lowers the EVMole and WABT cancellation-constructor baselines.
* fix(process): keep the provider's recovery for a native-thread exhaustion report
The shared classifier replaced the provider-supplied recovery with generic capacity advice whenever the JVM reported it could not create a native thread, dropping Apktool's launcher guidance and flipping retryable to true. The provider's recovery now wins; the capacity advice remains the fallback for providers that supply none.
* refactor(pwntools): classify pwntools failures through the shared provider classifier (#10782)
* refactor(pwntools): classify pwntools failures through the shared provider classifier
pwntools' per-site classifier is replaced by pwntoolsFailureDialect for classifyProviderFailure. Command outcomes go through classifyProviderOutcome, launch refusals through launcherRefusalOf instead of a 'spawn ' syscall prefix check, the input limit through the file dialect's fileSize, and the cleanup report through commandCleanup. The shared read cancellation now carries the reader's partial observation, and resource-signal reports accept provider detail.
* fix(pwntools): report a failed command's output once and expect resource_constraint for overflow
The shared outcome owner already attaches the command's output as capturedOutput, so the pwntools adapter fallback no longer copies the same text into diagnostics.stdout, diagnostics.stderr and diagnostics.captured_output; diagnostics keep the exit facts. Update the overflow expectations the dialect migration missed (layout provider cleanup and over-budget rows, pwntools cleanup) to resource_constraint.
* docs(adr): propose parsing native operation results once at the operation port
* docs(adr): accept parsing native operation facts once at the operation port
* refactor(pcap): read tshark JSON packets through one envelope owner
PcapOutput and the packet-capture reader each decoded tshark -T json and checked the packet envelope. The capture reader let SyntaxError escape and accepted any shape until its field schema ran. src/pcap/TsharkJsonPackets.ts now owns the envelope (malformed JSON, the packet array, each packet's _source.layers) as typed failures, and the pcap byte and expansion budget. Both consumers map those failures to their existing error kinds. An architecture guard keeps _source.layers reads in the owner.
* fix(captures): keep producer output on a refused packet reply
parsePacketOutput threw AnalysisOutputError itself, so the provider's failure mapping returned it without the captured stdout and stderr, both for envelope refusals and for unordered or out-of-limit frames. It now returns a Result with the reason and cause, and the provider builds the AnalysisOutputError with capturedOutput.
* feat(captures): bound packet-capture JSON expansion by an adjustable setting
The capture path parsed TShark's selected-field JSON without the expansion
estimate the pcap reader charges. It now reads through readTsharkJsonPackets
under capture.packet_json_expansion_bytes (REA_CAPTURE_PACKET_JSON_EXPANSION_BYTES).
The estimate charges at most 130 bytes per code unit and the output is one
Node string, so no admissible output reaches 69,793,215,440 bytes; the default
and maximum are 128 GiB, so the budget refuses only when lowered. A refusal is
an adjustable_budget resource constraint naming the setting and keeps the
producer output; a malformed launch seed is refused before the capture is read.
* refactor(browser): send only CDP commands with registered result schemas
* refactor(browser): open Playwright CDP sessions only through the registered-command owner
* fix(browser): keep the restore obligation for an applied override with a malformed reply
A schema-rejected success reply means the browser applied the Emulation
override, so the attached environment owner records the reset obligation
before reporting the protocol error. Only a rejected or unanswered command
leaves no obligation.
* refactor(browser): open verifier-script CDP sessions through the registered-command owner (#10710)
* fix(browser): register the CDP commands JavaScript verifiers send
The verifier scripts send Browser.close, Page.createIsolatedWorld, Page.navigate, Page.reload, Target.activateTarget, Target.closeTarget and Target.createTarget through CdpConnection, which now admits every reply through its command's result schema. Register those schemas, refuse an unregistered run-time name as a typed malformed reply instead of a TypeError, and check script send literals against the registry.
* fix(browser): keep Target.closeTarget's success member for the verifiers
* test(browser): characterize target_changed rules per consumer
* refactor(browser): decide target_changed from one document identity owner
* fix(browser): compare the selected document on every authorization read
Final-capture retries and the pre-capture frame-tree read now observe each
reply through the selected document tracker before scope, and a screenshot
command failure after a proven document change reports target_changed with
the failure as its cause.
* feat(cli): warn after mcp add writes Claude Desktop's configuration
rea mcp add resolves Claude Desktop from the client table, whose row
says a running app can rewrite its configuration from memory, but the
command never told the caller. Print that notice on stderr when it
writes the file. Backup notices now use the client's display name
instead of special-casing Amp.
* fix(cli): warn about Claude Desktop whenever mcp add wrote its file
Address review findings: a registration that failed after publication,
such as cancellation during verification, still left the file written,
so the running-app notice now follows the registration owner's written
fact instead of overall success. The CLI test compares the delivered
notice with its owner instead of freezing the sentence.
Each Kilo project-scope lookup ran about four short git rev-parse/worktree queries, each through runOwnedCommand with a token-owned process sweep that captures and verifies the whole process table. The probes now run through execFileBytes with a stop signal, which leads its own process group and stops it on cancellation or timeout, so the sweep is no longer paid per query. The ownership sweep remains for every other owned command.
* test(domain): bound folder collapse and echo search by work, not elapsed time
Two quadratic-regression tests in src/domain asserted elapsed time. Count
the comparisons and text searches instead, failing at a linear bound so a
pairwise or per-member search stops early instead of running on.
Refs #10513
* test(native): bound hostile ELF section admission by array work, not time
The 65,000-section case asserted it finished within 5 s, which a loaded
runner can miss and a fast one can pass while quadratic. Count array
elements walked through iteration and the element-visiting methods, and
stop at 64 per section; the parser walks about 19. A per-candidate rescan
of the admitted spans now fails at the bound.
* refactor(errors): read cleanup completion from dependencies instead of pushing it to dependents
AnalysisError.cleanupIncomplete becomes a getter over the error's own unresolved state and its bound dependencies. The WeakRef dependents graph and resolveCleanup's push propagation are removed, and subclasses mark their initial state through markCleanupIncomplete instead of overriding a public mutable field.
* fix(errors): settle bound dependencies before binding more
A composite whose dependencies had released but whose status was never read
stayed incomplete when a later dependency was bound.
* refactor(errors): let errors read a lease group's release instead of resolving them (#10759)
AnalysisCleanupLeaseGroup no longer calls resolveCleanup() on its error.
The error holds the group as a release claim and reads cleanupReleased when
asked, so completion follows the lease owner without a push.
* chore(verify): count resolveCleanup and bindCleanupDependencies calls so they only shrink (#10763)
* refactor(errors): let errors read a lease group's release instead of resolving them
AnalysisCleanupLeaseGroup no longer calls resolveCleanup() on its error.
The error holds the group as a release claim and reads cleanupReleased when
asked, so completion follows the lease owner without a push.
* chore(verify): count resolveCleanup and bindCleanupDependencies calls so they only shrink
* test(contracts): derive tool-name oracles from their contract owners
toolEffects.test.ts kept a hand-sorted copy of the destructive tools that
every effects change had to re-record; the effects-to-annotation properties
beside it already check what the copy asserted. adbMcp.test.ts listed the
ADB tools by hand; derive them from ADB_TOOL_CONTRACTS so a new ADB tool is
checked for advertisement without a test edit.
Refs #10511
* test(contracts): keep reviewed destructive tools in a shrink-only baseline
Deleting the hand list left no test failing when a tool lost its destructive
effect. Derive each tool's destructiveHint from its audited effects and
contract, and require every name in the reviewed, sorted baseline to stay
destructive; the derived set may grow, but a tool leaves only by an explicit
baseline edit.
Refs #10511
* fix(pcap): report a refused tool launch as an unavailable tool
When a capture read's tool launch was refused with EPERM, pcap reported it
as unreadable tool output (AnalysisProtocolError) instead of an unavailable
tool. Pcap also never told "not executable" apart from "missing".
Add classifyLaunchFailure in src/process as the one owner of which spawn
errnos mean missing (ENOENT, ENOTDIR) and which mean present but not
executable (EACCES, EPERM, ENOEXEC, EISDIR). Pcap, apktool and gdb used three
different errno lists; all three now use it.
* refactor(process): classify owned command outcomes through one owner
Add ProviderOutcome, read once from the supervisor's facts, and classifyProviderOutcome, the one mapping from an outcome to a typed error. Migrate pcap and the shared artifact producer command failure (TShark, sigrok). Add the rea/provider-failure-classifier lint rule with a shrink-only baseline of per-site (cause: unknown) => AnalysisError functions.
* refactor(process): classify Java and adb launch refusals with the shared owner (#10598)
The Ghidra and JADX Java probes and adb's version probe now take
missing-versus-not-executable from classifyLaunchFailure instead of their
own errno checks. ELOOP and ENAMETOOLONG join the owner's missing set, so
adb keeps its existing classification and both Java probes now treat a
non-directory path component as a missing runtime.
* fix(pcap): preserve command launch attribution
* test(pcap): expect an unknown failing stage when a limited capture tool exits unsuccessfully
* refactor(process): classify owned command outcomes through one owner
Add ProviderOutcome, read once from the supervisor's facts, and classifyProviderOutcome, the one mapping from an outcome to a typed error. Migrate pcap and the shared artifact producer command failure (TShark, sigrok). Add the rea/provider-failure-classifier lint rule with a shrink-only baseline of per-site (cause: unknown) => AnalysisError functions.
* test(lint): cover baselined files in the provider-failure classifier rule
* fix(apktool): report the decoder JVM's heap exhaustion as memory, not a target refusal
The supervisor now watches for a managed runtime's out-of-memory report while output streams (JVM java.lang.OutOfMemoryError, V8 'JavaScript heap out of memory'), so a retention budget cannot hide it. ProviderOutcome gains a runtime-exhausted kind and classifyProviderOutcome maps it once. Apktool decode opts in and moves to the shared classifier; Ghidra's single-purpose output token becomes the JVM runtime option and its area matching reuses the shared table.
* fix(process): keep the runtime's out-of-memory report on the memory error it supports
* fix(pcap): attribute a limited capture's exit through the prlimit wrapper
* fix(process): keep a shortened out-of-memory line marked as shortened
* fix(process): identify the refused executable for every launcher shape
Name the executable on errnos Node throws synchronously from spawn
(ENOEXEC, ENOTDIR, ...) and read the Windows native launcher's
executable-launch constraint, so pcap attributes those refusals to the
selected tool or limiter instead of an unknown stage.
* test(pcap): prove synchronous launch naming with ENOTDIR on every POSIX host
* fix(apktool): advise the JVM limit that governs the reported memory area
* fix(ghidra): read the supervisor's parsed OOM area instead of reparsing lines
* fix(sqlite,capture): read V8 heap exhaustion from streamed output, not retained stderr (#10724)
The SQLite worker and the Node capture decoders now declare runtime: "v8", so the supervisor observes V8's heap report as output streams. Both classifiers read the runtime-exhausted outcome instead of matching retained stderr, so a report beyond the retention budget, or one whose process then overflowed its output budget, is still reported as memory exhaustion.
* refactor(evm): classify EVMole worker outcomes through the shared provider classifier (#10716)
* refactor(evm): classify EVMole worker outcomes through the shared provider classifier
The EVMole worker's cancellation, deadline, output budget and CPU/file-size signals now go through classifyProviderOutcome; only the prlimit launch refusal and the reserved EFBIG exit remain in its dialect. The worker is opted into V8 exhaustion scanning so a heap report names its fixed 256 MiB heap. providerOutcomeOf now reads a launch refusal only from a spawn errno, so a private-storage or carrier read errno is no longer reported as a limiter launch failure.
* fix(evm): keep a stream failure's exit status and update Linux worker expectations
* refactor(process): read launch refusals through the launcher identity owner
* refactor(wasm): classify WABT command outcomes through the shared provider classifier (#10727)
* refactor(wasm): classify WABT command outcomes through the shared provider classifier
WABT's cancellation, command deadline, output budget, and inherited-limit signals now go through classifyProviderOutcome; its dialect keeps only the configuration-probe and launch-refusal unavailability and the unclassified-exit output error. A tool ended by SIGXCPU/SIGXFSZ or the memory limit is now a resource constraint instead of an output error, and a refused launch in the producer phase is reported as unavailable. The shared timeout context gains the analysis setting that selected the deadline, and the output-limit context may name its exceeded budget.
* fix(wasm): keep setup guidance when a WABT version probe's output cannot be collected
* refactor(process): build classified cancellations with cancelledBy and the carried caller reason
* chore(lint): count runOwned's close-failure mapper from main in the classifier baseline
* fix(process): carry the runtime's exhaustion report on every memory error that claims it
SQLite and historical-capture memory errors recognised a streamed V8 heap report but dropped it, so a report outside retained output reached callers without its source line or omission flags. One projection, runtimeExhaustionReport, now owns that shape for the shared classifier and both consumers. Also lowers the EVMole and WABT cancellation-constructor baselines.
* fix(process): keep the provider's recovery for a native-thread exhaustion report
The shared classifier replaced the provider-supplied recovery with generic capacity advice whenever the JVM reported it could not create a native thread, dropping Apktool's launcher guidance and flipping retryable to true. The provider's recovery now wins; the capacity advice remains the fallback for providers that supply none.
Input-contract and timeout failures already carry structured recovery
(recovery.kind, operation, timeout_ms) and retryable. Tests pinned the
remediation and message sentences beside those fields, and banned phrases
from earlier wording. Assert the structured facts instead; keep caller data
and protocol values such as operation names and tools/list.
Refs #10508
* refactor(browser): send only CDP commands with registered result schemas
* refactor(browser): open Playwright CDP sessions only through the registered-command owner
* fix(browser): keep the restore obligation for an applied override with a malformed reply
A schema-rejected success reply means the browser applied the Emulation
override, so the attached environment owner records the reset obligation
before reporting the protocol error. Only a rejected or unanswered command
leaves no obligation.
* refactor(browser): open verifier-script CDP sessions through the registered-command owner (#10710)
* fix(browser): register the CDP commands JavaScript verifiers send
The verifier scripts send Browser.close, Page.createIsolatedWorld, Page.navigate, Page.reload, Target.activateTarget, Target.closeTarget and Target.createTarget through CdpConnection, which now admits every reply through its command's result schema. Register those schemas, refuse an unregistered run-time name as a typed malformed reply instead of a TypeError, and check script send literals against the registry.
* fix(browser): keep Target.closeTarget's success member for the verifiers
* refactor(jadx): read the bridge identity from the bridge source it launches
* fix(jadx): read the bridge identity before launching the bridge
* fix(jadx): read the bridge identity from the outer class's own fields
* fix(jadx): read the bridge identity only from the ReaJadxBridge class body
* fix(jadx): return a malformed packaged bridge source as a read result
JadxRelease.ts is loaded by fixture scripts without a build, so it cannot import src/domain modules (the ParseFailure import broke the android-metadata job and the spawned JADX fixture). The identity read now returns a Result-shaped value; JadxProvider refuses a malformed packaged source as AnalysisCapabilityUnavailableError before acquiring resources, and tests and the fixture unwrap it.
* fix(test): serve fake Hopper segments in the bridge's reply contract
#10662 validates every Hopper bridge reply. The fake Hopper behind the darwin packaged-CLI baseline still served list_segments without permissions, provenance and sections, so 'rea analyze --provider hopper' failed as unreadable output on every PR that ran that lane. The fixture's replies move to fakeHopperReplies.mjs, list_segments takes the real bridge's shape, and a test checks every served reply against the contract table.
* test(hopper): derive the fake reply method list from one keyed factory object
The contract test iterated a hand-kept method list beside the reply switch, so
a reply added without updating the list would go unchecked. Replies are now
factories in one keyed object; fakeHopperReply looks them up by key and
FAKE_HOPPER_REPLY_METHODS is the object's keys, so every served reply is
validated against its contract.
* test(hopper): serve containing-procedure and reference replies from the keyed factories
#10553 moved the last caller of isUnresolvedPathErrorCode onto the shared
launch-failure classifier, leaving the export unused. knip in the deep
static checks lane fails on it, keeping main red. The errno set stays: the
selected-regular-file predicate still uses it.
* fix(artifacts): report ZIP bytes the central directory does not account for
* fix(artifacts): validate ZIP accounting and join traversal
* fix(artifacts): match an unsigned ZIP data descriptor whose CRC equals the signature
The optional descriptor signature only adds a candidate layout; both layouts are checked against the recorded CRC and sizes.
* fix(artifacts): keep ZIP close observable and account AES data descriptors
Close joins running traversal steps instead of returning the traversal, so a caller's next step after close still rejects as unavailable. Member accounting reads through BinaryReader, and a data descriptor whose directory reports no CRC (AE-2) is matched on its sizes.
* fix(pcap): report a refused tool launch as an unavailable tool
When a capture read's tool launch was refused with EPERM, pcap reported it
as unreadable tool output (AnalysisProtocolError) instead of an unavailable
tool. Pcap also never told "not executable" apart from "missing".
Add classifyLaunchFailure in src/process as the one owner of which spawn
errnos mean missing (ENOENT, ENOTDIR) and which mean present but not
executable (EACCES, EPERM, ENOEXEC, EISDIR). Pcap, apktool and gdb used three
different errno lists; all three now use it.
* refactor(process): classify owned command outcomes through one owner
Add ProviderOutcome, read once from the supervisor's facts, and classifyProviderOutcome, the one mapping from an outcome to a typed error. Migrate pcap and the shared artifact producer command failure (TShark, sigrok). Add the rea/provider-failure-classifier lint rule with a shrink-only baseline of per-site (cause: unknown) => AnalysisError functions.
* refactor(process): classify Java and adb launch refusals with the shared owner (#10598)
The Ghidra and JADX Java probes and adb's version probe now take
missing-versus-not-executable from classifyLaunchFailure instead of their
own errno checks. ELOOP and ENAMETOOLONG join the owner's missing set, so
adb keeps its existing classification and both Java probes now treat a
non-directory path component as a missing runtime.
* fix(pcap): preserve command launch attribution
* test(pcap): expect an unknown failing stage when a limited capture tool exits unsuccessfully
* refactor(process): classify owned command outcomes through one owner
Add ProviderOutcome, read once from the supervisor's facts, and classifyProviderOutcome, the one mapping from an outcome to a typed error. Migrate pcap and the shared artifact producer command failure (TShark, sigrok). Add the rea/provider-failure-classifier lint rule with a shrink-only baseline of per-site (cause: unknown) => AnalysisError functions.
* test(lint): cover baselined files in the provider-failure classifier rule
* fix(pcap): attribute a limited capture's exit through the prlimit wrapper
* fix(process): identify the refused executable for every launcher shape
Name the executable on errnos Node throws synchronously from spawn
(ENOEXEC, ENOTDIR, ...) and read the Windows native launcher's
executable-launch constraint, so pcap attributes those refusals to the
selected tool or limiter instead of an unknown stage.
* test(pcap): prove synchronous launch naming with ENOTDIR on every POSIX host
* refactor(process): build classified cancellations with cancelledBy and the carried caller reason
* chore(lint): count runOwned's close-failure mapper from main in the classifier baseline
* refactor(parsers): decode LEB128 through one owner
src/domain/leb128.ts owns LEB128 decoding: width, sign and whether padding may extend an encoding are stated per format, and overflow never wraps. ELF attributes, ELF unwind tables, Mach-O bind/rebase operands, Go build-info, protobuf captures and WABT custom names use it instead of six local loops.
* fix(macho): charge rebase operand bytes against the work budget while decoding
* refactor(process): own deadlines as one monotonic type with a finalize budget
Deadline (src/domain/deadline) holds a monotonic work budget and a separate finalize budget that starts when finalization begins. ProviderStartupDeadline is built on it and exposes finalize(), so a failed Ghidra startup drains launcher output under its own budget and the caller's signal instead of the expired startup signal. The six wall-clock deadline sums in src move onto Deadline; a counted guard keeps the one page-clock conversion and rejects new copies.
* refactor(process): share one finalize window across repeated beginFinalize calls
* fix(cutter)!: report undeclared output truncation and document generation as unknown
cutter_command filled a missing output_truncated with false and a missing
document_generation with the caller's expected_generation, so a reply that
did not declare them read as "nothing omitted" and "document unchanged".
Both are now null when the plugin did not declare them, with an Evidence
limitation for each. A command that never started still reports
output_truncated false, because there was no output to omit. The packaged
plugin now declares output_truncated on every command reply and the current
generation on its not-started replies.
BREAKING CHANGE: cutter_command's result fields output_truncated and
document_generation are now nullable. A caller that read document_generation
as a number must handle null by listing Cutter sessions again before choosing
an expected_generation. A caller that read output_truncated as a boolean must
treat null as "the output may be incomplete".
* test(cutter): cover the unknown-truncation limitation
* feat(cutter): identify the running plugin's source and refuse a modified one
The Cutter plugin now publishes the SHA-256 of its own file on its session
descriptor (an additive field). Discovery reports each session's
bridge_source as packaged, modified, or unknown (an older plugin without the
field). cutter_command refuses a modified plugin before sending anything and
names the packaged plugin to copy into Cutter's plugins/python directory.
* fix(cutter): report an unreadable packaged plugin instead of hiding the live session
* test(cutter): give the discovery fixture sessions their bridge source
* fix(artifacts): honor record-and-continue for lipo and FAT slice disagreement
* fix(artifacts): keep recovered slice names distinct, lipo field evidence, and unknown coverage without a FAT parse
* fix(artifacts): keep lipo's unnormalized CPU declarations when FAT ranges are authoritative
* fix(process): stop execFileOutput's process group, not only its root
* test(process): cover group escalation and abort the launcher on failed readiness
The session map, start reservations, the shared in-flight stop, retained cleanup failures, cleanup retry and close-all move into OwnedDebuggerSessionRegistry. Each debugger keeps its protocol, detach and stop-reason policy and supplies only its seal and stop hooks.
* fix(plist): record dictionary keys declared more than once before decoding collapses them
* fix(plist): bound repeated key names in the limitation and mark only real omissions
* fix(il2cpp): report the exceeded native-link maximum in the public refusal
* fix(il2cpp): report each native-link budget's maximum in the unit its producer counts
* refactor(elf): report pointer relocation failures through ParseFailure (#10748)