apw-onboardlisted

Use this skill to build and maintain a durable **codebase orientation** — the architecture map, entry points, build/test/run commands, and conventions a session needs to act competently — and to get oriented from it in a fresh session. It exists to fight one specific limitation: a model **re-derives its understanding of the codebase every session**, spending its opening context re-reading the same files and re-inferring the same structure, and sometimes guessing wrong. The fix is to write that orientation down once, where every future session already has it. The orientation lives in **`AGENTS.md`** — the always-on instructions file every session reads automatically — so the map is reused across sessions for free, with no separate file to remember to open. Because `AGENTS.md` is injected into *every* session, a stale orientation is misinformation everywhere; keeping it true is therefore part of the discipline, not an afterthought. This skill is **user-invoked only** — the agent must never reach for it on its own. Invoking it is a deliberate act, so the invocation *is* the authorization: the skill **writes the orientation to `AGENTS.md` directly** and the user reviews the result in the file (the diff), rather than approving wording in the chat first. This differs from `apw-learn`, which shares the file but proposes-then-confirms — there, the agent may self-invoke, so a chat gate is needed; here the user already chose to run it. `apw-learn` and `apw-onboard` both write to `AG