create-verification-skill

Pass

Audited by Gen Agent Trust Hub on Sep 6, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill analyzes the local repository codebase to derive verification steps, creating a surface where malicious instructions in the repo could influence agent behavior.
  • Ingestion points: The agent is instructed to 'Interview the repo' and observe the surface and codebase in Step 1.
  • Boundary markers: No specific delimiters or warnings to ignore embedded instructions are provided.
  • Capability inventory: The agent can perform file system writes and execute shell commands (launch, doctor, drive, cleanup) in Step 4.
  • Sanitization: The skill requires user confirmation before writing the generated harness to the project tree.
  • [DYNAMIC_EXECUTION]: The skill generates new script content and then executes it to validate functionality.
  • Evidence: Step 4 ('Prove it') requires the agent to run the newly generated 'launch', 'doctor', and 'drive' commands as a proof pass.
  • [COMMAND_EXECUTION]: The skill invokes shell commands to manage local application lifecycles.
  • Evidence: The 'External Side Effects' section explicitly states that the skill starts and tears down local app instances during the proof pass.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 6, 2026, 08:30 AM
Security Audit — agent-trust-hub — create-verification-skill