agentleFS
chengzi-iochengzi-io/pier/AGENTS.md

pier

AGENTS.md0 starsChanged yesterdayMITView on GitHub

pier = a skill + stdlib-only Python 3.9 CLI that turns agent Markdown files into pi argv, so a host agent delegates work to Pi on any model. The caller is an AI: every byte pier prints costs its tokens. - skills/pier/ is the delegation tool: SKILL.md is what the host agent reads, references/agent-authoring.md is the frontmatter reference (read only when writing agents), README.md documents the skill and CLI for humans. The root README.md covers the plugin as a whole.…

  • 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 32,671 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 repository3 other files it gives its agents.

The file

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

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    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:

    Test: python3 -m unittest discover -s tests (mock Pi, no model calls).

    line 50

≈ “Common” is an estimate: at least 1 in 5 of the 41,581 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

pier = a skill + stdlib-only Python 3.9 CLI that turns agent Markdown files into pi argv, so a host agent delegates work to Pi on any model. The caller is an AI: every byte pier prints costs its tokens.

Layout

  • skills/pier/ is the delegation tool: SKILL.md is what the host agent reads, references/agent-authoring.md is the frontmatter reference (read only when writing agents), README.md documents the skill and CLI for humans. The root README.md covers the plugin as a whole.
  • skills/orchestrate/SKILL.md is the plan → execute → evaluate loop for multi-part work. It uses pier; the pier skill covers only calling Pi and never refers to it.
  • skills/pier/scripts/pier: CLI entry (argparse, commands). pier_agents.py: frontmatter, lookup, validate. pi_runtime.py: RunSpec -> Pi flags, running Pi, output budget. pi_state.py: $PIER_HOME, sidecars, locks, worktrees. toon.py: TOON encoder.
  • skills/pier/agents/: bundled examples. mock/mock_pi: fake Pi for tests.
  • .claude-plugin/: plugin.json + marketplace.json, so the repo installs as a Claude Code plugin from its own marketplace.
  • hooks/hooks.json + scripts/<event>/<job>.py: plugin hooks for orchestrator mode, one job per script; today only scripts/session-start/orchestrator_reminder.py, silent when PIER_ORCHESTRATOR=off. Tool limits live in user settings (README, Orchestrator mode). Tests: tests/test_<job>.py.

Invariants

  • stdout = the reply only. stderr: pier: <ref> started (<agent>) first, then pier: <ref> running … progress lines while Pi runs (every 2 min), then pier: <ref> ok|failed (<agent>); a failure adds a final pier: error: … and a non-zero exit (124 when the run deadline passed).
  • Behaviour lives in the agent file; the CLI adds only --agent-file, --model, --worktree, --dry-run. Sessions are addressed by ref only. Keep it that way: a new flag needs a reason the agent file cannot cover.
  • Pi does not persist system prompt/tools/skills/extensions (or the run deadline): the sidecar stores the flags and send replays them.
  • Pi stays in pier's process group and is killed as a tree, on a deadline and on a signal alike. Every run holds a $PIER_HOME/locks flock slot (global + per caller session); a queued run says so once and keeps its deadline.
  • State only under $PIER_HOME; never write into the project or Pi's config.
  • Unreleased: change contracts freely, no compat shims.

Changing things

  • A CLI or frontmatter change updates skills/pier/SKILL.md, skills/pier/README.md, skills/pier/references/agent-authoring.md and tests in the same change.
  • A SKILL.md is loaded on every use: add a line only if it changes what the caller types or does, or prevents a retry.
  • Test: python3 -m unittest discover -s tests (mock Pi, no model calls); PIER_TEST_REAL_PI=$(command -v pi) adds a real-Pi argv check.

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.