Single tool with minimal definition quality. The tool 'get-selfie-analysis' has a basic description in Korean, a simple input schema with one string parameter, but lacks critical elements: no output schema documentation, no error recovery guidance, no parameter constraints, and the implementation is a hardcoded stub. The tool is directly registered via server.tool() with Zod schema, so definition is visible, but quality is poor across all dimensions.
이미지URL을 받아 분석 결과를 반환합니다.
No output schema documented. Tool returns hardcoded string 'RESULT="피부좋음"' without explaining what fields the LLM should expect or how to chain results downstream.
Description is in Korean (non-English) and lacks LLM-optimized structure. It does not explain WHEN to use the tool, what it returns, or any prerequisites. Baseline is 50-200 chars with clear action/context/result; this is ~25 chars and opaque to English-speaking LLMs.
Input parameter 'imageUrl' has a description, but no validation constraints (URL format, max length, allowed schemes). Parameter description does not state format expectations or handle invalid inputs.
Error handling is generic and does not guide recovery. Catch block returns 'error' as a string, not a structured error classification. LLM cannot determine if error is retryable, user-fixable, or fatal.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 33 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 6 | - | v1 |
Implementation is a hardcoded stub ('const RESULT = "피부좋음"'). Actual image analysis is commented out. Tool cannot perform its claimed function, making definition semantically incorrect.
Tool depends on a local API at 'http://localhost:2580' which is not running (code commented out). Server will not function in any real deployment.
Unused dependencies and contradictory codebase. Package.json includes express and multer (HTTP server), but server runs as STDIO. Dockerfile and package.json suggest a different architecture than what is implemented.