feat: pipeline orchestrator — shepherd agent PRs through the full lifecycle

[microsasa]

Problem

Agent PRs get stuck at multiple stages after creation. Individual workflows handle each stage (review-responder, quality-gate, ci-fixer) but nothing detects and fixes stalls. When Copilot never reviews, threads stay unresolved, or a PR falls behind main, it sits there until a human intervenes.

The old pr-rescue.yml (bash script) only handled rebasing and was brittle — tens of bugs across multiple review rounds.

Solution

Replace pr-rescue.yml with a pipeline orchestrator gh-aw agent that periodically sweeps open aw-labeled PRs and pushes each one forward.

Orchestrator logic per PR (first match wins)

1. No Copilot review → request review from @copilot
2. Unresolved threads → query real thread IDs via GraphQL, resolve threads where the responder (PAT owner) posted the last comment
3. Behind main → comment that rebase is needed (human or future automation handles it)
4. All clear → auto-merge should handle it, move on

Triggers

• schedule: */15 * * * * — catches stuck PRs between merges
• push: branches: [main] — fires when a PR merges
• workflow_dispatch — manual trigger

Tools & safe-outputs

• bash: ["gh:api:graphql", "gh:api:user"] for thread ID lookup and PAT owner detection
• add-reviewer, resolve-pull-request-review-thread, add-labels, remove-labels, add-comment
• No contents: write, no git access — pure reasoning agent

What this replaces

• Deletes pr-rescue.yml (106-line bash script with repeated bug-fix cycles)

Also in scope

Review-responder thread ID fix (#117): Add bash: ["gh:api:graphql"] to review-responder so it can look up real PRRT_ thread IDs before resolving, instead of hallucinating them.

Closes

• Enhanced PR Rescue: resolve threads + request reviews + rebase behind-main #116 — Enhanced PR rescue
• workaround: give review-responder bash tool for gh api graphql thread ID lookup #117 — Responder bash tool for thread IDs
• Stale PR cleanup: close aw-labeled PRs open >24h without approval #90 — Stale PR cleanup (orchestrator handles stuck PRs)

Related

• tracking: upgrade gh-aw MCP server to get real thread IDs for review-responder #114 — Upstream MCP server thread ID bug (workaround, not fix)
• tracking: gh-aw create-pull-request labels config fails to apply labels #108 — create-pull-request labels config bug (unrelated)

[microsasa]

Implemented in PR #130 (merged). Orchestrator is live. Some details in this issue are outdated (responder fix was instruction-only, not bash tool; orchestrator uses GITHUB_REPOSITORY_OWNER not gh api user).