Every agent skill.
Trust-scored.
Semantic search across the entire agent skill ecosystem. Security-scanned by Cognium. Ranked by trust and usage.
/
15K+
Skills indexed
<60ms
Search latency (p50)
6
Skill sources
3-tier
Confidence routing
Built for agents that need to trust their tools
Every skill is security-scanned, semantically indexed, and ranked by real usage signals.
Semantic Search
Natural language queries ranked by confidence, not keyword matching.
Trust Scores
Cognium Circle-IR scans every skill for vulnerabilities, license issues, and prompt injection.
3-Tier Intelligence
Vector search, cross-encoder reranking, and LLM fallback -- escalates automatically by confidence.
Multi-Source
One search across MCP Registry, ClawHub, GitHub, and human-curated sources.
Version Ranking
Ranked by trust x 0.7 + usage x 0.3. Most reliable version surfaces first.
Composition Detection
Multi-skill queries auto-decomposed. "Format code and check licenses" finds both.
Frequently asked questions
What Runics is, how it works, and how to use it.
What is Runics?
Runics is a semantic skill registry for AI agents. It indexes over 15,000 agent skills from sources like the MCP Registry, ClawHub, and GitHub, then ranks them by a security trust score computed by Cognium Circle-IR. Agents and developers can search by natural language and get results ranked by relevance, trust, and real usage.
How is Runics different from the MCP Registry and ClawHub?
The MCP Registry and ClawHub list skills but rely on keyword matching and do not score skills for security. Runics adds semantic search (vector embeddings plus cross-encoder reranking), vulnerability and prompt-injection scanning via Cognium Circle-IR, a continuously updated trust score, and confidence signals that tell an agent how certain a match is. Runics also aggregates skills from multiple sources in a single search.
What is a trust score and how is it computed?
A trust score is a 0–1 rating assigned to every indexed skill. It is produced by Cognium Circle-IR, which scans each skill for known vulnerabilities, license problems, unsafe permissions, and prompt-injection patterns. Scores update continuously as new scans run and as usage signals arrive. Versions are ranked by trust × 0.7 + usage × 0.3, so the most reliable version surfaces first rather than simply the newest.
What sources does Runics index?
Runics aggregates skills from the MCP Registry, ClawHub, GitHub repositories, and a human-curated source. All are normalized into a single search index, so one query returns results across every ecosystem.
How does the 3-tier intelligence search work?
Queries are routed through three tiers by confidence. Tier 1 is a fast vector similarity search; if the top results are clearly relevant, they are returned immediately with sub-60ms p50 latency. Tier 2 adds a cross-encoder reranker for medium-confidence matches. Tier 3 invokes an LLM for deep search and composition detection when the query is complex or ambiguous. Every response includes the tier used, confidence label, and latency.
Is the Runics API free?
Yes. Search is free and does not require authentication. Send a POST request to https://api.runics.net/v1/search with a JSON body containing a query and a tenantId. Rate limits apply for abuse prevention but are not publicly enforced for ordinary use.
What is composition detection?
Composition detection identifies queries that require more than one skill. For example, the query "format my code and check for license issues" is automatically decomposed: Runics finds a formatting skill and a license-scanning skill, and suggests composing them. This is returned as a structured composition object in the API response.
Start searching
$ curl -X POST https://api.runics.net/v1/search \
-H "Content-Type: application/json" \
-d '{"query": "web scraping", "tenantId": "public"}'