The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to search and load local files into the agent's context, which creates an indirect prompt injection surface. Ingestion occurs from metadata and file contents within the skills repository (defaulting to ~/.agents). While integrity verification via digests is mandated, there is no mention of sanitizing the instructions within those files for malicious patterns beyond manual applicability assessment.
[COMMAND_EXECUTION]: The skill relies on invoking local command-line tools including qmd, cm, and lev-skills. A provided Python script (skill-discovery.bak) uses subprocess.run to pass user-supplied task descriptions to these tools. While the use of list-based arguments mitigates shell injection, this tool-calling capability is a significant function of the skill.

skill-discovery