Mission Control — Current Strengths Inventory

Purpose: Honest inventory of what MC (Luci's dashboard, ~/workspace/mission-control/app.py, port 3001) does WELL today, before a Wingman-style UX redesign. Anything in "Must-Not-Regress" is a feature a polished black-box consumer agent (Wingman by Emergent) almost certainly lacks — do not lose it.

Research only. No code changed. Date: 2026-05-18.

1. Feature Inventory

MC is a single Flask app (app.py, ~10,966 lines) + models.py (~263KB query layer), Jinja2 templates, Alpine.js + HTMX, SSE + WebSocket, PWA service worker. ~150 routes. Surfaces:

Core ticket / worker surfaces

Board (/board, board() app.py:1569; board.html 17KB) — kanban ticket board, 30 cards + load-more, board-counts API (/api/v1/board-counts :5115), partial refresh (/partials/board-panel :1583).
Ticket detail (/ticket/<id>, app.py:1691; ticket.html 83KB) — per-ticket conversation, comments, tags, relationships, attempts, activity, vault context links, history replay.
Runtime Workbench (/runtime-workbench, app.py:1574; ticket_runtime_workbench.html 89KB + .js 145KB) — live tmux-backed worker session view: start/send/harvest/stop/retry/switch, status capture, glass-box-ish tool-call stream.
Tasks (/tasks, tasks_page() app.py:1719; tasks.html 30KB) + Task detail (/task/<id> :1800) + Task run log (/task-run/<run_id>/log :1854) — scheduler task list, run history, per-run logs, definition editor, schedule/notify/runtime-profile editors.
Activity (/activity :1593) — activity feed.
Team (/team :4373) + org-chart partial (/partials/org-chart :7395) — agent roster.

Real-time plumbing

SSE — broadcast_sse() app.py:393; global + per-ticket client buckets (global_sse_clients, ticket_sse_clients :308–353); stale-client pruner thread (_prune_sse_clients :655); SSE_MAX_TOTAL_CLIENTS cap. Per-ticket stream /api/v1/tickets/<id>/stream :9628, tail :9761.
WebSocket — flask-sock interactive ticket sessions (mc_interactive.py), ws-token mint :5720.
ticket_events table — granular stream-json events (tool_use, thinking, assistant) persisted per worker + replayed on page load (MC-747).

Health / observability

Dashboard (/, app.py:1001; dashboard.html 39KB) — health tiles: heartbeat age, worktree count + GB (gold>20/red>50, empty-count), persistent-Luci online badge + runtime, Data Feeds grid (20 feeds, freshness).
Server health / latency — /api/server-health :4986, /api/health :4996, /api/v1/health/latency :1220, /api/v1/health/data-freshness :5098.
Logs (/logs :2060) + /api/v1/logs/tail :5052 — scheduler/heartbeat log tail.
Worktrees — /api/worktrees :1200, /api/v1/dashboard/worktrees :1151.
Dispatch analytics — /api/dispatch-7d :1231, /api/v1/dashboard/dispatch :1122, /api/v1/dispatch/preview :5229.

Cost / provider control

Cost page (/cost; cost.html) — token/USD spend tables.
Usage — /api/v1/usage :1193, /api/v1/dashboard/usage :1135.
Provider switching — /api/provider/default :4905, /api/provider/larry :4927, runtime profiles CRUD /api/runtime/profiles :4945 + preflight :4972, per-task runtime-profile /api/task/<id>/runtime-profile :4792.

Content / knowledge surfaces

Reports (/reports :2193) — HTML research reports + DELETE, PDF convert (:2329, :2396).
Briefings (/briefings :2489) — audio/slides, audio-library serving :2563, slides API :2584.
Brain (/brain :2738) — SecondBrain file + DB browser (/api/v1/brain/db :2877, schema :2970, row :3028).
Wiki (/wiki :1908) + SB-wiki (/sb-wiki :1937) + Bible (/bible :2073) + md-view (:2664).
Skills (/skills :1685), Insights (/insights :4404), Timesheets (/timesheets :4429).

App / service dashboards (MC as hub)

Apps (/apps :3140) — external dashboards.
Spotify Radio (/spotify :3260) + Home Radio (/home-radio :3692) — playback control, mood, station toggles, mix rebuild.
F1 (/f1 :1880), Trading (/trading :4383), Price Watch (/api/price-watch/* :3170), Life Manager (/life-manager :1631).
Vibe Kanban sync — /api/vk/* :4278.

Telegram / persistent integration

Telegram bindings + events — /api/v1/telegram/* :8182–8291.
Persistent Luci controls — /api/v1/persistent_luci/{status,start,send,harvest,stop,switch} :8291–8350.

Ticket lifecycle APIs

Full REST: create/list/get/patch/claim/snooze, comments, tags, relationships, attempts, ticket_context (vault bridge), workflow templates, handoff_pack, runtime reattach.

PWA

Service worker (sw.js, CACHE_NAME='mctrl-v78-...'), app-shell route :1183, offline page, manifest.json, icons.

2. Must-Not-Regress Strengths

Each of these is real, working, and almost certainly absent from Wingman. Losing any in the redesign is a regression.

Top 5 (priority order)

Multi-worker, multi-host orchestration with live tmux-backed sessions. MC dispatches per-ticket background workers (mc_pickup.py), runs the Runtime Workbench against a real tmux session per active ticket (start/send/harvest/stop/retry/switch provider mid-run), AND remote-dispatches to Larry over SSH (elmar@46.225.208.1). Wingman is a single-agent consumer chat — no fleet, no remote hosts, no per-ticket session you can attach/reattach to. This is MC's core differentiator.
Provider switching + true cost/token accounting. MC routes work across Anthropic / GLM / Kimi / MiniMax via runtime profiles, with preflight checks, per-task profile pinning, and a Cost page + usage APIs recording real USD/token spend per run. A consumer product hides the model and the bill entirely — you cannot switch backends or see what a session cost. This is load-bearing for Luci's centralized cost control (CLAUDE.md runtime-profile rule).
Scheduled task engine with full transparency. scheduler.py tick runs everything in ~/workspace/tasks/; MC exposes the task list, run history, per-run logs, definition editor, schedule editor, notify-destination editor, and runtime-profile editor — all in-browser. Wingman has no cron, no recurring autonomous jobs, no run-history audit trail. The Tasks surface is an operational control plane, not a chat feature.
Deep health / observability instrumentation. Heartbeat age, worktree disk usage (count + GB + empty-count, color-thresholded), persistent-Luci online state, a 20-feed Data Freshness grid, server-health + latency endpoints, dispatch-7d analytics, MC canary. This is genuine ops monitoring — a consumer agent shows none of its own internals.
Full data ownership + glass-box ticket history. Everything lives in local mc.db (218MB SQLite, WAL) — tickets, comments, raw ticket_messages (thinking/tool_use/assistant/user), ticket_events (replayable stream-json), sessions FTS5 index, task_runs, heartbeats, ticket_context vault bridge. Nothing is in a vendor's cloud; every worker action is queryable, replayable, and auditable. Wingman is a black box — no export, no replay, no DB.

Other must-not-regress strengths

SSE live streaming with backpressure handling — global + per-ticket buckets, stale-client pruning, client cap. Real-time updates without polling; redesign must keep the SSE event contract.
dev-loop integration — workflow templates, handoff_pack, attempts tracking, council hooks. Tied to the mandatory dev-loop discipline.
ticket_context vault bridge (MC-462) — links tickets to the SecondBrain files consulted/created during work. An ops↔knowledge link no consumer app has.
Ticket relationships graph — blocks / duplicate_of / caused_by / child_of / relates_to, with validated types and a DELETE API.
PWA / offline — installable, app-shell cached, offline page. Keep.
MC-as-hub — Spotify, Home Radio, F1, Trading, Price Watch, Brain, Reports, Briefings all under one roof. The redesign should not silo these away.
Telegram + persistent-session integration — bindings, event ingest, persistent-Luci start/send/harvest/switch from the browser.
Token-based auth, gzip, partial/HTMX rendering — lightweight, fast, no SPA build step.

3. UX Weaknesses (where MC is genuinely behind a polished product)

Opaque worker log dumps. Workbench surfaces raw tmux/stream output; ticket_events exist but there is no clean per-tool-call card UI (collapsible input/output, status pill, duration). Comment in ticket_runtime_workbench.html:639 confirms tool_use entries "used to land in History" — no first-class glass-box tool panel like Wingman's step cards.
Template sprawl / inconsistency. 40+ templates, several huge (ticket.html 83KB, ticket_runtime_workbench.html 89KB, dashboard.html 39KB, spotify.html 34KB). Per-page inline <style> blocks (e.g. cost.html) — no shared component system; visual inconsistency across pages.
Empty states. No evidence of designed empty/zero states (no tickets, no tasks, first-run) — likely bare tables.
Mobile. Mobile-friendly is a hard requirement (CLAUDE.md) but dense tables (cost, tasks, board) and 145KB workbench JS suggest a desktop-first layout that degrades on phone.
Connection UX. SSE reconnect / "worker offline" / stale-session states are handled in plumbing but there is no polished connection indicator or graceful reconnect banner for the user.
Information density without hierarchy. Dashboard is a wall of tiles; no progressive disclosure, no "what needs my attention now" focal point. Wingman leads with a single clear next action.
No unified activity timeline per session — events are split across comments, messages, events, activity feed.

4. Tech Constraints (redesign must respect)

ASSET_VERSION / sw.js cache sync. ASSET_VERSION in app.py:49 ("20260517-workbench-ticket-view-rebuild") AND CACHE_NAME + ASSET_VERSION in static/sw.js:1,7 ('mctrl-v78-...'). Any static JS/CSS change MUST bump BOTH CACHE_NAME and ASSET_VERSION in sw.js plus ASSET_VERSION in app.py, or phones serve stale code (memory: project_mc_asset_version_sync).
systemd-managed. Service is luci-dashboard.service. NEVER spawn python3 app.py manually — orphan holds port 3001 and defeats systemd. Always systemctl restart luci-dashboard (memory: feedback_never_spawn_mc_manually).
Single Flask app. One app.py + Jinja2 + Alpine.js + HTMX + SSE + flask-sock. No SPA build pipeline. A React/Vue rewrite would be a major architecture change — keep server-rendered partials unless explicitly approved.
Port 3001, URL https://openclaw.tailb2ba18.ts.net / http://204.168.188.33:3001.
Mobile-friendly is mandatory (global CLAUDE.md — Elmar reads on phone).
mc.db is source of truth — 218MB SQLite WAL. Never commit it (gitignored). Schema changes go through migrate.py / migrations/.
SSE event contract — broadcast_sse(event_type, data) and the ticket_event SSE message shape are consumed by ticket_events replay; keep the contract or break live updates.
branch drift — mission-control tree drifts off master; verify git branch --show-current before commits (memory: project_mc_branch_drift).
MC canary — /api/v1/board-counts, create/list/route/comment/close ticket flow is exercised every 30 min; redesign must not break those endpoints.