Claude 설정관리 monitor — MCP App (카드 + 파일 diff history) + 로컬 watcher 하이브리드
24 tools with significant quality gaps. Tool names follow verb_noun convention (get_config, set_prefs, config_perm_add) but descriptions are inconsistent: many are terse Korean-language strings (e.g., 'Claude 설정 introspection') that lack English clarity and actionable context. Input schemas are present but sparse, most parameters lack type constraints (enums, ranges, patterns). Output schemas are undocumented. Error handling is minimal; tools shell out to Python scripts but responses are JSON-wrapped without recovery guidance. No pagination support despite tools returning lists (get_tracked, get_file_history). Parameter descriptions are often single-phrase Korean text without format/constraint details. Tool composition is reasonable (separate read/write tools) but parameter naming inconsistencies (e.g., 'project' parameter is overloaded, accepts folder name OR path, documented only in description, not as separate params). Security: no visible input validation, secrets not explicitly injected, no audit logging in tool definitions.
에이전트 추가
에이전트 제거 (.trash 로 이동)
훅 추가 (event + command)
훅 제거 (event + command substring)
항목 추가 (rule/output-style/workflow)
항목 제거 (.trash 로 이동)
출력 스타일 선택
권한 규칙 추가 (allow/deny/ask)
Descriptions are terse, Korean-language, and lack actionable context. E.g., 'Claude 설정 introspection (정규화 sections JSON)' does not explain WHEN to call this tool, WHAT it returns, or HOW to use the output. LLMs cannot infer intent from non-English, single-phrase descriptions.
Parameter 'project' is overloaded: accepts folder name OR path, but no separate parameters (project_name vs project_path). LLMs cannot disambiguate without explicit enum or separate params. Affects 11 config_* tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | 2026-07-28+ | v2 |
권한 규칙 제거
플러그인 활성화/비활성화
파일을 이전 리비전으로 복원
스킬 제거 (.trash 로 이동)
Claude 설정 introspection (정규화 sections JSON)
두 리비전 간 unified diff
파일 리비전 타임라인 (스냅샷 메타데이터)
UI 설정 (섹션 펼침/접힘, 보기 상태)
추적 중인 파일 목록 (phase: boot/refresh/poll)
대시보드를 브라우저에서 열기
UI 설정 저장 (섹션 펼침/접힘, 보기 상태)
UI 전용 도구 - 대시보드 표시
스킬 스캐폴드 생성
파일 watcher 시작
파일 watcher 상태 조회
파일 watcher 중지
No output schemas documented. Tools return JSON but LLMs cannot plan downstream calls without knowing response structure. E.g., get_config returns 'normalized sections JSON' but field names, types, and nesting are undocumented.
No input validation or error recovery guidance in tool definitions. Python scripts may fail (timeout, crash, bad JSON) but tool descriptions do not explain what to do on failure or how to retry.
No pagination support for list-returning tools (get_tracked, get_file_history). Large result sets will blow context window. Missing limit, offset, and total_count fields.