refining-game-prototypes
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from external sources, specifically player replays, flagged moments, notes, and proposals as described in
references/play-input.md. This content is used to generate findings and design decisions. - Ingestion points: Player-provided feedback, including notes and proposals processed according to the procedures in
references/play-input.md. - Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are defined to help the agent distinguish between data and instructions within the player input.
- Capability inventory: The skill instructs the agent to autonomously modify project files ('Change one thing' in
SKILL.md) and execute measurement procedures or simulators to verify changes. - Sanitization: There is no evidence of sanitization or validation of the untrusted data before it is integrated into the prompt for decision-making.
Audit Metadata