Update commit skill to four-pass method

[mikemindel]

Summary

• Replace the flat three-pass commit skill with the structured four-pass methodology
• Add WORKFLOW.md with detailed execution sequence (pre-flight, content, standards, final review, post-commit)
• Move atomic-git-commits.md into references/ directory
• Add chris-beams-commit-style.md reference

What changed

The old commit skill was a single SKILL.md pointing at a monolithic reference file. The new structure splits concerns across focused documents:

skills/commit/
├── SKILL.md                          # Overview, format reference, atomicity rules
├── WORKFLOW.md                       # Four-pass execution sequence
└── references/
    ├── atomic-git-commits.md         # Full methodology (moved from skill root)
    └── chris-beams-commit-style.md   # The 7 rules

Test plan

• Verify /commit skill triggers correctly
• Verify all cross-reference links resolve
• Run a test commit to confirm four-pass flow works end-to-end

[chriswickett]

I like the structure and the references cleanup, but I'm not sure about the multi-commit methodology for this use case.

To adopt this properly, we'd need to update the Actions skills (pr-start, pr-fix) to make multiple commits per task. That means reviewing a feature spread across several commits instead of one clean diff, which makes it harder for me to see what Claude did in a given pass.

For me, I wouldn't want that in this repo for how I like to work with my other clients. If you are married to small commits, then perhaps we need to start thinking about how agentic might implement config/preferences that can be changed per-client repo? Any ideas?

I'd also argue that SKILL.md and the reference file cover a lot of the same ground which would swell the token cost (something I need to keep an eye on!)

[chriswickett]

Also, I like the deterministic nature of Claude having to write one commit_msg.txt. It so far has made it easier to see if it succeeded or not to do the task.