agentleFS
CRELLIA-S-LCRELLIA-S-L/srs-dd/.claude/skills/srs-commit/SKILL.md

srs-commit

Skill0 starsChanged 32 days agoMITView on GitHub

Prepare a commit of finished work — hand it to the review and check procedures, establish what the message covers, put the generated files in the index, read the diff including untracked files, and write the message with the requirement identifiers the change implements. Invoke when a change is finished and a commit is the next step. It commits nothing itself: the commit is made with whatever git client the project is driven by.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    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. Very specific:

    Never reach for --no-verify, and never pipe the commit into anything: the exit code of a pipeline belongs to its last command, so a hook's …

    line 40
  2. Distinctive:

    Write it in English whatever the language of the conversation, and name no tool, agent or session in it.

    line 52

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “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
<!-- Copyright (c) 2026 CRELLIA S.L. — part of SRS-DD, distributed under the MIT License -->

Preparing a commit

The commit is the maintainer's act. This procedure is everything that has to be true before it: that the work was reviewed and checked, that the message covers what is actually being committed, and that what the gate compares is in the index.

Procedure

  1. Review it and check it first, both. .claude/skills/srs-review/SKILL.md closes on two consecutive passes that found nothing by different means, and .claude/skills/srs-check/SKILL.md derives what the change calls for from the requirements it touched and offers exactly that; neither runs anything unasked, which is SRS-DD-ART-030 of srs-dd/method.md. Neither stands in for the other: a review that closed clean is not a run of the suites the touched requirements list, and a green set is no judgement of the work. A commit is the moment after which a red check stops being yours. Where either already ran in this session and nothing has changed since, say which and do not run it again; where anything was edited after it finished, it runs again.
  2. Establish what is being committed.

    git status --short
    git diff --cached --stat
    git diff --stat

    git status --short is the only one of the three that shows an untracked file, and a change that is mostly new files is invisible in a diff: take the new files from its ?? lines and read them directly. Where anything is staged, ask whether the message covers the staged set alone or everything uncommitted, show both sides with their file counts, and wait — a message describing the wrong set is found out only after the commit exists. Where a merge, rebase or cherry-pick is in progress, stop and say so: that commit needs a different message.

  3. Put the generated files in the index.

    python3 srs-dd/tools/srs_gate.py

    It runs what the hook runs, and it compares every generated file with what git holds rather than with the working tree: specs/90-traceability.md, and arch/90-map.md and grounds/90-dashboard.md where the project carries those layers. A file regenerated and left unstaged is the commonest refusal there is — git add the path it named, then run the gate again. Never reach for --no-verify, and never pipe the commit into anything: the exit code of a pipeline belongs to its last command, so a hook's refusal read through grep looks exactly like a commit that was made.

  4. Read the diff, not your memory of the work. git diff --cached, or git diff HEAD where the message covers everything, plus the untracked files in full. Writing from intent is how these go wrong: the message says what was meant to happen, the diff is the only witness anyone consults later. Take out of it what the message owes its reader — the behaviour that changed, a contract broken for its callers, a migration, a new configuration key, anything deliberately left out or hardcoded. Flag, rather than describe, what should not be committed at all: a credential or anything else SRS-DD-ART-100 keeps out of a repository, a scratch file, a debug print, an edit that landed in the wrong file.
  5. Write the message. An imperative title; a body of plain sentences where there is something to say and none where there is not; and the requirement identifiers the change implements, bare, in a trailing parenthesis. SRS-DD-ART-060 asks the commit or the pull request for them, and this is the one place a record is named by its number alone: everywhere else AGENTS.md asks for the title, the file and the status, and here a citation would break the line, because a tool that prints a changelog or a log prints such a line whole. The identifiers are the ones the change was planned from, not the ones a search turns up afterwards. Write it in English whatever the language of the conversation, and name no tool, agent or session in it.
  6. Hand it back. Say which files are waiting, what the checks reported, and anything flagged in step 4; then give the message and stop. Nothing is committed, pushed or tagged here, and the maintainer takes it from there with the client they use — which may not be a console at all.

What this procedure does not do

Commit. Push. Tag. Stage anything but the generated files the gate compares. Decide that a red check is acceptable, or weaken one to get past it.

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.