research
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- Indirect Prompt Injection Surface: The skill ingests data from public platforms where content is controlled by external parties.
- Ingestion points: Public sources such as Reddit, X/Twitter, Hacker News, Stack Overflow, and YouTube comments (SKILL.md).
- Boundary markers: The skill lacks explicit delimiters or instructions to treat ingested research data as untrusted content separate from system instructions.
- Capability inventory: The skill possesses capabilities to read from internal data sources and execute local scripts.
- Sanitization: There are no defined filters or sanitization steps for the data retrieved from external URLs before processing.
- Dynamic Execution Pattern: The skill instructions specify running a 'preflight script' from the user context when shell access is available.
- This pattern is used for environment initialization but requires that the local script files are maintained in a secure state.
- Internal Source Access: The workflow includes searching internal tools like Slack and Notion. This behavior is consistent with the skill's UX research purpose and relies on the platform's existing connector permissions.
Audit Metadata