novel-to-game

Warn

Audited by Gen Agent Trust Hub on Oct 7, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill defines a build pipeline (game-build) that creates "whitebox candidates" and "full candidates." This process involves executing shell commands and build tools to generate playable game artifacts based on the project design.
  • [DYNAMIC_EXECUTION]: The orchestration logic in SKILL.md and pipeline-contract.md describes the generation and consumption of "executable models." These models are compiled or interpreted at runtime to verify game logic, representing the dynamic execution of agent-generated code.
  • [EXTERNAL_DOWNLOADS]: The intake-method.md file specifies that the agent should use network access to fetch information regarding engine capabilities, software licenses, channel restrictions, and service specifications. This introduces a dependency on external, potentially untrusted sources for decision-making data.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze novels in any language. These source materials are untrusted external data and could contain embedded instructions intended to manipulate the agent's behavior during the analysis, design, or build phases.
  • Ingestion points: The skill reads source novels during the intake and novel-game-analyze stages (referenced in SKILL.md).
  • Boundary markers: The skill uses a structured workspace and specific markdown files (PRODUCT_BRIEF.md, SOURCE_BIBLE.md) to isolate facts, but there are no explicit instructions to ignore natural language commands found within the novel content.
  • Capability inventory: The skill possesses file system write access for workspace management, network access for research, and the ability to execute build processes.
  • Sanitization: No evidence of sanitization or filtering for the input novel text is provided.
  • [DATA_EXFILTRATION]: The combined capability of reading user-provided novel content and performing network operations for engine research creates a potential surface for data exposure, as information derived from the novel could be included in outbound research queries.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 7, 2026, 09:37 AM