← ClaudeAtlas

add-roster-expertlisted

The end-to-end procedure for adding a Thalamus expert to the roster without breaking the tmux roster or the console's phone PWA. Use BEFORE creating any config/experts/ manifest or running `thalamus roster` with a new scope, when a roster/console surface misbehaves after a roster change (e.g. the PWA stuck at "connecting"), or before touching pin.py window mechanics or console server behavior.
Ybx-jp/thalamus · ★ 1 · AI & Automation · score 67
Install: claude install-skill Ybx-jp/thalamus
# Add a Roster Expert (without breaking the roster) **Custody:** any session that changes roster or console mechanics (pin.py windowing, the console server or its service worker) updates this skill in the same change. **Where things live.** This skill sits beside the agents it governs — in the same package that declares the experts and owns `pin.py` — so the manifest, the window mechanics, and the procedure for adding one all version together. The console it warns about is `src/thalamus/console/`, in this same package. ## Procedure 0. **Ask the operator what work is PLANNED in this scope, before consulting anyone.** Binding, and it comes first. Every expert can only see what has already happened; asked a question whose answer lies in the future, it will answer about the past and sound decisive doing it. **A count of past work sizes a corpus and never bounds a role** — the litmus below asks whether real sessions would be pinned here, and a session that is planned but not yet run counts. An `ml-systems` scope was cut on a measured 0-of-~10 corpus count that was really a measure of work not yet started; the operator's plans reversed it the same day. 1. **Roster decision first, not procurement drift.** Clear the roster discipline before writing any YAML: the granularity litmus (would real sessions be pinned here?), the null hypothesis (a scope that adds nothing a general session lacks should not exist), and the skill-vs-expert test (a procedure i