Back to directory

ROACH PI

An extension suite for the Pi coding agent that adds a structured engineering loop of clarifying a goal, driving verified execution, reviewing changes and searching code, implemented as plain, inspectable TypeScript and Markdown.

ROACH PI

ROACH PI bundles eight extensions into one disciplined engineering loop for the Pi coding agent. A clarify step forces a vague request into the open before implementation starts, asking one focused question and offering concrete choices; a goal step then creates or activates a durable goal, drives implementation, records evidence, and requires a verifier's pass before it lets the agent consider the work done, continuing automatically after a failed verification until it passes.

A review command checks non-trivial changes before merging, and further bundled tools cover fast code search, language-server-backed code intelligence across more than 40 languages, an MCP adapter, workspace memory across sessions, and support for nested AGENTS.md files. Everything, including commands, tools, hooks, agents and skills, is plain TypeScript and Markdown in the repository, so its behavior can be inspected directly. It installs as a Pi package and warns that it can conflict with a separately installed skill of a similar name.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.