arc-creating-evals

Pass

Audited by Gen Agent Trust Hub on May 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes local CLI tools (arc-skill-eval validate and arc-skill-eval run) to verify the integrity of the generated test files and execute smoke tests. These commands are standard for development workflows and are performed on local directories identified by the user.\n- [PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection by ingesting and interpreting data from external SKILL.md files. This is documented as a vulnerability surface (Category 8) due to the following structure:\n
  • Ingestion points: Target SKILL.md files processed in Phase 1 to derive test cases.\n
  • Boundary markers: Mandatory user confirmation required after each of the five phases before moving to the next set of actions.\n
  • Capability inventory: Ability to create directories, write JSON and fixture files, and execute local validation/testing commands.\n
  • Sanitization: Relies on human review of the derived summary and test cases rather than automated content filtering.
Audit Metadata
Risk Level
SAFE
Analyzed
May 13, 2026, 02:29 PM
Security Audit — agent-trust-hub — arc-creating-evals