screeps-world-ai
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external game state data which is untrusted and could be controlled by other players. \n
- Ingestion points: The agent reads "room objects, decisions, and telemetry" and "live state" (SKILL.md). \n
- Boundary markers: The skill does not provide specific instructions or delimiters to isolate game data from processing logic. \n
- Capability inventory: The agent has access to the "release process" and can "change the colony" (modify game code/state) (SKILL.md). \n
- Sanitization: No sanitization steps are documented for handling external inputs. \n
- [COMMAND_EXECUTION]: The skill assumes the agent will interact with local build and deployment tools. \n
- Evidence: The instructions direct the agent to "use the project's release process" and perform "required build and verification" (SKILL.md).
Audit Metadata