validating-performance-budgets

Pass

Audited by Gen Agent Trust Hub on Apr 3, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill ingests data from external JSON files which creates an indirect prompt injection surface.\n
  • Ingestion points: Performance metrics and budgets are loaded from metrics.json and budget.json in scripts/validate_budget.py and scripts/create_budget.py.\n
  • Boundary markers: No delimiters are used to separate untrusted metrics data from the agent's instructions.\n
  • Capability inventory: The agent can execute shell commands, write files, and perform network requests.\n
  • Sanitization: No input validation or sanitization is performed on the ingested metric strings.\n- [DATA_EXFILTRATION]: The skill can send performance reports to any URL provided by the user, representing a potential exfiltration vector for project metrics.\n
  • Evidence: The ViolationReporter.send_slack method in scripts/report_violation.py uses urllib.request to POST JSON data to an unvalidated URL argument.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 3, 2026, 12:19 PM