The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from external websites which presents a surface for indirect prompt injection. Malicious instructions embedded in audited pages could attempt to influence the AI agent when it parses the resulting audit report. * Ingestion points: External URLs and website content crawled via the squirrel audit command (referenced in SKILL.md). * Boundary markers: Audit reports are generated in a structured XML-based 'llm' format to establish data boundaries (documented in references/OUTPUT-FORMAT.md). * Capability inventory: The agent is authorized to use Bash, Read, Edit, Grep, and Glob tools to implement codebase fixes based on audit findings (specified in SKILL.md). * Sanitization: The output format applies XML character escaping to content to prevent structural injection into the report (noted in references/OUTPUT-FORMAT.md).

audit-website