dogfood-skill-testing

Pass

Audited by Gen Agent Trust Hub on Jun 17, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is a legitimate development tool for testing AI agent skills. It follows established security practices by using firewalled subagents for different testing phases (blind vs. informed).
  • [COMMAND_EXECUTION]: The skill uses claude CLI commands in its 'Probe A' phase to measure trigger precision. This is a standard and safe operation for its intended purpose of testing skill activation in a sandboxed or local environment.
  • [EXTERNAL_DOWNLOADS]: There are no external downloads or remote script executions. All resources, including the defect taxonomy and report templates, are bundled locally within the skill directory.
  • [DATA_EXFILTRATION]: No data exfiltration patterns were detected. The skill writes its findings to a local directory (docs/skill-dogfood/) and does not perform network operations to untrusted domains.
  • [PROMPT_INJECTION]: No prompt injection or instructions to bypass safety guidelines were found. The skill focuses on behavioral testing and quality assurance.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 17, 2026, 05:05 PM
Security Audit — agent-trust-hub — dogfood-skill-testing