Analyze conversation to detect which cognitive room you should be in. Returns room suggestion, confidence score, and switching signals. Use when user says things like "let's think bigger picture" or "time to ship".

Get current cognitive room status. Returns: current room, identity rules, memory palace anchor, recent switches, and terminal mapping.

Switch to a different cognitive room. Saves context from previous room, loads new room configuration. Returns memory palace anchor and identity rules for new room.

Open the HTML dashboard for a room in the browser. If no room specified, opens current room. The HTML auto-refreshes from state.json on tab focus.

Manage todos for a room. Actions: add, list, update, delete. Todos are stored in SQLite and synced to state.json for HTML display.

Send messages between rooms (flywheel coordination). Get unread input streams for a room.

Export current state to JSON file. HTML files read this on tab focus to refresh.

Detect which terminal Claude is running in, and which room it maps to.

Live 3-recent + 3-next overview computed from real git activity. Reads owned-surface globs from each .workflow/rooms/.html, groups commits in the last 14d by owning room (largest-fraction-of-files wins), filters auto-bumps, and returns: just_completed (3 most-recent meaningful commits, one per room), next_up (3 rooms scored by blocks_downstream × 2 + dormancy_days), and grid_12x12_colored_cells map. Spin-up DAG: voice → performer → network → operator → architect; foundation rooms (builder, laboratory, claudelab, vault, navigator) are independent.

Inspect and control the PMU XOR→ClaudBridge pipeline. Reads/writes data/pmu/pipeline/state.json (shared with the CLI driver, dashboard, and ClaudBridge mock). Use to: peek current intent/reality + last run sigma/friction (get-state), change inputs (set-intent, set-reality, set-threshold), trigger a full pipeline run end-to-end (run; optional stage to stop early), retrieve a specific run's receipt (get-run), list recent runs (list-runs), or inspect the cached axis/tile libraries (list-axes, list-tiles). The intent/reality control plane is the load-bearing surface — changing intent here makes the next run + dashboard render against the new coordinate without code edits.

Reproduce every PMU due-diligence claim — weld diff=0, T1 forgeries, σ-on-silicon, reef ρ. Runs scripts/pmu/verify-all.sh in the thetadrivencoach repo (the Rust daemon + canon are NOT bundled in npm; the repo is located via cwd / ../../, or set repo_root / THETACOG_REPO_ROOT). Includes the dossier-freshness gate (fails if the document drifts from the code).

Predict the next steps across all 9 rooms from THREE real sources — chat transcripts (read backwards, bounded), the repo + git logs per each room's owned file/dir surface, and the board (room punch lists). EVERY room is filled with a ≥2-sentence themed narrative (the direction + the actual area); the current room gets the exhaustive narrative + a numbered actionable to-do list. This is the subdivision substrate the mesh runner uses for inter-room contracts + task hand-off. Returns structured JSON; set render=true to also get TTS narration + HTML.