FleetHawk — Trust your AI fleet before you trust its output.

What is FleetHawk

FleetHawk is a CLI for AI fleet diagnostics, monitoring, and release safety. It works with OpenClaw, Claude Code, Codex, and any agent framework that leaves useful config, workspace, or session artifacts behind.

Why it exists

AI agents fail silently. Wrong models, stale configs, broken prompts, localhost DBs in production, and ghost monitoring entries all look harmless until they become incidents; FleetHawk catches them before that happens.

Quick start

npm install -g fleethawk
fleethawk doctor

All commands

fleethawk status

One-shot fleet health view with model names, last activity, last task, and colored status.

fleethawk status

Sample output:

🦅 FleetHawk Status

┌──────────┬──────────────────────────────┬────────────┬──────┬───────┬─────────┬────────────────────────────┬──────────────┐
│ Agent    │ Model                        │ Last Output│ Idle │ Files │ Commits │ Last Task                  │ Status       │
├──────────┼──────────────────────────────┼────────────┼──────┼───────┼─────────┼────────────────────────────┼──────────────┤
│ atlas    │ anthropic/claude-sonnet-4-6  │ 2m ago     │ 2m   │ 14    │ 2       │ Refactor auth module       │ ✅ active    │
│ nova     │ openai/gpt-5.4              │ 15m ago    │ 15m  │ 8     │ 1       │ Fix payment webhook        │ ✅ active    │
│ scout    │ google/gemini-2.5-pro        │ 1h ago     │ 1h   │ 3     │ 0       │ Research API docs          │ ⚠️ idle      │
│ forge    │ anthropic/claude-opus-4-6    │ 3h ago     │ 3h   │ 0     │ 0       │ Deep code review           │ 🔴 silent    │
│ pixel    │ ollama/qwen3.5-35b-a3b          │ never      │ n/a  │ 0     │ 0       │ —                          │ 🔴 no output │
│ sage     │ moonshot/kimi-k2.5        │ 45m ago    │ 45m  │ 2     │ 0       │ Summarize meeting notes    │ ⚠️ idle      │
└──────────┴──────────────────────────────┴────────────┴──────┴───────┴─────────┴────────────────────────────┴──────────────┘

fleethawk watch

Continuous monitoring with configurable thresholds and deduped alerts.

fleethawk watch --idle-threshold 20m --poll-interval 2m

fleethawk monitor

Combined watch + periodic doctor checks.

fleethawk monitor --poll-interval 2m --doctor-interval 30m

fleethawk report

Accountability report with table, markdown, or JSON output.

fleethawk report --format md
fleethawk report --format json

Sample markdown excerpt:

# FleetHawk Report

## atlas
- Model: anthropic/claude-sonnet-4-6
- Status: active
- Files modified: 14
- Git commits: 2
- Session messages: 8
- Last task: Refactor auth module
- Errors: none

## nova
- Model: openai/gpt-5.4
- Status: active
- Files modified: 8
- Git commits: 1
- Session messages: 5
- Last task: Fix payment webhook
- Errors: none

## scout
- Model: google/gemini-2.5-pro
- Status: idle
- Files modified: 3
- Git commits: 0
- Session messages: 12
- Last task: Research API docs
- Errors: none

fleethawk doctor

Full fleet diagnostic: model validity, provider mapping, stale gateway, agentDir existence, workspace collisions, identity bleed, ghost agents, repo hygiene, and DB env mismatch checks.

fleethawk doctor
fleethawk doctor --json

Sample output:

🦅 FleetHawk Doctor

✅ atlas — model anthropic/claude-sonnet-4-6 valid
✅ nova — model openai/gpt-5.4 valid
✅ scout — model google/gemini-2.5-pro valid
✅ forge — model anthropic/claude-opus-4-6 valid
⚠️ pixel — model ollama/llama-3.2:8b (local model, cannot verify remotely)
✅ sage — model moonshot/kimi-k2.5 valid
✅ Gateway running (pid 4821, port 18789)
✅ No workspace collisions detected
✅ No identity bleed between agents
⚠️ pixel workspace dir does not exist: ~/.agents/workspace-pixel
✅ DB env check passed — NEON_DATABASE_URL set, no localhost in production
✅ No ghost agents in monitoring config

Results: 10 passed, 2 warnings, 0 critical

fleethawk verify models

Probe each configured agent and compare configured vs responding model.

fleethawk verify models

fleethawk verify agent <id>

Deep inspection for one agent.

fleethawk verify agent atlas

fleethawk verify db

Database target safety check for the current repo.

fleethawk verify db

fleethawk release-check

Pre-merge safety gate for branch hygiene, SQL artifacts, and DB env risks.

fleethawk release-check

fleethawk audit security

Searches for common secret leaks, debug routes, missing rate-limit clues, and runs npm audit.

fleethawk audit security

fleethawk audit deps

Dependency health snapshot.

fleethawk audit deps

fleethawk fleet-map

Prints fleet topology in a clean table.

fleethawk fleet-map

fleethawk config validate

Validates openclaw.json for missing models, missing agent dirs, and workspace collisions.

fleethawk config validate

fleethawk init

Generate a starter fleethawk.config.yaml from a fleet directory.

fleethawk init --fleet ~/agents

Configuration

FleetHawk reads from ~/.openclaw/openclaw.json by default. You can also point it at a FleetHawk YAML config via --config, or a custom OpenClaw config via --openclaw-config.

Real issues caught

These are the kinds of failures FleetHawk is designed to catch:

• Model misrouting: an agent configured for GPT-5.4 silently running on a fallback model
• Broken production resources found via automated QA
• Ghost agents in monitoring config that no longer exist
• Workspace identity bleed between agents sharing directories
• Malformed database URLs (double postgresql:// prefix)
• Stale gateway state after config changes

Works with

• OpenClaw
• Claude Code
• Codex
• Any agent framework with readable config, workspaces, sessions, or repo artifacts

Contributing

PRs welcome, issues on GitHub.

License

MIT