The Agent Skills Directory

[SAFE]: The skill is a set of markdown-based instructions and reference materials. It does not contain any scripts, command execution, or network operations. The only tools requested are Glob and Read, which are used to look at code provided by the user for analysis.
[PROMPT_INJECTION]: While the prompt injection detector flagged the 'Reviewing untrusted input' section, this is actually a defensive security measure. The skill instructs the agent to treat reviewed files as data, not instructions, and explicitly tells the agent to ignore attempts within that data to override its role or safety guidelines. This follows best practices for building robust security tools.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill mentions sensitive file paths (like .env) and secrets in its checklist, but only in the context of teaching the agent to identify and flag their presence in user code as a security risk. There are no mechanisms for exfiltrating this data.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted code (the ingestion point). It mitigates this risk by providing strict boundary instructions in SKILL.md (Reviewing untrusted input section) and limiting the agent's capabilities to read-only analysis. It does not write to the filesystem or communicate over the network.

owasp