Clawdbot Security

Harden Clawdbot Gateway deployments by fetching current security practices and applying them.

Fetch Current Docs First

Before advising on security, fetch the latest official documentation:

# Primary security guide
web_fetch https://raw.githubusercontent.com/clawdbot/clawdbot/main/docs/gateway/security.md

# Specific topics
web_fetch https://raw.githubusercontent.com/clawdbot/clawdbot/main/docs/gateway/sandboxing.md
web_fetch https://raw.githubusercontent.com/clawdbot/clawdbot/main/docs/gateway/authentication.md
web_fetch https://raw.githubusercontent.com/clawdbot/clawdbot/main/docs/gateway/pairing.md

Quick Commands

clawdbot security audit           # Basic check
clawdbot security audit --deep    # Live Gateway probe
clawdbot security audit --fix     # Auto-apply safe fixes
clawdbot sandbox explain          # Debug sandbox/tool policy

Decision Tree

User needs help with security
├── Running audit or interpreting findings?
│   └── Run `clawdbot security audit`, explain findings, suggest fixes
├── Configuring access control (DM/group policies)?
│   └── Fetch security.md, explain pairing vs allowlist vs open
├── Setting up sandboxing?
│   └── Fetch sandboxing.md, explain mode/scope/workspaceAccess
├── Managing secrets in config?
│   └── Run scripts/extract_secrets.py or guide manual .env creation
├── Responding to incident?
│   └── Follow contain → rotate → audit workflow
└── General security question?
    └── Fetch security.md, consult references/quick-reference.md

Three-Layer Model (Quick Summary)

1. Sandbox — Where tools run (Docker vs host)
2. Tool Policy — Which tools are allowed
3. Elevated — Exec-only host escape hatch

For details, read [references/quick-reference.md](references/quick-reference.md).

Extract Secrets Script

Automate moving hardcoded secrets from config to .env:

python scripts/extract_secrets.py --dry-run  # Preview
python scripts/extract_secrets.py            # Execute

Audit Priority Order

1. Open groups + tools → Lock down with allowlists
2. Network exposure → Fix immediately
3. Browser control → Require token auth
4. File permissions → chmod 600 config, chmod 700 dirs
5. Plugins → Only load trusted ones
6. Model choice → Use instruction-hardened models (Opus 4.5)

Incident Response

1. Contain: Stop gateway, bind: "loopback", freeze policies
2. Rotate: gateway.auth.token, API keys, browser control token
3. Audit: Check logs (/tmp/clawdbot/*.log) and transcripts
4. Re-run: clawdbot security audit --deep