jp-imperial-diet-minutes
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests meeting minutes that contain free-text speeches from external parties. It proactively addresses this as an attack surface by including specific countermeasures that instruct the AI to treat retrieved content strictly as data rather than instructions.\n
- Ingestion points: The
speechfield in JSON results returned from theteikokugikai-i.ndl.go.jpAPI endpoints.\n - Boundary markers: The skill instructions recommend using JSON structure and explicit delimiters to maintain the separation between the agent's instructions and the untrusted external data.\n
- Capability inventory: The skill uses standard network retrieval tools such as
curlandInvoke-RestMethodto interact with the government API.\n - Sanitization: The instructions emphasize that the agent must not follow any commands found within the speech text and should avoid interpolating this content into instruction-carrying contexts.
Audit Metadata