| name | commit-changes |
| description | Commit uncommitted changes across all workspace repos and tag commits with ticket numbers. Each repo gets its own subagent that analyzes files individually, groups them by ticket, and makes multiple focused commits. Use when the user pastes issue text (with TICKET-XXX) and wants to commit, or says "commit with these tickets" / "commit my changes". |
Commit Changes Skill
Trigger conditions
Use this skill when the user:
- Pastes one or more issues (with
# Title and Identifier: TICKET-XXX lines) and wants to commit.
- Says "commit with these tickets", "commit my changes for TICKET-727", "commit everything and tag by ticket", or similar.
- Runs the /pwf-commit-changes slash command with pasted content.
What to do
Follow the workflow in commands/commit-changes.md exactly.
Summary of phases
- Parse — extract
{ id: "TICKET-XXX", title: "..." } for each ticket from the pasted text.
- Discover repos —
git -C <path> status --short for every workspace path; skip empty or non-git. Collect only the list of repos with changes (no diffs at orchestrator level).
- Spawn one subagent per repo (all in parallel) — each subagent (generalPurpose, model: fast) reads
skills/commit-changes-repo-worker/SKILL.md, fetches per-file diffs itself, classifies each file to the best ticket, groups files by ticket, and makes multiple targeted commits (one per group). Reports back a JSON result with a commits array.
- Summarize — parse JSON results from all subagents, print a results table (one row per commit), show total commit count.
Worker skill
Each per-repo subagent follows skills/commit-changes-repo-worker/SKILL.md.
The main agent passes these inputs to each worker:
REPO_PATH — absolute pathREPO_NAME — short nameTICKET_LIST — formatted list of TICKET-XXX: title
The worker fetches its own per-file diffs — no diff text is passed from the orchestrator.
Key difference from single-commit approach
The worker never runs git add -A. Instead, for each ticket group it runs:
git -C <REPO_PATH> add -- <file1> <file2> ...
git -C <REPO_PATH> commit -m "<ticket-specific message>"
This produces clean, reviewable git history with one commit per ticket per repo.
Commit message rules (rules/commits.mdc — always enforced)
[TICKET-XXXX] <emoji> <type>(<scope>): <subject>
| Rule | Detail |
|---|
| Language | English only |
| Ticket prefix | [TICKET-XXXX] when a ticket match exists; omit when NONE |
| Emoji | Always include — 🐛 fix, 🚀 feat, ♻️ refactor, 🎨 style, ⚡ perf, ✅ test, 📝 docs, 🔧 chore, 🔒 security, 🚧 wip |
| Type | One of: fix, feat, refactor, style, perf, test, docs, chore, security, wip |
| Mood | Imperative — "add", not "added" / "adds" |
| Subject length | ≤ 50 characters (not counting the [TICKET-XXX] prefix and emoji+type) |
| Ticket position | Prefix only — never inside the subject line itself |
Examples:
[TICKET-727] 🐛 fix(tasks): move back-to-table btn to right side
[TICKET-726] 🐛 fix(text-input): allow clicking inside widget
[TICKET-734] 🐛 fix(tasks): prevent stale WebSocket reconnect error
[TICKET-729] 🚀 feat(projects): demote lead adds them as contributor
🔧 chore(migrations): add project lead contributor column ← no ticket when unmatched
Constraints
- No branches — commit on the current branch of each repo; do not create or switch branches.
- No push — local commits only; remind user to push when ready.
- No bulk staging — always stage specific files per commit group, never
git add -A.
- Parallel subagents — all per-repo workers run simultaneously via the Task tool.
- No main-agent git ops — the main agent never runs
git add or git commit directly; all git work is delegated to the per-repo subagents.
- Worker contract — each subagent returns JSON:
{ repo, status, commits: [...], error }.
