agent-benchmarking

Pass

Audited by Gen Agent Trust Hub on Jun 27, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill documentation does not contain any malicious patterns, obfuscated content, or unauthorized data access commands.
  • [EXTERNAL_DOWNLOADS]: Recommends installing reputable benchmarking libraries such as 'swebench' and 'webarena' from standard public registries.
  • [COMMAND_EXECUTION]: Includes standard shell commands for package installation and running test suites (e.g., pip install, pytest, python) that are appropriate for the skill's context.
  • [PROMPT_INJECTION]: Identifies a potential attack surface for indirect prompt injection related to processing production bug reports.
  • Ingestion points: External bug report data ingested into the 'benchmarks/custom_regression/cases/' directory.
  • Boundary markers: None explicitly defined in the provided templates.
  • Capability inventory: The skill uses environment setup and test runner capabilities (pip, pytest).
  • Sanitization: The instructions explicitly recommend anonymizing production data to mitigate privacy and security risks.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 27, 2026, 04:56 PM
Security Audit — agent-trust-hub — agent-benchmarking