The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted user-generated content from App Store reviews, which creates a potential surface for indirect prompt injection where malicious reviews could attempt to influence the agent's behavior.
Ingestion points: The script scripts/app_review_local.mjs fetches external JSON data from Apple's public iTunes RSS and lookup APIs.
Boundary markers: While the analysis prompt in agent-prompt.md provides structure for the output, it lacks explicit delimiters or specific instructions for the agent to disregard potential commands found within the ingested review text.
Capability inventory: The local script performs network reads from itunes.apple.com and writes analysis results to multiple local files including reviews.json and evidence.html.
Sanitization: The script includes an escapeHtml utility to sanitize content for the HTML dashboard and truncates long review content to 700 characters.

qiaomu-app-review