Mission Control — Overview

Mission Control (MC) is Luci's web dashboard and operational hub. It manages tickets, workers, sessions, and provides a real-time conversation interface.

Basics

Code: ~/workspace/mission-control/app.py (Flask)
Service: luci-dashboard.service (systemd)
Port: 3001
Database: mc.db (SQLite, WAL mode)
URL: https://openclaw.tailb2ba18.ts.net or http://204.168.188.33:3001

Pages

Page	Route	Purpose
Overview	`/`	Health, stats, task status
Board	`/board`	Kanban-style ticket board (30 cards, load more)
Ticket	`/ticket/<id>`	Individual ticket conversation + actions
Activity	`/activity`	Activity feed
Life Manager	`/life-manager`	WhatsApp/email scan results
Tasks	`/tasks`	Scheduled task list and run history
Skills	`/skills`	Installed skills browser
Reports	`/reports`	HTML research reports served from ~/workspace/reports/
Radio	`/spotify`	Spotify Radio dashboard with mood controls
F1	`/f1`	F1 Predictor season dashboard
Team	`/team`	Agent team overview
Apps	`/apps`	External dashboards (Smart Money, Price Watch)
Logs	`/logs`	Scheduler and heartbeat logs

Tech Stack

Flask + Jinja2 templates
Alpine.js for reactivity
HTMX for partial updates
SSE (Server-Sent Events) for real-time streaming
WebSocket (flask-sock) for interactive ticket sessions
Service worker for PWA/offline support

SSE thread budget (gthread)

Gunicorn runs with -w 2 -k gthread --threads 16 → 16 handler threads per worker process. Caps are PER-WORKER (counters are module-level globals), so the sum of caps must fit one worker's thread budget with headroom for normal HTTP requests:

SSE_MAX_TOTAL_CLIENTS = 8 — global broadcast (/api/v1/stream) + per-ticket (/api/v1/tickets/<id>/stream) share this bucket.
SSE_MAX_TAIL_CLIENTS = 4 — worker log tail (/api/v1/tickets/<id>/tail?follow=1) has its own bucket so a noisy tail page can't starve the broadcast bucket or all page renders. Over-cap returns 503 with a retry: 30000 SSE hint.
Budget math: 8 + 4 = 12 SSE threads per worker, leaving ≥ 4 threads per worker for normal HTTP. Previous values (20 + 8 = 28) exceeded the 16-thread/worker budget and caused recurring deadlocks (2026-05-22 break-glass: API timeouts → ccgram restart loop → cycle).
Every stream caps max_lifetime ≤ 300s and heartbeats ≤ 30s so dead clients release threads quickly; browsers reconnect via the SSE retry hint.

Thread starvation symptom: listener accepts connections but every request hangs / times out, ss shows many CLOSE-WAIT sockets on the worker PIDs. Restart restores capacity; the caps prevent recurrence.

Database Tables (mc.db)

tickets — the core ticket table (id, identifier, title, status, priority, assigned_to, etc.)
ticket_comments — conversation history per ticket
ticket_messages — raw message stream (thinking, tool_use, assistant, user)
ticket_events — granular stream-json events for live activity panel (MC-747). Columns: ticket_id TEXT, seq INT, event_type TEXT, payload_json TEXT, ts TEXT. Populated by workers as they run; replayed on page load for any ticket.
ticket_tags — ticket categorization tags (auto-assigned on creation: bug, feature, infra, refactor, research, review)
ticket_relationships — ticket-to-ticket links with validated relationship types: blocks, duplicate_of, caused_by, child_of, relates_to, related_to
ticket_context — ticket↔vault bridge (MC-462). Links tickets to vault.db files that were consulted, updated, or created during work. Auto-populated by vault_mcp.py during worker sessions, also supports manual linking via API.
sessions — registered Claude sessions
sessions_index / sessions_fts — indexed session transcripts (FTS5). Canonical store for all sessions (both Luci and Lucienne).
queued_messages — messages waiting for worker pickup. Managed by queue_reaper.py which expires stuck messages (>30 min), retries failed ones, and sends Telegram alerts.
task_runs — scheduler execution log
heartbeats — health check history
vk_sync — Vibe Kanban external board sync mapping

Key Files

File	Purpose
app.py	Flask routes, API endpoints, SSE/WS handlers
models.py	Database queries, ticket/comment CRUD
auth.py	Token-based auth (env vars or hardcoded fallbacks)
mc_interactive.py	WebSocket interactive session manager
static/chat.js	Client-side conversation UI
static/app.js	Board, search, general UI logic
static/style.css	All styles

API Endpoints (ticket_context)

Method	Route	Purpose
GET	`/api/v1/tickets/<id>/context`	List vault files linked to a ticket
POST	`/api/v1/tickets/<id>/context`	Link a vault file to a ticket (manual or auto)
DELETE	`/api/v1/tickets/<id>/context/<ctx_id>`	Remove a ticket↔vault link

API Endpoints (ticket_events — MC-747)

Method	Route	Purpose
POST	`/api/v1/tickets/<identifier>/events`	Persist a stream-json event; broadcast via SSE
GET	`/api/v1/tickets/<identifier>/events?after=<seq>`	Replay ticket_events in seq order

Workers call POST to persist events during execution. The ticket detail page calls GET on load to replay history, then subscribes to the ticket SSE stream for ticket_event messages while the worker is live.

Key Takeaways

MC is the central nervous system — all tickets, workers, and conversations go through it
mc.db is the source of truth for tickets (separate from vault.db)
Tickets auto-tag on creation (bug/feature/infra/refactor/research/review based on content)
ticket_context bridges operations↔knowledge: vault files consulted during work are auto-logged
queue_reaper.py runs every 15 min to expire stuck messages and alert via Telegram
Board pagination: 30 cards default, "Load more" button
Interactive mode uses WebSocket; background workers use REST API + SSE
Notifications route to Telegram forum topics, not Elmar's DM
Live activity panel (MC-747): ticket detail page shows tool calls in real-time via ticket_events table + SSE. History replays on page load; collapsed by default, auto-opens when in_progress.

Wiki