Addresses a thirteenth automated review finding: `application/json;q=0`
explicitly refuses JSON, so it no longer counts toward the exact Accept ranges
a session takeover requires.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Addresses the twelfth automated review finding: bodyUsed only means a body
was disturbed, so a route that read one chunk of an upload left the rest
paused on a keep-alive socket. The close decision now follows the incoming
message's end of stream.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Addresses the eleventh automated review finding: dot segments such as
`/public/../mcp` were canonicalized to `/mcp` before routing. The bridge now
refuses any target whose path differs after parsing, so routing sees exactly
the path the client sent.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Addresses the tenth automated review round: the bridge joins an origin-form
request target onto the listener's origin so a network-path target stays a
path and absolute-form targets are refused; only a lone initialize can replace
an open session; and the shutdown hook records completion only after it
returns, so a hook that threw stays eligible for the next close.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Picks up upstream's test-runner configuration and the command-finalization
test fixes so the fork lane stops failing on unrelated flaky shards.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Interpret complete scalar source at the XML owner instead of accepting numeric prefixes, permissive date strings or nonempty booleans. Retain malformed content as explicit unknown scalars with original literals and reasons, while preserving the surrounding archive and independent numeric facts.
Share the representation across plist, keyed archive, Interface Builder and entitlement projection; remove the superseded numeric fallback and helpers. Parse date fields as UTC, retain real entity/CDATA semantics, and distinguish native integer/real whitespace without ambiguous whitespace backtracking.
Validation: 276 source tests and 60 compiled/public tests passed, with updated scalar regressions rerun; static, architecture and docs checks passed. Native macOS plutil controls and CLI/MCP source-hash/time-zone controls executed. No Linux/Windows or deep-provider qualification. Tiny completed CLI task costs recorded; no performance improvement claimed.
BREAKING CHANGE: Unresolved XML scalar values now use $plist_type unknown wrappers with element, literal and reason. This replaces the former real/null unknown marker and malformed scalar prefix fallback. Supported extended UTC years now preserve their timestamp.
Fixes#4852
Adapt the completed-quartet decoder from PR #4981 at 4cf6cac14. Bind raw source spans to validated DOM nodes before entity and line-ending interpretation can change data bytes. Preserve depth preflight, reject invalid data forms and allocate exact decoded output through one shared quartet grammar.
Fixes#4978
Co-authored-by: Avichal Pandey <avichal2018lko@gmail.com>
Addresses the ninth automated review round: a method Fetch refuses, such as
TRACE, is answered with 405 instead of escaping the request guard; the
unread-body decision now comes from the incoming message's own body
declaration, so a GET or HEAD carrying a body closes too; and the regression
uses a keep-alive client and checks that the socket actually closes.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Addresses the eighth automated review finding: a refused Host or Origin, an
unknown route or the startup 503 answered a POST while its body stayed unread
on the paused bridged stream; such replies now carry Connection: close so an
unread upload cannot keep a socket open.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Addresses the follow-on finding on the signal latch: the HTTP entry now owns
its exit status, so a 0 returned by startup after a recorded signal keeps the
signal's 130 or 143 while a real failure still wins; a process test sends the
signal as the listening line appears.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Addresses the sixth automated review finding: SIGINT and SIGTERM are recorded
from the moment the HTTP entry starts, the lifecycle owner acts on a recorded
signal as soon as it registers, and a second signal during startup exits at
once.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Addresses the fifth automated review round: the instance's shutdown hook runs
exactly once even when the listener closes without a live session; the
initialize probe no longer awaits cancelling its cloned body, which would wait
for the original request to drain; and the node:http bridge passes repeated
Set-Cookie values through as an array.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Addresses the fourth automated review round: a sessionless POST replaces a
session only when its Accept header lists application/json and
text/event-stream as exact media ranges, and a session whose transport
shutdown rejected stays owned so the next close retries it.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Addresses the third automated review round: loopback is classified from the
address the listener bound with the configured name added to the Host and
Origin allowlist; scoped IPv6 literals are refused before binding and a URL
that cannot be formed closes the listener; listener errors reach onerror only
after a successful bind; the MCP runtime keeps its factory owner as the single
owner of instance cleanup through `closeInstance: false`.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Verify the distribution Python embedding library during approved Linux Hopper
setup, conditionally repair the matching Ubuntu dependency, and confirm that
the library loads. Share fresh bounded observations with doctor, preserve
unknown configuration and caller cancellation, and join probe exit through
the existing process supervisor.
Validate clean Ubuntu installation, normal-user real Hopper CLI/MCP workflows,
independent ELF/SQLite oracles, cancellation and cleanup regressions, 154
focused checks and all 40 current-head CI checks. Preserve measured artifact
identity, failed attempts and local host-memory guard limitations.
Closes#5015.
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Addresses the second automated review round: a sessionless POST replaces a
session only when its media type proper is application/json, and the server
instance is created after the listener binds so a bind failure leaves nothing
to clean up and a factory failure releases the listener.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Addresses the automated review findings on the PR: refuse requests once the
session slot is closed and join an arming initialize during shutdown; retry
failed close stages instead of replaying the rejection; classify loopback
binds by address class and add the bind's own hostname to the Host and Origin
allowlist; let only a POST with the streamable headers replace an open
session; announce the bound URL at every log level; route `mcp -h --http` to
the HTTP entry.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Supply a real retained resource owner to every Android verifier prerequisite
check, sharing the existing Ghidra verifier cleanup scope. Match recovery
checks to typed Java or archive failures and applicable remediation, preserving
exact close retry authority and primary failures.
Assisted-by: Reimu (gensokyo-skills:reimu-incident-triage)
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.
Preserve the immutable 6.7.0 tag ancestry and synchronize version metadata after verified npm publication. Keep newer main performance changes and the full 6.6.0 changelog.
Add opt-in string_tables_only and CLI --string-tables-only to keep selected
strings once in typed table facts. Preserve existing defaults and legacy replies;
charge aggregate output capacity for the copies actually requested.
Validate shared CLI/MCP behavior with independent XML oracles, three real APKs,
94 source tests, nine compiled boundary tests, 36 completed MCP tasks and three
CLI parity tasks. Preserve full task estimates, negative timings, resource
admission refusal and verification limitations in the measured report.
Fixes#4982
Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
Use the pinned Vitest parser for coverage and shard flags. Keep ordinary CLI workers at 384 MiB and covered workers at the tested 512 MiB with unchanged concurrency. Retain complete coverage and add actual-config regressions for false, provider-only and spaced-shard options.
Validation: 27 focused cases; 16 covered cases with CLI preloaded; 5 docs checks; typecheck/lint; exact discovery of 1578 files. Measured covered control: 22.18s elapsed, 46.52s CPU, about 2.33 GiB sampled summed process-family RSS on macOS arm64. Full Linux CI pending. No runtime/device verification claimed.
Fixes#4960
Add rea mcp --http [--host] [--port]: the same server rea speaks over stdio,
served over MCP Streamable HTTP on a loopback listener by default, using the
SDK's web-standard transport already in the pinned @modelcontextprotocol/server
behind a small node:http bridge. One client and one session at a time; DELETE
ends the session without running provider teardown, a new client's initialize
replaces a session left open, foreign Host headers are refused on loopback
binds, and shutdown is signals-only. Stdio stays the default and registered
transport; server.json and client registrations are unchanged.
Refs morluto/rea#4933
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
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
Resolve lookup identity and selected/default origin together in the plist
source owner. Physical-file and conversion failures consume that fact
rather than reconstructing intent from omission of the path argument.
Selected plist artifacts resolve independently of readiness; unresolved
scope admission does not acquire an invented file-selection context.
Remove the superseded resolver and path/requested representation.
Cover direct-target CLI/MCP refusals, physical file conditions, permission
denial, default controls, unresolved scope context and cleanup retry for
both explicit path and selected-target workflows.
Fixes#4923. Related: #4849.
Validation: public CLI/MCP baseline3 failed8 controls passed; final55
passed across7files, check:fast, docs5, discovery1566 and source review.
Actual macOS26.3 plutil CLI/MCP; no Linux/Windows native qualification.
Read failed conversion diagnostics from both output streams, retaining
raw bounded excerpts, observed byte totals, exit status and truncation.
Remove only repeated path prefixes from display text and report a useful
reason when the command prints no diagnostic. Reuse native failure
capture bounds rather than expanding accepted nonzero output into errors.
Adapted from PR #4851. Fixes#4849.
Validation: six controlled boundary failures on baseline, including the
stdout-only empty-reason defect; final46 passed across6files, real macOS
CLI/MCP controls, static checks, docs5 and discovery1565. On macOS26.3
plutil uses stderr; stdout-only coverage is controlled output. Separate
selected-target classification defect is reproduced and tracked as#4923.
Co-authored-by: apitoagents <339136651+apitoagents@users.noreply.github.com>
Make XML numeric uncertainty explicit at its producing boundary, separate
from decoded IEEE NaN and signed infinities. Shared projection preserves
known values using canonical numeric labels; unresolved XML numeric
values remain explicit unknowns. Remove the superseded known-value
wrapper and spelling alias, keeping numeric interpretation with its owner.
Cover public CLI/MCP binary archives, lookalike dictionaries, authored
32/64-bit IEEE payloads, signed zero, and XML parsing/overflow uncertainty.
Document canonical labels and unrepresented NaN payload/sign details.
Fixes#4919. Related: #4906, #4852.
Validation: baseline 2 failures/5 controls through public CLI/MCP; stable
final 439 passed across25files, check:fast, docs:check and discovery.
Native macOS plutil and existing codesign controls executed; no
Linux/Windows run or real Interface Builder engine qualification.
Recognize complete CFPropertyList literals at the XML owner and retain
canonical NaN and signed infinity identity as typed scalars. Keep scalar
wrappers distinct from lookalike dictionaries, share limitation text and
document alias normalization. Incomplete markup retains its existing
compatibility fallback and limitation. Bound token matching before case
normalization using the accepted grammar.
Adapted from PR #4908. Fixes#4906.
Validation: public CLI/MCP regression baseline 2 failed, 4 passed.
Final focused validation 430 passed across 25 files, including controlled
entitlement response coverage in the following commit. Real macOS plutil
CLI, check:fast, docs:check and discovery passed. Binary value identity
remains tracked in #4919; malformed literal policy remains #4852.
Co-authored-by: Mert Kaya <47145541+mertkayacs@users.noreply.github.com>
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>
Replace configuration-specific write-file-atomic delegation with a verified
input snapshot, exclusive private staging, and an explicit canonical-parent
destination. Existing files publish by rename; first-run files use an exclusive
link so a newly created destination is retained. Both consumers read back the
publication owner's selected path.
Retain descriptor and staging cleanup capabilities through recovery and host
retry. Preserve mode and ownership, remove temporary permission handoffs, and
use positional restoration writes. Keep UTF-8 diagnostics with failed cleanup.
Validation: 284 focused tests passed, one skipped across 14 files; check:fast,
format:check, docs:check (including five consistency tests), two source reviews,
and 12 completed isolated public CLI setup/uninstall cost controls passed.
Native Windows/Linux, real editor concurrency, and providers were not verified.
Final-check/publication races and same-inode concurrent writers remain unexcluded.
Fixes#4904
Replace pathname writes with the shared configuration publication owner.
Reject replacement readback and restore through the exact staged descriptor.
Retain initial-read, readback and publication cleanup owners for retry. Share
nested cleanup handling with setup without granting authority to opaque
cancellation reasons. Separate host admission from nested file cleanup gates.
Preserve uninstall file modes, including readable read-only configurations.
Hold the exact staging permission owner through writable-descriptor handoff
and failed mode restoration. Remove the superseded writeText effect contract.
Validation: 206 focused tests across 9 files, check:fast, formatting and
docs:check passed. Three uninstall replacement controls fail on baseline
7e4eed9dd while thirteen setup controls pass. Twelve completed CLI uninstall
tasks preserve equivalent outputs, config, backups and modes0400/0640.
Two read-only source reviews completed. Native Windows/Linux and real editor
races remain unverified; in-place writer exclusion is still not provided.
Closes#4902
Retain the staged file descriptor through atomic publication, readback and
recovery. Refuse detected replacements or edits and restore through that
owned descriptor instead of reopening a pathname. First-run readback failure
now retains the file for manual recovery because pathname deletion cannot
prove exact ownership. Keep both descriptor owners when nested closes fail;
opaque close rejections do not transfer ownership.
Validation: 192 focused checks in 9 files, check:fast, formatting and docs:check
passed. Three replacement regressions fail on the baseline while six existing
recovery controls pass. Six completed setup tasks preserve equivalent config
and backups. Two read-only source reviews completed on macOS arm64 / Node24.
Native Windows/Linux and real concurrent editor behavior remain unverified.
State checks do not exclude an in-place writer after the last check.
Closes#4900
Project default and multiple selected localized string tables from one
immutable decode pass. Remove the need for repeated APK snapshotting,
digesting, version probing, apktool d invocations, and workspace enumeration.
- Contract: add optional locales array to decode_android_resources input
schema (bounded to 20 locales) and structured string_tables to output schema
with per-table status, strings, total_count, unparsed_count, and coverage.
Preserve legacy strings and locale fields for complete backward compatibility.
- Bounds: enforce 32 MiB aggregate XML read cap and 10,000 aggregate string
entries cap across batched projections. Handle missing, oversized, and
malformed tables per-table without failing or losing completed tables.
- Lifecycle: maintain strict descriptor cleanup and provider-owned workspace
removal in finally, preserving typed cancellation and failure details.
- Adapters: add --locales repeatable option to CLI and update MCP contract.
Validation: 10 service tests, 3 CLI boundary tests, and MCP boundary schema
tests pass. Fast typecheck/lint and turbo docs:check:built pass.
Fixes#4777
Plist inspection reported a serialized -0.0 real as the JSON number 0.
- The shared plist projection reports it as a typed real scalar whose value is the string -0, counted in a limitation for keyed archives and inspect_plist.
- The plutil JSON path applies the same value when it decodes -0.
- The XML decoder no longer returns -0 for a mixed-content integer.
Fixes#4834
Separate verified process cleanup from output closure. Bound additional retention from owned stop entry through cleanup retry, drain before detaching capture, and mark unreadable or abandoned output incomplete. Release owned event-loop references on drain timeout without changing the proven cleanup disposition; preserve borrowed handle reference state.
Cover held and failed cleanup, late tail, bounded drain, stream errors, borrowed authority, and a real POSIX inherited-stderr release barrier. The real pipe control drops its final ten bytes on the baseline and retains them with this change. Focused consumer checks passed, including simulated Windows ownership; native Windows and real provider failure scope remain unverified.
Android still constructs its startup error snapshot before retirement; that consumer correction remains separate work.
Fixes#4840
Translate actual OutOfMemoryError at the bridge boundary before upstream stringifies it. Preserve observed heap bytes, missing reasons, coroutine context and cancellation; retain SessionHolder ownership and load options. Restrict heap-specific remediation to known heap reasons and parse named bridge failure contracts once.
Coordinate bridge protocol v4, clarify doctor configuration versus target capacity, and preserve the memory primary across cleanup failure/retry. Verify controlled immediate/late/inline continuation outcomes with the pinned JAR and real JDK, focused tests, and tiny real CLI/MCP method workflows.
Actual heap exhaustion, the reported APK, pre-handshake failures, Linux and Windows remain unverified; no stress or profiling run.
Fixes#3517
Resolve smali using frozen raw method names and full DEX descriptors. Select
one complete block before truncation for missing Java, static initializers
and decompiler marker fallbacks. Reject ambiguous or incomplete blocks and
preserve upstream disassembly failure notes.
Bind method replies to the selected descriptor with bridge protocol 3.
Remove consumer placeholder/error-prefix interpretations and the superseded
blanket overload fallback rejection. Producer failures now use the shared
failed-tool path before truncation. Migrate protocol fixtures and consumers.
Adapt the direct selector guard from PR #3450; recover original bytecode
instead of inferring a holder from generated comments.
Validation: 122 focused tests in 9 files; real pinned-JAR baseline regression
fails and candidate passes, including overloads, aliases after generation,
branches/handlers, 16-byte output and negative selection controls. Nine real
CLI and nine MCP fixture calls pass with parity/identity/digest/no-mutation
checks. Static/docs checks and two source reviews pass. Six completed real
MCP cost workflows recorded; same calls with improved source, no speedup
claim. Tiny authored APKs, macOS/JDK 21 only; reported editor APK and native
Linux/Windows unverified. Heavy profiling and stress remain stopped.
Fixes#3443
Co-authored-by: Taksh <takshkothari09@gmail.com>
Freeze body-absence access flags with stable parsed method identity, before
code generation can rewrite method metadata. Skip source slicing for these
methods and return unavailable empty source with explicit native/abstract
markers, rather than attributing neighbouring members to the selected method.
Adapt the access-flag guard and tiny Smali fixtures from PR #3527. Extend real
metadata and public CLI/MCP regressions for native, static native, abstract,
concrete overloads and re-reading frozen facts after neighbouring generation.
Rename the expanded method-body verifier and remove its old entry point.
Validation: 72 focused tests in 3 files; real pinned-JAR baseline control
fails and candidate passes without triggering code generation; 6 real CLI
and 6 MCP fixture calls with parity, identity and unchanged APK; docs/static
checks and two source reviews. Six measured real MCP workflows, no speedup
claim. Tiny authored APK only; reported BinaryEye, Linux/Windows and browser
workflows unverified. Heavy profiling/stress remain stopped.
Fixes#3488
Co-authored-by: balkan16 <57933796+balkan16@users.noreply.github.com>
Resolve eight nested README links from their owning directory. Preserve the
shared README/website inventory contract by registering Malayalam with an
explicitly untranslated catalog and the existing visible, noindex English
fallback. Regenerate the shared website navigation and catalog references.
Existing parsed translations remain unchanged.
Validation: docs check, full website check (401 HTML pages, 27790 local
references, 19 tests), all 19 generated Malayalam fallback pages, and two
source reviews. No Malayalam website translation or browser layout claim.
Fixes#4821
Return tagged string-document parsing failures instead of fabricating one unparsed entry. Preserve XML failure reasons and publish unknown counts with partial coverage; count only examined unsupported entries in validated documents. Migrate every consumer, remove the superseded throwing XML wrapper, and retain completed failure Evidence through cancellation and failed cleanup.
Adapt the document-versus-entry distinction from PR #4804 using the shared Result and XML owner. Multi-table projection and its aggregate bounds remain under review in #4777.
Validation before incoming main rebase: 229 focused tests in 15 files, four baseline controls fail, static/docs checks and two source reviews. Twelve tiny actual MCP workflows and four CLI controls; authored launchers only, no real Apktool/JVM/Windows qualification.
Fixes#4820
Co-authored-by: Abhay Tiwari <abhaytiwari0821@gmail.com>
* docs: add malayalam translation
* docs: add ml hyperlinks to other docs
* docs: fix ml hyperlinks
* docs: fix ml hyperlinks
* docs: add english hyperlink in README_ml.md
* docs: fix README_ml.md
* docs: fix README_ml.md
* Update links and headings in README_ml.md
Carry the selected host into the version owner, disclose the Windows follow-up before 2.9.3, and bound the 3.0.x resource type-name validation warning before 3.0.2. Share warning formatting and preserve reported spelling, prerelease semantics and exact dirty-release handling.
Verify the upstream producing-boundary patch instead of assuming a file-writer sanitizer. Explicitly label inclusion of the tagged 3.0.0 release line as source-derived; version and host guidance do not establish engine contents or backports. Retain warnings with completed Evidence during cancellation and failed cleanup.
Validation: 222 focused tests in 15 files, 16 baseline controls fail, static/docs checks and two source reviews. Twelve tiny actual MCP workflows and six CLI controls with authored launchers; no real Apktool/JVM, native Windows, exploit or true cold verification.
Fixes#4818
Parse the launcher version once to own both compatibility and CVE-2024-21633 advisory facts. Inspection and decoding consume those facts, including completed Evidence retained during cancellation. Preserve source spelling, exact dirty-release handling, prerelease semantics, and existing version-only failure diagnostics.
Adapt the Apktool warning policy from PR #3440. State that a reported version does not establish engine contents or downstream backports; avoid recommending 2.9.2+ as generally safe. Other advisory applicability and release-range questions are tracked in #4818.
Validation: 194 focused tests in 15 files; final 87 source checks; five baseline controls fail; static/docs checks; two source reviews. Twelve tiny actual MCP workflow comparisons and four CLI controls; authored launcher only, no real Apktool/JVM/Windows or exploit verification.
Fixes#3438
Co-authored-by: Govindu Sasanka <31107592+GovinduSasanka@users.noreply.github.com>
Allow the 23-character script-region-variant representation emitted by Apktool, retaining the shared admission grammar and request aliases. Exact discovery and CLI/MCP selection consume the same bound.
Validation: 67 focused tests across four files; five controls fail on the baseline; static/docs checks and two source reviews. Twelve ordinary tiny CLI/MCP task comparisons preserve Evidence; two actual authored-launcher long-locale CLI/MCP calls select the table. No real Apktool/JVM or native Windows verification.
Fixes#4817
Inventory observed entries through the owned artifact traversal, preserve other configuration names, and disclose non-regular tables with partial coverage. Share request alias normalization with exact table classification and preserve filesystem failures.
Adapt the disclosure policy from #4738. Correct the fixture command separator so all advertised tables are actually created.
Validation: 164 focused tests; check:fast; test discovery; docs:check. Two regression controls fail on the baseline. Twenty-four tiny authored CLI/MCP comparisons; no real Apktool or Windows verification.
Fixes#4612
Co-authored-by: Oke Iyanuoluwa E. <oke.iyanuoluwa12@gmail.com>
Carry caller-selected target identity separately from launcher selection. Only fully collected ordinary nonzero decode exits represent target refusals; startup, preparation, capture and signal failures retain provider attribution. Reuse one command/preparation output projection for probe and decode errors.
Adapt the target-path correction from PR #3462 (08d3d0d3a0) and extend its producer-boundary classification and regression coverage.
Validation: 131 focused tests in 12 files, final 50 source checks, build, typecheck/lint and docs. Ten regression controls fail on baseline 4844ae812. Twenty-four tiny actual CLI/stdio MCP refusal tasks preserve every projected field except the corrected message/path. Authored launchers only; no real Apktool/JVM or native Windows claim. Heavy stress/profiling remained stopped.
Closes#3340
Co-authored-by: Peter <304086198+peter-research@users.noreply.github.com>
