Self-hosted AI agent workspace: multi-provider chat, MCP, skills, Model Council, and your own PC as an agent backend.
Agent Studio exposes 5 tools with significant quality gaps. create_artifact and update_artifact have reasonable schemas and descriptions (194-250 chars), but three 'skill' tools (activate_skill, read_skill_resource, run_skill_script) lack visible input schemas entirely in the provided source. Tool names are verb-based but descriptions for skill tools are minimal (20-50 chars, below the 50-200 char baseline). No parameter descriptions visible for skill tools. Error handling and output schemas are not documented. The artifact tools show better structure but skill tools appear to be placeholder/reserved definitions without full implementation details.
Activate and run a skill tool (reserved skill feature tool)
Create a rendered artifact the user can view in an interactive panel beside the chat. Use when you generate substantial standalone content worth displaying interactively rather than as plain text: complete HTML pages/apps/demos (kind='html'), SVG graphics (kind='svg'), Mermaid diagrams (kind='mermaid', the mermaid source without fences), or self-contained code files (kind='code') such as scripts or small programs the user may want to copy. Content must be COMPLETE and self-contained. Returns JSON {ok, artifactId, version, kind}. To revise it later call update_artifact with the returned artifact_id.
Read a skill resource (reserved skill feature tool)
Run a skill script (reserved skill feature tool)
Replace the full contents of an existing artifact created earlier in this conversation via create_artifact. Pass artifact_id and the COMPLETE new content (not a diff). Creates a new version visible to the user immediately. Use iteratively while refining a demo/diagram/app.
Three skill tools (activate_skill, read_skill_resource, run_skill_script) have no visible input schemas in source code.
Skill tool descriptions are 15-50 characters, well below the 50-200 char baseline. 'Activate and run a skill tool' and 'Read a skill resource' lack context for when/why LLMs should call them.
No parameter descriptions visible for any skill tools. LLMs cannot infer what parameters control without explicit descriptions.
Artifact tools lack documented output schemas. LLMs need to know what fields (ok, artifactId, version, kind) to expect in responses for downstream planning.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | 2026-07-28+ | v2 |
No error handling guidance documented. If artifact creation fails (content too large, invalid kind), LLMs have no recovery path.