Interfaces and permissions
Choose the supported UI, agent or MCP action without inventing API routes
Platform actions require a Semogram account with access to the skill's workspace. Workspace membership and a selectable project provide the relevant scope. Plugin ownership and read-only state still constrain editing.
Interface availability
| Action | Platform UI / local creation agent | Public HTTP /api/v1 | MCP |
|---|---|---|---|
| List/open packages | Installed Skills; inspect files | Not available | No general skill listing tool |
| Create | Manual, upload, catalog draft or approved agent artifact | Not available | No public create tool |
| Edit/save | Standalone package editor | Not available | No public update tool |
| History/compare/activate | Standalone version controls | Not available | No public version tools |
| Delete standalone | Governed Delete skill action | Not available | skill_delete |
| Change plugin skill | Parent plugin lifecycle | No skill CRUD route | Use supported parent plugin actions |
The platform's session-authenticated /api/skills family is an internal UI interface. It is not the public API-key interface. The creation agent's prepare/draft/revise/validate/create tools belong to that platform profile; they are not automatically exposed on the workspace MCP server.
MCP deletion example
This is destructive. Confirm that the selected standalone skill should be removed, check its consumers, and use an authenticated initialized workspace MCP connection. Replace both UUID placeholders with actual accessible IDs; a project-bound MCP endpoint supplies its project context itself.
{"id":"<STANDALONE_SKILL_UUID>","projectId":"<PROJECT_UUID>","confirm":true,"idempotencyKey":"retire-equipment-review-001"}Call the skill_delete tool with these arguments, not an HTTP /api/v1/skills URL. The key needs org:manage and an accountable permitted actor; plugin skills return a read-only conflict and must be managed through the parent plugin. Dependencies can block deletion.
Inspect the returned payload for deleted true and the actual skill ID. An assistant's summary alone is not confirmation of removal. The tool reference includes the discovered schema and response envelope.
Agent approval
A platform creation artifact is reviewable before saving. Its final creation action requires explicit approval of a validated revision and rechecks the package/name. Approving skill creation does not authorize a pipeline run, external write or prediction outcome update.