메뉴
Create, edit, review, evaluate, package, and optimize skills. Use when users ask to create a skill, revise skill instructions, review a skill, run skill evals, benchmark skill performance, package a skill, or optimize a skill description for trigger accuracy.
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
Review any artifact and output results in a Retrospective board format — what is well, what is bad, what should be improved, what to change. Use when the user says "review", "critique", "evaluate", "what's wrong with", "compare", or passes a skill, rule, doc, spec, code, diff, or pull request for assessment. Works on a single artifact, an artifact against a reference, a diff or PR, or two artifacts side by side.
Compare options and recommend a direction. Use for decision requests like "choose", "which option", "tradeoffs", "recommend", "should we", and option selection with criteria, risks, and reversibility.
Organize material into meaningful groups. Use for classification requests like "categorize", "group", "cluster", "sort", "taxonomy", "organize these", and grouping by criteria, priority, dependency, similarity, or abstraction level.
Explain any knowledge topic simply and accurately. Use for explanation requests like "explain X", "why/how/what is X?", concepts, science, definitions, code, design, architecture, and walkthroughs.
Investigate local repository, document, and attached-artifact context. Use for local investigation requests like "investigate", "find where", "understand this repo", "trace", and local-context research; do not use for web search.
Manage active work across people, agents, tasks, dependencies, blockers, and handoffs. Use for coordination requests like "manage this work", "lead this", "assign", "delegate", "track blockers", "status", "handoff", and multi-workstream execution.
| name | create-skill |
| description | Create, edit, review, evaluate, package, and optimize skills. Use when users ask to create a skill, revise skill instructions, review a skill, run skill evals, benchmark skill performance, package a skill, or optimize a skill description for trigger accuracy. |
| license | Apache-2.0 |
| tags | ["creator","skills","authoring"] |
| metadata | {"author":"Anthropic","version":"1.9.0","source":"github.com/anthropics/skills","catalog":"utility","category":"meta"} |
Create new skills, review and improve existing skills, evaluate outputs, optimize trigger descriptions, and package final skill folders.
Route the request. Read only the reference that matches the user's current task:
| User intent | Read |
|---|---|
| Create a new skill or revise skill instructions | references/authoring.md |
| Review a created or revised skill | references/review.md |
| Build eval cases, run iterations, benchmark outputs, or collect human feedback | references/evaluation.md |
| Optimize a skill description for trigger accuracy | references/description-optimization.md |
| Adapt the workflow for agents without subagents, Claude Code, generic CLIs, or Cowork | references/agent-compatibility.md |
| Validate eval YAML or grading, benchmark, and feedback JSON structures | references/schemas.md |
If the request spans multiple phases, read the references in workflow order: authoring, review, evaluation, description optimization, then agent compatibility only when platform details matter.
Clarify activation and behavior. Identify what the skill should do, which user phrases or contexts should trigger it, what output it should produce, and whether objective evals are useful.
Write or revise the skill. Name new skills using the <verb>-<subject>[-<variant>] convention or a concise <verb> format (e.g., code-tests, ask). Follow references/authoring.md for metadata, trigger descriptions, SKILL.md body format, reference file format, section delimiters, scan anchors, examples, helper scripts, portability, and validation. Always bump metadata.version using semantic versioning upon any material change to a skill's files.
Test behavior. Run this skill's scripts/validate.py against the target skill when available. For router skills, confirm routed eval cases use expect.routing.reference; for objectively testable skills, run skill-enabled outputs against a meaningful baseline.
Show evidence. Share validation output, eval results, benchmark summaries, and relevant diffs before making another revision.
Iterate deliberately. Continue until feedback is resolved or further changes stop improving behavior.
Package last. Package the final skill only after the user is satisfied with behavior and trigger accuracy.
scripts/, references/, assets/, or evals/ only when the skill actually uses them.scripts/run_eval.py, scripts/run_loop.py, and scripts/improve_description.pyscripts/validate.pyscripts/aggregate_benchmark.pyscripts/package_skill.pyeval-viewer/generate_review.pyagents/grader.md, agents/comparator.md, agents/analyzer.md, and agents/benchmark-analyzer.mdSKILL.md and reference edits follow references/authoring.md