Repository navigation
Releases: github/gh-aw
Release list
v1.0.1
🌟 Release Highlights
v1.0.1 focuses on safer pull request outputs, smoother compile and update workflows, and improved observability.
✨ What's New
- Lock-safe pull request outputs: code-producing workflows now use lock-safe PR outputs, with a suggested lock exclusion when workflow permission pushes fail (#72396, #72423).
- Actions-lock integration:
compilecan now optionally integrate with actions-lock (#72395). - Guided
pull_request_targetfixes: guided remediation with safe role restriction (#72412). - Interactive org and enterprise update rollouts (#72418).
- Native Actions expression grammar replaces legacy expression parsing (#72413).
- Richer unified sessions: safe-output, safe job/script, and memory-save outcomes are now recorded (#72146, #72138, #72133), plus harness retry telemetry in spans and usage artifacts (#71934).
- Copilot CLI upgraded to 1.0.103 with a reviewed checksum pin (#72099).
🐛 Bug Fixes & Improvements
- Preserve cancelled startup failure diagnostics and suppress noisy startup-failure reports for cancelled jobs (#72432, #72405).
- Fix privileged AWF environment forwarding on sudo-rs (#72409).
- Merge imported job outputs with main workflow precedence (#72407).
- Honor newer cached action pins in runtime setup (#72272).
- Trust same-repository bot PRs triggered by labels (#72074).
- Enforce
ubuntu-slimtimeout limits and upgrade default runners for longer jobs (#72131). - Fill model pricing gaps and accept
none/maxfor pinned Codex routing (#72414, #70952). - Harden AWF release download retries and fix work-queue provisioning recovery (#72157, #72400).
📚 Documentation
- New explanation of the activation gate and CAO integration (#72379) and a leaner troubleshooting guide (#72280).
Generated by 🚀 Release · copilot · auto · ~15 AIC · ⊞ 12.9K
What's Changed
- Trust same-repository bot PRs triggered by labels by
@pelikhanwith@Copilotin #72074 - Upgrade Copilot CLI to 1.0.103 with reviewed checksum pin by
@pelikhanwith@Copilotin #72099 - [blog] Agent of the Day – 2026-10-26 by
@github-actions[bot] in #72127 - Soften container pin update failure messages by
@dsymein #72093 - Use Gonum helpers for grader descriptive statistics by
@pelikhanwith@Copilotin #72100 - Correct map tool usage in built-in ledger smoke workflow by
@pelikhanwith@Copilotin #72095 - Make setup action reference tests deterministic by
@pelikhanin #72155 - Enforce ubuntu-slim timeout limits and upgrade default runners for longer jobs by
@pelikhanwith@Copilotin #72131 - Record memory save outcomes in unified sessions and conclusion data by
@pelikhanwith@Copilotin #72133 - Model routing: accept none/max for pinned Codex and add routed Codex smoke by
@SivaKesava1with@Copilotin #70952 - Preserve harness retry telemetry in spans and usage artifacts by
@pelikhanwith@Copilotin #71934 - Record safe job and script outcomes in unified sessions by
@pelikhanwith@Copilotin #72146 - Record privacy-preserving safe-output outcomes in unified sessions by
@pelikhanwith@Copilotin #72138 - Fix Aider Go test stub package declarations by
@pelikhanin #72164 - Harden AWF release download retries and diagnostics by
@pelikhanin #72157 - Preview Notices by
@salilsubin #72217 - Revert "Preview Notices (#72217)" by
@pelikhanin #72281 - Retry transient PR Sous Chef API failures by
@pelikhanin #72282 - [docs] docs: unbloat troubleshooting/common-issues.md by
@github-actions[bot] in #72280 - [log] Add debug logging to unlogged branches in cli package-manifest/actionlint code by
@github-actions[bot] in #72299 - [instructions] Sync instruction files with release v0.93.5 by
@github-actions[bot] in #72363 - Honor newer cached action pins in runtime setup by
@pelikhanwith@Copilotin #72272 - docs: explain the activation gate and CAO integration by
@pelikhanin #72379 - chore: update action and container pins by
@pelikhanin #72377 - build(deps-dev): bump fast-uri from 3.1.8 to 3.1.9 in /actions/setup/js in the npm_and_yarn group across 1 directory by
@dependabot[bot] in #71506 - build(deps): bump undici from 6.29.0 to 6.29.1 in /actions/setup/js in the npm_and_yarn group across 1 directory by
@dependabot[bot] in #72384 - build(deps): bump library/alpine from
294b683tobe26560by@dependabot[bot] in #72393 - Refactor arXiv paper memory to use a built-in ledger by
@pelikhanin #72394 - build(deps): apply pending Dependabot updates by
@pelikhanin #72397 - [docs] Update documentation for features from 2026-10-27 by
@github-actions[bot] in #72406 - Fix work-queue provisioning transport recovery by
@pelikhanin #72400 - fix: fill model pricing gaps and restore 1ES-compatible Vite by
@pelikhanin #72414 - feat: add optional actions-lock integration to compile by
@pelikhanin #72395 - Suppress startup-failure reports for cancelled agent jobs by
@pelikhanwith@Copilotin #72405 - Add guided pull_request_target fixes and safe role restriction by
@pelikhanin #72412 - Merge imported job outputs with main workflow precedence by
@pelikhanin #72407 - Add lock-safe pull request outputs and migrate code-producing workflows by
@pelikhanin #72396 - Add interactive org and enterprise update rollouts by
@pelikhanin #72418 - Replace legacy expression parsing with native Actions grammar by
@pelikhanin #72413 - Suggest lock exclusion for workflow permission push failures by
@pelikhanin #72423 - fix: refresh vendored Node image pin to patched build by
@pelikhanin #72382 - [blog] Agent of the Day – 2026-10-27 by
@github-actions[bot] in #72460 - Preserve cancelled startup failure diagnostics by
@pelikhanin #72432 - [dead-code] chore: remove dead functions — 3 functions removed by
@github-actions[bot] in #72463 - Refactor issue and comment Markdown into report trees by
@pelikhanin [https://github.com//pull/72415](https://github.com/github/gh-aw/p...
v1.0.0
🔧 Maintenance Release
v1.0.0 focuses on housekeeping: documentation refreshes, GitHub Actions version updates, improved debug logging, and expanded schema coverage.
📚 Documentation
- Added an AI Transparency Notice (#71520).
- Refreshed the glossary, architecture diagram, and synced instruction files.
🔧 Under the Hood
- Added debug logging to
agentdrainandactionpinsfor easier troubleshooting. - Added a schema coverage demo for the
work-queue-policyfield. - Updated the Safe Outputs conformance checker for recent spec changes.
Generated by 🚀 Release · copilot · auto · ~13 AIC · ⊞ 12.6K
What's Changed
- [docs] Update glossary - weekly full scan by @github-actions[bot] in #72068
- [architecture] Update architecture diagram - 2026-10-26 by @github-actions[bot] in #72054
- [instructions] Sync instruction files with release v0.93.5 by @github-actions[bot] in #72051
- [spec-review] Update Safe Outputs conformance checker for recent spec changes by @github-actions[bot] in #72004
- [log] Add debug logging to remaining unlogged functions in agentdrain and actionpins by @github-actions[bot] in #71993
- [schema-coverage] feat: Add schema coverage demo for
work-queue-policyfield by @github-actions[bot] in #72028 - [actions] Update GitHub Actions versions - 2026-10-26 by @github-actions[bot] in #72037
- AI Transparency Notice by @salilsub in #71520
Full Changelog: v0.93.6...v1.0.0
v0.93.6
🌟 Release Highlights
A focused release strengthening infrastructure dependencies, compile/audit reliability, and validation.
✨ What's New
- Updated runtime stack: MCP Gateway bumped to v0.5.0 and the firewall to v0.29.0.
- Better expression validation: missing step references in Actions/AW expressions are now detected at compile time.
- Resilient MCP tooling: compile and audit timeouts now return stage-aware partial results instead of failing outright.
🐛 Bug Fixes & Improvements
- Threat detection that is skipped no longer reports inherited agent tokens.
- Fixed a compile-progress file permission policy violation.
- Hardened fetch-script portability and structured lint reporting.
- Cached state is now read before workflow analysis, and title-parsing benchmarks are separated from file I/O costs.
Generated by 🚀 Release · copilot · auto · ~13 AIC · ⊞ 12.6K
What's Changed
- Separate workflow title parsing from file-I/O benchmark costs by @pelikhan with @Copilot in #71927
- Read cached state before workflow analysis by @pelikhan with @Copilot in #71878
- Detect missing step references in Actions/AW expressions by @pelikhan with @Copilot in #71926
- Return stage-aware partial results for MCP compile and audit timeouts by @pelikhan with @Copilot in #71886
- Fix compile-progress file permission policy violation by @pelikhan with @Copilot in #71947
- Prevent skipped threat detection from reporting inherited agent tokens by @pelikhan with @Copilot in #71933
- Harden fetch-script portability and structured lint reporting by @pelikhan with @Copilot in #71932
- Bump MCPG to v0.5.0 and firewall to v0.29.0 by @lpcox with @Copilot in #71957
- [blog] Weekly blog post – 2026-10-26 by @github-actions[bot] in #71984
- [community] Update community contributions in README by @github-actions[bot] in #71976
Full Changelog: v0.93.5...v0.93.6
v0.93.5
🌟 Release Highlights
This release focuses on hardening: stronger static analysis linters, safer credential handling, and more robust diagnostics.
✨ What's New
- Safe-output integration jobs can now use network credentials (#71833).
- Engine driver crashes are classified with secret-safe stderr diagnostics, making failures easier to triage (#71883).
- Pi CLI default updated to 1.4.0 (#71899).
- New and sharper custom linters:
readerstringconvanalyzer, parenthesizedFlushdiscard detection, missing goroutine recovery through local closures, and closed detection gaps (#71891, #71896, #71897, #71880). - Secret-leak analysis now tracks credential handoffs through step outputs (#71850).
🐛 Bug Fixes & Improvements
- Fixed generated conclusion references and fetch warnings; Redis image pinned (#71898).
- Guarded indexing boundaries and prevented cron-step overflow (#71889).
- Refreshed Node image pins and enforced patched runtime dependencies (#71879).
gh-aw-detectionis no longer treated as experimental (#71848).- Internal consolidation of permission scope expansion, GitHub Contents base64 decoding, and public API GET fallback transport (#71888, #71894, #71893).
📚 Documentation
- Documentation refreshed for recent features (#71867).
Generated by 🚀 Release · copilot · auto · ~14 AIC · ⊞ 12.6K
What's Changed
- Allow network credentials in safe-output integration jobs by @pelikhan in #71833
- [docs] Update documentation for features from 2026-10-25 by @github-actions[bot] in #71867
- [jsweep] Clean write_daily_aic_usage_cache.cjs by @github-actions[bot] in #71805
- fix: stop treating gh-aw-detection as experimental by @pelikhan in #71848
- [docs] Update Astro dependencies - 2026-10-25 by @github-actions[bot] in #71796
- Detect parenthesized Flush discards in uncheckedflushreturn by @pelikhan with @Copilot in #71896
- Track credential handoffs through step outputs in secret-leak analysis by @pelikhan with @Copilot in #71850
- Update Pi CLI default to 1.4.0 by @pelikhan with @Copilot in #71899
- Detect missing recovery in goroutines launched through local closures by @pelikhan with @Copilot in #71897
- Refresh Node image pins and enforce patched runtime dependencies by @pelikhan with @Copilot in #71879
- Close custom-linter detection gaps and exempt safe os error checks by @pelikhan with @Copilot in #71880
- feat(linters): add readerstringconv analyzer by @pelikhan with @Copilot in #71891
- Centralize missing permission scope expansion by @pelikhan with @Copilot in #71888
- Unify GitHub Contents base64 decoding by @pelikhan with @Copilot in #71894
- Consolidate public GitHub API GET fallback transport by @pelikhan with @Copilot in #71893
- Fix generated conclusion references and fetch warnings; pin Redis by @pelikhan with @Copilot in #71898
- Guard indexing boundaries and prevent cron-step overflow by @pelikhan with @Copilot in #71889
- Deduplicate AW import normalization in source policy lint rules by @pelikhan with @Copilot in #71887
- [dead-code] chore: remove dead functions — 2 functions removed by @github-actions[bot] in #71916
- Classify engine driver crashes with secret-safe stderr diagnostics by @pelikhan with @Copilot in #71883
Full Changelog: v0.93.4...v0.93.5
v0.93.4
🌟 Release Highlights
This release focuses on work-queue reliability, safer checkout credential handling, and a broad set of compiler and workflow fixes.
✨ What's New
- Work queues: pools and workers now install on first use, with clearer diagnostics for missing pools and privacy-preserving GraphQL errors.
- Safe outputs: standardized machine-readable error codes,
add_commentconstraints disclosed in the tool description, and visible staged-preview notices on PR and Claim guards. - Threat detection: matching-engine detection now inherits agent API targets, and validated status-only jobs are allowed after detection.
- Environments: workflow environments now propagate to activation jobs.
- Faster
update: fewer network fetches, plus verbose request timings.
🐛 Bug Fixes & Improvements
- Release checkouts no longer persist credentials; compiler-generated PR checkouts are narrowly exempted from artipacked warnings.
- Compiler-generated constants are emitted as quoted strings, and schema-supported frontmatter shapes are accepted.
- Copilot BYOK HTTP API targets are preserved.
- Unmatched
checkout.fetchrefs no longer block other fetches. - Fixed prompt tool scanner false positives,
mapdeletecheckwith asymmetric parentheses, and empty repo-memory wiki note substitution. - Fixed missing embedded session-parser and grader runtime dependencies.
- Workflow health checks now have a time budget and debug logging; log storage admission and observability audits are bounded.
- Behavior-defined engine session capture is now protected.
📚 Documentation
- MemoryOps pattern page streamlined and docs updated for recent features.
Generated by 🚀 Release · copilot · auto · ~14 AIC · ⊞ 12.6K
What's Changed
- [log] Add debug logging to repository package manifest import/resource/skill paths by @github-actions[bot] in #71541
- Report missing work queue pools with actionable recovery diagnostics by @pelikhan with @Copilot in #71495
- Prevent arXiv researcher Codex startup failure by @pelikhan with @Copilot in #71608
- [docs] Update documentation for features from 2026-10-24 by @github-actions[bot] in #71631
- Keep TYPE-016 conformance check compatible with added parameters by @pelikhan with @Copilot in #71615
- Require item-level verification in daily queue reports by @pelikhan with @Copilot in #71607
- Fix Update Astro dependency downloads and upgrade command by @pelikhan with @Copilot in #71619
- Fix mapdeletecheck detection with asymmetric parentheses by @pelikhan with @Copilot in #71618
- Strengthen ledger formal-spec coverage by @pelikhan with @Copilot in #71625
- Replace unavailable Copilot model in video analyzer by @pelikhan with @Copilot in #71620
- Disclose add_comment constraints in tool description by @pelikhan with @Copilot in #71614
- Add visible staged-preview notices to PR and Claim guards by @pelikhan with @Copilot in #71612
- [instructions] Sync instruction files with release v0.93.2 by @github-actions[bot] in #71598
- [dead-code] chore: remove dead functions — 1 functions removed by @github-actions[bot] in #71660
- [Safe Outputs] Standardize machine-readable error codes by @pelikhan with @Copilot in #71613
- Inherit agent API targets in matching-engine threat detection by @pelikhan with @Copilot in #71624
- Propagate workflow environments to activation jobs by @pelikhan with @Copilot in #71617
- Add detail-level A/B campaign to commit-changes-analyzer by @pelikhan with @Copilot in #71666
- Update CLI and scanner image versions by @pelikhan with @Copilot in #71616
- Approve libatomic container runtime license exception by @pelikhan in #71678
- fix: refresh verified TLA runtime checksum by @pelikhan in #71679
- Preflight agent assignment credentials before processing issues by @pelikhan with @Copilot in #71611
- Skip unmatched checkout.fetch refs without blocking other fetches by @pelikhan with @Copilot in #71622
- Fix external workflow sources for documentation jobs by @pelikhan in #71677
- Install configured work queue pools on first use by @pelikhan in #71675
- Fix missing work-queue diagnostic guidance by @pelikhan with @Copilot in #71665
- Fix logs storage admission and bound observability audits by @pelikhan in #71686
- fix(workflow): preserve Copilot BYOK HTTP API targets by @pelikhan in #71680
- Fix remaining CI diagnostic guidance gaps by @pelikhan in #71688
- Fix missing grader runtime error codes dependency by @pelikhan in #71689
- Improve privacy-preserving work queue GraphQL diagnostics by @pelikhan in #71676
- Allow validated status-only jobs after detection by @pelikhan with @Copilot in #71623
- Fix compiler integration fixtures for action pin linting by @pelikhan in #71692
- Fix Pi evals reading stale workflow prompts by @pelikhan with @Copilot in #71664
- Reduce update network fetches and add verbose request timings by @dsyme in #71687
- Restore safe-output handler and tool-description conformance by @pelikhan with @Copilot in #71609
- Fix remote repository API requests for experiments by @pelikhan in #71681
- fix: substitute empty repo-memory wiki note as a literal by @pelikhan in #71782
- Gate generated admission outputs on admission opt-in by @pelikhan in #71784
- Add a time budget and debug logging to workflow health checks by @pelikhan in #71786
- Fix first-use work-queue pool and worker registration by @pelikhan in #71787
- [docs] docs: unbloat MemoryOps pattern page by @github-actions[bot] in #71779
- Narrowly exempt compiler-generated PR checkouts from artipacked warnings by @pelikhan in #71783
- fix: avoid persisting release checkout credentials by @pelikhan in #71802
- Protect behavior-defined engine session capture by @pelikhan with @Copilot in #69586
- fix(workflow): gate conclusion stale-check output references by @pelikhan in #71808
- fix: emit compiler-generated constants as quoted strings by @pelikhan in #71803
- fix(workflow): avoid prompt tool scanner false positives by @pelikhan in #71813
- fix(workflow): accept schema-supported frontmatter shapes by @pelikhan in #71809
- Recognize verified compiler-owned checkout cleanup in artipacked lint by @pelikhan in #71806
- Fix missing embedded session-parser dependency by @pelikhan in #71822
Full Changelog: v0.93.3...v0.93.4
v0.93.3
🌟 Release Highlights
This release tightens workflow quality checks and hardens threat detection.
✨ What's New
- Workflow linting on by default –
gh aw compilenow lints workflows during compilation, surfacing problems earlier (#71349). - Work-queue smoke workflows – New daily smoke workflows with Bash 3-compatible validation improve confidence in work-queue behavior (#71363).
🐛 Bug Fixes & Improvements
- Work-queue Claim settlement now verifies discussion cleanup effects (#71361).
- Threat detection is pinned to v0.5.3 for reproducible, safer runs (#71501).
📚 Documentation
- Clarified directory restrictions for reusable agentic workflows (#71451).
Generated by 🚀 Release · copilot · auto · ~13 AIC · ⊞ 12.6K
What's Changed
- Add daily work queue smoke workflows and Bash 3-compatible validation by @pelikhan in #71363
- [instructions] Sync instruction files with release v0.93.2 by @github-actions[bot] in #71403
- [docs] Update documentation for features from 2026-10-23 by @github-actions[bot] in #71415
- feat: lint workflows during compilation by default by @pelikhan in #71349
- fix(work-queue): verify discussion cleanup effects during Claim settlement by @pelikhan in #71361
- docs: clarify directory restrictions for reusable agentic workflows by @pelikhan with @Copilot in #71451
- [dead-code] chore: remove dead functions — 2 functions removed by @github-actions[bot] in #71454
- [blog] Agent of the Day – 2026-10-23 by @github-actions[bot] in #71455
- Pin threat detection to v0.5.3 by @davidslater with @Copilot in #71501
Full Changelog: v0.93.2...v0.93.3
v0.93.2
🌟 Release Highlights
A reliability-focused release: Copilot CLI installs are more resilient, scanners and reports are more trustworthy, and Pi runs behave correctly on ARC/DinD runners.
🐛 Bug Fixes & Improvements
- Copilot CLI install fallback — when the live compatibility matrix selects a release without a trusted checksum pin, setup now falls back to the bundled default (within the compatibility window), with checksum verification still required. (#71339)
- Pi on ARC/DinD — Pi streaming logs now go to the writable engine-log directory, fixing log, summary and session export on ARC/DinD runners. (#71228)
- Custom ledger safe outputs — custom ledger tool schemas now require a
record, and invalid envelopes are rejected before queuing so agents can self-correct. (#71240) - Grype scan completeness — empty or malformed Grype output is no longer cached as a zero-finding success, and per-image scan status is tracked. (#71297)
- Report formatting — bold labels in collapsible report summaries now render correctly instead of showing literal
<b>tags. (#71273) - Release notes — fixed apostrophe encoding guidance so highlights no longer publish
u0027. (#71272) - Conformance checks — removed false positives for PTI-005 and SEC-005. (#71287, #71286)
- Windows CI — POSIX-only outcome API fixtures are skipped on Windows. (#71247)
📚 Documentation
- Added admission setup and server-size AIC policy guidance to Cost Management. (#71323)
Generated by 🚀 Release · copilot · auto · ~18 AIC · ⊞ 12.6K
What's Changed
- Fix Pi streaming logs on ARC/DinD runners by @pelikhan in #71228
- fix: enforce custom ledger record contract before queuing by @pelikhan in #71240
- fix: skip POSIX outcome API fixtures on Windows by @pelikhan in #71247
- Fix release highlights apostrophe encoding guidance by @pelikhan in #71272
- fix: preserve bold formatting in collapsible report summaries by @pelikhan in #71273
- fix: exempt display-only session metadata from SEC-005 by @pelikhan in #71286
- fix: accept integrity-first deadline guard in PTI-005 conformance check by @pelikhan in #71287
- [log] Add debug logging to webhook, model-routing, and JS taint-summary files by @github-actions[bot] in #71327
- Document admission setup and server-size AIC policies by @pelikhan with @Copilot in #71323
- fix: reject incomplete Grype scans and verify per-image completion by @pelikhan in #71297
- Fall back to pinned Copilot CLI when live compatibility targets lack trusted checksums by @pelikhan with @Copilot in #71339
Full Changelog: v0.93.1...v0.93.2
v0.93.1
🌟 Release Highlights
v0.93.1 focuses on engine reliability, MCP trust configuration, and audit tooling, plus many CI and workflow stability fixes.
✨ What's New
- Per-server custom CA trust for HTTP MCP servers — connect to MCP servers behind private PKI without weakening global trust (#71229). Learn more
- BM25 audit search — unified-session audit now supports search with automatic evidence grouping (#70998). Learn more
- Outcome evaluation moved into logs — the separate
outcomescommand is removed (#71218). - Detection gates enforced for custom workflow jobs (#70997) and hosted threat detection is isolated (#71244).
- Work-queue diagnostics included in health reports (#70965).
- MCP Gateway bumped to v0.4.37 (#71219).
🐛 Bug Fixes & Improvements
- Codex stages native TOML sub-agents with resolved models and fails closed before startup (#70945); Copilot SDK refuses silent model substitution (#70943).
- Model-routing
allowed-modelsnow expands aliases (#70942); Claude detection model aliases resolve before CLI launch (#70861). - Pi main/sub-agent requests are attributed with
x-awf-caller(#71043). - CLI proxy and MCP gateway containers honor runner proxy settings (#70347).
- Scalar
allowed-reposnormalized in GitHub guard policies (#70346); trusted-bot policy compilation covered and documented (#70343). - Footer AIC costs show as rounded-up estimates (#71193).
- Dry-run telemetry disabled via documented opt-outs (#71032).
- Fixed Ollama CPU warm-up, Aider command replay, reviewer checkout after branch deletion, and unavailable prompt job output detection.
📚 Documentation
Daily documentation, glossary, and instruction syncs, plus an unbloated audit reference.
Generated by 🚀 Release · copilot · auto · ~13 AIC · ⊞ 12.4K
What's Changed
- Preserve failed Claim delivery diagnostics for investigation by @pelikhan in #70971
- Copilot SDK: observe the main-session model and refuse silent substitution by @SivaKesava1 with @Copilot in #70943
- Stabilize smoke-copilot-sub-agents: mandatory delegation, experiment-independent contract checks by @SivaKesava1 with @Copilot in #70944
- Codex: stage native TOML sub-agents with resolved models and fail closed before startup by @SivaKesava1 with @Copilot in #70945
- feat(cli): include work-queue diagnostics in health reports by @pelikhan in #70965
- build(deps): apply docs updates available in 1ES by @pelikhan in #70996
- fix: bound daily AIC quota recovery and snapshot discovery by @pelikhan in #70968
- [docs] docs: unbloat reference/audit.md by @github-actions[bot] in #70991
- Update CLI and scanner pins with 1ES-compatible Claude by @pelikhan in #70999
- Add unified-session BM25 audit search with automatic evidence grouping by @pelikhan in #70998
- fix: detect and rewrite unavailable prompt job outputs by @pelikhan in #70973
- Fix authority-scoped work queue Issue mirror synchronization by @pelikhan in #70992
- Resolve Claude detection model aliases before CLI launch by @pelikhan with @Copilot in #70861
- Expand aliases in engine.model-routing.allowed-models by @SivaKesava1 with @Copilot in #70942
- Disable dry-run telemetry using documented environment opt-outs by @pelikhan with @Copilot in #71032
- Enforce detection gates for custom workflow jobs by @pelikhan in #70997
- [docs] Update documentation for features from 2026-10-22 by @github-actions[bot] in #71114
- [ci-coach] Fix modernize embedlit lint failures in pkg/cli by @github-actions[bot] in #71146
- [docs] Update glossary - daily scan by @github-actions[bot] in #71108
- [instructions] Sync instruction files with release v0.92.0 by @github-actions[bot] in #71091
- Fix Contribution Check tool-denial aborts with prefetched PR evidence by @pelikhan in #71162
- Fix Super Linter analysis and runtime failure recovery by @pelikhan in #71169
- fix: precompute Daily File Diet source metrics by @pelikhan in #71161
- Improve work queue failure diagnostics by @pelikhan in #71163
- [dead-code] chore: remove dead functions — 3 functions removed by @github-actions[bot] in #71155
- Normalize scalar allowed-repos in GitHub guard policies by @pelikhan with @Copilot in #70346
- fix: filter expired approval runs from health failure candidates by @pelikhan in #71164
- [blog] Agent of the Day – 2026-10-22 by @github-actions[bot] in #71154
- Attribute Pi main/sub-agent requests with x-awf-caller and preserve caller metadata by @SivaKesava1 with @Copilot in #71043
- build(deps): consolidate 1ES-supported Dependabot npm updates by @pelikhan in #71195
- fix: repair CI documentation contracts and regression fixtures by @pelikhan in #71194
- Show footer AIC costs as rounded-up estimates by @pelikhan in #71193
- build(deps-dev): bump @playwright/test from 1.64.0 to 1.64.1 in /docs by @dependabot[bot] in #70493
- build(deps): bump starlight-changelogs from 0.7.0 to 0.7.1 in /docs by @dependabot[bot] in #70489
- fix(ci): preserve explicit embedded struct literals by @pelikhan in #71209
- Fix Aider command replay and daily Go workflow completion by @pelikhan in #71211
- Fix reviewer checkout after merged PR branch deletion by @pelikhan in #71210
- build(deps): bump the pip group across 1 directory with 2 updates by @dependabot[bot] in #70876
- [actions-aw-linter] Fix workflow_run positive branch cancellation detection by @github-actions[bot] in #71189
- build(deps): bump dompurify from 3.4.16 to 3.4.17 in /docs in the npm_and_yarn group across 1 directory by @dependabot[bot] in #71216
- build(deps): bump geo-optimizer-skill from 4.18.3 to 4.19.4 in /.github/workflows by @dependabot[bot] in #70485
- Honor runner proxy settings in CLI proxy and MCP gateway containers by @pelikhan with @Copilot in #70347
- Bump MCPG to v0.4.37 by @lpcox with @Copilot in #71219
- Support per-server custom CA trust for HTTP MCP servers by @pelikhan in #71229
- Unify debugging bootstraps and verify instruction routing by @pelikhan in #71223
- Move outcome evaluation into logs and remove outcomes command by @pelikhan in #71218
- Cover trusted-bot policy compilation and document platform bot support by @pelikhan with @Copilot in #70343
- Fix CI instruction drift and registry-dependent negative tests by @pelikhan in #71241
- Fix code quality reviewer finalization sequencing by @pelikhan in #71242
- Fix Ollama CPU warm-up and isolate hosted threat detection by @pelikhan in #71244
Full Changelog: v0.93.0...v0.93.1
v0.93.0
🌟 Release Highlights
v0.93.0 focuses on security hardening, safer dry-run compilation, and more reliable safe outputs and sub-agent model resolution.
⚠️ Breaking Changes
- The missing-diagnostic safe-output tools are now named
report_missing_toolandreport_missing_data(#70738). Update any workflows or prompts that use the old names. Learn more - Admission jobs are now opt-in through frontmatter or
aw.json(#70763).
✨ Whats New
- Read-only health tool in the MCP server for checking server status (#70527).
- Artifact-first admission conclusion session delivery (#70622) and routing analytics recorded in
aw_session.jsonl, including model config, baseline estimate, upstream errors and harness identity (#70672). - Claude inline sub-agents are registered and Claude/Codex sub-agent models resolve before engine startup (#70671).
- Shared
/reviewalias for PR reviewer workflows (#70863). - New Go linters (
formatintitoa,osisnotexist) and a needs-consistency linter (#70801, #70546). - Firewall bumped to v0.28.70 and MCP Gateway to v0.4.36.
🔒 Security
- Dry-run compilation is hardened against hosted execution effects, and dry-run workflows use a read-only cache (#70851, #70788).
- Compiler bootstrap is hardened against compromised certificate authorities (#70549).
- OTLP token environment-file injection is prevented (#70597).
- Unsafe action pins are rejected (#70616), and fixable Node image CVEs were cleared (#70615).
- Redaction must now succeed for all agent artifact uploads (#70875).
🐛 Bug Fixes & Improvements
- Safe outputs honor target repos in invocation context (#70912) and fix discussion temporary IDs and missing terminal outputs (#70696).
- Copilot URL approvals follow the finalized runtime firewall policy (#70750).
- Runner-label indentation in custom jobs is fixed (#70840).
- Fixed webhook integer overflow in 32-bit builds (#70775).
- Better engine driver crash and secret-redaction failure classification (#70605).
- MCP log continuation cursors are preserved (#70856); AIC telemetry attribution is corrected (#70864).
- Work queue replay prefers local Git lookups (#70617).
📚 Documentation
- Debug playbook now covers empty-output failure codes (#70676), and the work queue reference was streamlined.
Generated by 🚀 Release · copilot · auto · 14.9 AIC · ⊞ 12.4K
What's Changed
- Fix PR reviewer safe-output payload quoting guidance by @pelikhan in #70913
- fix(health): ignore evidence-confirmed PR lifecycle interruptions by @pelikhan in #70915
- [safe-output-integrator] Add safe-output fixtures for report-missing-tool and report-missing-data by @github-actions[bot] in #70930
- Fix stale CI expectations and schema regressions by @pelikhan in #70918
- Reuse workflow metadata in OTLP attributes by @pelikhan with @Copilot in #70860
- fix: honor safe-output target repos in invocation context by @pelikhan in #70912
Full Changelog: v0.92.8...v0.93.0
v0.92.6
🌟 Release Highlights
v0.92.6 strengthens the work queue, episode-based log analysis, model-catalogue resilience, and static security linting.
⚠️ Breaking Changes
gh aw logsnow requires canonical episode identity and reports coverage (#70277). Update any tooling that relied on non-canonical episode identifiers.
✨ What is New
- Episode analytics: bounded canonical episode retrieval, usage joined with recorded outcome accounting, and better unified-session episode lineage (#70322, #70281, #70205).
- Work queue: guarded clear/hard-reset commands, deterministic expiration of unclaimed entries, better diagnostics, and stronger delivery recovery (#70022, #70025, #70027, #69951, #69790).
- Security linting: new AST linters for checkout trust, prompt expressions, secret-leak taint analysis, and history-backed zizmor checks (#70131, #69813, #70356).
- Provenance & redaction: agent provenance tracking, coarse issue-body redaction, and early activation failure for stale provenance workflows (#70032, #70348).
- Model resilience: consistent model alias resolution across engines, a Copilot SDK check against the live model catalogue, and a daily model-pin checker (#69487, #70320, #70295).
- Engines: OpenCode integration migrated to v2 (#69950).
- Pure graders: graders are now pure unified-session computations (#70130).
🐛 Bug Fixes & Improvements
- Custom safe-output jobs are gated on threat detection success; threat detection applies warn-mode policy when the detection engine fails (#69995, #70325).
- Placeholder issue titles are rejected before consuming
create_issuequota (#69994). - Fixed empty
configured_modelfor routed/fallback Copilot CLI runs (#69966). - Fixed OTLP placeholder masks suppressing boolean job outputs (#69996).
- Safe-output footers include Work and Claim references (#69865).
- Fixed Goose release redirects, Crush stdin startup, and Aider discovery (#70282, #70214, #70021).
- Fixed PR lifecycle races and merged-base-ancestor detection (#70141, #69998).
📚 Documentation
- Specified OIDC-based AW activation admission and aligned episode specs (#70286, #70279).
- Documented claim delivery diagnostics (#69987).
Generated by 🚀 Release · copilot · auto · 15.9 AIC · ⊞ 12.3K
What's Changed
- Allow narrowly gated Cloud Hypervisor agent acceptance fixture by @lpcox in #69777
- Surface sub-agent failures and replace unpriced kiwi workflow models by @pelikhan in #69788
- Finalize work-queue workers inline and recover stranded runs by @pelikhan in #69790
- Resolve model aliases the same way on every engine by @SivaKesava1 with @Copilot in #69487
- Fix Pi model availability in multi-device docs tester by @pelikhan in #69869
- Fix container scan analysis tool-permission mismatch by @pelikhan in #69867
- fix: use per-model SDK providers for Contribution Check delegates by @pelikhan in #69870
- fix: include Work and Claim references in safe-output footers by @pelikhan in #69865
- [eslint-miner] eslint-factory: add no-split-filter-trim-without-trim-map rule by @github-actions[bot] in #69831
- fix(workflows): report PR content integrity blockers accurately by @pelikhan in #69866
- Fix bounded CLI coverage and delegated-failure audit reporting by @pelikhan in #69868
- [dead-code] chore: remove dead functions — 2 functions removed by @github-actions[bot] in #69936
- Fix CI parser embedding and stale policy fixtures by @pelikhan in #69943
- Migrate OpenCode integration to v2 by @pelikhan in #69950
- Add extensible AST linting and secret-leak taint analysis by @pelikhan in #69813
- Fix work-queue delivery recovery and formalize settlement guarantees by @pelikhan in #69951
- Add daily Actions/AW AST linter factory by @pelikhan with @Copilot in #69972
- Fix empty configured_model for routed and fallback Copilot CLI runs, and isolate selection-persistence errors in the post-step by @SivaKesava1 with @Copilot in #69966
- Replace Linter Miner’s unavailable model by @pelikhan with @Copilot in #69990
- Document claim delivery diagnostics by @pelikhan with @Copilot in #69987
- Preserve AWF ranked choices and request/failure metadata in unified sessions by @SivaKesava1 with @Copilot in #69848
- Gate custom safe-output jobs on threat detection success by @pelikhan with @Copilot in #69995
- Fix rendering verifier dependencies and real-log harness by @pelikhan in #70017
- Fix work queue delivery transport continuity by @pelikhan in #70016
- Fix Windows Grower reusable workflow permissions by @pelikhan in #70020
- Fix Aider discovery from an empty file context by @pelikhan in #70021
- fix: give Copilot and Pi conformance probes bounded credit headroom by @pelikhan in #70023
- Reject placeholder issue titles before consuming create_issue quota by @pelikhan with @Copilot in #69994
- Prevent OTLP placeholder masks from suppressing boolean job outputs by @pelikhan with @Copilot in #69996
- Render complete issue body ASTs once by @pelikhan in #70019
- Fix stale Linter Miner model contract in CI by @pelikhan in #70030
- Add deterministic expiration for unclaimed work queue entries by @pelikhan in #70025
- Add guarded work-queue clear and hard reset commands by @pelikhan in #70022
- Add fast authored JavaScript analysis with reviewed API summaries by @pelikhan in #70029
- Improve work-queue diagnostics for agents by @pelikhan in #70027
- Add missing compiler threat linters and reference threat rule IDs by @pelikhan in #70026
- [docs] docs: unbloat artifacts reference by @github-actions[bot] in #70108
- Preserve AWF model inventories with null endpoints and live reasoning metadata (G-7) by @SivaKesava1 with @Copilot in #70013
- Recognize merged base ancestors when the base tip advances by @pelikhan with @Copilot in #69998
- [blog] Weekly blog post – 2026-10-19 by @github-actions[bot] in #70124
- Replace retired MAI picker with available MAI-Code-1.1-Flash by @pelikhan in #70133
- Add AST threat linters for checkout trust and prompt expressions by @pelikhan in #70131
- fix(queue): bound deployment metadata retries by @pelikhan in #70129
- Track agent provenance and support coarse issue-body redaction by @pelikhan in #70032
- [spec-review] Update Safe Outputs conformance checker for recent spec changes by @github-actions[bot] in #70150
- Handle queued and in-flight PR lifecycle races by @pelikhan in #70141
- Make graders pure unified-session computations by @pelikhan in #70130
- fix(work-queue): recover transient metadata read failures by @pelikhan in #70200
- Preserve HTML image layout and code fidelity in issue reports by @pelikhan in #70158
- Fix Crush smoke engine startup on open stdin by @pelikhan in #70214
- Fix CI contracts and custom grader download failures by @pelikhan in #70217
- [archivx] Update workflow diagram — 2026-10-19 by @github-actions[bot] in #70210
- [spec-enforcer] Enforce specifications for fileutil, github, githubapi by @github-actions[bot] in #70237
- [docs] Update glossary - weekly full scan by @github-actions[bot] in #70231
- [schema-coverage] docs: add schema coverage demo for docker-image-pull-policy by @github-actions[bot] in #70183
- [instructions] Sync instruction files with release v0.92.0 by @github-actions[bot] in #70207
- Improve outcome evaluation and unified session episode lineage by @pelikhan in #70205
- fix(goose): allow GitHub release asset redirects by @pelikhan in #70282
- Fix work-queue guidance and WASM golden CI failures by @pelikhan in #70283
- docs: align episode specifications with supported behavior by @pelikhan in #70279
- Fix claim-scoped engine conformance noop checks by...