Skip to content

About

Simple Agentic Swarms: A Skill for your agent to Shepherd (via herdr) many instances of Pi

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

36 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Shepherd's Pi

Shepherd's Pi logo — a pi symbol shepherding three sheep inside a circular badge

Combine Herdr and Pi

Like peanut butter and chocolate, herdr and pi are two great tastes that taste great together. Herdr gives you the multiplexed panes and the socket API to drive them; pi gives you a small, fast, model-agnostic agent to put in each pane. A herdr tab full of idle pi panes is a flock — ready for a shepherding agent to dispatch work into.

A skills repo, in the style of ponytail: plain-text instructions for standing up a herdr workspace, wiring pi to an inference backend of your choice, and raising a flock of idle helper panes ready for another agent to shepherd. Bring your own backend, your own model, your own repo.

Codex shepherding two pi agents through parallel reviews in Herdr

Codex dispatches two pi helpers and reads their progress. Recorded at 2.5× speed.

What is Herdr?

Super TMUX. As in Terminal Multiplexer. You have one terminal with many sub-terminals "multiplexed" together — an MCP server (and CLI) that agents can use to pilot multiple shell programs: split panes, launch commands, read output, wait for output, rename and label panes, all scriptable. github.com/herdrdev/herdr

What is Pi?

A minimalist agentic coding harness — Pi Coding Agent. It is so minimal it is actually hard for a human to use. That minimalism reduces distractions for the LLM inside it and makes it subtly more effective: fewer built-in opinions, less ceremony, a smaller surface to get confused by.

In this repo

Setup — get pi and herdr running:

Skill What it does
skills/pi-install Install pi and confirm it runs
skills/pi-inference-backend Point pi at any OpenAI-compatible inference endpoint and set a model as the CLI-wide default
skills/herdr The herdr CLI itself — workspaces, tabs, panes, and agent control, straight from the tool's own authority (herdr --help)
skills/herdr-helpers-tab Recreate a herdr tab with a controller pane plus N idle pi panes, ready for another agent to dispatch into
skills/herdr-flock Run one project as a herdr flock: shepherd, out-group reviewer and four helpers, with on-disk lanes and a gated review-and-merge pipeline
skills/tmux-drive Drive an interactive TTY/TUI program hands-free from a non-interactive agent, on an isolated tmux server

Shepherd guidance: brief pi helpers, inspect their work, and steer corrections.

These skills are for the controlling agent. Send each helper a focused brief with the relevant guidance, then check its artifacts and evidence. See Guiding pi helpers for the dispatch and review loop.

Skill What it does
skills/tdd-red-green-blue Guide pi through Red, Green, and Blue; verify test evidence and correct drift
skills/functional-cohesion Supporting doctrine: group code by job and expose a narrow interface
skills/three-cs Guide pi to separate domain data from mechanisms without speculative frameworks
skills/unix-philosophy Brief a single tool contract and verify pi's output composes with its consumer
skills/concision Brief pi on audience and purpose; review its draft for clarity without losing meaning

Deeper reference material that doesn't belong inline in a skill lives in docs/:

  • docs/shepherding.md — how the shepherd assigns, checks, and corrects pi work; each shepherd skill links its longer doctrine
  • docs/copilot-doctrine.md — the opposite posture: one agent advising one human through work on a system where a wrong command is expensive and attributable. The human runs everything
  • docs/pi-config-reference.md — pi's three config files (models.json, settings.json, auth.json) and how they interact
  • docs/installing.md — per-agent skill install commands (Claude Code, Codex, pi, and a generic fallback)

Installing the skills

Each skill is a self-contained SKILL.md (some carry a small bundled tools/ script) — plain markdown, no framework dependency. See docs/installing.md for the symlink command for Claude Code, Codex, pi, and any other agent that reads SKILL.md files.

License

MIT for everything authored in this repo — see LICENSE. skills/herdr and skills/tmux-drive are mirrored from Shawn Hartsock's newt-agent under its Apache-2.0 license. skills/three-cs and skills/functional-cohesion are adaptations under the same license. Their source notices identify the changes; the three-Cs doctrine reference also retains its attribution.

About

Simple Agentic Swarms: A Skill for your agent to Shepherd (via herdr) many instances of Pi

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages