The Agent Skills Directory

[SAFE]: The skill implements a defensive execution strategy that includes risk classification, phased planning, and explicit user approval steps before performing file modifications.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-defined tasks and instructions, creating an interface where untrusted input could attempt to influence the agent's behavior. * Ingestion points: User task descriptions provided at runtime in SKILL.md logic. * Boundary markers: None defined to isolate user instructions from the planning instructions. * Capability inventory: The allowed-tools list includes Bash, Read, Grep, and Glob for environment interaction. * Sanitization: No specific filtering or escaping is applied to the incoming user requests.

safe-task