jillesme/pi-herdr-squad — explained in plain English
Analysis updated 2026-05-18
Spin up one to four read-only Pi agents to investigate different parts of a codebase in parallel.
Compare frontend and backend logic or trace a bug across code, tests, and configuration at once.
Configure which AI model each investigation squad uses through project or global config files.
Run codebase investigations with a guaranteed read-only boundary that can never modify the checkout.
| jillesme/pi-herdr-squad | cschanhniem/clawping | cub3y0nd/entropic | |
|---|---|---|---|
| Stars | 24 | 24 | 24 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 3/5 | 3/5 | 3/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires an existing Pi session running inside a Herdr-managed pane before the package can be installed and used.
This repository is a package called pi-herdr-squad, an extension for the Pi coding agent that creates temporary teams of read only investigation assistants inside Herdr, a tool for managing Pi sessions. When a parent Pi agent needs to investigate something across a codebase, this package spins up a dedicated Herdr tab holding between one and four child Pi agents, gives each of them its own separate area of the codebase to look at, waits for each one to send back a structured report, and then hands the parent everything it needs to combine those findings into an answer. To use it, you install the package from npm into a Pi session that is already running inside a Herdr managed pane, then reload Pi or start a new session. From there you can start a squad with a simple command specifying either an automatically chosen number of agents or an exact count, describing what should be investigated, such as comparing frontend and backend validation logic or looking into checkout failures across the code, tests, and configuration. Plain language requests that clearly ask for a Herdr squad or parallel subagents will also trigger it. Which AI model each child agent uses follows a clear order of precedence: a model requested for that specific squad, then a trusted per project configuration file, then a global configuration file, and finally Pi's own normal default model if nothing else is set. Configuration files are read fresh every time a squad starts, so edits take effect immediately without needing a reload. The read only boundary is enforced directly through tool access: every child agent is only ever given tools to read files, search text, list directories, and submit its report, with no ability to run shell commands, edit files, or write anything except its own private report file, so an investigation squad can never modify the shared codebase it is looking at. The package has no third party runtime dependencies and is released under the MIT license.
A Pi coding agent extension that spins up temporary, strictly read-only teams of investigation agents inside Herdr to research a codebase and report back.
Mainly TypeScript. The stack also includes TypeScript, Pi, Herdr.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.