Goal
Track in-flight and planned CI improvements for the repo — pipeline performance, test-pipeline unification, build/publish validation, workflow security, and a sub-track of agentic automation to reduce manual CI monitoring.
Pipeline performance
Test pipeline unification
Build & publish validation
Workflow security
PR testing / infra
Agentic CI automation
Reduce hand-monitoring of CI: agents spot, analyze, file issues, and open PRs for humans to review/approve, then validate their own work. Builds on the repo's existing gh-aw setup and CI plumbing.
Failure visibility / monitoring
Quarantine automation
CI health
Guardrails for the agentic items: confidence thresholds before acting; keep rolling pass-rate even when a rerun recovers; quarantine budget + enforced de-quarantine; cost/recursion caps; global /agent-stop; every code or quarantine mutation goes through a human-approved PR.
Goal
Track in-flight and planned CI improvements for the repo — pipeline performance, test-pipeline unification, build/publish validation, workflow security, and a sub-track of agentic automation to reduce manual CI monitoring.
Pipeline performance
microsoft-aspirepipeline wall-clock from ~121 min to ~57 minmain(currentlyNO-MERGE)Test pipeline unification
run-tests-core.yml)Build & publish validation
Workflow security
PR testing / infra
pr-testingto handle and test infra changes (branchankj/pr-testing-infra-skill, not yet pushed)Agentic CI automation
Reduce hand-monitoring of CI: agents spot, analyze, file issues, and open PRs for humans to review/approve, then validate their own work. Builds on the repo's existing gh-aw setup and CI plumbing.
Failure visibility / monitoring
ci-brokenissue when the internal AzDO build breaks onmain/release/*ankj/workflow-failure-issues, not yet pushed) — adapt gh-awci-doctor+aw-failure-investigator; dedup; consumes AzDO-filed issues tooQuarantine automation
docs/unquarantine-policy.md(relates to [tests] Quarantined tests dashboard #8813)CI health
repo-pulsepattern)Guardrails for the agentic items: confidence thresholds before acting; keep rolling pass-rate even when a rerun recovers; quarantine budget + enforced de-quarantine; cost/recursion caps; global
/agent-stop; every code or quarantine mutation goes through a human-approved PR.