icp-website-review
Pass
Audited by Gen Agent Trust Hub on May 5, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from external websites.
- Ingestion points: Website content fetched via
WebFetchand search results fromWebSearchas described inSKILL.md(Step 2: Crawl Target Pages). - Boundary markers: The instructions do not explicitly provide delimiters or warnings to the agent to ignore any natural language instructions found within the fetched website content.
- Capability inventory: The skill has the ability to read local files (
personas.json), search the web, fetch web pages, and write results to the local filesystem (clients/<client-name>/icp-reviews/). - Sanitization: There is no explicit sanitization or filtering of the fetched content before it is used to prompt the persona reactions.
Audit Metadata