synthetic-scenarios
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to ingest and analyze external data sources, including an agent specification document and a database or corpus, which introduces a potential vector for indirect prompt injection. \n
- Ingestion points: The coding agent is directed to read an external specification file and query a SQLite database or similar corpus to generate scenario data (SKILL.md). \n
- Boundary markers: The procedure lacks instructions for using delimiters or boundary markers to isolate processed data from the agent's core instructions. \n
- Capability inventory: The agent possesses capabilities to execute SQL queries and eligibility functions against the provided data source (SKILL.md, Step 5). \n
- Sanitization: There are no explicit requirements or steps provided for sanitizing or validating content retrieved from the external sources before it is incorporated into the scenario generation pipeline. \n- [EXTERNAL_DOWNLOADS]: The README.md file contains a link to the author's GitHub repository (github.com/ai-evals-course/evals-course-revamp) for accessing course context and materials, which is a reference to a vendor-controlled resource.
Audit Metadata