agentleFS
tbhbtbhb/vale-ai-tells/AGENTS.md

vale-ai-tells

AGENTS.md118 starsChanged 37 days agoMITView on GitHub

Guidance for AI coding agents working in this repository. Read it alongside the per-tool documentation and any memory files the harness loads. Write Conventional Commits (type(scope): subject) with a DCO Signed-off-by trailer, and keep the subject under 80 characters. AI assistance is credited with a kernel-style Assisted-by: AGENT:VERSION [TOOL] trailer placed before the sign-off. Never credit a model through Co-authored-by. Draft every message in a repo-root COMMIT_AGENTMSG file before you run git commit. A gitignore entry keeps that file out…

The file

We split this AGENTS.md into the rules it gives an agent

4 rules. About 0 are common, 1 is distinctive, and 3 are 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.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Write Conventional Commits (type(scope): subject) with a DCO Signed-off-by trailer, and keep the subject under 80 characters.

    line 7
  2. Very specific:

    Write the full message (subject, body, and trailers) to COMMIT_AGENTMSG.

    line 11
  3. Very specific:

    Run mise run lint-commit-msg and resolve whatever it reports.

    line 12
  4. Distinctive:

    Commit the validated draft with git commit -s -F COMMIT_AGENTMSG.

    line 13

≈ “Common” is an estimate: at least 1 in 5 of the 42,283 AGENTS.md 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
AGENTS.md

Agent instructions

Guidance for AI coding agents working in this repository. Read it alongside the per-tool documentation and any memory files the harness loads.

Commit messages

Write Conventional Commits (type(scope): subject) with a DCO Signed-off-by trailer, and keep the subject under 80 characters. AI assistance is credited with a kernel-style Assisted-by: AGENT:VERSION [TOOL] trailer placed before the sign-off. Never credit a model through Co-authored-by.

Draft every message in a repo-root COMMIT_AGENTMSG file before you run git commit. A gitignore entry keeps that file out of history, so it serves purely as a scratchpad:

  1. Write the full message (subject, body, and trailers) to COMMIT_AGENTMSG.
  2. Run mise run lint-commit-msg and resolve whatever it reports.
  3. Commit the validated draft with git commit -s -F COMMIT_AGENTMSG.

The commit-msg stage runs four hooks from the shared devtools repository: commitlint (the Conventional Commits format and length bounds), commit-trailers (the trailer format and order), vale-commit-msg (prose, under this repo's own ai-tells and ai-tells-commits styles), and cspell-commit-msg (spelling). Run mise run devtools:prek-install once so the hooks run on every commit.

That hook stage is the real gate. mise run lint-commit-msg only previews it, so a clean recipe run predicts a clean commit without replacing the hook.

Prose lint output

The toolchain defaults to the agent template: mise run lint-prose, mise run lint-messages, and the vale pre-commit hook all pass --output=ai-tells-agent.tmpl. Pass the flag yourself only when invoking vale directly. The template prints one self-contained line per finding (location, severity, rule, the exact matched text, and the replacement parameter when the rule defines one) plus a totals line, so you can apply fixes without re-reading context through separate commands. Empty output means a clean run, and the exit code reports the result.

ai-tells.zip includes the template, tracked here at styles/config/templates/ai-tells-agent.tmpl, so a repository syncing the core style alone can pass the flag. A vale sync puts it under StylesPath/config/templates/, where vale looks up an --output name.

Treat the output format as a published interface. The prose-fix skills in devtools parse it: they count findings with grep -c '^[0-9]' and read the replace_with= field to apply a correction. Reshaping a line or renaming a field breaks those skills and anything else reading the format, so a change there is a breaking change for consumers rather than a local edit.

Drafting a document

Each recipe named so far runs one checker over many files, so a draft that clears mise run lint-prose can still fail spelling and structure afterwards, which is how a short document turns into four rounds of linting. mise run lint-draft <file> runs vale, cspell, and rumdl over one document and reports all three at once.

It also probes the path before accepting a clean run. Vale matches a path against the sections in .vale.ini exactly, and a path outside every section has an empty style list. Vale then reads that file and exits 0 without printing anything, which is what a clean document produces too. The recipe sends known-bad text through vale under the target's own path and reports unscoped-path rather than a pass. Silence from that probe means a draft under styles/, tmp/, or .claude/ has no style bound to it. Move it to a path .vale.ini names.

mise run fix-prose-replacements <file> then applies the findings whose rule defines a correction. The Google and proselint rules define one, ai-tells.HouseStyle does too, and the template prints it as replace_with= on the finding line. The script refuses any finding whose span no longer contains the quoted text, and it skips a pronoun contraction that follows a preposition, because a verdict from that is narrower breaks when the contraction goes in. Whatever it leaves behind needs a decision.

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.