cm-dashboard — Multi-Agent Mission Control

Visual + API control plane for tasks dispatched to Codex, Google Antigravity, Cursor, Gemini CLI, and Claude Code. Runs locally at http://localhost:6969.

When to use

• You are juggling work across multiple AI coding agents and need one source of truth.
• You want a kanban (backlog → in-progress → review → done) per project.
• You need to dispatch a task to a specific agent and capture its progress.
• A CLI agent (Codex, Gemini, etc.) needs to report its conversation lifecycle to a shared board.

Supported agents

(Also: windsurf, cline, copilot, manual.)

Quick start

# 1. Start the dashboard (auto-starts on first cm command, or run explicitly)
cm dashboard         # opens http://localhost:6969

# 2. Health-check
curl -s http://localhost:6969/api/projects | jq

Core workflows

A. Create + dispatch a task to an agent

# Create
TASK_ID=$(curl -s -X POST http://localhost:6969/api/tasks \
  -H "Content-Type: application/json" \
  -d '{"title":"Refactor auth middleware","agent":"codex","skill":"cm-execution","priority":"high"}' \
  | jq -r .id)

# Dispatch (writes .agent-tasks/<id>-*.agent-task.md inside the project)
curl -s -X POST http://localhost:6969/api/tasks/$TASK_ID/dispatch | jq

The dispatcher emits a per-agent CLI command:

• codex --task "<file>"
• antigravity -p "$(cat <file>)"
• cursor --task "<file>"
• gemini run --task "<file>"
• claude --task "<file>"

B. Move a task across columns

curl -s -X PUT http://localhost:6969/api/tasks/$TASK_ID/move \
  -H "Content-Type: application/json" \
  -d '{"column":"in-progress"}'   # backlog | in-progress | review | done

C. Auto-sync from an agent conversation

Use this from a running Codex/Gemini/Cursor/Antigravity session to report status — upserts by conversationId.

curl -s -X POST http://localhost:6969/api/tasks/auto-sync \
  -H "Content-Type: application/json" \
  -d '{
    "conversationId":"codex-2026-05-11-001",
    "title":"Refactor auth middleware",
    "status":"in-progress",
    "agent":"codex",
    "skill":"cm-execution"
  }'

Status values map to columns: pending|todo → backlog, started|active|in-progress|idle → in-progress, review → review, completed|done → done.

D. Get agent suggestions for a skill

curl -s 'http://localhost:6969/api/agents/suggest?skill=cm-debugging' | jq
# → { skill, domain, agents: ["claude-code","codex","cursor","antigravity"] }

E. Live activity & per-agent badges

The web UI shows:

• Agent badges with color + task count per agent (Codex teal, Antigravity green, Cursor blue, Gemini gold, Claude purple).
• Active Agents sidebar — currently working on in-progress tasks.
• Activity feed — task created/moved/dispatched/transitioned events in real time (WebSocket).

API reference (most-used)

Full surface: see src/dashboard.ts.

Recommended per-agent usage

Troubleshooting

• `{"error":"not found"}` — endpoint typo; check the table above. The fallback returns 404 for any unknown /api/*.
• Port 6969 already in use — another dashboard is running. lsof -i :6969 to find it.
• `NO_AGENT` on dispatch — task has no agent set; PUT /api/tasks/:id with {"agent":"codex"} first.
• `NO_PROJECT_PATH` — set the project's path so dispatch can write into .agent-tasks/.

Files of authority

• src/dashboard.ts — HTTP API
• src/agent-dispatch.ts — per-agent file + CLI generation
• src/judge.ts — agent affinity per skill domain
• public/dashboard/app.js — UI badges and colors