worldview-game-scarce-inventory-triage
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (item definitions, UI layouts, and route graphs) to derive game mechanics, creating a vulnerability surface for indirect prompt injection.\n
- Ingestion points: Specified in SKILL.md as project paths, asset definitions, and route graphs.\n
- Boundary markers: The skill instructions emphasize separating project facts from proposed decisions, though it lacks explicit technical delimiters for isolation.\n
- Capability inventory: The skill performs file system read/write operations and source code modification when a runtime is available.\n
- Sanitization: Employs a robust transaction-based 'Ownership and Recovery Lock' model to ensure data integrity during state transitions.\n- [DYNAMIC_EXECUTION]: The skill generates and implements source code in a project's existing runtime based on authored mechanic contracts.\n
- Evidence: SKILL.md specifies delivering a 'working implementation in the project's existing runtime' and 'playable entry point'.\n
- Risk Mitigation: Instructions explicitly state to work within harness boundaries and prefer local project code and tools.
Audit Metadata