The tool description string conflicted only because this branch adds its write-plan sentence beside the coverage text; both are kept.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A response summary's envelope member longer than the description sample
was cut and printed raw, so a newline or separator in it could read as
more of the tool description. The shown prefix is now written as a JSON
string, as shorter names are, before the truncation note.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The tool description string conflicted only where main added Frida session status and this branch describes document reads; both are kept.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Resolve relative pointers using field-owned ABI encodings. Keep null, raw, rebased, imported and unknown targets distinct, and retain imported conformance rows with exact Mach-O binding provenance. Decode legacy rebase membership with source-derived memory and work bounds, preserving validated prefix facts when streams are incomplete.
Validation: check:ci, docs:check and build passed; 158 source tests and 20 schema/acceptance tests passed. Ten source-built Apple dispatch variants and four issue-specific CLI/MCP workflows passed without executing targets. Paired cold/warm CLI/MCP cost observations retained locally; model tokens and maximum-sized input runtime were not measured.
Closes#5683Closes#5685
BREAKING CHANGE: Swift conformance type_name and protocol_name can be null when names cannot be resolved; consumers must use typed type_binding and protocol_binding evidence. MachoImportBinding replaces the former Objective-C-specific binding type and schema.
Compare explicit bundle-resource CLI and MCP results at the positive empty-inventory offset. Preserve a separate binary-bound MCP assertion with the exact fixture executable digest and complete metadata payload.
Validation: reproduced the stale root-versus-binary parity failure on current main, then passed the macOS public CLI/MCP acceptance case across all five NIB fixtures. Focused lint, formatting and diff checks passed. No target code was executed.
Preserve load-command paths as evidence while passing a separate root-relative lookup path to the bounded tree resolver. Apply the same containment rule to absolute install names and rpath bases, including caller-selected root symlink spellings.
Validation: 140 source tests, full check:ci, docs and runtime build passed. Compiler-built literal, rpath and relative fixtures passed CLI/stdio MCP parity, preserved both dependency edges and left all bytes unchanged. External system paths stayed outside-target; relative control was unchanged. Fixtures were not executed, and actual dyld execution was not qualified.
Closes#5637
A served write of a missing entity also looked at the write tool's own
recordings, so a delete that recorded 200 null for a successful
deletion answered a missing target with success. A 2xx null is now an
absence only when the call is a read, as the gate takes it from item
reads; a write still answers with a recorded 400, 404 or 410, or is
refused.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Match memory failures to the exact source identity instead of graph ordering. Preserve CLI parse, file-evidence, partial-result, later-source and unchanged-input checks across valid memory-pressure outcomes. Use a controlled parent heap envelope with a real owned analyzer to prove module retention and graph rejection independently of host GC timing.
Validation: full check:ci passed; 17 focused tests passed, with the final 4 changed tests rerun successfully. Independent test-strength and isolation review passed. The earlier hosted failure exposed missing graph-only fields; its exact alternate memory stage was not captured.
Own CDP frame applicability and identity at the parsed event boundary. Preserve malformed or incomplete main-frame facts, recognize replacement roots, and carry the same document identity through captures, screenshots, timelines, and execution windows.
Validation: full check:ci and docs:check; 763 browser and dispatch tests passed, 6 skipped. Independent source and fixture-lifecycle review passed. Local real-Chrome behavior was not exercised.
Closes#5665
Four places where the gate and serving still disagreed:
- the gate applied a document write to the held document whatever path
it wrote, while serving refuses any other path; a write at another
path is now unmodelled;
- a create took the first held collection its parent value matched,
so with acme1 and %61cme1 both held a create for acme1 could land in
the encoded one; the verbatim spelling now wins, as for item calls;
- the gate accepts a 2xx JSON null as a recorded absence, but serving
answered a missing entity only with a recorded 400, 404 or 410; the
null answer is now served too;
- data/add measured only the document it was given, so one that omits
types and the state Evidence id could load under 32 MiB and grow past
it as data/get writes it back; such a document is now refused.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Count each admitted unique class or metaclass object at its parsing owner. Preserve emitted metadata, independent class-list scan bounds, partial failures, and truncation.
Closes#5706
Count examined and decoded vtable slots in the same population, preserving all returned slots and partial descriptor failures. Document the coverage unit and retain malformed-sibling and unresolved-slot checks.
Closes#5681
Keep raw physical read diagnostics alongside explicit operation provenance on the same cleanup owner. Preserve provenance after successful close retry for opted-in readers, restore legacy error projections, and prevent borrowed cleanup-shaped failures from authorizing foreign resource retries.
Refs #5686
Reuse the exact selected advertised descriptor for SDK output validation instead of compiling the entire cached catalog for every fresh parity-test client. Retain actual catalog discovery, schema validation, and all result assertions under the existing worker heap limit.
Parse only documented compound otool field labels so valid segment and section names are retained at the owning boundary. Verified real Apple clang fixtures through compiled CLI and stdio MCP without executing the targets.
Closes#5680
Tag physical read outcomes at the owning boundary and preserve opaque caller cancellation reasons through JSON pipelines and reference imports. Share named bounded-read options across consumers and retain failed cleanup ownership for retry.
Closes#5686
Join provider class observations only when selected-image identity and exact import bindings establish equivalence. Preserve distinct observations and unsupported coverage.
Refs #5607
Preserve exact selected-image identity and descriptor ownership across preparation, final cleanup, and retries. Keep physical failures and failed cleanup leases observable through provider and launcher boundaries.
Closes#5688
Keying checked lines by the raw path skipped a read serving still
answers: with the list recorded at /api/orgs/%61cme1/books, a later
/api/orgs/acme1/books read names the same line through replay's decoded
fallback, so it went unchecked while serving listed the encoded
collection, and a delete it contradicted stayed served.
The gate now ranks a view's checked lines for a read as replay ranks
them for its call: every parameter the read's value verbatim or once
percent-decoded and the same query, then the most verbatim parameters,
then recorded order. A list is compared from the collection of the line
it answers from, as serving lists it, in the check, the taint checks and
the insert retry.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Checked lines now keep the path as spelled and canonicalise only the
query: replay prefers a verbatim path spelling, so two spellings of
one parent (acme1 and %61cme1 under one templated tool) stay separate
lines and each read is compared with its own line's envelope.
- A slug rule a later call shows is checked against the values earlier
calls recorded (the slug of the source they sent against the slug
they showed), not only their member names, so a call that resent an
unchanged name no longer hides the tool; an increment still conflicts
with an earlier call that showed an integer field unchanged.
The view fixture takes template rules so the gate can be tested over
templated parents.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Five places where the gate approved a write serving would answer
differently:
- list and document reads are matched to checked lines as serving
matches calls (path and query values decoded, the query in any
order), so a reordered query no longer hides a contradiction;
- an increment or slug applies to every call, so a plan is
unexplained-change when one call changes such a field and another
leaves it as it was, in either order;
- an item read confirming a create no longer stops a later list from
choosing another insert candidate; only a list check fixes the rule;
- the gate runs serving's document body check (accepted members, JSON
types from the held and initial document), so a recorded write that
serving would refuse leaves the plan unserved;
- a served create decodes its path values from the collection the call
names, as the gate decoded the recorded path, whichever spelling the
call used.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Keep malformed or missing CDP opcodes and payload facts unknown, retain observed zero and empty payloads, and avoid false capture differences while preserving independently known facts.
Validation: 59 source tests; compiled public CLI and stdio MCP fake-CDP acceptance; check:fast; docs:check; changed-file formatting. Real Chromium malformed-event behavior was not tested.
BREAKING CHANGE: WebSocket event opcode is nullable and numeric values are restricted to the four-bit protocol range.
Fixes#5666
* test(cli): keep IPC trace input under the Linux spawn limit
The listener-callsite CLI builds one trace argument from the full
application Evidence document. On 4 KiB-page Linux that string exceeds
MAX_ARG_STRLEN, so spawn throws E2BIG before the command starts. Keep the
document in a file, require the spawned argv to fit the single-argument
budget, and lock that budget through the test process registry.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* test(process): size the spawn ceiling from ARG_MAX and MAX_ARG_STRLEN
A payload at the single-string limit still fails execve with E2BIG when
the stack shrinks ARG_MAX below the rest of the image. Accept only the
bytes that fit both budgets, and skip that success spawn when the fixed
argv and environment already fill ARG_MAX.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* style(test): format files rejected by oxfmt in static checks
check:ci:fast failed format:check on the CLI fixture import and, after
merging main, on the admission test's new command-abort case.
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 coverage branch writes captured field names in response summaries as
JSON strings, which changes the replay card's tool descriptions. The pin
moves to the 12,818-byte card; removing the quotes reproduces the old
12,618-byte digest exactly, so nothing else in the card changed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add a read-only app resource session shared by CLI and MCP. Preserve resource-root provenance without inventing executable identity, reject binary-only snapshots and operations, and retain failed cleanup owners across switches and retries.
Keep filesystem admission failures typed and preserve cancellation ordering from observed cleanup facts. Document resource routing in the packaged investigation skill.
BREAKING CHANGE: resource-only observations may have a null Evidence subject and null executable target digests. Clients must use resource_root and reported per-resource provenance instead.
Refs #5589
* fix(observation): keep incomplete cancellation observations
Service projection treated a retained observation that failed the completed
schema as a malformed completed result. Cancellation became unreadable_output
and the partial was dropped. Incomplete observations now stay on the owner
error, including the exact abort reason.
* fix(observation): drop non-JSON incomplete observations
Keep a retained partial only when projectAnalysisError can serialize it.
Clear a non-JSON partial and record an AnalysisOutputError as a secondary
failure so cancellation and timeout stay the owner error.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(observation): validate incomplete partials without copying
jsonValueSchema.safeParse cloned a JSON-safe partial that projection
copies again. Use jsonValueValidationIssue so the gate only checks.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* test(ida): expect non-retryable unresolved procedure lookups
Main stops treating invalid_request as retryable. An unchanged missing
procedure stays invalid_request and is not marked retryable.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* test(observation): share incomplete-observation assertions
The new non-JSON cases repeated the same owner-error checks and pushed
duplication over the 3.5% limit. One expectation helper keeps those
checks without a second copy.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* test(observation): keep the assertion helper out of the package
The helper imports Vitest. Rename it to a fixture so the production
build excludes it and dist does not import a dev dependency.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(observation): clear non-JSON partials on cleanup primaries
Projection can return the cleanup primary after retry. Clearing only
the wrapper left that primary's partial in place, so projectAnalysisError
threw. The gate now clears every non-JSON partial in the chain.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(observation): freeze retained incomplete observations
A JSON partial stayed a live provider object, so a later mutation could
make projection throw. The gate now keeps one frozen snapshot and still
drops a non-JSON partial without copying it.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* refactor(observation): share web runtime partial retention
Observe and inspect used the same promote-or-keep path. One helper
owns that sequence so the two operations cannot drift.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(observation): snapshot a shared partial once
A cleanup wrapper and its primary can hold the same partial object.
The gate now reuses the first frozen snapshot for that object.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(observation): parse a kept partial once
jsonValueSchema.safeParse validates and copies only a JSON partial.
The gate freezes that copy and no longer walks the same value first.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* test(observation): share repeated observation fixtures
The incomplete-observation tests repeated the same provider setup and
an empty Electron coverage tail, which pushed duplication over 3.5%.
One coverage fixture and shared assertions bring the tree back under
the threshold.
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 state-writes spec now says what the review fixes changed: document
writes are served only at the path the state holds the document at and
type-check their members; a value data/add would refuse, or growth past
the 32 MiB intake limit, is refused; only a 2xx write is modelled; the
gate compares the full response serving would give; an update's
transformed member stores the value its call recorded; and an unfiltered
list answers from a line of the collection its path arguments name.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(errors): preserve caller abort reasons at admission
Keep the opaque caller reason on reverse, pcap, and SQLite cancellation,
and leave a recognized physical digest or command failure in place when a
late abort is also observed.
Fixes#5670
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(reverse): keep a foreign digest ABORT_ERR as an adapter failure
Treat ABORT_ERR as cancellation on the command boundary, where that code
already meant an aborted provider process. A digest or stat rejection with
the same code stays a physical adapter failure unless the caller signal
is aborted. Cover aborted command EACCES, ETIMEDOUT, and ABORT_ERR with
and without captured output.
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>
Served writes now refuse, before changing the state:
- a document write with members or JSON types its plan did not accept
(the same body check entity writes use), or at a path other than the
one the state holds the document at;
- a body or resulting value data/add would refuse (a number inexact as
data/get writes it, a __proto__ member, nesting past the limit), so
data/get output always loads back, increments included;
- a create, update or document write that would take the state past the
32 MiB data/add intake limit (resource_constraint), with the state's
serialized size kept incrementally rather than reserialized.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The write gate confirmed effects the served environment would not show:
- a list check let a member the state no longer holds pass, so a
filtered list still showing a deleted entity confirmed the delete; it
now compares the whole list serving would return;
- item and list checks compared only the entity, so a wrapper member
that changed after the write (a revision, say) went unnoticed while
serving kept the base recording's; reads are now compared as the full
response rendered into the item view's or the line's base envelope;
- a later update's constant rule kept no value, so serving stored the
first observation's; each constant rule now carries the value the
call that showed it recorded, and the gate and serving read it alike;
- a redirected or unanswered write could become the example or a
modelled effect; only a 2xx write is modelled, others taint as
unmodelled writes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An unfiltered list call chose the first paging-only line of the tool
whatever its path arguments named, so with /api/orgs/1/books?limit=50
and /api/orgs/2/books?limit=50 behind one tool, a call for org 2 got
org 1's books. The line must now be of the collection the call's path
arguments name (matched as item calls match, chosen as replay chooses);
a parent with no such line answers from its recording.
The serving guide and the evidence-workflow reference still said state
mode leaves write tools out and that no write has an effect; they now
describe confirmed write tools being served and what a served write
does not model.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Reuse the shared object-member owner for literal spreads, retain known property names under unresolved membership, and distinguish complete, partial, unknown, and absent projections. Keep historical coverage omission unchanged and preserve own keys across prototype setters.
Refs #5075
* test(acceptance): split competing loader ownership across files
The loader matrix was one 578-second file, which set the floor for any
shard that received it. Three files keep every CLI and MCP case and let
workers overlap them.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* perf(ci): run eight fast built shards under a ten-minute budget
Charge each Vitest project by its resolved worker width so serial
process-boundary files spread as tail time. Fast feedback uses eight
built shards. The npm download cache stays; node_modules is too large
to ship as a shared artifact.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(ci): keep each scheduled file on one worker lane
Dividing a file by maxWorkers dropped its own wall-time floor and could
pack a parallel file onto a serial tail. Place each file whole on a lane,
overlap peers only inside a sequence group, and add later groups after.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* docs(testing): separate the shard budget from workflow time
The ten-minute figure is the built shard job. Classification, build, and
the portable-test merge are measured separately on run 39451548945.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* docs(testing): match fast shard count to lane scheduling
Main still described five shards and half-duration weights. Fast
qualification uses eight shards, and each file keeps its full duration
on one worker.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(ci): seed split loader durations from pre-split history
Timings published before the file split still name one 578-second path.
Until a new file appears in that artifact, share the recorded duration
by the pre-split case counts.
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>
Keep POSIX symlink-parent traversal at the shared JSON input boundary and forward the selected browser executable without a second lexical normalization. Verify Electron, browser, and Evidence-file consumers through registered commands and the built CLI.
Refs #5103
* ci: keep npm download caches from Turbo snapshot growth
Per-commit static and portable Turbo snapshots had filled the 10 GB Actions
cache, which is what makes warm npm ci cheap. Trusted static runs now keep two
generations and one download-cache writer per host.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(ci): declare cache retention types for strict checks
Typecheck imports cache-retention.mjs and needs a sibling declaration.
The generation count stays module-private so knip does not report it.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(ci): keep compatible caches and qualification read-only
Npm dedupe deletes a record only when another has the same key, version,
and ref. Cache deletion moves to a follow-up workflow so reusable
qualification can keep its read-only token.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(ci): retain caches after failed default-branch runs
Static and build save Turbo snapshots before later lanes finish. Retention
now follows a failed, timed-out, or cancelled trusted run, not only a green one.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* test(cli): pass application trace input as a file
The CLI already reads workflow JSON from a path. Putting the full
application Evidence document on argv exceeds the runner argument limit.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(ci): retain deep and Knip snapshots without deleting a repeated id
Fast and deep static checks publish rea-static-deep-v1 and knip-meta-v1
entries per commit. Retention prunes those families with the same two-generation
rule. A cache id listed twice is one record; inconsistent repeats abort
before any delete.
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>
