发现优质的 AI Agent 技能
聚合 Claude Skills、LangChain、AutoGPT 等优质资源，助力开发者快速构建智能应用
Security Expert Personas
Activate the appropriate persona based on current phase.
Ralph Ultra — 1,000 Iterations (~4-8 hours)
Deep-dive security audit with thorough coverage across all attack vectors.
Ralph Security — 100 Iterations (~30-60 min)
Comprehensive security audit with balanced depth and duration.
Ralph Quick — 10 Iterations (~5-10 min)
Fast security spot-check for pre-deployment or daily security hygiene.
Security Proxy Pattern
**Use case:** Isolate high-risk operations from main agent context (API calls to untrusted services, experimental features, sandbox testing)
Proxy Recovery Pattern
**Use case:** Define what "dying cleanly" means for a sacrificial security proxy, and how core responds when a proxy crashes — adversarially or otherwise.
Skill Vetter 🔒
Security-first vetting protocol for AI agent skills. **Never install a skill without vetting it first.**
OpenClaw Security Policy
Default stance: least privilege. Do not chain high-risk actions unattended.
OpenClaw Configuration Reference
Reference normalized against: - `https://docs.openclaw.ai/cli/config` - `https://docs.openclaw.ai/gateway/configuration` - Individual CLI pages for models, channels, agents, hooks, cron, security, secrets
OpenClaw CLI Reference
Source: `docs.openclaw.ai/cli` + individual subcommand pages. Verified: 2026-02-27. High-risk commands require `OPENCLAW_WRAPPER_ALLOW_RISKY=1`. See `security-policy.md`.