菜单
Deep user interview for pivotal plan/design uncertainties. Auto-suggest when user judgment would materially change scope, approach, or success criteria. Scoped mid-task grilling for subtask-pivotal forks. NOT for micro-reversible questions or codebase-discoverable facts.
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
Authorized pentest planning with mandatory ROE scope gate. Synthesizes phase checklists and findings. NOT for static audit (security-scanner), CTF labs (ctf-*), or C2/webshell tooling.
Operate MCPHub groups, endpoints, compression, and CLI from repo registry. Use for hub preflight, group picking, tunnel vs local exposure. NOT harness sync.
Use for session, scoped, PR, range, full audit, simplification, and source/provenance reviews with evidence-first findings. NOT for feature implementation or benchmarking.
Deep multi-source research with reviewable plans, source-support auditing, and confidence scoring. Use for technical, academic, market, fact-checking investigation. NOT for code review or simple Q&A.
Extract clean article text and metadata from URLs or HTML with trafilatura CLI. Use for single-page extraction, piped/local HTML, bounded discovery. NOT for research synthesis (research), PDFs (docling), raw fetch (fetch), video (yt-dlp).
Plan cross-harness skills sync dry-run steps before any live install. Use when reconciling harness skill inventory. NOT for --apply or npx installs.
| name | grill-me |
| description | Deep user interview for pivotal plan/design uncertainties. Auto-suggest when user judgment would materially change scope, approach, or success criteria. Scoped mid-task grilling for subtask-pivotal forks. NOT for micro-reversible questions or codebase-discoverable facts. |
Repo overlay for Grok Build. Canonical routing policy lives in instructions/global.md Depth routing.
Interview relentlessly about every aspect of the scoped branch (or full plan when upfront) until we reach a shared understanding. Walk down each branch of the design tree within scope, resolving dependencies between decisions one-by-one. Do not re-grill settled plan areas listed in resolved context.
For each question, provide your recommended answer. Ask the questions one at a time.
If a question can be answered by exploring the codebase, explore the codebase instead.
Mid-task triggers: subagent returns blocked-user-pivotal; orchestrator re-enters Uncertainty Gate; mid-wave fork classified as subtask-pivotal.
