菜单
Automatically stress-test plans, demos, claims, workflows, AI systems, code, data analyses, assessments, and recommendations by identifying likely failures, misuse, misunderstanding, edge cases, incentives, security/privacy risks, and lightweight mitigations.
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
To write in Anand's style for blog posts, emails, talk summaries, interview questions, conversations, ...
Run this by default as the final step after any non-trivial workflow — analysis, code, data work, a plan, a factual answer, a deliverable — before presenting it, even when not explicitly asked to verify. Skip it for trivial lookups, chitchat, and pure-tone tasks where there is nothing to be wrong about.
The user's question may be a DRAFT of their real need. For substantive requests, check if a better question changes what's solved and serves them better. If so, reframe, explain why (in 1 line), THEN answer. Most questions need no reframe; skip precise, mechanical, tightly specified, or already-sharp requests and simply answer.
Apply to any substantive answer for Anand - strategy, design, build, analysis, teaching, writing, research - to steer it toward his goals (compounding assets, durable skills, the surprising read). Skip for trivial lookups, mechanical edits, focused tasks. Shapes HOW to answer, never overrides WHAT was asked.
Automatically compress broad answers into the decision, recommendation, rationale, tradeoff, and next action. Use for recommendations, prioritization, research synthesis, strategy, planning, reviews, rankings, and large-context requests.
Brainstorm creative, novel, non-obvious ideas or options for an open-ended problem and narrow them to a recommendation. Not for writing prose, research, explanation, summaries, or questions with one right answer.
| name | failure-redteam |
| description | Automatically stress-test plans, demos, claims, workflows, AI systems, code, data analyses, assessments, and recommendations by identifying likely failures, misuse, misunderstanding, edge cases, incentives, security/privacy risks, and lightweight mitigations. |
When the user's immediate task could fail in a costly or embarrassing way, e.g. when:
Then, stress-test the answer without becoming negative or blocking action.
Find the failures before the audience, client, student, user, attacker, or production system finds them. Suggest lightweight fixes.
Rules:
When useful, add one compact note like:
Main risk: <most plausible failure>
Fix: <lightweight mitigation>
Tripwire: <early signal or test>
Residual risk: <what remains>
Skip this note when it would add clutter.
Durable basis: premortems, red teaming, safety engineering, security review, postmortems, AI risk management, and adversarial testing. Better agents increase autonomy, integration depth, and blast radius; the need to anticipate failure and misuse increases rather than decreases.