Provides a policy enforcement layer that sits between clients and tool execution, applying allow/deny/step-up approval rules to every tool call. Supports both standalone mode with sandboxed shell and filesystem tools, and proxy mode that routes to downstream servers with namespaced tool merging. Features include 2-minute TTL step-up approvals via local CLI, append-only JSONL evidence logging, separation of duties enforcement, and Docker-based sandboxing with dropped capabilities and network isolation.
Cognium trust score
50%
Tier
Unverified
Composite of vulnerability cleanliness, spec conformance, provenance, stability, and usage signals — scanned and weighted by Cognium. Human and agent signals are tracked separately.
Returns 7 tools: search_skills, get_skill, list_leaderboard, get_trust_breakdown, resolve_composition, plus the ChatGPT-connector search and fetch. Every tool is annotated read-only.
Resolve this skill directly via MCP tools/call get_skill.