菜单
AI safety review: alignment, misuse prevention, bias, fairness, and responsible AI. Invoke for: "AI safety review", "bias check", "fairness audit", "alignment", "responsible AI", "misuse prevention", "AI ethics", "safety guardrails", "could this AI system be misused".
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
Accessibility audit against WCAG 2.1 AA standards. Invoke for: "accessibility", "a11y", "WCAG", "screen reader", "aria labels", "color contrast", "keyboard navigation", "accessibility audit", "ADA compliance".
Write Architecture Decision Records (ADRs) to document significant technical decisions. Invoke for: "write ADR", "document this decision", "architecture decision record", "why did we choose X", "decision log", "record this choice", "ADR for".
Design and implement multi-agent orchestration systems. Invoke for: "multi-agent system", "orchestrate agents", "agent pipeline", "agent workflow", "how should agents communicate", "agent coordination", "task delegation to agents", "build an agent system".
LLM and AI agent security: prompt injection, jailbreaks, agent defense, guardrails. Invoke for: "prompt injection", "LLM security", "agent security", "jailbreak defense", "AI safety audit", "system prompt leakage", "adversarial inputs", "AI pipeline security", "tool call validation", "LLM guardrails", "model security", "is my prompt safe".
Improve algorithmic efficiency — find better algorithms and data structures. Invoke for: "algorithm improvement", "better algorithm", "O(n²) is too slow", "data structure choice", "optimize this loop", "algorithmic complexity", "time complexity".
Design clean, consistent REST or GraphQL APIs with OpenAPI specs. Invoke for: "design this API", "API spec", "OpenAPI", "REST endpoints", "API review", "API versioning", "endpoint design", "HTTP API", "what endpoints do I need".
| name | ai-safety |
| description | AI safety review: alignment, misuse prevention, bias, fairness, and responsible AI. Invoke for: "AI safety review", "bias check", "fairness audit", "alignment", "responsible AI", "misuse prevention", "AI ethics", "safety guardrails", "could this AI system be misused". |
| argument-hint | AI system or model to audit for safety |
| allowed-tools | Read, Grep, Glob, WebSearch |
Category: AI/ML Research
Audit AI systems for safety issues: misuse potential, bias, fairness, alignment with intended use, and responsible deployment.
## AI Safety Audit — <system> — <date>
### Intended Use
### Misuse Risks (ranked)
### Bias Assessment
### Guardrails: ✅/⚠️
### Human Oversight: ✅/⚠️
### Recommendations
/ai-safety audit the code-review agent — check for bias and misuse potential