jx-hxxx
UserThe vibe-coding seatbelt for Claude Code — catches AI coding mistakes · 바이브코딩 안전벨트, AI 코딩 실수를 막는 플러그인
Categories
Indexed Skills (6)
docs-keeper
Maintains the four living documents of a hi-vibe project: CLAUDE.md (lean project context — constraints, pitfalls, rationale), per-folder MODULE.md (detailed design), handover.md (session-to-session context log), CHANGELOG.md (substantive change history, Keep a Changelog). Use for /hi-vibe:init, :handover, :log, :recall, :welcome, and whenever the user asks to 문서 정리, 인수인계 기록, 체인지���그 작성, CLAUDE.md 갱신, "이 폴더 문서 만들어줘", or after a structural change that must be reflected in docs. Also use (mode: recall) when the user asks about PAST decisions or work — "예전에 왜 이렇게 했지?", "저번에 뭐까지 했더라", "이거 언제 바꿨지", "왜 이 방식으로 결정했었지", why did we choose X — search the handover/archive/ CHANGELOG records instead of guessing from memory.
grounded-answers
Grounding discipline for decisions and factual claims. Use when facing an ambiguous decision (여러 해석이 가능한 요청, 요구사항 빠짐, 되돌리기 어려운 작업), or when about to state numbers, limits, prices, API behavior, library/framework usage, version facts, or how an external platform/service currently behaves or why — 얼마야, 제한이 몇이야, 지원돼?, 왜 이렇게 동작해?, 정책 바뀌었어?, 스펙, 공식 문서, pricing, rate limit. Check context7 / official docs first; never answer library, API, or platform-behavior facts from memory (training data goes stale). Applies in casual chat and troubleshooting too, not only coding tasks. Prevents silent unilateral decisions and fabricated authoritative figures.
guards-setup
Machine-enforced quality guards installer. Use for /hi-vibe:gate or when the user asks to 린트 설정, 타입 체크 강제, 순환의존 검사, CI 세팅, eslint/ruff/mypy 설정, complexity limit. Detects project language(s) and installs lint/type/cycle guards and optional CI — always asking before writing, always merging with existing configs.
repo-xray
Evidence-based repository structure analysis for Python and JavaScript/TypeScript (js/jsx/ts/tsx) projects. Runs a bundled scanner script and answers only from its JSON output — never from guessing. Use this skill whenever the user asks about duplicate code, unused/dead code, "does function X already exist?", cleaning up or refactoring the codebase, oversized files, or repo structure health — including Korean phrasings like "중복 코드 찾아줘", "안 쓰는 코드/죽은 코드 정리", "이런 함수 이미 있어?", "구조 분석해줘", "정리할 거 알려줘", "리팩토링 뭐부터 해?". Also use it before writing a new helper function when there is a real chance an equivalent one already exists in the repo.
write-gate
Pre-write and review gates for code changes. Pre-write: before creating any new function, helper, type, or file — 만들어줘, 추가해줘, 구현해줘, 새 파일, 리팩토링, new helper/component. Post-write: after finishing a change — 다 했어, 리뷰해줘, 검토, review my change. Prevents duplicate reimplementation and enforces the review checklist including run-verification (scaled to change size/runtime surface — no full verification for small mechanical changes) and doc sync.
root-cause-first
Root-cause discipline for debugging and error handling. Use whenever fixing a bug or error, writing try/except/catch, adding a fallback or default value, or the user says 에러 나, 버그, 고쳐줘, 왜 안 되지, fallback, 방어코드, "일단 돌아가게", error, crash, exception. Prevents symptom-masking fixes that silently swallow failures.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.