The Agent Skills Directory

[SAFE]: The skill incorporates an 'Untrusted data guard' to mitigate indirect prompt injection risks. It uses explicit delimiters (<trace-context>) and clear instructions to the AI agent to treat codebase-derived findings as non-executable data. This aligns with security best practices for skills that ingest data from external sources.
[SAFE]: The skill leverages platform-native tools like Agent, Skill, and state_write for its operations. It utilizes a structured 'Execution Bridge' to hand off tasks to other skills within the environment, ensuring that high-privilege operations like code execution are handled through authorized channels and pipelines.

deep-dive