agentleFS
vib-zvib-z/stampede/AGENTS.md

stampede

AGENTS.md9 starsChanged 22 days agoMITView on GitHub

These rules apply to every agent that edits this repo: Claude, Codex, Cursor, Hermes, Grok Bot or any other. The rules for running a job with stampede are not here: they live in SKILL.md and the skills it names. The design is in docs/superpowers/specs/2026-10-21-stampede-lean-design.md. Read it before your first change. The README is the front door. It covers every point at least briefly, then links to the doc with the depth.

  • Reads credentials
  • 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 31,700 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 repository6 other files it gives its agents.

The file

Show its rules and text

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

14 rules. About 0 are common, 13 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈13

    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. Distinctive:

    Every claim, command and number in the README is true of the current code.

    line 12
  2. Distinctive:

    Every command in it was run, against a scratch job folder with a fake Herdr when it would otherwise touch real panes.

    line 13
  3. Distinctive:

    Roadmap and status: update them at every milestone: what works, what was tested, what is next.

    line 15
  4. Distinctive:

    Every front door: a Herdr chat, Claude, Codex or Cursor, Grok Bot (preferred bot), Moshi with any of them, or others.

    line 25
  5. Distinctive:

    Every agent CLI the lead can start: Claude, Codex, Cursor.

    line 26
  6. Distinctive:

    Configure: every key in config.example, with its default.

    line 32
  7. Distinctive:

    Safety: every gate that needs the owner's yes, what never happens without asking, and how to reopen.

    line 33
  8. Distinctive:

    Reference: every script, its flags and its exit codes.

    line 35
  9. Distinctive:

    Keep only images a doc uses.

    line 46
  10. Distinctive:

    One context per agent: every skill, guide and brief stands on its own.

    line 60
  11. Very specific:

    No secrets, .env files, keys or tokens.

    line 68
  12. Distinctive:

    No personal paths, names, hostnames, owner tags or private repo or skill names.

    line 69
  13. Distinctive:

    Commit only your own change.

    line 77
  14. Distinctive:

    Do not touch the owner's private skills repo from here.

    line 78

≈ “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

Rules for agents working on stampede

These rules apply to every agent that edits this repo: Claude, Codex, Cursor, Hermes, Grok Bot or any other. The rules for running a job with stampede are not here: they live in SKILL.md and the skills it names.

The design is in docs/superpowers/specs/2026-10-21-stampede-lean-design.md. Read it before your first change.

1. The README is the product's front page. Keep it current at every step

  • Same commit: any change to behaviour, commands, scripts, skills, front doors, the layout or the status updates README.md in that commit. It also updates every doc under docs/ and remote/ that the change affects.
  • Incomplete otherwise: a change without its README update is not done.
  • True today:
    • Every claim, command and number in the README is true of the current code.
    • Every command in it was run, against a scratch job folder with a fake Herdr when it would otherwise touch real panes.
    • Anything not built yet goes under the roadmap, marked as planned.
  • Roadmap and status: update them at every milestone: what works, what was tested, what is next.
  • The cross-client suite: any change to behaviour, a front door, a client or a worker kind updates, in that same commit, the suite (tests/test_clients.py, tests/clients_run.py) and docs/CLIENTS.md (its matrix, what differs, and the live runbook). A front door or a client is a row in CLIENTS and in that doc; a worker kind is an entry in WORKERS under each backend that takes it, with its refusal in REFUSED for one that does not, and in HANDED when a Core job hands it to a Herdr TUI pane. A client that does worse gets a finding there and a test marked as an expected failure.
  • Standard: short, true and complete. A stranger can understand, install, use, configure, troubleshoot and extend stampede from the README and its docs alone, with no questions to anyone.

What the README and docs must cover

Shape the sections to fit the content; no template is fixed. Together, the README, docs/, remote/ and the skills answer all of these:

  • What and why: what stampede is, the problem it solves, and who it is for.
  • See it: real screenshots of a run, with captions.
  • How it works: the flow from ask to report and reopen, the roles (owner, front door, lead, workers, reviewer), and what lives where.
  • Every option:
    • Every front door: a Herdr chat, Claude, Codex or Cursor, Grok Bot (preferred bot), Moshi with any of them, or others.
    • Every agent CLI the lead can start: Claude, Codex, Cursor.
  • Install:
    • Prerequisites, with versions.
    • A copy-paste quick start.
    • A full setup, and the setup per front door.
  • Use: a walkthrough of one job, start to finish, with the actual commands and their output.
  • Configure: every key in config.example, with its default.
  • Safety: every gate that needs the owner's yes, what never happens without asking, and how to reopen.
  • Troubleshoot: known failures with their fixes, and an FAQ.
  • Reference: every script, its flags and its exit codes. Also the repo layout.
  • Contribute: how to add a front door or an agent CLI, and how to run the tests.
  • Roadmap, status and license.

The README is the front door. It covers every point at least briefly, then links to the doc with the depth.

Imagery and video

  • Images go in docs/images/, 1800 px wide.
    • Each is either a real screenshot or rendered from an HTML source in docs/images/src/, using system.css and render.sh, so it can be re-rendered.
    • Each has alt text that says what it shows.
    • Each caption says what is real, what is a mock-up, and the date.
    • Keep only images a doc uses.
    • Product marks go in docs/images/logos/ as the owner's own SVG, each credited in its SOURCES.md.
  • Screenshots: take them from real runs. Crop out personal details first.
  • Video: the film assets stay as they are for now. The owner decides when to re-cut them, after significant progress.
    • The poster and the teaser go in assets/.
    • The master goes in a GitHub release, never in git.

Before each commit that touches the README or docs

  • python3 -m unittest discover tests passes. This includes test_shape.py: links, anchors, line limits, frontmatter, no em or en dashes, and no personal details (home paths, private paths, private names), in docs/superpowers too.
  • Render the README the way GitHub does (gh api markdown -F text=@README.md) and look at it. Fix overflow, stacked badges and broken images.

2. The skills and the playbooks

  • Where they live: the playbooks are skills/ (the four skills) and remote/ (the front-door guides).
  • The four skills are copies. skills/stampede, skills/herdr-fleet, skills/grokbot-herdr and skills/share-plan come from the owner's own skills, which are their source. Change a skill there first, then copy it here and scrub it (section 3); never edit it here first. Only what makes the copy public (a scrubbed name or path, a setting with a default) differs from the source.
  • One context per agent: every skill, guide and brief stands on its own. An agent gets only its own file plus its brief, never the whole repo.
  • Line limits: a skill's SKILL.md at most 140 lines, every other playbook file at most 160 (test_shape.py).
  • Credit: a linked role or skill from someone else names its owner, its license and a pinned commit.
  • No history or reasoning in playbooks. Git keeps it.
  • One place per rule: each rule lives in one file. Others link to it.

3. Open source hygiene

  • Nothing personal or secret:
    • No secrets, .env files, keys or tokens.
    • No personal paths, names, hostnames, owner tags or private repo or skill names.
    • The three skills herdr-fleet, grokbot-herdr and share-plan may be named (the owner's call, 21 Oct 2026).
  • Machine-specific settings are environment variables listed in config.example. Each has a default.
  • Scan before every push for keys and personal details.
  • The repo is private until the owner says yes to making it public.

4. Git

  • Work on a branch and open a draft PR. Merging to main is the owner's call.
  • Commit only your own change. Never commit caches, scratch files or media over 10 MB.
  • Do not touch the owner's private skills repo from here; it is changed only in its own checkout, by the owner's own jobs.

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.