Cinema-grade command UI
Glass panels, scanline HUD, Orbitron chrome — built as a real product surface, not a bootstrap admin. Weather, clock, BTC, Today / Overdue / Upcoming, and a task stream that feels like a mission board.
J.A.R.V.I.S. // PRODUCT PITCH
MISSION CONTROL
A live command surface for humans and AI agents on the same board. Not a chat log. Not a sticky-note app. A multi-engine ops HUD where tasks are the source of truth, agents claim work by contract, and smoke never lies.
Live API SoT
Multi-agent fleet
Dual seats
Token-aware
Operator-verified
The board that actually runs the shop
Human-readable HUD. Machine-readable API. Same tasks, same statuses, no dual books.
Agent status machine
pending → in_progress → awaiting_review → done, plus needs_promotion when work is unauthorized. Badges flash for SMOKE and Specs so operators know what needs eyes.
Field-projected live API
api.php is task-state SoT. Agents request only the fields they need
(id, status, project, pr_id…) — no full store dump into context.
Claim, reconfirm, close. Fail closed on conflict.
Specs + Checklist on every card
Collapsible Specs and Checklist (default closed). Checkboxes persist. Agents load a hot specs index first; full pages only when needed. Done work archives cold so the hot path stays thin.
Completed audit trail
Full History view with breadcrumbs, agent stamps, smoke notes, and verified_by (including Operator). Who finished it, who signed it off, and when — not a silent checkbox.
Multi-project classification
Work / Personal lanes; project chips for Todolist, MESK8, Lightwave; ops tags like automation, needs-operator, orchestrator-attention. Filters that agents and humans both understand.
Autobots, Orchestrator, dual seats
One claim per process. One overseer pass per tick. Two machines that talk without WinRM.
One-task implementers
Scheduled (or manual) agents claim one authorized PR, implement, self-check, update Mission Control, exit. Grok Build · Codex · Claude — same contract, different engines. No second task. No freestyle deploy.
Snapshot · Decide · Report
Read-only overseer. One PowerShell snapshot builds a compact packet: queue, smoke, schedules, quota, seat bus. Decision is hold / serial / dual — never invents work, never claims tasks.
Heavy + Secondary
SuperGrok Heavy on the Alienware for Design / Dev / deep overseer. SuperGrok on the laptop for scheduled Autobots and light orch. Same Mission Control wall; different horsepower.
DreamHost coordination channel
Heartbeats, pending requests, schedule health between Heavy and Secondary — public GET, authenticated POST. No remote desktop required for “is the other seat ready?”
Open-row authorization
Agents only implement what is on the Open board (project-qualified project:PR-####). Missing Open row → needs_promotion. Humans promote; agents don’t invent roadmap.
Smoke hold + Operator stamp
SMOKE_HOLD = 10 pauses batch while review stacks up. Product first. Operator can complete from the UI with verified_by = Operator — human sign-off is first-class, not a footnote.
Need-to-know context, not novel-length preload
Agents that re-read entire boards burn money. Mission Control is designed so PowerShell and projections do the heavy lifting before the model ever sees a packet.
Compressed orchestrator packet
One script, one JSON object: counts, samples (top N), meters, seat bus. Light vs full profiles. Measured shrinks on the order of 80%+ vs naive dumps.
Hot index / cold archive
Active acceptance lives in specs/index.json. Done PRs archive out of the
hot path. Agents never pull proposals novels for a single checklist.
Compact multi-engine skills
Shared skill contracts for Grok, Codex, and Claude — same claim/close rules, archived novel protocols off the runtime path. Heavy means more quality loops, not more documents.
Agent JSON projections (next)
Planned: public/agents/*.json — queue, open rows, meters, goals —
so agents stop scraping human HTML entirely. Local boards become human mirrors only.
See the fleet. Spend on purpose.
Reports, meters, IDs, and cost feedback — the control room behind the control room.
Board analytics HUD
Weekly plan meters, project schedule matrix, agent matrix, spend panels. Navigate week/month. Copy markdown tables. Live refresh without hard reload. TV dock (PR-0201): plan pies stay pinned in a persistent footer so high-screen viewing keeps usage in line of sight while you scroll matrices.
Plan meters + cost feedback
SuperGrok Heavy / Secondary, Codex, Claude — pie meters with reset heatmaps. Orchestrator cost feedback can recommend cheaper tiers or pause Autobots when burn is hot. Schedule XML is never auto-edited.
Atomic DIP / PR reservation
Global 4-digit IDs across projects. UI Next ID badge + Reserve copies to clipboard. Server-side lock so concurrent agents never double-claim a number.
Share the data stream
Download Markdown or send via email from the board. Human handoff without screenshot archaeology. Directives leave the HUD in a form agents and teammates can reuse.
Default runtime flow
Live MC API
Snapshot / claim query
Open-row auth
Implement + self-check
Smoke / Operator
Done + History
// BOTTOM LINE
Mission Control is where the agents clock in.
A shared wall for humans and machines: authorized work only, live state, bounded context, and a smoke queue that keeps the fleet honest.