The earendil-works/pi project is a modular agent harness designed to facilitate the development and execution of self-extensible coding agents. The ecosystem is composed of several specialized TypeScript packages, including @earendil-works/pi-agent-core, which manages agent runtime, tool calling, and state management, and @earendil-works/pi-ai, which provides a unified API for multiple LLM providers such as OpenAI, Anthropic, and Google. Additional components include @earendil-works/pi-coding-agent for interactive CLI operations, @earendil-works/pi-durable for managing conversation and task persistence, and @earendil-works/pi-tui for differential rendering in terminal UIs. Because the system lacks a built-in permission model and operates with the permissions of the host user, the documentation recommends sandboxing via Docker, OpenShell, or the "Gondolin" extension, which routes tools into a Linux micro-VM. To ensure supply chain security, the project utilizes strict dependency management, including pinning direct dependencies to exact versions, using package-lock.json as the ground truth, and generating a shrinkwrap file for the published coding-agent CLI to pin transitive dependencies. The build process supports offline model data and includes automated checks for TypeScript compatibility and dependency integrity.