auto-onboardlisted

# auto-onboard Repository discovery. Builds bounded project truth from evidence, not guessing. First action: run `node .agent/.automaton/scripts/get-context.mjs` from the project root. ## Preamble auto-onboard builds bounded project truth from repository evidence, not training data, not conversation, not guessing. It does not write code or produce specs. Loading discipline: keep REPO-MAP.md under 150 lines; stop scanning once you have enough. ## Quality Gate Before writing steering artifacts: - Separate observed, inferred, and unknown facts. - Cite paths for repo-shape claims. - Treat artifact writing as expensive: write only durable project truth and immediate blockers, not scratch notes. - Stop scanning once the next action is clear. - Read `references/quality.md` when artifacts turn into broad inventory. ## Do ### Detect State Choose the smallest valid path: - First-time or scaffold-level steering: scan and write the starter artifacts. - Real steering, no refresh requested: do not rewrite; report what exists and route active work to `auto-resume`, otherwise `auto-office-hours`. - Targeted refresh: read only evidence relevant to the requested steering update and rewrite only affected artifacts. ROADMAP.md stays restrained. Do not create roadmap phases on a first run. On refresh, write phases only when strong repo evidence shows an existing or ongoing roadmap and the user confirms importing or refreshing it in chat; then use `.agent/.automaton/references/ROADMAP-CO