web-reader
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process data from arbitrary external URLs, creating a surface for indirect prompt injection.
- Ingestion points: External web content is retrieved via the
read_urltool and provided to the agent in Markdown format. - Capability inventory: The skill documentation mentions the ability to use
bash + requestsas a fallback, and agents typically possess file system and shell execution capabilities. - Boundary markers: The instructions do not define clear delimiters or specific instructions for the agent to ignore any commands found within the retrieved web content.
- Sanitization: No sanitization or filtering of the retrieved content is performed to remove potential malicious instructions before processing by the AI.
- [DATA_EXFILTRATION]: The skill transmits the full target URL, including potentially sensitive query strings or parameters, to the third-party service
r.jina.ai(Jina Reader). - The documentation includes a warning advising users not to provide URLs containing credentials, tokens, or private addresses because they will be sent to the external service.
Audit Metadata