clawdbot-security

📁 spanishflu-est1918/clawdbot-security 📅 Jan 26, 2026
1
总安装量
1
周安装量
#55379
全站排名
安装命令
npx skills add https://github.com/spanishflu-est1918/clawdbot-security --skill clawdbot-security

Agent 安装分布

windsurf 1
clawdbot 1
opencode 1
claude-code 1
gemini-cli 1

Skill 文档

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.

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