The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-generated content from app reviews and testing feedback.
Ingestion points: Untrusted review text is ingested via the get_reviews tool in SKILL.md.
Boundary markers: There are no explicit boundary markers or instructions to the agent to treat review content as untrusted data during analysis.
Capability inventory: The skill has file-writing capabilities to generate findings.json and report.html as specified in the Emit section of SKILL.md.
Sanitization: The skill implements safety measures by requiring report.html to be self-contained with no external network requests, and the use of a JSON schema for findings.json ensures structured output.

feedback-analysis