Serve write tools whose later reads confirm the effect, and keep main's
state-file ownership and part-wise reply measurement.
Squashed rebase of #5600 onto the rebased #5352 tip.
* fix(mcp): name response-byte and startup ceilings in recovery
Name REA_MCP_MAX_RESPONSE_BYTES, its effective value, the 10485760 byte
floor, and process restart when the stdio response budget is hit, and
state that the setting does not raise Node's string ceiling. Name fixed
MCP startup phase budgets when those deadlines fire, and keep Codex's
30 second startup figure as client guidance.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(mcp): distinguish string ceilings and keep dossier recovery
Name REA's reserved string limit separately from Node's ceiling, restore
detail:summary discovery for an oversized complete evidence bundle, and
apply the doctor deadline sentence when binary_session itself times out.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(mcp): keep retained SDK errors inside the text budget
Quoted SDK validation diagnostics expand when recovery JSON is escaped
into MCP text. Shorten only that preview so the retained-evidence
reference still arrives as parseable error content.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(mcp): keep export advice when a summary exceeds the response budget
An oversized get_evidence_bundle summary no longer repeats detail: summary.
export_evidence_bundle still writes the retained bundle outside one MCP frame,
including when a larger byte budget cannot raise the string ceiling.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* style(mcp): format the SDK diagnostic preview return
oxfmt rejected the fitted-preview ternary in EvidenceMcpServer.ts, which failed Static checks and the CI required gate.
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>
* fix(browser): name web-bundle budgets and recover web deadlines
Name the existing heap, per-script, and total web-bundle phases, and reject a per-script deadline above the total. Promote capture, CDP command, source-map decode, and source-map fetch deadlines at today's defaults. Name DevTools timeoutMs, and state that discovery, cleanup grace, byte budgets, and Electron hook retention have no runtime override.
Refs #5910
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(browser): apply selected web deadlines to the owning phase
Keep HTTP and browser-socket discovery on the fixed discovery deadline so a
short command_timeout_ms starts at the handshake. Apply scenario
command_timeout_ms to launch startup and the primary connect session, and
keep service-worker restoration on the fixed 5000 ms grace. Omit
command_timeout_ms from recovery text when the operation does not accept it.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(browser): retain scenario commands that finish after their deadline
A timed-out scenario command stays owned so cleanup can close a late
browser or context and restore a mutation that lands after the deadline.
Child storage replies use that deadline, scenario recovery names
command_timeout_ms in the capture JSON, and the session test connection
implements beginCleanupGrace.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(browser): keep child command timeouts as the public failure
A nested command such as Page.enable that is delivered but never answered
was stored as protocol_error, hiding the command name, selected deadline,
and recovery. Store the typed timeout and assert that public error.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(browser): keep runtime cleanup on the fixed grace
Switch runtime restoration to the cleanup grace before the first send so a
short command_timeout_ms cannot fail an otherwise successful observation.
Stop advertising command_timeout_ms on list_browser_targets, which never
bounds a post-discovery command.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* test(browser): prove a short command deadline cannot fail runtime cleanup
Drive restoration through the runtime fixture. A cleanup command that
outlives command_timeout_ms still succeeds, and a stalled restoration
names the fixed 5000 ms grace.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(browser): name the effective source-map decode deadline
codecBudgetLimitation reported a fixed 20-second deadline after a
caller selected decode_timeout_ms. Result and Evidence limitations
now name that effective millisecond deadline.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(browser): separate handshake and target-attach deadlines
A browser-level Target.attachToTarget no longer shares the WebSocket
handshake budget. Blocked service-worker attachment keeps discovery on
its fixed deadline and applies command_timeout_ms to the later handshake
and attach.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(browser): name the web-bundle resource each runtime exhausts
Bun ignores the V8 old-space flag. A Windows Job reports the process
commit limit, and other Bun hosts no longer suggest raising max_heap_mb.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(browser): cite the browser-socket discovery message budget
WebSocket discovery enforces the 64 MiB protocol budget. Timeout
recovery no longer advertises the 8 MiB HTTP document budget.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(browser): bound the blocked service-worker check before discovery
The first registration check has its own fixed 10s deadline. Discovery,
handshake, and Target.attachToTarget stay on their separate deadlines.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(browser): name the operand that stopped web-bundle admission
Host heap pressure no longer claims the fixed result budget is exhausted.
A Windows Bun allocation before the commit limit still names that limit.
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>
* feat(environment): extract the initial state a capture's reads observed
Add the environment-state analysis view. From the parents the environment
manifest accepts, it extracts the data the capture's reads observed before
the first related write: entity types with an identity field (proved by a
path parameter or inferred from an identity name), entities with their
recorded values and provenance, documents for written non-entity paths, and
a render plan per read tool that is kept only when it reproduces the tool's
latest initial recordings from the state; every other read tool is a
recorded view with the reason. The Evidence is always inferred.
Response-body decoding moves to a domain owner shared with replay, exact
JSON parsing moves to the owner of the exact-number policy, the manifest and
state views share parent admission and record grouping, and
projectAnalysisView takes its accumulators as one object.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): bound state at redacted writes; infer identity only from items
A scenario write whose URL was redacted has no request line, so it was
dropped and reads after it counted as initial state. Such a successful
write now joins the timeline with an empty path, whose root relates to
every type, so it ends every type's state; a limitation counts them.
A list whose recordings were all empty passed the "every listed item
carries it" check vacuously and published an identity field nothing
showed; it now needs at least one item, else the view stays recorded
as no-identity.
A value that differed between two request lines of one item tool, such
as a query variant, was called volatile, and the gate then substituted
each recording's own value; only repeats of one request line now show
volatility, so differing lines go through fields-differ.
List membership is compared only while a list can still be the whole
collection, and merged entities are indexed by type and collection once,
so paged captures no longer cost pages times entities (8,000 pages of
40,000 entities: 5.7 s to 0.23 s).
Docs and the tool contract say scenario state keeps the capture's
redaction markers and how redacted writes bound it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): keep first sightings, one list pointer and linear field unions
Entities took the order of the first plan observed, and item plans run
before list plans, so a complete list read before its detail reads came
out reordered and lost "all" membership; each entity now keeps its
earliest recorded sighting.
A list plan reused the first recording's pointer for later recordings,
so a later response that was itself an entity wrapping a list was read
as that list; every recording must now locate the same list.
The field union scanned its ordered list for every key, quadratic in an
object's width; a set now answers membership.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): keep post-write values and partial lists out of the initial state
A value that changed between repeats of one request line was named
volatile whenever no write under the type's own root lay between, so a
write under another root (for example POST /api/dashboards/db before a
second GET /api/search) left its effect in the initial state. Any
recorded write between the two observations now leaves every tool of the
type recorded with reason fields-differ instead.
A list was published with membership "all" even when a total or count
envelope member exceeded the number it listed. Such a counter now makes
the membership "recorded".
Also drop a stale doc comment on ObservedEntity.order.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): scope item-shaped and deleted written paths out of documents
State documents are held only for list-shaped paths a POST, PUT or PATCH
tool writes. Reads of item-shaped written paths and of paths that are only
deleted stay recorded views; the guide now says so and the spec defers them
to a later increment of #4933.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): count writes in flight when a value changes between reads
The volatility check compared write start positions with the two reads,
so a write already in flight when the earlier read started was missed and
its effect was named volatile. A write now counts when its interval
overlaps the span from the earlier read's start to the later read's end.
The check also ran once per changed field and scanned every write. It now
runs once per pair of observations, answered by a binary search over the
writes in timeline order and the latest end among each prefix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): catch cross-tool changes across writes; read only item totals
A value that changed across a write was caught only between repeats of
one tool's request line, so a list read before a write and an item read
after it put the post-write value into the initial state. Changes are
now classified per entity in timeline order after observation: across
any recorded write, a change between repeats of a line or between two
lines of the type rewrites the type (every tool recorded as
fields-differ); with no write between, a repeat's change is volatile.
The classification lives in environmentStateMerge.ts.
Any numeric member whose name contained "total" or "count" counted as
an item total, so total_pages, account_id or page_count made complete
lists partial. Item totals are now members whose name tokens include
total or count and none of page, pages or per; a string of digits counts
as a number; and a set has_more, next or next-page token member also
marks the list partial.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): keep post-write observations out of initial state
* fix(environment): retain field provenance and bound state inference work
* feat(environment): serve tool calls from the environment state
`environment-serve --state <state.json>` serves an environment from the
initial state an environment-state view extracted. Read tools whose views
are state-backed answer from a live state store: item views render the held
entity into the recorded envelope, list views render the collection or the
recorded members, and document views return the held value; a missing entity
or one lacking a required field is refused, and requests the state cannot
answer fall back to their recordings with the reason. The /agentenv data
plane implements data/reset, data/add (a rea-state/v1 document as a data
part, base64 bytes, a file:// path or an https URL, validated and replaced
atomically) and data/get, and the card advertises those operations and an
intake declaration. Stage 1 is read-only: write tools are not served in
state mode. Without --state nothing changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): record verification after merging the state view's review fixes
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(environment)!: serve the environment state by default
environment-serve derives the environment state at startup by running the
environment-state view in-process on the capture with the manifest's
template rules, so an environment comes with its state and a user supplies
nothing extra. The derived Evidence is the same one inspect-analysis-view
produces by hand. --state is removed; --replay-only serves exactly the
stateless replay of main. When the state cannot be derived, or its catalog
does not fit the discovery budget, the server starts in replay mode, logs
the reason and reports it in the stop summary's `state`. Exported bundles
serve the state by default too: export derives it the same way, so the
verifier's tool count and checked answer match what is served, and the
README describes the served mode.
BREAKING CHANGE: environment-serve answers read tools from the derived
state and does not serve write tools unless --replay-only is given; the
stop summary reports `state` instead of `state_evidence_id`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): bound data-plane replies, keep batch order, cancel intake on stop
- One /agentenv response is held to 64 MiB: replies are measured as they
are produced and one past the budget becomes a -32000 error stating it.
- Every message of a batch enters the arrival-order queue on arrival, so
another request's reset never lands between them.
- Stopping the server, or a caller leaving, cancels the active download
and answers queued calls `cancelled` without running them.
- file:// URIs name only this host: another host or a network share is
refused before anything is opened.
- A loaded view value may not give an entity another identity.
- List answers refuse a member lacking a field every recording carried,
as item answers do.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): keep bodies too deep for the state's nesting out of the state
A recorded body at the JSON depth limit was accepted by extraction, then
nested a few levels deeper as a document or entity value, so sealing the
state Evidence threw and environment-serve and environment-export stopped
instead of falling back. stateBodyJson, the one rule extraction and
serving share, now treats a body nesting deeper than the limit minus 16
levels as unusable, so its tool stays a recorded view.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): measure data-plane replies before building them
83fcc739 bounded one /agentenv response to 64 MiB, but it still
serialized each reply whole before measuring it, so a batch of data/get
calls on a large state built every refused reply in full only to drop
it. Each reply is now measured in bounded JSON parts (the shared
jsonParts encoder), its parts kept only while it fits; a reply past the
budget is counted for its exact reply_bytes but never assembled. A test
checks that a refused read is never handed to JSON.stringify whole and
that its reply_bytes is exact.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): release batch replies as they are encoded; judge exact numbers by value
answerBatch kept every settled reply in its pending array until the
batch ended, so a batch alternating data/add and data/get held each
loaded state alive; each reply now leaves the queue as it is encoded.
The exact-number policy judged a number by its recorded spelling, so a
state loaded with 1e20 came back from data/get as
100000000000000000000, which data/add refused, breaking the promise
that data/get output loads back unchanged. A number is exact only when
both its recorded spelling and the one serialization writes are safe,
so integers from 2^53 below 1e21 are refused in any spelling; replay
keeps such bodies as recorded text. A domain test checks that every
accepted number is accepted again once written back.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): keep room for refusals and framing in the reply budget
The response budget counted accepted replies only: a budget error that
replaced a reply, and a batch's brackets and commas, were not charged,
so near 64 MiB later refusals pushed the body past the advertised
limit. Every reply still to come now keeps room for its own budget
error, and a batch keeps room for its framing, so a refusal always fits
and the body never passes the budget. Reply ids and which messages are
answered come from one place shared by the replies and the reserve.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): refuse a loaded state that data/get would write past the intake limit
data/add measured only the document it was given, while data/get
writes the state back with the members a document may omit (types, the
state Evidence id, view fields and record citations). A document just
under 32 MiB could therefore load and be reported larger than any intake
accepts, breaking the data/get then data/add round trip. replace now
measures the state as data/get writes it, part by part, and refuses it
past the limit, leaving the state unchanged. The limit moves to the
state document module, so the application store can read it; the
intake re-exports it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): keep merged export checks on the shared test graph
Main's export tests installed module mocks for the whole services
project, which reuses one module graph. A tarball read failure then
missed its importer, and a one-shot heap mock could be taken by another
file. Read those through bindings the tests replace for one case.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(environment): keep a state file whose close failed owned until it closes
A file:// data/add that could not close its descriptor dropped the handle,
and an admission failure whose cleanup close failed dropped its owner, so
nothing could retry the close or report it.
The read now runs through withOwnedRegularFile, and a failed close is
retained in a server-lifetime ArtifactResourceScope. The next file read
retries it first and answers -32000 while it is still open; inline and
https documents still load. Closing the served environment retries it again
and rejects with the unfinished close.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: gensokyo-dev <gensokyo-dev@users.noreply.github.com>
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: morluto <morluto@users.noreply.github.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(native): recover macOS command deadlines
Promote native_command_timeout_ms (default 60000), native_ui_deadline_ms
(default 180000), and pwntools_timeout_ms (default 30000) as validated
per-call settings. Keep cpuSeconds, byte allowances, LLDB grace, and plist
probes fixed, and say those caps have no runtime override.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(native): report the deadline that actually expired
Pass the selected native command timeout through xcrun resolution, thread
the pwntools deadline into pwndbg failures, and state the selected UI
deadline apart from the fixed wait-step cap. SIGXFSZ and EFBIG recovery
names the file-size allowance.
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(native): honor the selected deadline on helper stages
The UI helper compile and capture subprocesses used fixed 60s and 30s
exec timeouts, so a longer native_ui_deadline_ms still failed as a generic
provider error. Preparation timeouts reported the 60000 ms default instead
of the selected native_command_timeout_ms.
Memory resource-constraint recovery names memory again, and the Darwin
plist CLI expectation matches the byte-counted input allowance.
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>
Bind artifact lifecycle and analysis calls to their advertised named SDK
output schemas, sharing discovery per connection and rebinding on reconnect.
Reject schema overrides and ambiguous catalogs, preserve request options,
and retain primary failures while attempting all owned cleanup.
Add real-stdio boundary regressions for malformed output, discovery reuse,
reconnection, schema ambiguity, override rejection and timeout preservation.
Assisted-by: Reimu (gensokyo-skills:reimu-incident-triage)
Assisted-by: Kogasa (gensokyo-skills:kogasa-surprise-testing)
* fix(browser)!: keep malformed console argument inventories unknown
An omitted, null, or non-array Runtime.consoleAPICalled args inventory now
reports argument_types as null, text capture as unavailable, and
console_events as invalid_protocol_value instead of a complete empty
inventory. Malformed argument elements keep their CDP position as
"unknown", so retained text values keep their original argument_index.
Call-site attribution now uses the frame at position zero: a malformed top
stack frame is no longer replaced by its caller. Console events are
excluded and request initiator locations are null, both reported as
invalid_protocol_value.
BREAKING CHANGE: console event argument_types may now be null, and
console text_capture.status may now be "unavailable", when the CDP
producer omits or malforms the argument inventory.
Fixes#5934
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(browser): report malformed WebMCP registration stacks as incomplete
topCallFrame now owns the omitted-versus-malformed decision for the whole
stack value: an omitted stack has no call site, while a non-object stack,
non-array callFrames, or a malformed top frame is malformed. WebMCP
discovery keeps the tool with a null registration_source but records
webmcp_tools invalid_protocol_value, so capture comparisons stay unknown
instead of confirming a changed source. Console and initiator stacks use
the same rule.
Refs #5934
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Remove shared-module mocks for input reads, heap statistics, and tarball streaming. Exercise actual file changes and real V8 heap admission in fresh CLI processes; keep only selected-descriptor syscall fault injection for the pathless EIO case.
Reproduced the four hosted source failures with a preloaded module graph and found the same flaw in the adjacent tarball test. The preloaded control now passes 17 cases; 32 focused tests, all 1,739 service tests (2 skips), and check:ci pass. Snapshot, cleanup, identity, and error projection coverage is retained.
Binary plist object tables may list one object offset more than once;
plistlib reads each entry independently. The scalar scanner marked any
repeated offset as incomplete metadata, which made the number projection
drop every exact decimal in the archive. A keyed archive whose entries
10 and 11 share an Int64 9007199254740993 reported both as observed
9007199254740992 and rounded an unrelated 2^60+3 sibling, with only the
generic incomplete-metadata caveat.
A repeated valid offset carries the same source facts as its first
visit, so observe it once and continue. Offsets that fail to parse still
mark metadata incomplete. Shared 16-byte integers and dates are rewritten
once in the decoder copy, which every entry then reads; the overlap plan
already treats one shared offset as one span. Keyed archives and
Interface Builder decoding share this scanner.
Distinguish CDP session detachment, observed crashes, and transport loss. Preserve the observed window alongside the original typed coverage-stop failure and unconfirmed cleanup.
Validated with 40 focused boundary tests, check:ci, docs:check, and the real Chrome public CLI/MCP verifier on macOS. Browser timing comparison remains incomplete because unchanged-main fixture readiness was intermittent.
Fixes#5756
Decode native binary plists from bounded owned source bytes before conversion can erase scalar values. Share scalar preflight and isolate overlapping decoder records while retaining exact integers, date uncertainty, and original bytes. Refuse numeric payloads that read structural table bytes, and avoid retaining projection metadata when callers only need structural validation.
Closes#5701Closes#5703Closes#5705Closes#5782Closes#5783Closes#5792
Move ELF, PE and Mach-O container parsing into a shared native owner. Go alone selects and validates Go build-info sections, so unrelated malformed Go metadata no longer blocks Rust dependency or IL2CPP inspection. Remove the superseded Go-specific container representation and migrate all consumers. Adapted from draft PR #5510 without asserting human PR readiness.
Validation: 51 compiled CLI/MCP acceptance tests across Rust, Go and IL2CPP; 186 shared-reader tests; full check:ci passed. Three public synthetic-file probes reproduce the baseline Rust refusal and validate the fixed SDK-advertised MCP output, artifact digests and unchanged bytes. The exact zero-address nonallocated ELF case now succeeds for Rust while Go still reports invalid_format. No target execution or real compiler-artifact qualification was performed; no performance improvement is claimed.
Closes#3398
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.
Join provider class observations only when selected-image identity and exact import bindings establish equivalence. Preserve distinct observations and unsupported coverage.
Refs #5607
Sample the browser monotonic clock with observation already subscribed, filter historical lifecycle and network facts, and preserve scope and termination handling when event time is unknown.
Bound pre-arm retention using transport-owned wire bytes. Fail closed when the relevant-event buffer fills, and keep unrelated protocol payloads outside that budget.
BREAKING CHANGE: observe_web_session timeline timestamps can be null when the producer supplies no comparable timestamp. Consumers must handle unknown time explicitly.
Refs #5609
Use the shared command supervisor for installation probes and retain exact
process cleanup owners through doctor, profile resolution, and session health.
Preserve exit status, bounded captured output, cancellation reasons, and live
cleanup facts. Block subsequent inspection until retained cleanup is retried.
Generalize the existing inspection resource scope and migrate its consumers,
including verifier scripts; remove the verifier's duplicate raw version probe.
Validation: 116 distinct focused tests passed, including built CLI doctor with
inherited output, cancellation/finalization, cleanup retry, and fake-provider
verifier workflows. Build, check:fast, and test discovery passed. No real Kuna
distribution or analysis target was executed.
Synthetic Darwin inspection timing: first-call median 707ms versus 5.79ms;
same-process repeat 37.9ms versus 3.08ms. First use includes ownership-helper
preparation. Family CPU and actual peak family memory remain unmeasured.
Fixes#5557
Have method, property, ivar, and adopted-protocol list owners flag real budget omissions on returned class, protocol, and category records. Preserve known empty lists and existing decode reasons.
Validated 59 source tests, two built CLI controls, five static inspections of an Apple-clang fixture, and typecheck/lint. The target was never executed. Fixes#5543.
Use the public rename API for an explicit empty existing name, retain omitted names and raw listing values, and verify fresh unnamed creation and explicit mutations by readback. Preserve independent verifier restorations and their primary failures.
Refs #5475. Focused bridge regressions and the complete real-Hopper suite passed on macOS arm64 with Hopper 6.6.0-demo 42.6316032. The verifier used a temporary unavailable Ghidra override to meet its sole-candidate prerequisite; reporter Hopper 6.1 remains unverified.
Link IL2CPP metadata names to native addresses and types through shared CLI/MCP workflows, with bounded queries and atomic Ghidra seed export.
Preserve the original implementation by 8rulerstar and follow-up correctness, resource, provenance, and CI dependency repairs. Verified against real Windows/Android artifacts, independent references, actual Ghidra, and focused CLI/MCP tests.
Fixes#5112.
* feat(il2cpp): read member names, string literals and metadata versions 24 and 27
Adds field, property, event and parameter names with their owners, managed string literals, and the version 24 and 27 header layouts to inspect_il2cpp_metadata, with filtered pages for the new collections. Member decoding yields between row batches like the other row readers.
* fix(il2cpp): bound metadata expansion and query parameter owners
* test(il2cpp): cover nonempty property and event tables
---------
Co-authored-by: gensokyo-dev <gensokyo-dev@users.noreply.github.com>
Separate independent exclusion, partial-source, credential-preflight and descriptor-failure workflows so their costs do not accumulate under one test deadline. Reuse one worker-owned MCP connection with explicit awaited teardown, preserving sensitivity isolation and retained-Evidence checks.
Assisted-by: Reimu (gensokyo-skills:reimu-incident-triage)
* feat(environment): extract the initial state a capture's reads observed
Add the environment-state analysis view. From the parents the environment
manifest accepts, it extracts the data the capture's reads observed before
the first related write: entity types with an identity field (proved by a
path parameter or inferred from an identity name), entities with their
recorded values and provenance, documents for written non-entity paths, and
a render plan per read tool that is kept only when it reproduces the tool's
latest initial recordings from the state; every other read tool is a
recorded view with the reason. The Evidence is always inferred.
Response-body decoding moves to a domain owner shared with replay, exact
JSON parsing moves to the owner of the exact-number policy, the manifest and
state views share parent admission and record grouping, and
projectAnalysisView takes its accumulators as one object.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): bound state at redacted writes; infer identity only from items
A scenario write whose URL was redacted has no request line, so it was
dropped and reads after it counted as initial state. Such a successful
write now joins the timeline with an empty path, whose root relates to
every type, so it ends every type's state; a limitation counts them.
A list whose recordings were all empty passed the "every listed item
carries it" check vacuously and published an identity field nothing
showed; it now needs at least one item, else the view stays recorded
as no-identity.
A value that differed between two request lines of one item tool, such
as a query variant, was called volatile, and the gate then substituted
each recording's own value; only repeats of one request line now show
volatility, so differing lines go through fields-differ.
List membership is compared only while a list can still be the whole
collection, and merged entities are indexed by type and collection once,
so paged captures no longer cost pages times entities (8,000 pages of
40,000 entities: 5.7 s to 0.23 s).
Docs and the tool contract say scenario state keeps the capture's
redaction markers and how redacted writes bound it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): keep first sightings, one list pointer and linear field unions
Entities took the order of the first plan observed, and item plans run
before list plans, so a complete list read before its detail reads came
out reordered and lost "all" membership; each entity now keeps its
earliest recorded sighting.
A list plan reused the first recording's pointer for later recordings,
so a later response that was itself an entity wrapping a list was read
as that list; every recording must now locate the same list.
The field union scanned its ordered list for every key, quadratic in an
object's width; a set now answers membership.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): keep post-write values and partial lists out of the initial state
A value that changed between repeats of one request line was named
volatile whenever no write under the type's own root lay between, so a
write under another root (for example POST /api/dashboards/db before a
second GET /api/search) left its effect in the initial state. Any
recorded write between the two observations now leaves every tool of the
type recorded with reason fields-differ instead.
A list was published with membership "all" even when a total or count
envelope member exceeded the number it listed. Such a counter now makes
the membership "recorded".
Also drop a stale doc comment on ObservedEntity.order.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs(environment): scope item-shaped and deleted written paths out of documents
State documents are held only for list-shaped paths a POST, PUT or PATCH
tool writes. Reads of item-shaped written paths and of paths that are only
deleted stay recorded views; the guide now says so and the spec defers them
to a later increment of #4933.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): count writes in flight when a value changes between reads
The volatility check compared write start positions with the two reads,
so a write already in flight when the earlier read started was missed and
its effect was named volatile. A write now counts when its interval
overlaps the span from the earlier read's start to the later read's end.
The check also ran once per changed field and scanned every write. It now
runs once per pair of observations, answered by a binary search over the
writes in timeline order and the latest end among each prefix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): catch cross-tool changes across writes; read only item totals
A value that changed across a write was caught only between repeats of
one tool's request line, so a list read before a write and an item read
after it put the post-write value into the initial state. Changes are
now classified per entity in timeline order after observation: across
any recorded write, a change between repeats of a line or between two
lines of the type rewrites the type (every tool recorded as
fields-differ); with no write between, a repeat's change is volatile.
The classification lives in environmentStateMerge.ts.
Any numeric member whose name contained "total" or "count" counted as
an item total, so total_pages, account_id or page_count made complete
lists partial. Item totals are now members whose name tokens include
total or count and none of page, pages or per; a string of digits counts
as a number; and a set has_more, next or next-page token member also
marks the list partial.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(environment): keep post-write observations out of initial state
* fix(environment): retain field provenance and bound state inference work
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Co-authored-by: gensokyo-dev <gensokyo-dev@users.noreply.github.com>
* feat(artifacts): read Godot PCK packs
Add a Godot Engine PCK artifact reader for standalone packs and packs
embedded behind an executable trailer (formats 2 and 3). Members are
uncompressed byte ranges verified against their declared MD5 on read.
Encrypted directories, sparse bundles, patch removal entries, unknown
flags and format 1 fail closed as unavailable; encrypted members are
reported without being read. inspect-artifact and extract-artifact
recognize standalone packs by their GDPC magic.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(artifacts): bound Godot PCK metadata and own member streams
Cap member paths at 4096 bytes and the retained directory at 64 MiB
before allocation. Read members with positional chunk reads through
pipeline so cancellation is typed, abandoned consumers release their
source, and the shared descriptor stays reader-owned. Retain the root
descriptor for godot-pck inventory, as for ZIP and universal Mach-O.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* feat(artifacts): complete Godot PCK facts, integrity and selected views
* fix(artifacts): preserve atomic-root inventory Evidence
* fix(artifacts): bound aggregate PCK member expansion
---------
Co-authored-by: Samucahub <151155051+Samucahub@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
* feat(il2cpp): inventory IL2CPP global-metadata images, types and methods
Add inspect_il2cpp_metadata (CLI inspect-il2cpp-metadata), a read-only
in-process decoder for unencrypted Unity global-metadata.dat version 39.
It reports header sections, serialized index widths, images, type
definitions (nesting from the nestedTypes section) and method names.
Unrecognized magic and other versions are unsupported; damaged headers,
overlapping sections, inconsistent record sizes, out-of-range indices and
invalid strings fail with a reason code and no partial success. MCP
detail summary retains the complete Evidence and pages rows through
inspect_analysis_view kind il2cpp-metadata.
Adds an opt-in verify:il2cpp:metadata -- --metadata PATH check.
* feat(il2cpp): deliver metadata summaries by default over MCP
A 112,232-method file produced a 10.5 MB complete response, above the
10 MiB stdio budget. Default MCP delivery is now the retained summary;
detail complete stays available, and the one-shot CLI requests it.
* fix(il2cpp): reject shifted strings, nesting cycles and method owner drift
Review fuzzing showed that moving the strings section offset by one byte
silently renamed every image, type and method. Name indices must now start
a string, which also bounds total decoding by the section size.
Nesting cycles and methods whose declaring type disagrees with the owning
type range are rejected instead of returned. Adds service, shutdown and
index-width coverage, and moves environment-specific measurements out of
the public guide and tool description.
* feat(il2cpp): decode global-metadata versions 29 and 31
Unity 2022.3 writes version 29 or 31 metadata: a 256-byte header of
(offset, size) pairs, fixed 4-byte indices, type records with
elementTypeIndex, and method records that gained returnParameterToken in
version 31. Record counts come from section sizes and must be whole.
Sections report a null count for these versions and index widths report
fixed-width-version. Image, type and method rows keep the same shape, and
the string, nesting and method-ownership checks apply to every version.
* fix(il2cpp): bound report size before serializing and tighten coverage
Rows share decoded names, so a 130 KB file could expand into gigabytes of
JSON before the 64 MiB budget check, ending in an internal RangeError. The
provider now refuses from a lower bound on the serialized rows first.
The decoder also rejects non-empty sections inside the header, methods
outside their declaring type's range and overlapping image type ranges, and
keeps a leading byte order mark in names. Layouts now carry one indexing
mode instead of two coupled flags, and the tool description explains
nested_type_start.
* fix(il2cpp): refuse oversized metadata before decoding rows
Address review findings on the IL2CPP metadata inventory:
- check a lower bound from header row counts before any row object is
allocated, so a large rejected file no longer builds every row first;
- validate sealed parent Evidence and count its rows once per report
instead of on every inspect_analysis_view page;
- decode version 39 fixtures with 2- and 4-byte type-definition,
generic-container and parameter indices.
* docs(il2cpp): attach the decoder comment to readIl2cppMetadata
* fix(il2cpp): deliver complete rows by default like the other inspection tools
inspect_il2cpp_metadata defaulted to detail summary, the only inspection
tool to do so. Omitted detail now returns every row inline, as
inspect_artifact, decode_interface_builder and the capture tools do;
summary stays one argument away and the guide, CLI reference and skill
reference say to select it first for large games, whose complete MCP
delivery is refused at the stdio budget with retained Evidence.
* test(il2cpp): cover narrow-width -1 sentinels and fix the skill's default
An empty image stores its type start as the all-ones value of a one- or
two-byte index; the decoder reads it as -1, but no fixture exercised it,
so removing either conversion passed the suite. The fixture can now append
an empty image, and both widths decode it. The skill still described a
summary default; it now says to select summary first, as the guide does.
* docs(il2cpp): request summary explicitly in the summary example
* chore(il2cpp): hash the verifier metadata through owned file resources
* fix(il2cpp): count escaped name length without building the escaped string
* fix(il2cpp): keep the verifier's check failure and small-table last page
* feat(domain): count JSON string length without building the escaped string
* fix(il2cpp): keep descriptor cleanup owners and share the JSON length count
* test(domain): give the backslash case real backslashes
* fix(il2cpp): select the CLI path and finish the command like the other tools
Follow main's absoluteCallerPath (#4067) and pass the command signal to finishAnalysisCommand.
* fix(il2cpp): select the verifier input path like the CLI
Use absoluteCallerPath (#4067) so a path with a symlink followed by .. names the same file the CLI and MCP inspect.
* feat(il2cpp): add compact composable metadata queries
Retain source-indexed Evidence for filtered batch and field queries. Bound representation growth before serialization, yield decoding for cancellation, and share CLI Evidence publication with desktop inspection. Verify public Unity metadata against independent byte and managed-DLL oracles and exercise actual stdio SDK workflows.
* fix(il2cpp): retain the validated metadata header magic
* test(il2cpp): remove obsolete execution wording exception
* fix(artifacts): preserve export selection and completed source facts
---------
Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Build an agent environment directly from retained capture_browser_scenario Evidence through the existing manifest, serve and export commands. Preserve request ordinals, source pointers, redaction markers, unavailable bodies, provenance and the default stdio MCP transport.
Followups admit copied/transformed text before body validation or decoding, avoid request-text allocation in replay, normalize portable file inventories, select dependent CI lanes, retain the author's redaction/body-budget corrections, and count body coverage over represented requests.
Validation: full pinned-toolchain static gate, source and public CLI/MCP regressions, canonical catalog checks, actual Chromium interactive and live GitHub API captures with exact HTTP replay and stdio parity, the generated non-root Docker image, and its smoke task through real agent-env 0.9.1334. Verification scopes, resource measurements and platform limits are recorded in the PR and feature verification document.
Co-authored-by: gensokyo-dev <gensokyo-dev@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Serve retained capture Evidence as a bounded MCP replay environment with discovery.
Preserve default stdio, caller-selected facts and complete source Evidence.
Fix numeric fidelity, argument identity, startup cancellation and bounded discovery,
query/path matching and output metadata. Verify real recorded API data and public
CLI/MCP workflows; preserve explicit gaps for Windows and third-party hosts.
Co-authored-by: Edgar Arakelyan <edgar.arakelyan@scale.com>
Co-authored-by: gensokyo-dev <gensokyo-dev@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Keep a run-scoped evidence owner from acquisition through settlement and
cleanup retry. Bind token observations to native generations and retain
related generations and unresolved identity lookups across table omissions,
reparenting and process-group changes without treating PID hints as authority.
Use native effective UID facts for withholding decisions, corroborate Linux
procfs absence, and share the Linux stat parser with sampling and debugger
identity checks. Signal only after current token and group revalidation.
Complete swept cleanup across all groups with one bounded final family check;
retain unresolved owners and successful-peer completion across retries.
Validation: npm run check, docs:check and test discovery passed. Focused
ownership/consumer coverage totals 513 passed and 1 skipped across 35 files,
including the corrected Electron cancellation test rerun independently. Darwin helper compilation, real
owned Node child cleanup, CLI/MCP capture and retained cleanup retries ran.
Real Electron applications, native Linux races/PID reuse and Windows cleanup
were not qualified. The original Linux reports remain open for that work.
Fixes#5238
Refs #4322, #5206
Scan provider output before retention and keep bounded confirmed token facts
through output completion. Exclude identifier prefixes, pre-checkpoint context,
and unconfirmed tails; preserve invalid UTF-8 and directional truncation facts.
Make confirmed OOME observations authoritative for projection. Retained raw
prefixes cannot establish an OOME, and unrelated output cannot supply its
resource kind or recovery advice. Incomplete no-OOME output preserves the
primary error and explains uncertainty. Current checkpoint coverage remains
separate from historical budget loss.
Adapt the retention approach from #4774 with independently reproduced scanner
and projection counterexamples corrected before integration.
Validation: 268 focused tests, npm run check, npm run docs:check, and test
discovery (1639 files exactly once) pass. Controlled Node pipes/socket and fake
Ghidra JVM only; real Ghidra and native Linux/Windows behavior are unverified.
Fixes#4682Fixes#5220
Co-authored-by: reason-machines[bot] <293529679+reason-machines[bot]@users.noreply.github.com>
Co-authored-by: Saim Shafique <msaimshafique4664@gmail.com>
Parse Pickle extents once at the ASAR boundary and carry finalized signature
facts through root and nested classification. Scan JSON whitespace within its
declared extent and decode budget without retaining the whitespace prefix.
Keep prefix storage bounded independently of stream chunk ownership.
Fill fixed ASAR prefixes and DMG trailers across short reads, preserve
cancellation through admission and stream EOF, and retain existing cleanup
ownership. Cover public CLI/MCP selection, root/nested inventory, malformed
JSON whitespace and padding, limits, native/XML consumers and descriptor faults.
Validation: 24 focused files / 170 tests; full check; docs check; test discovery.
Official inert ASAR fixtures and controlled descriptor faults were exercised;
real Electron execution, Windows and natural OS short-read frequency remain
unverified. Equivalent compact CLI tasks returned identical results and reply
token estimates; repeated timings do not establish a speedup.
Fixes#3574Fixes#5188
* feat(analysis-view): project an environment manifest from web network captures
Add the environment-manifest view kind to inspect_analysis_view for retained
HAR Evidence: the primary origin's recorded request lines become a
deterministic tool surface (names, path templates, parameters with their
rule and basis, observed query names, value-free body shapes, record
pointers), with excluded groups, three fixed unknowns and the shared
projection budget. The four format rules always apply; after-uid-literal,
opaque-token and sibling-variance are opt-in and labelled, and literal
fold-in never touches a format-rule parameter. The api-surface accumulator
moves onto shared route, media-type and budget helpers without behaviour
change.
Refs morluto/rea#4933
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* test(web-network): check the manifest refusal text like the api-surface refusal
The real-capture verifier stringified the whole MCP result for the
environment-manifest refusal, so a declared "[" literal matched the result
envelope itself; compare the parsed refusal as the api-surface branch does.
Refs morluto/rea#4933
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* docs(specs): record the hosted web-network lane run for the manifest view
Refs morluto/rea#4933
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* docs(specs): reconcile the manifest view verification scope note with the hosted run
Refs morluto/rea#4933
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* fix(analysis-view): bound body, route and name work in the environment manifest
Addresses the automated review findings on the PR: read request bodies at the
top level only instead of walking nested values; admit every literal route
part before materializing it; compare at most 32 segments in sibling-variance
and say so; keep stop words literal after a uid literal; keep the next
collision suffix per tool-name base; preserve a templated exclusion when two
groups print the same row.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* fix(analysis-view): keep manifest folding, naming and media-type reads linear
Addresses the second automated review round: fold literal siblings through
set membership, key collision suffixes by the truncated name they share, and
charge a request media type before it is scanned.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* fix(analysis-view): charge the manifest group key before building it
Addresses the third automated review round: the composite key copies the
method, origin, template and rules, so that copy is reserved before the key
is stringified.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* fix(analysis-view): accept a +json suffix only on application media types
Addresses the fourth automated review round: `image/example+json` or
`text/example+json` no longer counts as JSON for origin selection, exclusion
or body shapes, matching the documented `application/json` and
`application/*+json` contract.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* fix(analysis-view): require the capture parent's exact subject for a manifest
Addresses the fifth automated review round: the manifest cites the report
path and digest, so a parent whose subject is absent or names another path or
SHA-256 is refused at admission, as the capture view already does; the
api-surface view keeps its existing admission.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* fix(analysis-view): share truncated suffix allocation work
Track suffix progress by the actual truncated prefix and suffix width. Distinct full names can collide only after suffix truncation; sharing that progress prevents quadratic rescanning while preserving shorter free names and existing allocation order.
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
* fix(analysis-view): merge only identifier-like siblings under sibling-variance
Addresses the sixth automated review finding: a bucket merged whenever one
literal carried a digit, folding plain words such as `mine` into the
parameter. The rule now partitions the bucket, merges the identifier-like
siblings and leaves the other literals as their own tools.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* fix(analysis-view): combine qualifying sibling cohorts
Keep coded identifiers and a qualifying plural hyphenated cohort in one eligibility Set before checking the minimum merge size. Read each literal once and reuse the same membership at the consumer, preserving non-identifier routes and the three-hyphen minimum.
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
* docs(analysis-view): clarify opt-in sibling folding
State the additional sibling-variance selection in the guide, acceptance specification and MCP parameter description. Preserve the existing single-rule behavior and caller-selected inference rules.
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Update public-boundary expectations after the explicit XML unknown-scalar
and native UI deadline provenance changes. Retain raw malformed literals,
independent exact integers, expanded-year ISO dates, and partial UI reports.
Assert successful cleanup releases its lease and failed cleanup retains it
until close retries. Keep malformed-input and true date-range coverage.
The shared number fixture owns CLI/filesystem/MCP expectations. Native
plutil controls confirm integer entities are invalid plist literals, even
though they are valid XML. No runtime behavior or CI gates change.
Validation: 187 source cases, 138 focused public cases, final 97-case
rerun after fixture typing/group changes, full static/architecture checks,
and test discovery (1591 files).
Normalize shared test fixture prefixes using the hygiene verifier ownership rule. Preserve caller labels, existing owned names, and third-party caches; cover retained fixtures and exact cleanup through subprocess regressions.
Interpret source-bound ELF64 NT_FILE descriptors once before debugger
launch. Bound counts and pathname coverage by the descriptor, retain exact
path bytes, and refuse relative names or dot segments rather than letting
GDB resolve them against the host. Pin GDB's pathname semantics to Unix.
Preserve the decoded partial observation on refusal, basic-only recovery,
and pre-admission cancellation. Share an exact mapping fixture and replace
the old fixture boolean modes with explicit report producers.
Validation: final series has 33 passing Vitest cases plus five Python
producer controls; 38 Linux x64 provider cases are explicitly skipped on
macOS. Typecheck/lint, docs checks and test discovery pass. Real GDB16.3
and strace reproduce the relative-path escape using only an owned fixture;
compiled admission admits a real absolute-path core and refuses that
relative-path core. Full native pwndbg / Linux x64 remain unverified here.
Task token, CPU and peak-family-memory costs are unmeasured; no speed claim.
Fixes#4942
Share schema keyword roles and preserve reference bases for explicit local pointer targets and uncertain intermediate scopes. Retain literal values and canonical validation. Refs #4936.
Request epoch timestamps from capinfos and normalize them once at the
metadata boundary, preserving nanosecond precision and explicit unknowns.
Bound numeric parsing to the producer representation and cover negative
fractions, calendar limits, malformed output and MCP projection.
Adapted from PR #4909. Fixes#4907.
Validation: 47 passed, 24 platform-specific skips across five focused
files; check:fast and test discovery passed. No live Wireshark or
Linux/Windows capture execution was performed on this macOS host.
Co-authored-by: Mert Kaya <47145541+mertkayacs@users.noreply.github.com>
Transfer canonical leases through opening, session and storage boundaries. Keep response deadlines separate from actual completion, drain independent owners, and join late opening failure cleanup before confirming recovery. Remove error copies in favor of canonical primary/cleanup composition.
Preserve exact caller reasons, completed observations and independent physical failures through final-close cancellation and retry. Closes#4869.
Validation:155 tests pass across15 focused source, service, CLI subprocess and MCP files; static/docs/discovery1551 pass. Two original-revision controls fail as expected. Bounded authored-session public CLI comparison preserves equivalent Evidence and1282 reply token estimates; no real Chromium qualification or forced-termination recovery claim (#4659).
Share exact command/helper cleanup adoption across Java probes. Preserve complete live resource labels and explicit unknowns through inspection leases, and retain unresolved ownership when a cleanup capability is unavailable.
Validation: focused cleanup, provider/doctor and Kuna consumer checks; shared Ghidra lifecycle and public doctor regressions. Native Windows/Linux provider behavior remains unverified.
Record the first transport failure before SDK teardown, then capture terminal stderr after owned retirement. Preserve exact opaque cancellation reasons and keep foreign cleanup ownership separate through retries.
Validation: 119 focused tests across eight files; five selected baseline regressions; check:fast, docs:check, test discovery; nine real CLI and nine real MCP tiny-APK inspections. Native Windows/Linux startup failures and actual JVM OOM remain unverified.
Closes#4844
Cursor Agent