finish-work

>- FMC finish-work workflow: run bun run check, fix failures, commit with a user-facing English message by default, and only show a draft when the user clearly did not want a commit. Use for finish work, run checks, wrap up, land changes, or create commits.

First seen Jun 22, 2026

Installation

$ npx skills add fixmyberlin/fixmyskills --skill finish-work

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Not declared

Cursor Not declared

Codex Not declared

GitHub Copilot Not declared

Windsurf Not declared

Gemini CLI Not declared

Cline Not declared

OpenCode Not declared

Repository health

License LICENSE

Default branch main

Open issues 0

Status Active

Package contents

Files included with this skill beyond the listing page.

skill md SKILL.md 3,510 B

docs SUMMARY.md 273 B

History

Jun 22, 2026 First seen on skills.sh

Sep 9, 2026 First recorded snapshot · 46 installs

SKILL.md

Finish work

Run in the project root, or in each changed monorepo package (app/, processing/, etc.).

Workflow

Verify: read scripts.check in package.json, run bun run check, fix the root cause, and rerun until green. lint/format may rewrite files.
Keep fixes together: stage lint/format changes with the functional changes; never make a separate "lint" or "format" commit.
Write the commit message using the format below.
Default: commit. Run the commit path below unless the user clearly did not want a commit.
Draft only when no-commit intent is clear. Show the message and stop; do not run git commit.

No-commit intent includes: explicit deferral ("don't commit", "draft only", "message only", "what would the commit be"), check/fix-only asks ("run check", "fix lint", "fix CI" with no wrap-up), review or question-only turns, or the user saying they will commit themselves.

When intent is ambiguous, prefer committing — especially after implementing changes or when the user invoked finish-work to wrap up.

Commit path: git status, git diff, and git log -5 in parallel; stage relevant files only; git commit via HEREDOC; git status to verify.

Safety: no push unless asked. No --no-verify. No amend of pushed commits. Do not stage secrets or unrelated dirty files.

Already committed and verify only fixed lint/format: git commit --amend --no-edit when HEAD has not been pushed; otherwise stage and ask before amending.

E2E is not in check. For UI/routes/auth changes, also run bun run e2e (or check-full when defined). See [playwright-skill](../playwright-skill/SKILL.md).

Script names: [package-json-scripts.md](../tech-stack/references/package-json-scripts.md).

Commit message

<Topic>: <Desc>

- <user-facing outcome bullet>
Ping https://github.com/org/repo/pull/123

Topic: scope (Map, Auth, Processing, Dev, etc.). FMC internal-only changes (deps, CI, tooling) use Dev.
Desc: imperative outcome, not filenames.
Body: bullets of what users/operators see, do, or get after this lands. Include why when the conversation established it. Stay at outcome level; do not list files, symbols, refactors, or diff steps. Bodies feed [user-changelog](../user-changelog/SKILL.md), so they must be readable without the patch.
Ping: if user cited a PR/issue, add one Ping <full-url> line per item.

Example:

Region map: clamp initial viewport to bounds

- Users: map fits selected region on load instead of generic Germany view (confusing deep links)
- Preserve explicit zoom in URL so shared links do not jump
Ping https://github.com/FixMyBerlin/tilda-geo/issues/5678

Internal-only (no user-visible change): maintainer bullets ok; omit user bullets when none apply.

[close-worktree](../close-worktree/SKILL.md) (land worktree → develop; calls this skill if dirty) | [unslop-code](../unslop-code/SKILL.md) (post-LLM cleanup; calls this skill per phase commit) | [playwright-skill](../playwright-skill/SKILL.md) | [tech-stack](../tech-stack/SKILL.md) | [review-dependabot](../review-dependabot/SKILL.md) / babysit | [user-changelog](../user-changelog/SKILL.md)