prinova/pi-agent-codebase-workflows

pi-agent-codebase-workflows is a package of skills and prompt templates for a tool called pi, built to help AI coding agents work safely and systematically with unfamiliar or complex codebases rather than guessing their way through them. The package provides three main capabilities. The first, called codebase-recon, is a workflow that builds a structured set of documentation about a project, covering architecture, data invariants, dependency rules, a risk register, and change guidance, and writes it into a docs/agent folder plus a top level AGENTS.md file, so an AI agent has a durable written understanding of the codebase before it starts making changes. The second, arch-code-review, reviews a current code diff against that documented architecture, checking it against invariants, the data model, dependency rules, known risks, and tests. The third, safe-change, is a step by step workflow for making a code change, covering a preflight check, a design or diagnosis phase, implementation, validation, and an update to the documentation afterward. Each capability is broken into smaller prompt templates, such as slash commands like recon-01-inventory or feature-design, that can be run individually rather than as one large workflow. In large monorepos, the reconstruction and review prompts can be scoped to a specific module, package, app, or directory using an optional focus argument, and the tool then writes hierarchical, scoped documentation for that area alongside the repository-wide documentation. You would use this when working on a large or unfamiliar codebase with an AI coding assistant and wanting to reduce the risk of the AI making changes that break existing contracts, violate architectural decisions, or miss edge cases the agent was never shown. It is distributed as a package for pi and installed with a pi install command from git or npm.