Mission Control

Standalone service + OpenClaw plugin for managing autonomous AI agent swarms. Provides task management, a real-time command center dashboard, knowledge management with graph-aware retrieval, service health monitoring, and agent tools.

Orchestrates an autonomous coding pipeline: Linear ticket → triage → plan → step-by-step execution → verify → PR → review → done.

Architecture

Mission Control runs as a standalone HTTP service (port 18790) independent of the OpenClaw gateway. OpenClaw connects via a thin proxy plugin — MC stays up even when the gateway restarts.

┌─────────────────────────────────────────────────────────────┐
│ Mission Control (standalone, port 18790)                     │
│  ├── SQLite (tasks, activities, deliverables, workspaces)    │
│  ├── HTTP API (/ext/mission-control/api/*)                   │
│  ├── Dashboard UI (/ext/mission-control/)                    │
│  ├── Service Health Monitor                                  │
│  └── Knowledge Review UI                                     │
├─────────────────────────────────────────────────────────────┤
│ Context Fabrica (PostgreSQL + pgvector)                      │
│  ├── Knowledge records (218+ entries)                        │
│  ├── Embeddings (3072-dim Gemini)                            │
│  ├── Entity relations (645+ edges)                           │
│  └── Graph-aware retrieval                                   │
├─────────────────────────────────────────────────────────────┤
│ OpenClaw Gateway (port 18789)                                │
│  ├── Proxy plugin → MC API                                   │
│  ├── mc_* agent tools                                        │
│  ├── memory plugin → Context Fabrica                         │
│  └── Slack / Telegram channels                               │
└─────────────────────────────────────────────────────────────┘

How It Works

Linear (ticket created)
  │
  ▼
Linear Sync ─────────────────── pulls tickets with a configured label
  │                              into Mission Control as tasks
  ▼
Bridge (triage) ─────────────── asks codebase-aware questions,
  │                              self-answers from codebase when possible,
  │                              extracts knowledge from ticket + Notion,
  │                              posts to Linear, waits for answers
  ▼
Planner (Sonnet) ────────────── generates structured execution plan
  ▼
Agent (Claude Code) ────────── works in git worktree with GSD:
  │                              /gsd:plan-phase → /gsd:execute-phase
  │                              → /gsd:verify-work (GUARDRAIL)
  ▼
Review Loop ─────────────────── self-review + codex review + re-verify
  ▼
PR Created ─────────────────── GSD artifacts saved as deliverables,
  │                              completion comment posted to Linear,
  │                              knowledge distilled from artifacts
  ▼
Watch-PR-Reviews ────────────── polls GitHub for human review comments,
  │                              relaunches agent on feedback
  ▼
Done ────────────────────────── Linear sync detects completion

Knowledge System

Knowledge is stored in Context Fabrica (PostgreSQL + pgvector), replacing the previous LanceDB backend. All knowledge consumers — OpenClaw chat, bridge triage, agent tools — read and write from the same database.

Knowledge Sources (Auto-Extraction)

Source	When	Stage	Trigger
Codebase scanning	On-demand / every 10 min	staged	`mc-explore` CLI / repo-watcher
Linear ticket descriptions	During triage	staged	Bridge auto-extract
Notion page content	During triage	staged	Bridge auto-extract
Task completion (GSD artifacts)	Task → review	canonical	Bridge + knowledge-distill
Research Q&A	Research task done	canonical	linear-sync
OpenClaw chat (Slack/Telegram)	User interaction	canonical	memory plugin
Manual injection	Anytime	canonical	`mc_add_knowledge` tool / CLI
Cross-repo architecture	On-demand	canonical	`mc-explore --cross-repo`

Knowledge Tiers

Developer Notes (human-injected, 1.5x scoring boost, always surface)
Procedural Skills (auto-created from complex tasks, full content injected)
Atomic Facts (auto-distilled one-liners, compact)

Staged Review

Auto-extracted knowledge enters as staged and must be reviewed before it surfaces in agent recall. The MC dashboard includes a Knowledge Review panel for approving/rejecting staged entries.

Entity Relations & Graph Scoring

Knowledge entries include entity relations (DEPENDS_ON, OWNS, CALLS, IMPLEMENTS, USES, etc.) stored in memory_relations. Query-time scoring blends:

50% semantic (BM25 + embedding similarity)
30% graph (entity relation proximity, multi-hop traversal)
12% recency (temporal freshness)
8% confidence (source credibility)

Codebase Explorer

# Scan a repo, extract architectural knowledge
mc-explore myorg/backend-api

# Focus on specific area
mc-explore myorg/backend-api --focus "API endpoints"

# Monorepo — auto-detects packages
mc-explore myorg/platform

# Specific monorepo package
mc-explore myorg/platform --package api

# Preview without storing
mc-explore myorg/backend-api --dry-run

# Trust mode — store as canonical (skip staged review)
mc-explore myorg/backend-api --trust

Repo Watcher

Background service (every 10 min) that detects code changes and incrementally updates knowledge:

Compares git SHAs, pulls latest, analyzes diffs
Extracts new architectural facts from changed files
Supersedes outdated knowledge (old records get valid_to, new records link via supersedes)
Skips non-architectural changes (lock files, tests, CI configs)

Service Health Monitoring

MC monitors 9 services with auto-refresh every 30s:

Service	Type	Purpose
Mission Control	HTTP (18790)	Task management, API, dashboard
OpenClaw Gateway	HTTP (18789)	Agent communication, Slack/Telegram
Bridge	launchd (60s)	Triage, planning, dispatch
Linear Sync	launchd (300s)	Ticket sync from Linear
Watch PR Reviews	launchd (120s)	GitHub review comment detection
Review PRs	launchd	PR review orchestration
Check Agents	launchd (120s)	Agent health, tmux validation
Repo Watcher	launchd (600s)	Codebase knowledge updates
PostgreSQL	pg_isready	Context Fabrica database

Health API: GET /ext/mission-control/api/services/health

Auto-restart: MC uses KeepAlive: true in launchd. Bridge uses KeepAlive.SuccessfulExit: false (restart on crash, not on normal exit). Health monitor script alerts via Telegram/Slack after 2+ consecutive failures.

Dashboard

The command center dashboard (http://localhost:18790/ext/mission-control/) provides:

Services panel — real-time health of all 9 services with status badges
Pipeline bar — INTAKE → TRIAGE → DISPATCH → SWARM OPS → REVIEW → COMPLETE
Task card grid — status badges, agent indicators, priority highlights
Task drawer — description, triage Q&A, deliverables (PLAN.md, VERIFICATION.md), activity log
Knowledge panel — Browse/Review/Canonical tabs, approve/reject staged entries
System stats — CPU, memory, agent concurrency

API Endpoints

All endpoints served under /ext/mission-control/api/:

Tasks

Method	Endpoint	Description
GET	`/tasks`	List tasks (query: `status`, `workspace_id`, `assigned_agent_id`)
POST	`/tasks`	Create task
GET	`/tasks/:id`	Get task by ID
PATCH	`/tasks/:id`	Update task
DELETE	`/tasks/:id`	Delete task
GET	`/tasks/:id/activities`	List task activities
POST	`/tasks/:id/activities`	Log activity
GET	`/tasks/:id/deliverables`	List task deliverables (GSD artifacts)
POST	`/tasks/:id/deliverables`	Add deliverable

Knowledge

Method	Endpoint	Description
GET	`/knowledge`	List entries (query: `stage`, `project`, `repo`, `scope`, `limit`)
POST	`/knowledge`	Add knowledge entry
DELETE	`/knowledge/:id`	Delete entry
POST	`/knowledge/:id/promote`	Promote staged → canonical
POST	`/knowledge/:id/reject`	Reject (delete) staged entry
PATCH	`/knowledge/:id`	Update entry text/domain

Services

Method	Endpoint	Description
GET	`/services/health`	Health status of all services

Other

Method	Endpoint	Description
GET	`/agents`	List agents
GET	`/agent-status`	Live agent status with tmux detection
GET	`/workspaces`	List workspaces
GET	`/board`	Dashboard board data
GET	`/system-stats`	CPU, memory, concurrency

Setup

Prerequisites

Node.js 20+ with tsx
PostgreSQL 17 with pgvector extension
Python 3.12+ with context-fabrica installed
OpenClaw gateway (optional, for Slack/Telegram integration)

Install

# Clone
cd ~/GitProjects/openclaw-plugins
git clone https://github.com/jimmdd/mission-control.git
cd mission-control
npm install

# Set up PostgreSQL
brew install postgresql@17 pgvector
brew services start postgresql@17
createdb context_fabrica
psql -d context_fabrica -c "CREATE EXTENSION IF NOT EXISTS vector;"

# Install context-fabrica
pip install context-fabrica
python -m context_fabrica.bootstrap_cli --dsn "postgresql://$(whoami)@localhost/context_fabrica"

# Configure environment
cat >> ~/.openclaw/.env << 'EOF'
CONTEXT_FABRICA_DSN=postgresql://$(whoami)@localhost/context_fabrica
GOOGLE_GENERATIVE_AI_API_KEY=your-gemini-key
ANTHROPIC_API_KEY=your-anthropic-key
MISSION_CONTROL_URL=http://127.0.0.1:18790/ext/mission-control
EOF

Start Standalone Service

# Start MC
npm start
# Or via launchd (auto-restart on crash):
launchctl load ~/Library/LaunchAgents/ai.openclaw.mission-control.plist

# Verify
curl http://localhost:18790/health
open http://localhost:18790/ext/mission-control/

Register with OpenClaw (optional)

The plugin proxies to the standalone MC service. Set dbPath to the PostgreSQL DSN:

{
  "plugins": {
    "entries": {
      "mission-control": {
        "config": {
          "dbPath": "postgresql://user@localhost/context_fabrica"
        }
      }
    }
  }
}

Initial Knowledge Extraction

# Scan all repos in a project
for repo in $(ls ~/GitProjects/myorg/); do
  mc-explore myorg/$repo
done

# Review staged entries in the dashboard
open http://localhost:18790/ext/mission-control/

Project Structure

mission-control/
├── server.ts                        # Standalone HTTP server entry point
├── index.ts                         # OpenClaw plugin (thin proxy to standalone MC)
├── openclaw.plugin.json             # Plugin manifest
├── package.json
├── public/
│   ├── index.html                   # Dashboard (services, tasks, knowledge review)
│   └── space.html                   # Space page (operational view)
├── src/
│   ├── db.ts                        # SQLite schema, CRUD operations
│   ├── routes.ts                    # HTTP API + dashboard serving
│   └── tools.ts                     # Agent-callable tools (proxy to MC API)
├── health/
│   ├── service-health.py            # Service health checker (9 services)
│   └── mc-health-check.sh           # Health monitor with alerting
└── swarm/                           # Orchestration scripts
    ├── bridge.py                    # Triage, plan, dispatch, Q&A loop
    ├── linear-sync.py               # Linear ticket sync + research distillation
    ├── knowledge-manage.py          # Knowledge CRUD via context-fabrica
    ├── knowledge-distill.py         # Extract skills + facts from completed tasks
    ├── knowledge-review.py          # Review staged knowledge (promote/reject)
    ├── knowledge-feedback.py        # Track recall/helped counts
    ├── mc-explore.py                # Codebase knowledge extractor CLI
    ├── mc_explore_common.py         # Shared utilities for explore + watcher
    ├── repo-watcher.py              # Background repo change → knowledge updater
    ├── repo-watcher-precheck.sh     # SHA comparison precheck (zero tokens)
    ├── watch-pr-reviews.sh          # GitHub PR review comment detection
    ├── check-agents.sh              # Agent health monitor
    ├── spawn-agent.sh               # Create worktree, register agent, start tmux
    ├── run-claude.sh                # Launch Claude Code agent
    └── run-codex.sh                 # Launch Codex agent (fallback)

Dependencies

context-fabrica — Knowledge storage (PostgreSQL + pgvector + entity relations)
OpenClaw — Agent gateway (optional, for Slack/Telegram/tool registration)
better-sqlite3 — Task database
Gemini API — Embeddings (gemini-embedding-001) + knowledge extraction (gemini-2.5-flash)
Anthropic API — Agent model (claude-sonnet-4-6) + planning

License

MIT

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Mission Control

Architecture

How It Works

Knowledge System

Knowledge Sources (Auto-Extraction)

Knowledge Tiers

Staged Review

Entity Relations & Graph Scoring

Codebase Explorer

Repo Watcher

Service Health Monitoring

Dashboard

API Endpoints

Tasks

Knowledge

Services

Other

Setup

Prerequisites

Install

Start Standalone Service

Register with OpenClaw (optional)

Initial Knowledge Extraction

Project Structure

Dependencies

License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Name		Name	Last commit message	Last commit date
Latest commit History 37 Commits
health		health
public		public
src		src
swarm		swarm
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
index.ts		index.ts
openclaw.plugin.json		openclaw.plugin.json
package-lock.json		package-lock.json
package.json		package.json
server.ts		server.ts
tsconfig.json		tsconfig.json

Folders and files

Latest commit

History

Repository files navigation

Mission Control

Architecture

How It Works

Knowledge System

Knowledge Sources (Auto-Extraction)

Knowledge Tiers

Staged Review

Entity Relations & Graph Scoring

Codebase Explorer

Repo Watcher

Service Health Monitoring

Dashboard

API Endpoints

Tasks

Knowledge

Services

Other

Setup

Prerequisites

Install

Start Standalone Service

Register with OpenClaw (optional)

Initial Knowledge Extraction

Project Structure

Dependencies

License

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages