Skill studio
Capabilities the mesh offers today, plus the review-gated skill registry: candidates are hashed, scanned, and sandbox-checked, and nothing installs without explicit human approval.
Loading…
Demo: review-gated lifecycle
Each step is explicit and confirmed. The registry forces the candidate to start as "submitted"; approval and installation are separate human actions. Skills cannot declare their own review status, capabilities outside the deployment allowlist, or permissions.
Run a research workflow
Three agents collaborate: the research agent gathers fixture documents, the verification agent checks them, and the news agent summarizes. All data is deterministic fixtures; the run is recorded and traceable.
Loading…
Note: SKILL.md content is untrusted instructions — hashed and scanned, never executed. Sandbox checks run in an isolated vm with no host access and a hard timeout.