Public repo and local stdio MCP workflow for turning UI briefs into React + shadcn delivery with proof, review, and acceptance before ship.
Critical failures across all dimensions. Tool definitions are inferred from a registry file (tool-registry.ts) with no visible schema definitions, parameter types, or descriptions in the provided source. The 18 tools lack explicit input/output schemas, parameter documentation, and error handling guidance. Descriptions are present but generic (10-50 chars), far below the 50-200 char LLM-optimized baseline. No evidence of enum constraints, validation rules, or recovery guidance. STDIO transport caps protocol readiness at 50, but definition quality is the primary blocker.
Apply file changes to the workspace
Build acceptance test pack
Build review bundle for code review
Convert React components to shadcn/ui compatible format
Detect shadcn/ui installation paths and configuration
Embed content into components
Generate UI components from brief descriptions
No input schemas visible for any tool. Tool definitions are inferred from registry file with no explicit JSON Schema, parameter types, or constraints.
Descriptions are generic and under 50 characters (e.g., 'Detect shadcn/ui installation paths and configuration' = 54 chars, but most are 25-30 chars). LLM-optimized descriptions should be 50-200 chars with WHAT, WHEN, and prerequisites. Current descriptions lack context for tool selection.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 35 | 2026-07-28+ | v2 |
List available AI models for generation
Create a complete React page with shadcn components
Run Next.js smoke tests
Plan changes to the codebase
Run quality gates and checks
Refine generated UI components based on feedback
Generate repository workflow summary
Review UI/UX quality and accessibility
Scan workspace for project structure and dependencies
Ship a complete feature flow
Ship a React page to production
No parameter descriptions visible. LLMs cannot infer parameter meaning from names alone. Every parameter needs a description explaining what it controls, expected format, and constraints (e.g., 'The project key (2-10 uppercase letters)').
Destructive/irreversible tools (shadcn_brief_ship_react_page, shadcn_brief_ship_feature_flow, shadcn_brief_apply_files) lack confirmation or dry-run support. Agents make mistakes, these tools should support a confirmation step or return 'input_required' for user approval before executing.
No error handling guidance visible. Tools must tell LLMs what to do next on failure: 'User not found. Try search_users() with a partial name.' Raw errors or stack traces give agents nothing to act on.