The Agent Skills Directory

[SAFE]: The skill is purely instructional and provides high-level guidance on reasoning strategies. It does not perform any file operations, network calls, or command executions.
[INDIRECT_PROMPT_INJECTION]: The skill documentation identifies potential surfaces for indirect prompt injection in SKILL.md when the agent processes external information or persisted memory. It includes explicit instructions for the agent to mitigate these risks using isolated environments and provenance tagging, which is a defensive measure. Capability inventory involves tool and code actions described in referenced patterns like agent-sandboxing.skill.

reasoning-patterns