run-iteration-eval

Pass

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute local bash scripts and Python scripts for syncing caches and running evaluation pipelines, such as bash scripts/run-content-evals.sh and python3 scripts/grade-iteration.py.
  • [PROMPT_INJECTION]: The skill processes evaluation test cases stored in evals/content/v2/evals-v2.json. This presents an indirect prompt injection surface if the test cases contain malicious instructions designed to subvert the agent's behavior during the evaluation process.
  • Ingestion points: evals/content/v2/evals-v2.json
  • Boundary markers: Not specified in the instructions.
  • Capability inventory: Shell script execution (bash) and Python execution (python3).
  • Sanitization: No sanitization or validation of the input data is described.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 16, 2026, 03:21 AM
Security Audit — agent-trust-hub — run-iteration-eval