improve-game
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data (the user's game codebase) which could contain embedded instructions intended to manipulate the agent.
- Ingestion points: The skill reads numerous files from the local directory, including
package.json, all files insrc/(core, scenes, systems, entities, ui, audio),index.html, and thetests/directory. - Boundary markers: There are no boundary markers or explicit instructions provided to the agent to disregard potential instructions found within the data it reads.
- Capability inventory: The skill can create or modify any file in the project directory and execute shell commands via
npm run buildandnpm test. - Sanitization: There is no mention of sanitization or validation of the files' contents before the agent processes them, allowing potentially malicious code comments or strings to be interpreted as instructions.
- [COMMAND_EXECUTION]: The skill executes shell commands based on the local environment's configuration.
- Evidence: The instructions direct the agent to run
npm run buildandnpm testas verification steps. If the localpackage.jsonhas been tampered with, these commands could execute arbitrary malicious scripts.
Audit Metadata