The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process untrusted data (other agent skills), which creates an inherent surface for indirect prompt injection. Ingestion points: Target skill directories, SKILL.md files, and referenced documentation/scripts. Boundary markers: No explicit delimiters or ignore-instructions are defined for the agent when reading target content. Capability inventory: File system reads, directory structure validation, and line count commands. Sanitization: No content validation or sanitization instructions are provided.
[COMMAND_EXECUTION]: The instructions direct the agent to use the wc -l utility to verify line counts of the target skill files during the review process.
[SAFE]: The skill acts as a quality assurance guide and workflow. It does not exhibit network exfiltration, credential access, or persistence mechanisms. Potentially harmful code examples like 'rm -rf /' are included only for illustrative purposes in the best-practices reference documentation.

skill-review