loop-execution-evaluator
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes
metadata.jsonandspec.mdto determine track types and dispatch evaluators. This creates an ingestion point for untrusted data that could contain instructions designed to influence the agent's logic or behavior. - Ingestion points: The skill reads
metadata.jsonandspec.mdduring the 'Dispatch Workflow' to determine keywords. - Boundary markers: There are no explicit instructions or delimiters defined to separate untrusted data content from the agent's instructional context.
- Capability inventory: The skill utilizes
read_file,write_file, subprocess execution (npm run build), and the ability to dispatch/invoke other specialized evaluation skills. - Sanitization: No sanitization, validation, or escaping of the content found within the ingested files is mentioned.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute
npm run buildas a structural check to verify build integrity. While standard for development workflows, this involves executing shell commands which could be manipulated if the build configuration (e.g.,package.json) is untrusted.
Audit Metadata