Multi-Agent Autonomous Startup System for Claude Code

Transform a PRD into a fully deployed, revenue-generating product with zero human intervention.

Loki Mode is a Claude Code skill that orchestrates 37 specialized AI agents across 6 swarms to autonomously build, deploy, and operate a complete startup. Just say "Loki Mode" and provide a PRD.

PRD → Competitive Research → Architecture → Development → Testing → Deployment → Marketing → Revenue

eng-frontend eng-backend eng-database eng-mobile eng-api eng-qa eng-perf eng-infra

ops-devops ops-sre ops-security ops-monitor ops-incident ops-release ops-cost ops-compliance

biz-marketing biz-sales biz-finance biz-legal biz-support biz-hr biz-investor biz-partnerships

data-ml data-eng data-analytics

prod-pm prod-design prod-techwriter

growth-hacker growth-community growth-success growth-lifecycle

review-code review-business review-security

SKILL.md              # ← THE SKILL (required) - contains YAML frontmatter
references/
├── agents.md         # Agent definitions
├── deployment.md     # Deployment guides
└── business-ops.md   # Business workflows

1. Go to Releases
2. Download loki-mode-X.X.X.zip or loki-mode-X.X.X.skill
3. Go to Claude.ai → Settings → Features → Skills
4. Upload the zip/skill file

The zip has SKILL.md at the root level as Claude.ai expects.

1. Go to Releases
2. Download loki-mode-api-X.X.X.zip (note: the -api- version)
3. Go to console.anthropic.com → Skills
4. Upload the zip file

The API version has SKILL.md inside a loki-mode/ folder as the API requires.

Option A: Download from Releases

# Download the Claude Code version
cd ~/.claude/skills

# Get latest version number
VERSION=$(curl -s https://api.github.com/repos/asklokesh/claudeskill-loki-mode/releases/latest | grep tag_name | cut -d'"' -f4 | tr -d 'v')

# Download and extract
curl -L -o loki-mode.zip "https://github.com/asklokesh/claudeskill-loki-mode/releases/download/v${VERSION}/loki-mode-claude-code-${VERSION}.zip"
unzip loki-mode.zip && rm loki-mode.zip
# Creates: ~/.claude/skills/loki-mode/SKILL.md

Option B: Git Clone

# For personal use (all projects)
git clone https://github.com/asklokesh/claudeskill-loki-mode.git ~/.claude/skills/loki-mode

# For a specific project only
git clone https://github.com/asklokesh/claudeskill-loki-mode.git .claude/skills/loki-mode

Option C: Minimal Install (curl)

mkdir -p ~/.claude/skills/loki-mode/references
curl -o ~/.claude/skills/loki-mode/SKILL.md https://raw.githubusercontent.com/asklokesh/claudeskill-loki-mode/main/SKILL.md
curl -o ~/.claude/skills/loki-mode/references/agents.md https://raw.githubusercontent.com/asklokesh/claudeskill-loki-mode/main/references/agents.md
curl -o ~/.claude/skills/loki-mode/references/deployment.md https://raw.githubusercontent.com/asklokesh/claudeskill-loki-mode/main/references/deployment.md
curl -o ~/.claude/skills/loki-mode/references/business-ops.md https://raw.githubusercontent.com/asklokesh/claudeskill-loki-mode/main/references/business-ops.md

# Check the skill is in place
cat ~/.claude/skills/loki-mode/SKILL.md | head -5
# Should show YAML frontmatter with name: loki-mode

Use the autonomous runner - it handles everything:

# Run with a PRD (fully autonomous with auto-resume)
./autonomy/run.sh ./docs/requirements.md

# Run interactively
./autonomy/run.sh

The autonomous runner will:

1. Check all prerequisites (Claude CLI, Python, Git, etc.)
2. Verify skill installation
3. Initialize the .loki/ directory
4. Start status monitor (updates .loki/STATUS.txt every 5s)
5. Start Claude Code with live output (see what's happening)
6. Auto-resume on rate limits or interruptions
7. Continue until completion or max retries

No more staring at a blank screen! Claude's output is displayed in real-time:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  CLAUDE CODE OUTPUT (live)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

[You see Claude working in real-time here...]

Monitor task progress in another terminal:

# Watch status updates live
watch -n 2 cat .loki/STATUS.txt

Output:

╔════════════════════════════════════════════════════════════════╗
║                    LOKI MODE STATUS                            ║
╚════════════════════════════════════════════════════════════════╝

Phase: DEVELOPMENT

Tasks:
  ├─ Pending:     10
  ├─ In Progress: 1
  ├─ Completed:   5
  └─ Failed:      0

If you prefer manual control:

# Launch Claude Code with autonomous permissions
claude --dangerously-skip-permissions

# Then say:
> Loki Mode

# Or with a specific PRD:
> Loki Mode with PRD at ./docs/requirements.md

Environment variables to customize the autonomous runner:

# Example with custom settings
LOKI_MAX_RETRIES=100 \
LOKI_BASE_WAIT=120 \
LOKI_MAX_WAIT=7200 \
./autonomy/run.sh ./docs/requirements.md

./autonomy/run.sh prd.md
         │
         ▼
┌─────────────────────┐
│ Check Prerequisites │ ← Claude CLI, Python, Git, etc.
└─────────────────────┘
         │
         ▼
┌─────────────────────┐
│ Initialize .loki/   │ ← State, queues, logs
└─────────────────────┘
         │
         ▼
┌─────────────────────┐
│ Run Claude Code     │◄─────────────────┐
└─────────────────────┘                  │
         │                               │
    Exit code?                           │
         │                               │
    ┌────┴────┐                          │
    ▼         ▼                          │
 Success   Rate Limit                    │
    │         │                          │
    ▼         ▼                          │
 DONE!    Wait (exponential backoff) ────┘

If you stop the script (Ctrl+C) or it crashes:

# Just run it again - state is saved automatically
./autonomy/run.sh ./docs/requirements.md

See autonomy/README.md for detailed documentation.

Every task goes through 3 reviewers simultaneously:

IMPLEMENT → REVIEW (3 parallel) → AGGREGATE → FIX → RE-REVIEW → COMPLETE
                │
                ├─ code-reviewer (opus)
                ├─ business-logic-reviewer (opus)
                └─ security-reviewer (opus)

When running, Loki Mode creates:

.loki/
├── state/          # Orchestrator and agent states
├── queue/          # Task queue (pending, in-progress, completed, dead-letter)
├── messages/       # Inter-agent communication
├── logs/           # Audit logs
├── config/         # Configuration files
├── prompts/        # Agent role prompts
├── artifacts/      # Releases, reports, backups
└── scripts/        # Helper scripts

# .loki/config/circuit-breakers.yaml
defaults:
  failureThreshold: 5
  cooldownSeconds: 300

# .loki/config/alerting.yaml
channels:
  slack:
    webhook_url: "${SLACK_WEBHOOK_URL}"
    severity: [critical, high]

Test the skill with these pre-built PRDs in the examples/ directory:

# Example: Test with simple todo app
claude --dangerously-skip-permissions
> Loki Mode with PRD at examples/simple-todo-app.md

The skill includes a comprehensive test suite:

# Run all tests
./tests/run-all-tests.sh

# Run individual test suites
./tests/test-bootstrap.sh        # Directory structure, state init
./tests/test-task-queue.sh       # Queue operations, priorities
./tests/test-circuit-breaker.sh  # Failure handling, recovery
./tests/test-agent-timeout.sh    # Timeout, stuck process handling
./tests/test-state-recovery.sh   # Checkpoints, recovery

• Claude Code with --dangerously-skip-permissions flag
• Internet access for competitive research and deployment
• Cloud provider credentials (for deployment phase)
• Python 3 (for test suite)

Optionally integrate with Vibe Kanban for a visual kanban board to monitor Loki Mode's agents:

# Install Vibe Kanban
npx vibe-kanban

# Export Loki tasks to Vibe Kanban
./scripts/export-to-vibe-kanban.sh

Benefits:

• Visual progress tracking of all 37 agents
• Manual intervention/prioritization when needed
• Code review with visual diffs
• Multi-project dashboard

See integrations/vibe-kanban.md for full setup guide.

Contributions welcome! Please read the skill and open issues for bugs or feature requests.

MIT License - see LICENSE for details.

• Inspired by LerianStudio/ring subagent-driven-development pattern
• Built for the Claude Code ecosystem

Keywords: claude-code, claude-skills, ai-agents, autonomous-development, multi-agent-system, sdlc-automation, startup-automation, devops, mlops, deployment-automation