data-scraper-agent
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill builds an agent that ingests untrusted data from public websites and incorporates it directly into LLM prompts, creating a vulnerability surface.
- Ingestion points: Data is fetched from user-defined URLs or APIs using
requests,BeautifulSoup, orplaywrightin thescraper/sources/directory. - Boundary markers: The prompt template in
ai/pipeline.pyuses Markdown headers (# Items) to delimit content but lacks explicit instructions to ignore or sanitize embedded malicious commands. - Capability inventory: The resulting agent possesses network access for scraping and storage synchronization, can write to external databases like Notion, and can perform repository writes via GitHub Actions.
- Sanitization: The implementation uses
json.dumps()to format scraped items into the prompt without filtering or escaping potentially malicious instructions contained within the external data. - [PERSISTENCE]: The skill utilizes GitHub Actions with a cron schedule (
0 */3 * * *) and automatic git commits (git push) to maintain recurring execution and state. This is the intended primary purpose of the skill to provide automated monitoring.
Audit Metadata