agentleFS
vadimvitvickiyvadimvitvickiy/vv-agent-kit/skills/committing-changes/SKILL.md

committing-changes

Skill0 starsChanged 50 days agoMITView on GitHub

Use when writing a commit message, opening a pull request, titling and describing a change for someone else to review, or deciding how to integrate a finished branch.

  • Commits and pushes

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 4 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    The repo's convention wins: Read git log --oneline -20 first.

    line 12
  2. Distinctive:

    ~50 characters, never past 72: No trailing period.

    line 28
  3. Distinctive:

    Call out what needs a human decision: a schema change, a new dependency, a user-visible behavior change, a deliberate workaround.

    line 67
  4. Very specific:

    Never --force on your own.

    line 85
  5. Distinctive:

    Remove only worktrees you created.

    line 87

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Committing changes

The diff shows what changed. Everything you write is for the why.

Ground rules

  • The repo's convention wins. Read git log --oneline -20 first; follow its prefixes, types, case and trailers. What follows is the default where it has none.
  • Never commit or push unasked. git add, git commit and git push wait for an explicit request. "Finish the feature" is not one.
  • Check the branch first. On main or master, branch — unless history shows the project commits there directly with no merges or PRs, and the human asked. Then say so.
  • One logical change per commit. A rename mixed with a behavior change makes bisect land on an unreadable diff — vvkit:debugging-systematically.

Subject line

<type>(<optional scope>): <description>
  • Imperative: completes "If applied, this commit will ___". add retry, not added.
  • ~50 characters, never past 72. No trailing period.
  • Lowercase after the colon, unless a proper noun or identifier.
  • Scope is a noun naming a section of the codebase: fix(parser):.
TypeFor
featA capability the codebase did not have
fixA defect repaired
refactorBehavior identical, structure different
perfBehavior identical, measurably faster or smaller
testTests added or changed, with no production change
docsDocumentation only
build, ciBuild system, dependencies, pipeline
choreHousekeeping that fits nothing above

Really both feat and fix means really two commits.

Body

  • Blank line after the subject, wrapped at 72 — git log indents four spaces, so 72 fits 80 columns.
  • In order: what was wrong or missing; why this approach, and the alternative rejected — the most useful sentence for whoever revisits it; what a reader would find surprising — a measured number, a constraint found partway, a deliberate temporary workaround.
  • Omit the body when the subject says it all. Ceremony on trivial commits trains readers to skip bodies.
  • Breaking changes: feat!:, or a BREAKING CHANGE: <what breaks> footer. Footers are Token: value, one per line, at the end.

Pull request

  • Title: subject-line rules, and readable with no other context — it shows in release notes and merge queues. fix: race in session cleanup that returned 502s under load, not Fix bug.
  • Description, in order:
    1. What changed — grouped by area, not file. Name the entry point to start reading from.
    2. Why — the problem, and the approach chosen over which alternative.
    3. What was verified — commands run and what they said — vvkit:verifying-changes. "Tests pass" without a command is an assertion.
    4. Out of scope — deferred work, and what a reviewer would otherwise flag as missing.
  • Call out what needs a human decision: a schema change, a new dependency, a user-visible behavior change, a deliberate workaround.
  • Keep it reviewable. Past ~400 changed lines reviewers find fewer defects per line. Split it, or say why not and point at the parts that deserve attention.

Integrating a finished branch

Integration is the human's decision — present it, do not take it:

  1. Run the full suite on the tree about to be integrated. Red: report and stop.
  2. Confirm the base branch from the plan, the conversation or the upstream — or ask.
  3. Offer the options: merge locally, push and open a PR, or keep the branch. Wait.
  • A local merge is verified by the suite on the merged result before anything is deleted; on failure the branch and worktree stay. A PR keeps its worktree for review fixes.
  • Discard only when asked in so many words, after the human confirms the exact list: branch, commits, worktree.
  • Worktree removal refused over uncommitted files: those files exist nowhere else — show them and ask. Never --force on your own.
  • A rejected push means the remote moved: investigate; force-push only on explicit request.
  • Remove only worktrees you created.

Never

  • AI attribution, unless the project's instructions or its history's trailers carry it.
  • A ticket id in code comments — a PR footer is its home.
  • "misc fixes", "wip", "address feedback" as a final message.
  • A description restating the diff file by file.

Before opening

  • The subject completes "If applied, this commit will ___"
  • Each commit is one logical change
  • The body says why, not what
  • The title reads correctly with no other context
  • The description names what was verified, with the command
  • The diff is small enough to review, or the split is explained

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.