worldview-game-death-loop-persistent-clues
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to generate game logic and implementation code.
- Ingestion points: The skill accepts a creative brief via the
/worldview-game-death-loop-persistent-cluescommand and reads existing project source code as specified inSKILL.md. - Boundary markers: While the skill establishes a conceptual boundary for the input, it lacks explicit technical delimiters or sanitization for the processed project data.
- Capability inventory: The agent can write implementation code and verification logic to the file system, and execute it within the project's authorized runtime environment.
- Sanitization: No explicit instructions for the agent to sanitize or escape user-provided narrative text before using it to generate game logic or state ledgers are described.
- [SAFE]: The skill mandates a high standard of data safety, explicitly prohibiting behaviors common in horror games that could be misinterpreted as malicious, such as faking file corruption, simulating system errors, or deleting user progress.
- [SAFE]: The methodology uses a structured ledger to isolate sensitive system and profile data from resettable game state, preventing unauthorized modifications to platform-level configuration.
Audit Metadata