A Meta-Skill Engine for AI Agents: Intelligent Routing, Global Discovery, and Visible Security Guard.
When your AI Agent (like OpenClaw) encounters a missing capability or a complex long-chain task, Skill-Safefinder takes over as the central orchestrator. To prevent LLM hallucinations, it strictly enforces the Skill First principle: It breaks down complex intents, diagnoses local component gaps, automatically retrieves missing extension packages from global open-source communities (ClawHub, MCP Market, etc.), and runs a strict backend visible security code pre-check and risk warning before presenting them to you!
To orchestrate tools dynamically and perform global searches, this skill explicitly declares the following requirements:
- Runtime Required:
Node.js(npm/npx) andPython 3must be installed. - Filesystem Access: Read access to
~/.agents/skillsto evaluate the current baseline. - Network Access: Internet access to hub APIs (ClawHub, MCP, etc.) for skill discovery.
- Note on Security: Before downloading and recommending any external code, this skill enforces a strict local security scan using
npx agentguard scanvia a secured python wrapper.
-
🧭 Intelligent Routing It parses ambiguous language into a structured pipeline (e.g.,
[Search Web] -> [Analyze Data] -> [Send Email]) instead of doing everything in a single fragile prompt. -
🔍 Active Diagnostics Automatically scans your
.agents/skillsdirectory and recommends essential starting kits (Web Browsing, File Operating, Calendar) if you are a new user. -
🌍 Global Discovery Directly connects to major hubs (ClawHub, MCP Market, Smithery, Glama) to fetch whatever capability you lack.
-
🛡️ Visible Security Guard Integrates seamlessly with AgentGuard. Before recommending any external code, it runs a deep static scan (
security_check.py). If it detects malicious intents like stealing keys, it drops the skill and immediately reports the blockade to you, so you know exactly why it was rejected. -
💬 Language Mirroring No hardcoded UX texts. The status updates and safety prompts will automatically mirror the exact language you used in your initial prompt.
Copy and send the following complete instruction to your AI Agent:
"Please install skill-safefinder: run
npx --yes clawhub install skill-safefinder. After the terminal confirms success, it is recommended to read itsSKILL.mdand execute the[AGENT POST-INSTALL HOOK]inside it for initialization. Then report back to me."
- Download the Package: Download the
Skill-Safefinder.skillarchive from the Releases page or ClawHub. - Mount to your Agent: Copy the folder into your global agent directory (e.g.,
~/.agents/skills/Skill-Safefinder). - Tell your AI: After mounting, say to your Agent: "I just installed Skill-Safefinder. Read its SKILL.md and run the
[AGENT POST-INSTALL HOOK]."
If you wish to modify its behavior, you can directly edit:
SKILL.md: The main entry and triggering boundary (what NOT to intercept).references/routing.md: Execution rules.references/discovery.md: External repo search priorities and AgentGuard strict hooks.
License: Standard MIT License.