# Nw Buddy Project Reading

> Evidence-first project reading protocol based on durable authorities, Git, tests, and installed surfaces.

- Skill: `nwave-ai/nw-buddy-project-reading-2` (Agent Skill)
- Install (CLI): `npx skillmds@latest add nwave-ai/nw-buddy-project-reading-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/nwave-ai/nw-buddy-project-reading-2/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: nWave-ai (https://skillmd.com/u/nwave-ai)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/nwave-ai/nw-buddy-project-reading-2

---


# Project Reading for the Buddy Agent

Answer from current evidence and cite repository-relative paths and lines.

## Cheap-to-expensive order

1. Read project instructions and the user's named artifact.
2. Read durable product authorities for jobs, journeys, vision and KPI meaning.
3. Read durable design authorities and ADRs for boundaries and decisions.
4. Read Request-derived execution facts only through the resident software
   when they are necessary.
5. Read the oracle, source and tests only when the question requires them.
6. Use `des code-fact` for structural facts and degrade LOUD when approximate.
7. Use Git and terminal command evidence for completion; use an installed
   runtime probe for distribution claims.

Do not infer progress from a directory, markdown status, model narration or
process exit alone. A delivery is complete only when the resident runner reports
its terminal result after applicable review, source-blind EXAMINE and finalization.

When evidence conflicts, name the owning authority and the first falsifier.
Never reconstruct state from a retired feature artifact or workflow mode.