funding-signal-monitor
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill aggregates data from several untrusted external sources, including social media platforms and web search results, to identify funding announcements. This ingestion of external content creates an attack surface for indirect prompt injection, where malicious instructions could be embedded in the source data to influence the agent's analysis or reporting behavior.
- Ingestion points: Data is collected from the WebSearch tool and results from Twitter, Reddit, and Hacker News scrapers (SKILL.md and scripts/search_funding.py).
- Boundary markers: There are no specific instructions or delimiters provided to the agent to distinguish between the fetched content and instructions, increasing the risk of the model following instructions embedded in the external data.
- Capability inventory: The skill has access to web-search, apify-search, and standard network operations via the requests library. It also allows saving the resulting report to a user-specified file path.
- Sanitization: The helper script scripts/search_funding.py uses regular expressions to extract specific fields like funding stages and dollar amounts, which helps validate the structure of the data processed for those specific attributes.
- [EXTERNAL_DOWNLOADS]: The skill specifies a dependency on the 'requests' library, which is a standard and well-known Python package for HTTP communication.
- Evidence: The Setup section in SKILL.md explicitly requires the installation of the library using 'pip3 install requests'.
Audit Metadata