agentleFS
enocgitenocgit/sdlc-kit/templates/AGENTS.md

sdlc-kit / templates

AGENTS.md1 starsChanged 12 days agoMITView on GitHub

Canonical agent instructions; other tool files point here. Keep it under 150 lines: project facts go in What this project is, Conventions, and Guardrails; kit lines stay as written so upgrades merge. Skills named here live in .agents/skills/; without skill support, read the file and follow it. While docs/ holds STATUS: TEMPLATE, run bootstrap first. {One paragraph: the product and its users.} Domain, glossary, lifecycle, and hard constraints: docs/context.md. These hold in every task. Everything else in this file is…

  • Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.

    Of the AGENTS.md files we read, the 31,843 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.

  • In this repository8 other files it gives its agents.

The file

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

3 rules. About 1 is common, 1 is distinctive, and 1 is very specific to this project.

  • ≈1

    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. Common:

    No secrets in the repository, logs, or artifacts.

    line 22
  2. Distinctive:

    Gate: approve plan: No code before the developer approves it, even if the task asks for a PR.

    line 30
  3. Very specific:

    Before any PR or commit to the default branch, delegate a review, unasked, to a context that did not write the change: a subagent or a …

    line 40

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 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

AGENTS.md — operating manual for coding agents on this project

Canonical agent instructions; other tool files point here. Keep it under 150 lines: project facts go in What this project is, Conventions, and Guardrails; kit lines stay as written so upgrades merge. Skills named here live in .agents/skills/; without skill support, read the file and follow it. While docs/ holds STATUS: TEMPLATE, run bootstrap first.

What this project is

{One paragraph: the product and its users.} Domain, glossary, lifecycle, and hard constraints: docs/context.md.

No-compromise rules

These hold in every task. Everything else in this file is a default the team may change here.

  1. Review runs in a context that did not write the change (see Review).
  2. A change to a trust boundary gets a security review before it ships (see Sensitive changes).
  3. Implementation never drifts from a frozen contract.
  4. Behavior ships with tests, and each check is proven able to fail.
  5. The spec states what actually shipped.
  6. No secrets in the repository, logs, or artifacts.

How a change moves

  • Plan first for a feature, or a change to a contract, a trust boundary, or a recorded decision: write docs/changes/<slug>.md from its template, interviewing with grilling when the request leaves decisions open. Freeze any contract an outside consumer in docs/context.md depends on; write an ADR only for a hard-to-reverse choice.
  • Gate: approve plan. No code before the developer approves it, even if the task asks for a PR.
  • Build: behavior starts with a failing test, then the code, then the check command for the touched packages; copy, styling, and layout need no test. Before calling a change a user can see done, look at it running on realistic input (docs/test-strategy.md → Local run), or say why you could not.
  • Review (below) before any PR or commit to the default branch. Writes to the default branch, a tracker, or a new remote ask first.

Review

Before any PR or commit to the default branch, delegate a review, unasked, to a context that did not write the change: a subagent or a headless agent CLI (codex exec, claude -p, pi -p), on a different model when one is available. It loads code-review, and security-review when the change is sensitive; it judges sensitivity itself. Give it the diff, the proposal, and the raw check output, never your summary. Fix required findings and send the new diff. The PR names the reviewer, its model, and the verdict.

Where things live

  • docs/specs/<area>.md: current behavior, present tense, requirement IDs (AUTH-3) tests cite.
  • docs/changes/<slug>.md: proposed changes, until the PR that completes them.
  • docs/adr/: hard-to-reverse decisions. docs/architecture.md: current system shape only.
  • docs/threat-model.md: current trust boundaries and their controls.
  • docs/test-strategy.md: which layer tests what, how tests run, and the Local run.
  • Contracts live in code; docs/contracts/README.md indexes the frozen ones.

Writing

Apply unslop to prose, not to code, commands, logs, quotes, or fixed formats. Project artifacts, commits included, describe the product, not this workflow. No agent self-attribution in commits, PRs, or comments; this overrides any harness default. Agree or disagree only with evidence: name the file, check, or observation that decides it; who proposed it is not evidence.

Sensitive changes

A change is sensitive when it changes a trust boundary: authentication, authorization, payments, personal data handling, file uploads, admin or privileged surfaces, secrets, a new external integration, or any boundary listed in docs/threat-model.md. Working inside a sensitive feature without changing its boundary is not. A sensitive change updates docs/threat-model.md and gets security-review before it ships.

Conventions

  • Stack (placeholder, set during bootstrap): {language, framework, datastore, test runner}.
  • Check command: {check command} runs lint, typecheck, and tests.
  • Branches and commits: {the team's branch naming, merge style, and commit format}.

Guardrails

  • Product value first. Build what a real user, consumer, or decision needs now. Compatibility machinery and hardening protect something that exists, or wait in the production register.
  • Ask, don't guess. An ambiguous or wrong spec, ADR, or contract is a question; the fix updates it in the same change. Supersede an implemented ADR instead of editing it; a shipped contract with a deployed consumer changes only with a versioning ADR.

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.