data-refresh-eval

Pass

Audited by Gen Agent Trust Hub on Feb 28, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the Bash tool to execute local CLI commands for dataset management and evaluation, specifically running bun src/index.ts within the vendor's repository.
  • [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it processes untrusted customer message data retrieved from the Front API during the evaluation process.
  • Ingestion points: Untrusted conversation content is fetched from the Front API and stored in local files such as data/eval-dataset.json.
  • Boundary markers: No boundary markers or specific instructions to disregard embedded commands were found in the workflow documentation.
  • Capability inventory: The skill leverages Bash(skill:*) for script execution and data manipulation using tools like cat and jq.
  • Sanitization: There is no evidence of sanitization or escaping of external content before it is analyzed by the agent response quality workflow.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 28, 2026, 10:33 PM