feat(skills): add a release-captain skill for the weekly release cadence #4433

User Story

As an OpenShell release captain, I need a Monday and Tuesday checklist that an agent can run, so I can get a qualified pre-release out as a stable release every Tuesday without relying on knowledge that only exists in Slack.

Problem Statement

RFC 0014 sets a Tuesday stable cadence, and maintainers are rotating release captain duties: keep pre-releases passing, coordinate the release with contributors, and cut the release. The automation (Release Auto-Tag, Release Tag qualification, advisory compatibility review, Release Canary) is documented in CI.md, but the weekly human steps are not written down:

• what to check on Monday, and where qualification, compatibility, and security reports live;
• how to collect ready PRs and set a merge cutoff before the release;
• how to select and verify the final candidate on Tuesday, tag it, and confirm publication;
• how to start the next train's -pre.1;
• what to do when a release run is stuck or fails, without moving a published stable tag.

The v0.1.3 release (October 7–9) needed a pulled tag and two retries, and the v0.1.4-pre.3 compatibility review produced likely false positives that nobody owned triaging.

Impact / Why This Matters

New release captains, including next week's, can run the release from a repeatable checklist with copyable gh/git commands. That reduces Tuesday slippage and keeps tag pushes behind explicit maintainer approval.

Proposed Design

Add an internal contributor skill, .agents/skills/release-captain/SKILL.md (metadata.internal: true), with:

• Monday: identify the current train and latest -pre.N; read its Release Qualification summary, advisory compatibility report, and security results; check Release Canary; triage findings; draft the coordination post listing ready PRs and the merge cutoff.
• Tuesday: after the cutoff, confirm the final candidate's qualification matches its tag and commit; get maintainer approval; tag that exact commit vX.Y.Z; watch Release Tag; verify the GitHub Release and published artifacts; start the next train.
• Recovery: stuck or failed runs, rerun versus a new pre-release, and the rule against moving a published stable tag.
• Guardrails: the agent never pushes or deletes tags, dispatches release workflows, or posts to Slack without explicit maintainer approval.

This complements the tagging mechanics in #4335 (release-openshell) and does not change workflows. Register the skill in CONTRIBUTING.md and the sync-agent-infra maintenance map.

Acceptance Criteria

• .agents/skills/release-captain/SKILL.md exists with metadata.internal: true, Monday, Tuesday, and recovery sections, and copyable commands verified against a real Release Tag run.
• Every command that mutates tags, workflows, or Slack requires explicit maintainer approval.
• CONTRIBUTING.md lists the skill, and the sync-agent-infra maintenance map includes it for release workflow changes.
• No workflow behavior changes.

Alternatives Considered

• Expand release-openshell in fix(release): qualify prerelease tag pushes #4335. Kept separate so the weekly cadence and the tagging mechanics can change independently.
• Add mise helper tasks for Monday and Tuesday. Deferred; the gh commands in the skill avoid new scripts to maintain.