* fix(setup): write the recommended tool call timeout into dsh registrations
rea setup --client dsh wrote registrations without toolCallTimeoutMs, so
DeepSeek Harness applied its 60-second default while a native provider
startup alone can exceed it. Setup now derives the deadline from the same
provider allowance plus operation headroom used for Kimi and Codex,
preserves a positive caller-selected value, and doctor reports a
registration without a usable timeout as stale.
ProviderRequestBudgetResolver declares an optional budgetSettings read
set, so the fail-closed budget gate rejects unresolved Cordis expressions
only for settings the derivation may read plus the timeout inputs; an
unrelated expression such as GHIDRA_INSTALL_DIR still refreshes, and an
expression in REA_IDA_MCP_CONFIG fails closed.
* test(setup): record the dsh tool call deadline in the NUL repair fixture
The NUL refusal scenario rewrites a valid dsh registration and expects
doctor to report it healthy. That fixture omitted toolCallTimeoutMs,
which doctor now reports stale, so the repaired registration records the
deadline a current dsh registration carries.
* test(setup): separate dsh timeout repair coverage
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(evidence): derive default authority from subject binding
evidenceEnvelopeForAdmission defaulted a missing authority to shipped-artifact
for every observation, including ones with no artifact subject and ones
produced by a live debugger session. An unbound record therefore claimed
authority over an artifact it cannot name, and live GDB, Rizin and Cutter
observations could never qualify a live-observation contract because that
contract accepts only controlled-replay or external-service Evidence.
Default to analyst-inference when no subject is bound and keep
shipped-artifact when one is, so the default matches what the record can
actually support. Declare controlled-replay at the eight live debugger
observation sites (GDB 6, Rizin 1, Cutter 1) so those records describe the
controlled replay they performed and satisfy the documented contract.
* fix(gdb): label retained process-exit observations as controlled replay
The GDB command process-exit branch was the only unbound retention path that
named no authority, so this branch's new default derived it as
analyst-inference. That label cannot support the live-observation contract,
and every sibling path (timeout, cancellation, protocol failure, startup)
already states controlled-replay.
The authority and the unavailable-subject reason now live beside the
provider identity so the retention paths read them from one place instead
of repeating the literals and drifting apart.
* style(evidence): apply repository formatting
* fix(cutter): classify command observations as external-service
The bridge drives the documented already-running Cutter GUI, whose
document state REA neither owns nor resets, so its command
observations claim no controlled replay. Label them external-service
like the comparable IDA and Frida bridge evidence: live-observation
obligations still accept it, while controlled-replay-only verifier
contracts and bundle replay scenarios no longer include these
observations.
* fix(adb): keep device observations at external-service authority
The admission default now derives authority from the subject, so a subjectless ADB observation fell to analyst-inference. A device is a service REA attaches to rather than drives, so the observation supports external-service and must stay eligible for live-observation contracts.
Carry an explicit authority through AnalysisObservation so a producer that knows its own authority can name it, and set it for every ADB operation.
* fix(adb): keep retained observations at external-service authority
The admission default derives authority from the subject, and a subjectless
observation falls to analyst-inference. createAnalysisExecution never carried
an authority field, so every observation the ADB adapter built through it was
committed at the subject-derived default instead of the service's.
Name the authority once in the domain model, stamp it on every successful
execution at the adapter boundary, and route the cleanup-failure and
cancellation-retention sites through one helper. Open the execution options
field so other providers can name their own authority.
Verified: a push whose device digest does not match now retains its receipt at
external-service; removing the stamp fails that assertion. check:fast passes.
* fix(adb): keep the retained verification receipt at device authority
The verificationFailed branch built its retained receipt with createEvidence
directly, so the admission default derived the authority from a subjectless
target and labelled an observed push as analyst-inference. Pass the device
authority through the same way the neighbouring digest-mismatch partial does,
so the receipt can still support live-observation verification.
* style(adb): apply oxfmt formatting to the authority fix
The previous commit wrapped two expressions that oxfmt keeps on one line, so
npm run format:check failed on Static checks.
* fix(decompile): repair a torn journal on Windows before resuming
Windows opens an O_APPEND handle with FILE_APPEND_DATA only, so ftruncate
through the journal's append handle fails with EPERM and every resume after
an interrupted write was refused. Truncate through a separate write handle
first, then open the append handle.
* docs(decompile): say why the journal opens without following symlinks
fix(javascript): keep wrapper redeclarations from inventing application owners
A top-level lexical declaration that reuses a CommonJS wrapper parameter
cannot evaluate, and its recovered syntax is complete. Counting that
diagnostic as dropped syntax marked the entrypoint as an unexamined loader
and erased another package's proven loadFile root.
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: morluto <morluto@users.noreply.github.com>
The Windows-only Fx case still read mcp.rea, but a new Fx profile is written
under mcpServers. Two Fx cases assert POSIX modes or a private-directory
close that Windows never performs; they now run off Windows, beside the
existing Windows case. Two release-exposure assertions spelled paths with a
POSIX separator; one now accepts either separator and one uses the reader's
host-file/member spelling.
Inspect local Maven/Gradle metadata as Evidence with bounded report/XML admission, preserved caller choices and owned cleanup.
Refs #8828; broader comparison and resolution work remains outside this change.
Assisted-by: Suika (gensokyo-skills:suika-scatter-gather-planning)
Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>