Pi for customizing a coding-agent harness

How it works

The monorepo exposes packages for model access, the terminal interface and the coding agent. Extensions give developers a way to adapt the harness and its tools for a particular workflow.

What you need

Start with the terminal agent package or the library layer you intend to embed. Keep extensions and model-provider settings small enough to reproduce in another checkout.

What to try first

Use a simple extension in a disposable checkout, then compare how it changes the tool calls available to a new session. Keep the base task small enough to debug without the extension.

Add an extension command that only emits diagnostic output. It gives you an inexpensive way to learn how extension state and session state interact.

Where to look for failures

Extensions run within the agent's environment. Review their code and permissions before using them, and track which customizations a project depends on when sharing configuration.

Related Franklin pages

Compare how each agent exposes tools and where extensions execute. Start from an empty extension set and add one capability at a time so access stays understandable.

• Give a coding agent access to editor context, workspace files and approved tools. Cline for agent-assisted development in an editor
• Run a terminal coding assistant with project context and connected tools. Crush for terminal coding with configurable providers

For a library of development workflows that can complement a coding agent, see Franklin’s Superpowers profile. Superpowers

Project documentation

• Current project README
• README at the reviewed revision