ai-cockpit

AI Cockpit Template

Universal session lifecycle primitives for Claude Code workspaces.

Hop in. Fly. Land.

What This Is

A cockpit is a Claude Code workspace with a rhythm:

1. /takeoff — Boot sequence. Reads your last bookmark, checks what changed, shows priorities.
2. Work — Use your domain-specific skills. The cockpit tracks state.
3. /touch-and-go — Mid-session checkpoint. Saves state, compacts context, keeps flying.
4. /can-i-close — Pre-close audit. Checks 3 contracts (workspace, session, conversation).
5. /land — Park sequence. Gated by can-i-close. Commits, pushes, bookmarks.

This template provides the universal primitives that every cockpit needs. Fork it, add your domain-specific skills, and you have a cockpit.

Installation

pip install ai-cockpit

Or install from source:

gh repo clone eidos-agi/ai-cockpit-template
cd ai-cockpit-template
pip install -e .

Then tell cockpit where your workspaces live:

cockpit config --add-scan-dir ~/my-cockpits
cockpit scan

Claude Code Plugins

The cockpit pairs well with Eidos plugins for Claude Code:

cockpit marketplace

Or install the marketplace directly:

claude plugins marketplace add eidos-agi/eidos-marketplace
claude plugins install resume-resume
claude plugins install ike
claude plugins install visionlog

Quick Start

# Create a new cockpit
cockpit new ~/repos/my-ops-cockpit

# Launch it
cockpit my-ops-cockpit
# > /takeoff
# > ... do your work ...
# > /touch-and-go  (mid-session checkpoint)
# > ... more work ...
# > /can-i-close   (pre-close check)
# > /land

Or from a GitHub template:

gh repo create my-cockpit --template eidos-agi/ai-cockpit-template --public
cd my-cockpit
cockpit add .

CLI Commands

The cockpit command manages your fleet of cockpits:

Every command supports --help.

What's Included

Skills (Universal Primitives)

State Management

• state.json — Watermarks, counters, last-run timestamps. Skills read/write this to enable incremental operations.
• Bookmarks — Written to ~/.claude/bookmarks/ by /land. Read by /takeoff on next session. This is the bridge between sessions.

Cockpit Architecture

your-cockpit/
├── .claude/
│   └── skills/
│       ├── takeoff/              # Boot sequence (from template)
│       ├── land/                 # Park sequence (from template)
│       ├── cockpit-status/       # Instrument panel (from template)
│       ├── pre-flight/           # Situational scan (from template)
│       ├── cockpit-repair/       # Diagnostics (from template)
│       ├── clean-sweep/          # Workspace sweep (from template)
│       └── your-domain-skill/    # Your additions
├── tools/
│   └── learning-browser/         # Persistent browser research (from template)
├── bin/
│   └── update-from-template      # Pull skill updates from upstream
├── skills_manifest.json           # SHA256 hashes for update tracking
├── state.json                    # Session state & watermarks
├── CLAUDE.md                     # Role context + instructions
└── ...                           # Your domain-specific structure

Fleet Sync (opt-in)

For multi-repo organizations, /takeoff can sync your entire fleet before boot. Add these fields to state.json:

{
  "cockpit": {
    "org": "your-github-org",
    "repos_dir": "~/repos-your-org"
  },
  "fleet": {
    "project-a": { "display_name": "Project A" },
    "project-b": { "display_name": "Project B" }
  },
  "pilots": {
    "alice": { "git_names": ["Alice Smith", "alice"], "git_emails": ["alice@example.com"] },
    "bob": { "git_names": ["Bob Jones"], "git_emails": ["bob@example.com"] }
  }
}

When configured, /takeoff will:

1. Discover repos via gh repo list <org>
2. Fetch and pull all local fleet repos
3. Build a pilot activity map (who committed where in the last 7 days)
4. Surface fleet health in the terminal, takeoff.md, and cockpit.html

Single-repo cockpits (no org or repos_dir) skip fleet sync entirely — no config needed.

Design Principles

1. Cheap boots — /takeoff reads 2 things: state.json and latest bookmark. Git state and CLAUDE.md are already in session context. No redundant I/O.
2. Session contracts — Every session has a lifecycle: boot → work → checkpoint → audit → land. Three contracts must pass before closing: workspace (git), session (bookmarks/tasks), conversation (promises/decisions).
3. Drift detection — If the branch changed, files were modified, or commits landed since your last bookmark, /takeoff tells you.
4. Domain-agnostic — The template knows nothing about your project. It only knows about sessions, bookmarks, and state.
5. Composable — Add as many domain skills as you want. The primitives stay the same.

Keeping Cockpits in Sync

For cockpit owners (pull updates into your cockpit)

cd your-cockpit
./bin/update-from-template              # pull latest skills from upstream
./bin/update-from-template --dry-run    # preview what would change
./bin/update-from-template --check      # just check if updates are available

This is the recommended approach. Each cockpit pulls from its upstream template. No central registry needed — the cockpit knows its template from state.json.

• Only updates template skills — never touches your domain-specific skills
• Detects local customizations via SHA256 hashing — won't clobber your changes
• Updates skills_manifest.json and state.json with the new version

Requires: gh (GitHub CLI) and jq.

For template maintainers (push updates to known cockpits)

cd ai-cockpit-template
./bin/sync-skills              # push all skills to all registered cockpits
./bin/sync-skills land         # push just one skill
./bin/sync-skills --dry-run    # preview without changes

Reads bin/cockpit-registry.txt for local cockpit paths. Useful when you maintain multiple cockpits on one machine.

Release workflow (for template maintainers)

# 1. Make your skill changes
# 2. Generate the manifest
./bin/generate-manifest v1.3.0

# 3. Commit and tag
git add skills_manifest.json
git commit -m "release: v1.3.0"
git tag v1.3.0
git push --tags

# 4. Downstream cockpits can now: ./bin/update-from-template

Documentation

In the Wild

Cockpits built from this template:

• Eidos Cockpit — Multi-pilot planning & mission control for Eidos AGI
• AIC Director of AI — Email-centric command post across 4 sub-roles
• Greenmark Planning — Waste management leadership planning hub
• Reeves Cockpit — Personal assistant command post
• (Add yours here)

License

MIT. Built by Eidos AGI — free software for coding agents.