Pi Finder + Librarian Playbook

This repo now includes three bundled skills:

• pi-finder-subagent
• pi-librarian
• pi-context-pipeline (orchestrates both)

Why this helps

Large coding tasks often waste context on broad file reads. This pipeline narrows discovery first, then hands coding agents a strict file shortlist.

Setup

npm i -g @mariozechner/pi-coding-agent
pi install npm:pi-finder-subagent
pi install npm:pi-librarian
gh auth login   # needed only for librarian mode

Recommended flow

1. Run pi-context-pipeline with a concrete query.
2. Keep run_librarian=false unless external GitHub patterns are actually needed.
3. Use artifacts/.../context-pack.md and codex-kickoff.txt as the coding handoff.
4. Only expand scope if the context pack still has unresolved blockers.

Guardrails

• Keep Finder output to max 12 local files.
• Require file paths + line ranges in reports.
• Prefer explicit unknowns over speculative guesses.
• Avoid starting implementation before discovery artifacts are written.

Fast commands (manual mode)

Local discovery only:

pi --no-session --tools read,grep,find,ls,bash -e npm:pi-finder-subagent -p "Use finder to locate auth entrypoints and token validation code. Return max 12 files with line ranges."

GitHub discovery:

pi --no-session --tools read,grep,find,ls,bash -e npm:pi-librarian -p "Use librarian to find robust webhook signature verification patterns in known OSS repos. Include cited files and line ranges."