/safety-evaluation
Design and run a safety evaluation suite for an AI model or feature across harm categories — refusals on disallowed content, robustness, over-refusal vs helpfulness, groundedness/truthfulness — with rubrics and pass/fail thresholds. Use to measure an AI system's safety,
$ npx -y skills add jassics/awesome-claude-security --skill safety-evaluation --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
- Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.
- You can call itInvoke it directly when you want it.
- Slash command
/safety-evaluation
Context preview
The summary Claude sees to decide when to auto-load this skill.
Design and run a safety evaluation suite for an AI model or feature across harm categories — refusals on disallowed content, robustness, over-refusal vs helpfulness, groundedness/truthfulness — with rubrics and pass/fail thresholds. Use to measure an AI system's safety,
A Claude Code plugin marketplace for the full cybersecurity & GenAI-security lifecycle — from recon and threat modeling to detection engineering, GRC, and CISO-level strategy. A pentester knows which OWASP test bends a broken-access-control endpoint.
Repo: jassics/awesome-claude-security
Other skills on awesome-claude-security.
- /a2a-security-review
Review agent-to-agent (A2A) / multi-agent-system trust: peer identity and authentication, message integrity, capability-negotiation trust, and delegation-chain privilege narrowing. Use when an orchestrator dispatches to sub-agents or peer agents (same or different trust domain)
Open skill - /agent-harness-review
Test the agent execution harness/runtime itself — LangChain/LangGraph, AutoGen, CrewAI, custom ReAct-style loops, or computer-use/browser-use agents — for intermediate-state poisoning, unscoped action spaces, and missing resource limits. Use when the agent isn't built on Claude
Open skill - /agent-security-review
Assess an autonomous / tool-using AI agent for security end-to-end: tool privileges, autonomy and approval boundaries, excessive agency, memory/state poisoning, and multi-agent trust. Use when reviewing the security of an AI agent or agentic workflow.
Open skill - /autonomy-boundary-test
Test what an AI agent will actually do without human confirmation, including under injected-goal / prompt-injection scenarios, to validate its autonomy and approval boundaries. Use on an authorized agent to confirm excessive-agency controls hold in practice.
Open skill - /mcp-security-review
Review the security of MCP (Model Context Protocol) servers/clients an agent uses: server trust tier, tool/resource description and result poisoning, confused-deputy risk, consent/scope UX, and supply-chain pinning. Use when an agent or assistant integrates one or more MCP
Open skill - /tool-permission-audit
Inventory the tools/functions an AI agent can call and audit their privileges, side effects, and approval requirements to find excessive-agency and least-privilege gaps. Use when reviewing an agent's tool/function surface.
Open skill