claude/skills/commit-change/SKILL.md

name, description, allowed-tools

name	description	allowed-tools
commit-change	Analyze all pending changes (staged, unstaged, untracked) and create atomic commits grouped by logical unit, retracing the work. Any git repository. Triggers: "commit my changes", "smart commit", "auto commit", "commit everything", "analyse et commit", or any variation of committing messy pending work intelligently.	Bash Read Grep Glob Agent AskUserQuestion

/commit-change — propose → confirm → apply dispatcher

Grouping and committing both run on the sonnet-pinned commit-changer subagent (dispatch makes the pin effective). No inline reflection happens in this dispatcher to protect, so there is no model gate. This dispatcher owns the two approval gates that used to live inside the subagent: commit-plan approval and capitalize approval — the subagent never asks; MODE: propose only proposes, MODE: apply only executes what this dispatcher confirms. Never auto-commit blind — a wrong group is harder to undo than not committing.

STEP 0 — Pre-flight (STOP conditions, before any dispatch)

git rev-parse --abbrev-ref HEAD                        # "HEAD" = detached
git status --porcelain=v1 | grep -c '^UU\|^AA\|^DD'     # unmerged conflicts
git status --porcelain=v1 | wc -l                       # nothing pending?
git config user.email

• Detached HEAD → STOP, report the state, do not dispatch.
• Any unmerged conflict entries (UU/AA/DD) → STOP, tell the user to resolve conflicts first, do not dispatch.
• Nothing pending (git status --porcelain empty) → STOP, tell the user there's nothing to commit.
• git config user.email empty → STOP, ask the user to configure identity first, do not dispatch.

On a protected base (main/develop) the subagent runs the gitflow aiguillage itself inside MODE: propose (its Phase 0) and branches to chore/* before drafting the plan — code never lands directly on a protected branch.

STEP 1 — Propose

Agent(subagent_type="commit-changer")
prompt: "MODE: propose
$ARGUMENTS"

Read the returned COMMIT PLAN + EDGE CASES + CAPITALIZE CANDIDATES, terminated by READY TO APPLY — awaiting dispatcher confirmation.

The subagent reported BLOCKED: unresolved merge conflicts... instead of a plan (a race with STEP 0) → STOP, surface it, do not proceed.

STEP 2 — Gate 1: commit-plan approval

Show the COMMIT PLAN and any EDGE CASES verbatim, then:

AskUserQuestion:
  Approve the commit plan? (all / <numbers> / edit <n> / skip)

• all → every step in the plan is approved as-is.
• <numbers> (e.g. 1,3) → only those steps are approved; the rest stay uncommitted for a later run.
• edit <n> → collect the corrected message/grouping for step N from the user, redraw the plan (this dispatcher owns the text, no re-dispatch needed), show it again and re-ask.
• skip → exit cleanly, no commits created, no MODE: apply dispatch.

STEP 3 — Gate 2: capitalize approval

If the STEP 1 output said CAPITALIZE: nothing to log, skip this gate — treat the capitalize entries as none and go straight to STEP 4.

Otherwise show the CAPITALIZE CANDIDATES block, then:

AskUserQuestion:
  Valider les entrées mémoire ? (all / <IDs> / skip)

• all → every candidate entry is approved verbatim.
• <IDs> (e.g. BDR-041,LRN-019) → only those entries are approved.
• skip → no memory write; MODE: apply still runs for the code commits.

STEP 4 — Apply

Agent(subagent_type="commit-changer")
prompt: "MODE: apply
APPROVED PLAN: <the STEP-2-approved steps — numbers, messages, files,
  exactly as confirmed, including any edits>
APPROVED CAPITALIZE ENTRIES: <the STEP-3-approved entries verbatim, or none>"

Parse the COMMIT-EXEC REPORT:

• STATUS: DONE → report the COMMITS + MEMORY hashes to the user.
• STATUS: BLOCKED → surface the blocker verbatim and stop. Do not retry automatically — a blocked step (e.g. one file needs an interactive git add -p split) needs a human decision.