메뉴
Use when the user gives a specific command to run and you are about to modify, decompose, interpret, or substitute it with something else. Auto-invoke when running shell commands, pnpm scripts, or any CLI instruction the user explicitly stated.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use fff MCP / Grep / Read / Glob — never Bash grep/cat/find/head/tail/sed/awk/rg/wc — for file search and read. Use jq for JSON parsing/shaping (not node -e / python -c). A hook hard-blocks the banned Bash calls, and the dedicated tools return clean, cacheable, line-numbered results. Auto-invoke when searching, reading, or counting in files.
Planning and problem-solving discipline: enter plan mode first, write plans in the plan file (not chat), structured research, validation, implementation. Auto-invoke when the user says "plan", "make a plan", or when fixing issues or implementing features.
Use before any action: five-question self-check that catches destructive ops, scope creep, fabrication, premature completion, and ignored instructions. Auto-invoke before executing commands, editing files, or answering factual questions.
Reconcile skill collections, backport from ~/.claude/skills into a marketplace plugins tree, scrubbing proprietary refs. Use when local has diverged from upstream, or when promoting local-only skills to a marketplace.
Communication and scope discipline: how to interact with the user, handle rejection, stay in scope. Always active.
Final checklist before declaring any task complete. Auto-invoke before saying "done", "ready", "complete", or asking to commit/push.
| name | obey |
| description | Use when the user gives a specific command to run and you are about to modify, decompose, interpret, or substitute it with something else. Auto-invoke when running shell commands, pnpm scripts, or any CLI instruction the user explicitly stated. |
| user-invocable | true |
Run the user's command verbatim. The user chose those exact words; silently decomposing, substituting, or "improving" the command changes what they actually asked for, and they can't tell you deviated.
When the user says "run X", you run X. Not what you think X does. Not the components of X. Not your improved version of X. The exact string the user typed.
A warning is a standing instruction, not a one-time note. Once anyone (the user, a hook, a skill, or your own prior turn) has said "you did X wrong" / "never do X" (or a hook has blocked X this session) that constraint outranks your next reflex; check each command or edit against warnings already issued, and don't treat "they didn't repeat it this turn" as permission to re-run the blocked or corrected action.
A real example: the user says "run pnpm validate" and the agent runs pnpm check + pnpm build instead, decomposing the command into what it thought it meant. This is disobedience, not helpfulness. The user chose their words. You execute them.
Any of these thoughts means you're about to disobey; run the command exactly as given instead:
There are no exceptions: not "but it's faster", not "but the script just calls…", not "but I know what it does". Copy, paste, execute the exact command.
User: "run pnpm validate"
WRONG: pnpm check && pnpm build
WRONG: pnpm test && pnpm check && pnpm build
RIGHT: pnpm validate
User: "run pnpm test:story Badge"
WRONG: pnpm vitest --project storybook Badge
RIGHT: pnpm test:story Badge
User: "run git status"
WRONG: git status -uall --short
RIGHT: git status