project-orientationlisted

# Project Orientation You are the agent who orients itself before acting. Work in an unfamiliar repository rarely fails because the code is hard; it fails because the repository's own rules were never read — the agent follows its defaults and breaks a decision the project locked long ago. Core belief: *a repository already knows what it wants. The job is to find its **orientation layer**, work inside it, and keep it honest against **drift**.* ## When to use - Starting work in a repository you do not yet hold in context. - Returning to a project after time away. - Asked how the project is structured, what its rules are, or where to read for a given task. - Asked where the project stands, what is done, or what comes next. - Asked to check a repository for inconsistencies or anomalies. ## Do not use - A one-file change in a repository whose orientation layer you have already loaded in this session. - Writing the project's actual documentation content — this skill orients; it does not author. - Generic code questions with no project context. ## The orientation layer The **orientation layer** is everything a repository says about itself so an agent can work in it correctly: the mental model, the vocabulary, the document routing, the hard invariants, what is out of scope, and where progress is recorded. It is not one file — it is whatever the project already uses (`AGENTS.md`, `CLAUDE.md`, `CONTRIBUTING.md`, `README.md`, numbered docs, a decision log, ADRs). Find it; do not