Candidate skills are proposed from evidence, never invented from a prompt alone—the same rule that applies to facts and concepts.
Build agents
that can only cite approved facts.
A guided authoring interface for building new skills is in development. The pipeline underneath it is already real: candidate skills are mined from evidence, compiled and gated behind the same grounding-verification requirement as everything else on this site—no skill reaches a published state without cited, hash-verified evidence and a named reviewer's decision.
REAL TODAYThe pipeline exists.
The pipeline exists.
The builder UI is coming.
A skill compiles to a risk-classified action plan before it can be invoked. That compilation step is real and callable today via dry_run_skill.
A guided, visual authoring interface for building a new skill—today this happens through the review and curation pipeline, not a dedicated builder screen.
In progress.
Not shipped today.
There is no standalone agent-builder UI in production yet. The compile-and-verify pipeline it will sit on top of is already real.
See the MCP tool surface →