orchestrator-updatelisted

# /orchestrator:update Manage the orchestrator's config evolution. Opt-in: nothing auto-applies. ## When to use - You run `/orchestrator:update` explicitly - Session-start check found entries in `~/.claude/PENDING-UPDATES.md` with `status: pending` and you want to review - After any significant model or config change, to catch regressions early ## Workflow ### Step 1 — Load registry Read `~/.claude/PENDING-UPDATES.md`. Parse the entries into a list. For each pending entry, capture: ID, title, change type, blast radius, expected eval delta, effort. ### Step 2 — Present options Show a numbered list of pending entries with ID, title, change type, effort, and expected eval delta. Format: ``` Pending orchestrator updates (N): 1. UPD-001 (efficiency, ~30min) — MEMORY.md tier structure Expected: -300 tokens per session boot, no behavioral regression 2. UPD-002 (efficiency, ~15min) — ORCHESTRATOR.md rehydration short-circuit Expected: -2-3 min per boot on known projects, no token change 3. UPD-003 (quality, ~10min) — Fanout ceiling heuristic Expected: better behavior on 8+ unit tasks, no regression Options: - Pick one or more numbers to apply (e.g., "1" or "1,3") - "prune" to dismiss stale entries (60+ days pending) - "all" to apply all pending (not recommended without review) - "cancel" to exit ``` ### Step 3 — Mandatory baseline eval (non-bypassable) Before applying anything: - If `~/.claude/evals/` exists: run the tokenize script `python3 ~/.claude/evals/t