[vitek-karas]

Upgrade all six agentic workflows from gh-aw v0.86.2 to v0.89.21, the latest stable, non-prerelease release as of 2026-10-08, and regenerate their lock files and action/container cache.

Use gh-aw's default dependency versions rather than maintaining repository-specific version overrides. The six main MCP gateways and holistic-review's CLI proxy now all use v0.4.25, up from the original v0.4.9. There is no custom aw.json gateway mapping or pin-specific instruction file. Compiler-generated immutable action and container references are retained.

The companion validation workflow's setup action is also updated to v0.89.21 while retaining its existing CLI version, and the dispatcher references matching upstream authoring documentation. The breaking-change documentation workflow now declares its base-repository checkout explicitly to satisfy the new compiler's validation while keeping repository skills available. Workflow-specific triggers, schedules, job permissions, the shared import, and configured output limits are preserved.

Remove the handwritten GPT-6.1 pricing overlays from the scanner and feedback workflows. The upgraded runtime discovers pricing from Copilot's model catalog at startup; the compiler's bundled table alone does not contain this model. Both workflows retain gpt-6.1-sol, their 2,500-credit limits, and their existing Responses routing. Daily accounting and cost summaries consume the runtime's reported credits rather than requiring our own model-price table.

Also fixes a pre-existing Skill Validation failure exposed by the current nightly skill validator. The validator rejects <T> as an XML-like tag in skill descriptions, so .github/skills/vectorization/SKILL.md now says "the generic Vector type" instead of Vector<T> in its description. The skill's technical guidance and examples are unchanged. This check validates all skills when an agent file changes, which is why the unrelated error surfaced in this PR.

Validation

Using the verified official v0.89.21 release executable as $GH_AW:

• "$GH_AW" compile --strict --schedule-seed dotnet/runtime --force-refresh-container-pins --validate --validate-images — passed, all six workflows; repeated regeneration produced identical lock files and pin cache.
• "$GH_AW" compile --strict --schedule-seed dotnet/runtime --validate --validate-images — passed after removing the pricing overlays; another regeneration left all six lock files and the pin cache unchanged.
• docker run --rm --network none --entrypoint /app/awmg ghcr.io/github/gh-aw-mcpg:v0.4.25@sha256:9be0a86220e807a0ecc89e53d7453468f7a53fbc6b3d1efd2299025ffe01d086 --version — passed, reports MCPG Gateway v0.4.25.
• Offline smoke test against the released AWF v0.28.23 and gh-aw v0.89.21 modules — passed with synthetic provider-catalog responses: startup discovery resolves the otherwise unlisted model, the 2,500-credit cap is enforced, daily totals and reported costs agree, and incomplete pricing is rejected. This did not query the live Copilot catalog or perform inference.
• actionlint -no-color -format '{{json .}}' .github/workflows/*.lock.yml .github/workflows/validate-pat-pool.yml — not clean: 40 pre-existing diagnostics. A per-workflow comparison against the original lock files verified identical diagnostic types and messages, with no new diagnostics.
• actionlint -no-color .github/workflows/validate-pat-pool.yml — passed.
• git diff --check — passed.

Previously recorded validation of the retained skill-description adjustment, using the current nightly skill-validator:

• skill-validator check --skills .github/skills --allow-repo-traversal --verbose — passed, all 23 skills.
• skill-validator check --agents .github/agents --allow-repo-traversal --verbose — passed, all 3 agents.

The Skill Validation CI run also passed for the earlier commit 1f7c7a3748c.

Local assertions also verified every gateway consumer uses the compiler default v0.4.25 and checked the preserved workflow settings against the original baseline. No live agentic workflow runs were triggered.

Note

This pull request was prepared with GitHub Copilot.

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).
15 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[PureWeen]

Holistic Review

Motivation: Updating the compiler, runtime dependencies, and generated workflows together is justified.

Approach: The migration preserves the inspected Responses configuration and publication boundaries. One existing test consumer still assumes the old detector execution layout.

Summary: ⚠️ Needs Changes. Please update the scanner and feedback Responses-routing assertion alongside the external-detector migration. I independently reproduced 28/28 passing tests at the base and 26/28 at this head; both failures are the stale execution-step count. No production routing regression was verified, and this is not a claim that hosted CI newly fails at that assertion.

Detailed Findings

⚠️ Tests — preserve routing coverage for the external detector

The inline finding identifies the generated-layout change and the unchanged test consumer. Identify the agent and detector separately and retain the Responses environment check on each, rather than reducing the expected count.

Note

This review was prepared with GitHub Copilot assistance.

[kotlarmilos]

LGTM, thanks for cleaning up sol-6.1 config.