geo-brand-mentions
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from public platforms such as Reddit, YouTube, and Quora to perform sentiment analysis and calculate authority scores. This content is processed by the agent to generate reports, creating a surface for indirect prompt injection if the external data contains instructions intended to manipulate the agent.
- Ingestion points: Social media and community site content retrieved via WebFetch.
- Boundary markers: The instructions do not specify any delimiters or warnings to treat the retrieved content as untrusted data.
- Capability inventory: The agent has access to Bash, WebFetch, and Write tools.
- Sanitization: No sanitization or filtering of the platform data is described before it is used to generate the final report.
- [DYNAMIC_EXECUTION]: The skill provides a procedure to generate and execute a Python script using the Bash tool to interact with Wikipedia APIs.
- Evidence: A script template in SKILL.md interpolates a [Brand_Name] variable into the code. While this is a standard method for API interaction within this skill's context, it involves dynamic generation and execution of code.
- [SAFE]: The skill fetches data from well-known and authoritative sources. It performs API queries to Wikipedia and Wikidata to verify brand entities. These external references are restricted to trusted domains and are necessary for the skill's primary functionality.
Audit Metadata