菜单
Use when a request is underspecified and the user's actual goal, constraints or success criteria must be elicited before implementation.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Governs AI models through inventory, purpose, risk tier, approval, monitoring, drift, kill-switch and retirement. Use when introducing, changing or operating models in production or agent workflows.
Applies an Asimov-inspired hierarchy of AI safety laws when material AI or instruction risk is present. Use when harm, unlawful requests, authority conflicts, or safety trade-offs need an explicit law hierarchy. Also known as apply_laws_of_AI.
Use when implementing input validation, output filtering, tool restrictions, policy enforcement, secure coding controls, destructive-tool guardrails, approval gates, or safe agent behaviour.
Use when starting a session or deciding which local skill applies; acts as the meta-skill for discovering and loading the smallest relevant skill.
Use when measuring coding-agent quality, RAG evaluation, regression risk, latency, cost, reliability, safety, or production performance for agent systems.
Designs ontology-driven knowledge graphs for explainable KG-RAG. Use when modelling semantic graphs, ontologies, or graph schemas for retrieval.
| name | requirements-elicitation |
| description | Use when a request is underspecified and the user's actual goal, constraints or success criteria must be elicited before implementation. |
| aliases | ["interview-me","requirement-discovery","elicit-requirements","elicit-requirement"] |
Use this skill when the request is ambiguous, has multiple plausible interpretations, or is missing user, scope, constraint or acceptance details.
Extract enough intent to make the next implementation decision complete and low risk.
If the user says make the agent safer, ask which risk matters most before planning: tool permissions, prompt injection, data leakage, approval gates or audit evidence. Offer concrete options and recommend the lowest-risk default when enough context exists. Stop interviewing once the next action and acceptance evidence are clear.
