This was generated by AI during triage.
Split out from #462 (item 3). Create a .claude/skills/steward/SKILL.md so any session driving agent PRs (the fix worker, a session babysitting a PR, remote CI-fix sessions) shares the pipeline's conventions. Notably, Claude Code remote sessions already look for .claude/skills/steward/SKILL.md on a PR's head branch before acting on CI and review events — today that lookup finds nothing.
To settle in a grill-with-docs session before implementation:
- What the steward skill carries vs. what stays in
docs/agents/autonomic-issues.md (the skill should distill conventions, not duplicate the playbook)
- Conventions to encode: commit/PR format (
create-pr skill), self-review expectations, how to respond to maintainer review feedback, when to stand down vs. push
- How proactive the steward should be on CI failures and review comments on agent PRs
- Relationship to the existing
continue-pr and pr-review-fixer skills
Marked ready-for-human for the design pass; the skill-writing itself could then be delegated (writing-for-agents skill applies).
Split out from #462 (item 3). Create a
.claude/skills/steward/SKILL.mdso any session driving agent PRs (the fix worker, a session babysitting a PR, remote CI-fix sessions) shares the pipeline's conventions. Notably, Claude Code remote sessions already look for.claude/skills/steward/SKILL.mdon a PR's head branch before acting on CI and review events — today that lookup finds nothing.To settle in a grill-with-docs session before implementation:
docs/agents/autonomic-issues.md(the skill should distill conventions, not duplicate the playbook)create-prskill), self-review expectations, how to respond to maintainer review feedback, when to stand down vs. pushcontinue-prandpr-review-fixerskillsMarked
ready-for-humanfor the design pass; the skill-writing itself could then be delegated (writing-for-agentsskill applies).