The Agent Skills Directory

[PROMPT_INJECTION]: The skill mentions the phrase "Ignore previous instructions" in skill.yaml. Upon analysis, this is a false positive. The text is contained within the capabilities section to describe that the tool performs "Pattern matching for common injection vectors". It is an informative description of the tool's functionality, not an attempt to bypass agent safety filters or override behavior.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing external files (skill_path). While this is an ingestion point for potentially untrusted content, the current definition is limited to metadata and documentation. The capability description indicates it performs heuristic and semantic analysis, which is a standard security auditing function.

prompt-injection-scanner