commit-pipeline

name	commit-pipeline
description	Orchestrate a complete commit flow: use commit-planner to define batches, stage the selected batch, generate the message with commit-messages, and execute git commit safely.

Commit Pipeline

Run a full plan -> stage -> message -> commit flow. Use this skill when the user wants to finish commits end-to-end instead of only planning or only writing messages.

Skills Used

• $commit-planner: define coherent commit batches
• $commit-messages: draft a structured commit message for the staged batch

Core Rule

Never commit files that are outside the selected batch. If staging state does not match the selected batch, fix staging first before generating the message.

Workflow

1. Capture working tree state.
   • git status --short
   • git diff --name-only
   • git diff --staged --name-only
2. Run $commit-planner.
   • Produce batch options with goal, file list, and type/scope.
   • Select one batch as the next commit.
3. Decide whether living docs belong in this batch.
   • Think in terms of the repository's memory system, not just a few default docs.
   • Framework memory may include memory-docs/OVERVIEW.md, STATUS.md, HISTORY.md, CONVENTIONS.md, and GLOSSARY.md.
   • Routing memory is memory-docs/DIRS.md; detailed memory includes memory-docs/detail_mem/MAP.md, PROGRESS.md, and DECISIONS.md.
   • Session and historical memory usually do not belong in the same commit by default, but include relevant memory-docs/SHORT_MEMORY/... or memory-docs/archive/... files when the batch intentionally captures handoff context or a retrospective.
   • Follow the repository's docs conventions and keep claims aligned with implemented reality.
   • Use $update-living-docs when it helps, but only for docs that are genuinely relevant to the selected batch.
   • If docs should be separate, make that an explicit follow-up batch instead of silently deferring them.
4. Stage only the selected batch.
   • If index is dirty with unrelated files, unstage first: git restore --staged <path>.
   • Stage selected files only, including any relevant living-doc updates: git add <paths...>.
   • Verify: git diff --staged --name-only must match the chosen batch.
5. Run $commit-messages on staged changes.
   • Generate title + Why/What/Risk/Tests/Live Docs.
   • Ensure claim discipline: staged facts only.
6. Commit with the generated message.
   • git commit -m "<title>" -m "Why:\n- ...\n\nWhat:\n- ...\n\nRisk:\n- ...\n\nTests:\n- ...\n\nLive Docs:\n- ..."
7. Verify commit result.
   • git show --stat --name-only --oneline -1
   • Confirm no accidental files were included.
8. Repeat from step 2 for the next batch until done.

Output Format

Use this structure when reporting execution:

Pipeline Run:
- Selected batch: ...
- Staged files: ...
- Commit title: ...
- Commit hash: ...
- Remaining changes: ...

Safety Checks

• Stop if git diff --staged --name-only includes non-batch files.
• Stop if message claims behavior not present in staged diff.
• Do not auto-add untracked files unless they are explicitly part of the selected batch.

Bad Patterns

• running $commit-messages before staging is finalized
• committing all changes with git add . in a mixed working tree
• combining unrelated hygiene and feature slices into one commit without rationale

Output Checklist

Before finishing, verify:

• each commit corresponds to one batch from $commit-planner
• staged files exactly match the batch before message generation
• commit message comes from staged reality
• commit hash and remaining work are reported