Michael D'Angelo
fc51c1a42a
fix(plugin): require thread attachment for legacy scan continuations ( #2428 )
...
* fix(plugin): scope legacy scan continuations to the attached scan
* fix(plugin): rejoin attached legacy deep scans by target
* fix(plugin): keep attached deep scan results with their conversation
2026-11-10 11:12:55 -08:00
Michael D'Angelo
0a12733373
fix(windows): verify scan output directory ownership and ancestry ( #2432 )
...
* fix(windows): verify scan output directory ownership and ancestry
* fix(windows): read access entries within the ACL buffer
* test(windows): read the native rename access result
* fix(windows): check directory reparse access during output selection
* fix(windows): preserve output link diagnostics and verify native permissions
* test(windows): exercise empty directory removal and runtime admission
* fix(windows): keep known checkpoint reads scoped to selected files
* fix(windows): preserve selected output directories during preparation
* fix(runtime): verify selected POSIX output parents before resolving the leaf
2026-11-09 08:41:41 -08:00
Michael D'Angelo and Codex
1d586acc2b
fix: keep workbench lifecycle results scoped to the selected scan ( #2421 )
...
* fix: keep workbench lifecycle results scoped to the selected scan
* test(plugin): bind lifecycle fixtures to their scan thread
* fix(plugin): bind rejoined scans and load assigned remediation context
---------
Co-authored-by: Codex <codex@users.noreply.github.com >
2026-11-09 08:41:33 -08:00
Michael D'Angelo and mldangelo-oai
884b7b657a
refactor!: consolidate scan, workbench and release rules ( #2452 )
...
* refactor(workbench): share schema repair and saved finding projections
* refactor(cli): resolve the command route once
* refactor(settings): share Deep Scan configuration and restore saved inputs
* refactor(imports): reuse scanner projections at the file boundary
* refactor(publish): share pending request and receipt handling
* refactor(telemetry): classify publication failures in the recorder
* refactor(package): share plugin inventory and archive checks
* refactor(release): reuse verified provenance and release decisions
* fix(ci): reconcile release labels with an empty array on Bash 3.2
* ci: report failed source checks after cancelled PR edits
---------
Co-authored-by: mldangelo-oai <269034524+mldangelo-oai@users.noreply.github.com >
2026-11-09 05:02:21 -08:00
Michael D'Angelo
228003abbd
refactor(deep-scan): stop passing host CPU counts to worker setup ( #2482 )
...
Deep Scan selects workers from per-run settings, legacy TOML, or the bundled default of four. The former automatic setting also resolves to four. Stop sampling host CPUs in the MCP client, setup inspection, and workbench launch. Preserve validation for callers that still send the legacy argument.
Verified workbench startup and the standalone helper with omitted, zero, one, and sixteen as the CPU count. Project setup, saved settings, coordinator resume, and the MCP store integration suite passed on both revisions. The SDK types and format checks and portable plugin checks also passed.
2026-11-09 04:08:25 -08:00
Michael D'Angelo and Codex
977e659713
fix: clear Deep Scan consolidation progress when a scan stops ( #2433 )
...
Co-authored-by: Codex <codex@users.noreply.github.com >
2026-11-07 00:30:57 -08:00
Michael D'Angelo
4d83ce9195
refactor: share workbench arguments and simplify SDK and eval workflows ( #2325 )
...
* Simplify workbench dispatch, SDK helpers, and their test fixtures
* test(sdk): keep distinct progress matrix helpers
* test(sdk): expect canceled lifecycle for aborted component scans
2026-11-05 01:28:44 -08:00
Michael D'Angelo and Codex
3a51de0d54
fix(telemetry): preserve usage coverage and operation outcomes ( #1841 )
...
* fix(telemetry): preserve usage coverage and operation outcomes
* fix(telemetry): reconcile worker receipts and preserve failure outcomes
* fix(telemetry): reconcile cumulative usage and protect exporter settings
* docs(telemetry): clarify hosted remediation ownership
* fix(telemetry): scope coverage recovery and protect helper configuration
* fix(telemetry): keep worker accounting in protected runtime state
* fix(telemetry): preserve output paths and large usage receipts
* fix(telemetry): avoid replaying archived session events
* fix(telemetry): classify budget stops and refresh partial totals
* fix(telemetry): reconcile scope gaps and collection races
* fix(telemetry): retain observed worker receipt totals
* fix(packaging): include telemetry artifacts and update smoke contracts
* fix(telemetry): consume complete compressed streams
* fix(telemetry): preserve recovered usage and coverage provenance
* fix(telemetry): reconcile native usage throughout completion
* fix(telemetry): retain partial usage and publication outcomes
* fix(telemetry): preserve uncertain Linear publication outcomes
* fix(telemetry): reconcile resumed and reused operation receipts
* fix(telemetry): retain measured worker cache-write provenance
* docs(telemetry): clarify inline owner accounting
* fix(hosted): retain foreign execution request identities
* test: retain native preflight observations before shutdown
* fix(telemetry): retain completed patch usage on invalid results
* fix(telemetry): retain patch usage before child shutdown
* fix(telemetry): preserve component measurement observers
* fix(validation): retain usage received before shutdown failures
* test: share Deep Scan worker usage fixtures
* fix(telemetry): retain the first partial usage timestamp
* test: let skill process fixtures accept Codex arguments
* ci: retry managed Code Quality after GitHub installation rate limit
* Preserve stored scan costs when reconciling canceled usage
* Use the fixture error class in the native coordinator test runner
* Remove stale coordinator error imports after test extraction
* Finalize discovery attempt receipts after durable worker acceptance
* Align operation outcomes and worker usage after budget integration
* Classify native skill startup cancellations in operation telemetry
* Distinguish skill deadlines from explicit child cancellation
* Retain workbench coverage warnings in terminal usage receipts
---------
Co-authored-by: Codex <codex@users.noreply.github.com >
2026-11-04 16:56:20 -08:00
c9cb6b4bca
fix(cost): enforce scan budgets from verified usage ( #465 )
...
* fix(cost): enforce verifiable scan spending limits
* fix(cost): reject incomplete final session usage
* fix(cost): verify delegated worker completion and ownership
* fix(cost): require completed root evidence for scan budgets
* fix(cost): preserve definitive budget overages
* fix(api): retain definitive budget failure classification
* fix(cost): retain overages discovered during cleanup
* fix(cost): retain highest usage through budget cleanup
* fix(cost): retain verifiable per-session accounting
* fix(cost): respect proven worker ownership
* fix(cost): exclude unrelated historical accounting evidence
* fix(cost): reverify late final accounting
* fix(cost): retain fresh usage from missing-session reads
* fix(cost): preserve unpriced completed usage
* fix(cost): preserve finalization and usage boundaries
* fix(cost): reject unusable owned accounting
* fix(cost): accumulate reset-aware session usage
* fix(cost): preserve final accounting and worker shutdown
* fix(sdk): exclude internal helpers from public declarations
* fix(cost): align whole-counter reset accounting
* fix(cost): exclude inherited fork usage
* fix(cost): preserve full UUIDv7 fork ordering
* fix(cost): reject incomplete worker session metadata
* Attribute incomplete rollouts through trusted session ownership
* Verify ownership of every unidentified budgeted rollout
* Account for equal-total token counter resets
* Verify complete scan accounting before budget finalization
* Attribute fork-only workers during live budget tracking
* fix: avoid double-counting custom validation usage
* fix(cost): refresh the cached accounting plugin
* fix(cost): preserve owned accounting and completed worker cancellation
* fix(cost): preserve selected runtime accounting and late cache usage
* Preserve native SQLite environment precedence
* Format the merged scan ownership and worker controls
* Read worker invocation before checking SQLite forwarding
* Adapt ownership checks to current environment and test helpers
* Format the current ownership test interface
* Resolve the native executable only when ownership fallback needs it
* test: retain cancellation while completed workers drain
* test: type completed worker callback fixture
* fix(cost): verify native owned rollout accounting
* fix(cost): verify rollouts archived after polling
* fix(cost): reconcile live archives and native SQLite precedence
* test(cost): type SQLite override fixtures
* fix(policy): reuse the launched environment for usage ownership
* fix(cost): bind owned rollouts and preserve unbudgeted completion
* test(cost): supply owned sessions in live budget fixture
* fix(workers): preserve shutdown behavior for unbudgeted scans
* fix(cost): include custom-validation session ownership
* fix(cost): retain retry ownership and consistent SQLite paths
* test(cost): check native worker directory argument
* test(workbench): derive current schema checks from migrations
* fix(cost): preserve effective SQLite settings during ownership checks
* fix(cost): preserve native paths and cumulative accounting
* fix(cost): align persisted usage and worker storage
* fix(cost): preserve late cache classification across usage snapshots
* fix(cost): retain partial cache refinements across sessions
* fix(cost): retain separately reported cache categories
* fix(sdk): retain selected settings in scan resume recipes
* fix(cost): reconcile final receipts before budget callbacks
* test(ci): print the random seed before running each shard
* fix(deep): interrupt native workers and preserve final accounting
* test(deep-scan): exercise graceful worker shutdown in CI fixtures
* fix(deep-scan): translate cyber settings for the native protocol
* test(api): await socket closure after cancellation resets
* test(runtime): create private session ownership fixtures
* fix(deep): preserve native worker warning diagnostics
* Use Bun runtime with the Windows event-loop fix in CI
* Preserve inherited and overridden SQLite worker settings in tests
* fix(cost): verify compressed owned session rollouts
* Exercise log retries after recording worker ownership
* Decode complete compressed receipts on supported Node releases
* test(package): exercise native Deep worker lifecycle and decoder artifact
* test(cost): align budget fixtures with retained diagnostics and ownership checks
* test(logs): isolate Node capability fixture compilation
* test: preserve app-server worker launch coverage
* test: keep preflight fixture independent of SDK source
* test: grow tracked session before exercising read retries
* fix(usage): retain late legacy cache counts across turns
* test(api): align budget recovery with canonical completion
* test(deep-scan): split worker lifecycle coverage
* test(deep-scan): verify provider settings over app-server RPC
* fix: retain worker profiles and graph ownership after main integration
* test: align scan fixtures with merged runtime
* test: preserve portable policy and log fixtures
* fix(sdk): read session logs from the beginning
* test: align scan budget fixtures with native worker contracts
* test: preserve bundled worker helpers and normalized scan settings
* test(ci): allow time for the serial Windows baseline
* test(cost): retain budget checks with optional owner reports
* test(ci): preserve existing production Bun qualification
* test: stop polling worker markers after the fixture deadline
* style: format the worker marker condition
* test(evals): include SDK decoder types in the eval build
* test(workers): exercise the app-server directory and scope cleanup on Windows
* test(deep-scan): isolate worker fixture phase markers
* test(sdk): restore Unicode separators in scan input fixtures
* test(cost): keep simulated workers active through live polling
* fix(cost): distinguish fresh and resumed worker retries
* fix(cost): retain unverified worker attempts across later retries
* test(workbench): pin checkpoint upgrade fixtures to version 57
* test(workbench): restore the historical Deep Scan schema accurately
* test(workbench): upgrade frozen Python checkpoint fixtures from their release schema
* test(workbench): keep fixture updates outside assertions
* ci: retry managed Code Quality after GitHub installation rate limit
* Bound the live budget approval fixture with a single watchdog
* test(sdk): expect canceled lifecycle for aborted component scans
(cherry picked from commit 61d32357f4 )
* test: import the shared coordinator failure fixture
(cherry picked from commit 925b70af34 )
* test(sdk): expect cancellation from an aborted component scan
* Attribute cumulative cache tokens only to the reported categories
* Keep contradictory cumulative measurements out of verified usage
* Retain genuine over-capacity cache warnings and remove stale test imports
* Remove the local dependency symlink from the plugin checkout
---------
Co-authored-by: mldangelo-oai <269034524+mldangelo-oai@users.noreply.github.com >
Co-authored-by: Codex <codex@users.noreply.github.com >
2026-11-04 15:59:18 -08:00
Michael D'Angelo
5c0be2444b
refactor: clarify Python contracts and simplify scan recovery and reporting ( #2340 )
...
* docs: document Python module and function contracts
* refactor: simplify scan recovery and reporting
* test(deep-scan): restore the coordinator stop error import
2026-11-04 15:55:14 -08:00
Michael D'Angelo
366093f77b
fix(deep-scan): report completed audits and batch consolidation ( #2305 )
...
* fix(deep-scan): report completed audits and batch consolidation
* fix(deep-scan): include measured usage in the completion receipt
* docs(deep-scan): use the completion receipt for final reporting
* ci: retry managed Code Quality after GitHub installation rate limit
* test(deep-scan): restore the coordinator stop error import
(cherry picked from commit 7630eea34c )
2026-11-04 15:30:51 -08:00
Michael D'Angelo
aba412dbc8
fix(workbench): preserve archived deep scans and streamline progress reads ( #2314 )
...
* fix(workbench): preserve archived deep scans and streamline progress reads
* fix(workbench): recover archived paths written by older versions
* fix(workbench): retain the full scan read for existing callers
* test(workbench): archive a completed scan through the packaged SDK
* ci: retry managed code scanning after GitHub rate limiting
* perf(workbench): read only progress fields when polling scans
* ci: retry managed Code Quality after GitHub installation rate limit
* Assert the canceled lifecycle emitted by an aborted component scan
2026-11-04 15:27:11 -08:00
8bede3b191
fix(deep-scan): avoid worker dispatch for empty targets ( #2054 )
...
* fix(deep-scan): avoid worker dispatch for empty targets
* fix(deep-scan): honor requested paths during inventory preflight
* fix(deep-scan): preserve scoped preflight compatibility
* fix(deep-scan): respect inherited ignore rules during preflight
* fix(deep-scan): preserve preflight target semantics
* fix(deep-scan): align preflight with shared inventory
* fix(inventory): align scoped tracked enumeration
* fix(deep-scan): reject physically empty fresh targets
* fix(deep-scan): coordinate empty-target admission
* fix(deep-scan): preserve legacy recovery and abort responsiveness
* fix(deep-scan): isolate directory probe protocol
Preserve native filename bytes while keeping directory admission responsive to cancellation and discovery deadlines. Transport probe results over a dedicated IPC pipe so inherited stdout and stderr instrumentation cannot corrupt success or filesystem errors.
Raw-byte filename handling and the one-entry buffer probe adapt ideas from Dipesh Babu’s PR #2181 : https://github.com/openai/codex-security/pull/2181
---------
Co-authored-by: Muzzy5150 <210135147+Muzzy5150@users.noreply.github.com >
Co-authored-by: mldangelo-oai <269034524+mldangelo-oai@users.noreply.github.com >
Co-authored-by: dipeshbabu <59379458+dipeshbabu@users.noreply.github.com >
2026-11-04 14:18:15 -08:00
b6d739b99a
fix(deep-scan): stop on known policy refusal messages ( #1035 )
...
* fix(deep-scan): recognize the complete cybersecurity refusal message
Refused turns report the provider's complete message, which follows
"This content was flagged for possible cybersecurity risk." with Trusted
Access guidance. The exact-match refusal set listed only the first
sentence, so refused Deep Scan workers were resumed and retried as
transient errors instead of being retired. Add the complete message and
cover it in the classifier, executor, and coordinator tests.
Refs #276
* fix(deep-scan): stop on terminal cybersecurity refusals
* test(deep-scan): avoid scheduling race in refusal regression
* test: reuse accepted source path coverage
* refactor(deep-scan): share terminal worker error handling
* test: reuse Deep Scan worker request fixtures
* fix(deep-scan): honor terminal worker failures across scheduler races and handoff
* fix(deep-scan): persist a terminal worker failure before executor cleanup
* test(deep-scan): compare scheduler dispatch counts at the discovery deadline
* fix(deep-scan): avoid resending failed worker thread metadata
---------
Co-authored-by: mldangelo-oai <269034524+mldangelo-oai@users.noreply.github.com >
Co-authored-by: Michael D'Angelo <mdangelo@openai.com >
2026-11-04 14:09:11 -08:00
ianw-oai and mldangelo-oai
cb92c4fb41
feat!: deep scan upgrade ( #2118 )
...
* refactor(deep-scan): merge repeated Standard scans in the host
* test(deep-scan): cover comparison turns and resumed discovery
* refactor(deep-scan): simplify finding counts and receipt handling
* test(deep-scan): separate comparison and runtime test registration
* test(deep-scan): exercise supported turns when replaying saved profiles
---------
Co-authored-by: mldangelo-oai <269034524+mldangelo-oai@users.noreply.github.com >
2026-11-03 17:48:49 -08:00
Aismit
6c33be7b0f
feat(sdk): preserve the severity rubric across standalone scan attempts ( #2258 )
...
* feat(sdk): preserve the severity rubric across standalone scan attempts
* test(sdk): allow time for disk-backed scan retries
2026-11-02 19:44:57 -08:00
Michael D'Angelo
223dea6390
fix(plugin): inherit the launching turn's Deep Scan speed ( #2231 )
...
Persist the service tier supplied by the host when a scan starts and use it for fresh and resumed discovery and reducer workers. Keep the configured fallback when the host omits the tier.
2026-11-02 16:57:57 -08:00
Michael D'Angelo and mldangelo-oai
f615be2e70
refactor(workbench): share saved coverage logic and checkpoint fixtures ( #2185 )
...
Co-authored-by: mldangelo-oai <269034524+mldangelo-oai@users.noreply.github.com >
2026-11-01 01:38:05 -07:00
Michael D'Angelo
5c1bd494c6
fix(plugin): retain the reviewed surface count for completed Deep scans ( #2161 )
2026-10-30 13:02:31 -07:00
ianw-oai
621bad47e6
refactor!: simplify scan completion and recovery ( #1762 )
2026-10-26 17:19:54 +00:00
Michael D'Angelo
d329d8dce5
fix(deep-scan): finalize runs, preserve findings, and honor project budgets ( #1912 )
...
* fix(deep-scan): preserve results and honor saved project budgets
* fix(deep-scan): recover results after a target caller detaches
* fix(deep-scan): preserve completion ownership and recovery across upgrades
* fix(deep-scan): keep retained findings compatible with direct reducer tools
* fix(deep-scan): return the scan ID when finalization fails
2026-10-25 19:05:34 +08:00
Michael D'Angelo
3e0bf51bba
fix(deep-scan): retain unfinished deadline checkpoints ( #1710 )
2026-10-23 19:18:47 +00:00
Michael D'Angelo
57f698ebc1
fix(deep-scan): recover checkpoints at discovery deadlines ( #1594 )
2026-10-24 01:29:36 +08:00
Michael D'Angelo
2c35052cf5
fix: preserve complete scan failure diagnostics ( #1884 )
2026-10-24 01:23:40 +08:00
Michael D'Angelo
4a550a6041
fix(workbench): read excerpts from the scanned source ( #488 )
2026-10-20 04:44:03 +08:00
Michael D'Angelo
c380befb9e
fix(scans)!: simplify scan outcomes ( #1633 )
...
Simplify scan outcomes across the CLI, SDK, and GitHub Action. Completed report-only scans return 0, an opt-in findings threshold returns 1, and execution or required-output failures return 2. Remove partial and unknown outcomes while retaining coverage diagnostics and saved results.
Require the current plugin for Deep Scan and canonical fields for new findings. Preserve readability of historical sealed reports. Update documentation and tests, and remove obsolete compatibility paths.
2026-10-14 11:32:39 -07:00
Michael D'Angelo
ecc637beee
fix(deep-scan): preserve selected scope through recovery ( #1602 )
...
* fix(scan): normalize scoped findings and clean worker inputs
* test: match synchronous worker artifact server configuration
* fix(scan): preserve original scoped path kinds through recovery
* fix(scan): preserve recipes without selected file paths
2026-10-12 18:55:13 -07:00
Michael D'Angelo
ddffc29eea
fix(deep-scan): prevent replaced coordinators from overwriting scan results ( #916 )
...
* fix(plugin): fence stale Deep publication and preserve replay
* test(plugin): preserve legacy Deep publication replay
* Preserve Deep parent drafts and reducer publication order
* Read reducer sequence from its worker directory
* fix(deep): preserve terminal aggregates during completion replay
* Preserve partial Deep setup drafts before orchestration
* fix(deep-scan): retain terminal legacy publication during replay
* test(deep-scan): retain actual SQLite replay crash boundaries
* fix(deep): preserve selected legacy publications during completion replay
* fix(deep): preserve legacy progress and selected-result replay
* style(deep): format selected publication check
* fix(deep): retain parent publication after legacy lease claims
* fix(deep): align publication with recovered reducer selection
* test(workbench): track the current migration history
* fix(deep): retain progress after publication selection
2026-10-12 18:30:10 -07:00
Michael D'Angelo
5a7788b54a
fix(scan): enforce Deep Scan scope across execution adapters ( #1557 )
...
* refactor(scan): run CLI Deep Scan through the shared coordinator
* fix(package): include the direct Deep Scan driver
* fix(scan): refresh permissions through native Deep Scan resume
* fix(scan): retain isolated Python workbench startup
* fix(scan): preserve complete worker usage in direct scans
* fix(scan): preserve standalone Python dependency lookup
* fix(scan): preserve Bun scan compatibility
* fix(scan): preserve Deep Scan inputs across execution adapters
* test(scan): allow Windows helper paths in Bun routing checks
* fix(scan): preserve access advisories and completed responses
* fix(scan): preserve explicitly selected MCP Node runtimes
* ci: allow Windows SDK shards to finish
* fix(scan): enforce Deep Scan finding scope on resume
* fix(scan): keep worker scope lists off command lines
* test(scan): await bundled worker artifact setup
* Stabilize inherited-pipe fixture readiness
2026-10-11 15:10:36 -04:00
mldangelo-oai and Codex
826c556238
fix(deep-scan): match user context before reusing discovery ( #1048 )
...
* fix: preserve severity assessments for each scan
* Verify coordinator claims preserve the current schema version
* Require scan provenance when migrating severity assessments
* fix(deep-scan): match user context before reusing discovery
* test(deep-scan): separate request identity regressions
* fix(deep-scan): retain original discovery context
* fix(deep-scan): repair recorded discovery context migration
* test: preserve unrelated migration records in severity fixture
---------
Co-authored-by: Codex <codex@users.noreply.github.com >
2026-10-08 15:23:06 -04:00
mldangelo-oai
1163374f6c
fix(cli): show active reviews and result consolidation during Deep Scan ( #1253 )
...
* fix(cli): show durable Deep Scan progress
* Keep Deep Scan consolidation progress after terminal polls
* Resume discovery progress when new Deep Scan reviews start
2026-10-07 22:51:36 -04:00
mldangelo-oai
f82ae5be89
fix: simplify action cleanup and stabilize provider isolation tests ( #1421 )
...
* fix: simplify action cleanup and resolve code quality warnings
* test: remove repeated native startup from provider isolation checks
2026-10-07 22:42:38 -04:00
ff450186e1
fix(plugin): load saved feedback and align scan completion instructions ( #1264 )
...
* fix(plugin): align scan instructions and generated reports
* fix(sdk): align custom validation with the updated scan workflows
* test(plugin): cover literal backticks in report paths
* fix(plugin): escape reachability fallback once
---------
Co-authored-by: Codex <codex@users.noreply.github.com >
Co-authored-by: mldangelo-oai <269034524+mldangelo-oai@users.noreply.github.com >
2026-10-07 15:16:44 -04:00
mldangelo-oai
c39bbaf9dc
refactor(plugin): simplify workbench persistence ( #1226 )
2026-10-04 15:07:32 -07:00
mldangelo-oai and Codex
37cd0ed191
refactor(plugin): simplify workbench owner calls ( #1205 )
...
The Python workbench carries forwarding functions and repeated data handling around existing owners. Call those owners directly and consolidate equivalent operations.
Co-authored-by: Codex <noreply@openai.com >
2026-10-03 15:06:54 -07:00
mldangelo-oai and Michael D'Angelo
455544132a
fix(plugin): close saved review tasks and simplify recovery ( #1036 )
...
* fix(plugin): finalize diff scans and preserve resolved work
* fix(plugin): restore Windows recovery and simplify checkpoint handling
* fix(plugin): bind diff digests during finalization
* docs(plugin): explain workbench digest binding
* refactor(plugin): close saved review tasks by ID
* fix(plugin): preserve parent authority during recovery
* fix(plugin): preserve sealed digests through later recovery
* refactor(plugin): simplify saved review reconciliation
* fix(plugin): persist worker surface IDs before checkpoints
* fix(plugin): preserve draft identities and reopened evidence
* fix(plugin): retain subsequent progress after reopening
* fix(plugin): preserve accepted candidate outcomes during progress
* refactor(plugin): simplify saved review reconciliation
* fix(plugin): preserve pending task updates during recovery
* fix(plugin): preserve accepted review state during publication recovery
* fix(plugin): normalize recovered tasks and candidate closures
* chore(test): remove unused runtime test import
* fix(plugin): preserve legacy review identities and file-authored coverage
* fix(plugin): keep rejected inventory drafts repairable
* fix(plugin): keep raw drafts without promoting rejected heads
* fix(plugin): roll back rejected completion state and preserve legacy surfaces
* chore(plugin): keep workbench within portable source limit
* fix(plugin): retain selected worker checkpoint coverage
* fix(plugin): read recovery timestamps with their snapshots
* fix(recovery): retain valid tasks after malformed updates
* fix(recovery): retain legacy surface identities during closeout
* fix(recovery): preserve legacy work across interrupted closeout
* fix(recovery): keep reopened work incomplete and preserve task ownership
* fix(completion): validate the selected checkpoint before sealing
* fix(recovery): restore selected checkpoint semantic context
* fix(recovery): validate reported outcomes and selected coverage
* fix(recovery): preserve ambiguous legacy surfaces during replacement
---------
Co-authored-by: Michael D'Angelo <269034524+mldangelo-oai@users.noreply.github.com >
2026-10-02 23:36:22 -07:00
Adam Hupp and Codex
2aab4e90ab
fix(plugin): create private scan output directories ( #987 )
...
* fix(plugin): create private scan output directories
* Create completion lock directories privately
---------
Co-authored-by: Codex <codex@users.noreply.github.com >
2026-09-27 13:36:17 +01:00
Dane Schneider
fc28739f28
fix(deep-scan): remove coverage from aggregation ( #767 )
...
Make Deep scan publication depend on coordinator completion rather than model-authored worker coverage. Keep reducers focused on findings and preserve partial results when the overall scan is stopped or cut off.
2026-09-03 12:48:02 -07:00
akosiakov-oai
0bc42c69d3
feat(sdk): expose deep scan progress ( #710 )
...
* feat(sdk): expose deep scan progress
* fix(sdk): avoid blocking on deep progress
* fix(sdk): allow deep progress package files
* test: cover deep progress maximum
2026-08-28 12:42:44 -07:00
soyeon-oai
42a60ff473
refactor(plugin): generate the bundled runtime from canonical source ( #675 )
...
* refactor(plugin): establish canonical source tree
* build(plugin): generate the npm runtime payload
* ci(plugin): reject tracked runtime payload
* feat(plugin): add patch risk assessment
* fix(plugin): reconcile portable runtime contracts
* test(plugin): add deterministic triage evals
* ci(plugin): validate portable source contracts
* fix(package): preserve bundled MCP launcher permissions
* fix(plugin): reconcile scan runtime and recovery contracts
* fix(plugin): align finding evidence schemas and scan guidance
* fix(plugin): resolve completed checkpoint follow-ups
* fix(plugin): restore preflight and Windows build contracts
* fix(plugin): preserve portable build and recovery contracts
* fix(plugin): restore portable CI contracts
* fix(ci): install ripgrep for MCP tests
* fix(ci): type workflow job lookup
* test(plugin): preserve resumable completion failures
2026-08-27 11:40:42 -07:00