agentleFS
mannobeatsmannobeats/hyper-taste/skills/agents-md/SKILL.md

agents-md

Skill1 starsChanged 4 days agoMITView on GitHub

Keep agent instruction files (AGENTS.md, CLAUDE.md, GEMINI.md) small and in sync. Use when creating, auditing, slimming or splitting one, before adding a rule or memory to one, when an agent keeps repeating a mistake, or when deciding whether guidance belongs in AGENTS.md, docs/ or a skill.

37 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use jira, how and 9 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • MicrosoftSkill

    jira

    Jira issue workflows for search, issue updates, transitions, comments, field discovery, and interactive credential setup via the Jira REST API. Use when you need to configure Jira access, search with JQL, inspect an issue, create or update work items, move an issue between statuses, post comments, or discover required fields for issue creation.

    Also in 3 projects that use this

  • jaydubya818Skill

    how

    Use for runtime, ownership, layering, code-walkthrough, and placement questions.

    Also in 3 projects that use this

  • asterkinSkill

    adr

    Create Architecture Decision Records (ADRs) following project standards. Guides template usage, process, and best practices for documenting architecturally significant decisions.

    Also in 3 projects that use this

  • jaydubya818Skill

    why

    Use when design rationale must be recovered from source history and available records.

    Also in 3 projects that use this

  • photon-groveSkill

    unslop

    Detects and rewrites generic, overly polished, or AI-sounding prose while preserving meaning. Use when the user asks to "unslop", humanize, de-AI, tighten, or review whether writing sounds generated.

    Also in 4 projects that use this

  • AWSSkill

    wait-what

    Stop. That last message did not land: re-pitch it.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 33 edited versions we read keep these.

  • 3

    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:

    Ask of each line: does it apply to nearly every task in this directory?

    line 46
  2. Distinctive:

    Delete getting-started sections, duplicates, file paths, enforced and generic rules.

    line 75
  3. Very specific:

    Don't have an agent read the repo and generate one: that produces a long, generic file.

    line 85
  4. Distinctive:

    Every installed skill's name and description also load on every request.

    line 109

“Common” is measured: kept by at least 6 in 10 of the 33 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

AGENTS.md / CLAUDE.md

AGENTS.md loads in full into every request, so every line costs tokens and attention on tasks it has nothing to do with. Keep it tiny, and let it point to detail that loads only when needed.

The scripts are in scripts/ next to this file. Run them from the repo root with bash.

1. One file, no drift

AGENTS.md is the source of truth. CLAUDE.md and GEMINI.md are symlinks to it, so an agent writing to either edits the same content.

bash <this skill>/scripts/sync.sh .            # every AGENTS.md in the tree gets a CLAUDE.md link
bash <this skill>/scripts/sync.sh . --gemini   # also GEMINI.md

In each directory (skipping node_modules, .git, dist, build, .venv): a CLAUDE.md with no AGENTS.md is renamed to AGENTS.md, then linked; one identical to AGENTS.md becomes the link; one that differs prints the diff and exits 1, so merge its unique lines into AGENTS.md and rerun (--force moves it to CLAUDE.md.bak instead). Safe to rerun after adding a nested AGENTS.md.

Where links can't work (Windows without Developer Mode, or a CLAUDE.md that needs Claude-only lines), CLAUDE.md holds @AGENTS.md on its first line and those lines below it. A CLAUDE.md that only says "see AGENTS.md" drifts: agents append rules to it instead. Recent Claude Code reads a lone AGENTS.md directly, unless a CLAUDE.md or CLAUDE.local.md sits in the same tree; the link covers older versions and other tools.

2. Audit

bash <this skill>/scripts/audit.sh .

Flags files over 50 lines or ~1,500 tokens, onboarding sections, task-specific headings, sections over 15 lines, paths that no longer exist, and CLAUDE.md or GEMINI.md that aren't links. A CLAUDE.md with no AGENTS.md is audited too. Treat the output as a to-do list, not a verdict.

3. What goes where

Ask of each line: does it apply to nearly every task in this directory?

ContentHome
What the project is in 1–2 sentences, non-standard tooling (pnpm, an unusual script), top-level layoutAGENTS.md
Conventions for one area: CSS, testing, API, databasedocs/<area>.md, linked with its trigger. Claude Code only: .claude/rules/<area>.md with paths: loads on its own
A workflow that happens sometimes: a PR, a release, a delivery checklistdocs/<workflow>.md, linked with its trigger
Rules for one package in a monorepoThat package's own AGENTS.md (plus its link)
A procedure reused across repos, or one that ships scriptsA skill: only its name and description cost context until used
Install steps and onboarding for humansREADME. The agent works in a repo that's already set up
File paths, folder trees, what package.json or a config already saysNowhere: the agent can look, and copies go stale

An @path import loads the whole file every session, so it saves no context; use a plain link with a trigger for anything on demand.

4. Write a line

  • Trigger first: "Before writing CSS, read docs/css.md." The trigger tells the agent when to load it.
  • Say the target, not the ban: "Use pnpm" beats "Don't use npm". Keep a prohibition only for a real guardrail, paired with what to do instead.
  • One meaning, one place. A rule written twice drifts.
  • Delete what the agent already does ("write clean code", "be careful") and what the linter or the code enforces.

5. Fix an existing file

  1. sync.sh ., merging any diverged content.
  2. audit.sh . and read the whole file.
  3. Delete getting-started sections, duplicates, file paths, enforced and generic rules.
  4. Move each task-specific section verbatim to docs/<topic>.md and leave one line: - Styling: read docs/css.md before writing CSS.
  5. Monorepo: move package-only sections into apps/<pkg>/AGENTS.md with a one-line description of the package at the top, then sync.sh . again.
  6. Rerun audit.sh . until only intentional warnings remain, then check every link resolves: grep -o 'docs/[^ )]*\.md' AGENTS.md | xargs ls.

6. Start a new file

Don't have an agent read the repo and generate one: that produces a long, generic file. Start with this, then sync.sh .:

# <Project>

<One or two sentences: what it is and who uses it.>

- Package manager: pnpm
- Layout: `apps/web` (React frontend), `apps/api` (FastAPI backend), `packages/` (shared libs)

## Conventions

It grows only from real mistakes: when an agent gets something wrong, write the correction where the table puts it (usually a doc) and add the doc's trigger line here if it isn't linked yet.

Loading

  • Claude Code loads CLAUDE.md (or a lone AGENTS.md) from the launch directory and every parent at startup, and a subdirectory's file when it reads files there.
  • Codex and other AGENTS.md agents read the chain from the repo root down to the launch directory, so launching at the root doesn't load apps/api/AGENTS.md. The root file says "Working in apps/api? Read apps/api/AGENTS.md." for agents that don't load nested files.
  • Every installed skill's name and description also load on every request. Remove skills you don't use.

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.