tinyworld-auto-batching
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process game board data (8x8 cell snapshots) as input for model inference. This creates a potential surface where malformed board state data could influence the model's output.
- Ingestion points: The skill uses
snapshotCells()to capture current board state as JSON for model context. - Boundary markers: No explicit prompt delimiters or boundary markers are defined to isolate the board data from instructions.
- Capability inventory: The skill interacts with browser local storage (
AI_LS) and suggests using browser evaluation for development validation. - Sanitization: The skill includes a requirement to use
normalizeAutoActionto validate model output before it is applied to the game state.
Audit Metadata