The Agent Skills Directory

[SAFE]: The skill does not contain any malicious code, obfuscated content, or unauthorized access patterns. It is a set of instructions and reference materials for improving agent-system design.
[PROMPT_INJECTION]: The skill explicitly addresses prompt injection by instructing agents to treat tool metadata and documentation as untrusted input, providing strategies for validation and pinning to prevent attacks.
[DATA_EXFILTRATION]: The guidelines include robust heuristics for preventing data exfiltration, such as implementing credential brokers, token exchange mechanisms, and ensuring that at least one part of the lethal trifecta (private data, untrusted content, and exfiltration path) is always broken.
[COMMAND_EXECUTION]: A shell script is provided for local static analysis during development. It uses standard path resolution to call a local Python script and does not execute remote code or perform unauthorized system modifications.

ax