create-verification-skill

Warn

Audited by Gen Agent Trust Hub on Aug 15, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill instructions require the agent to generate a new project-specific skill and immediately execute it. Step 4 ('Prove the generated skill before handing it over') mandates running the generated 'launch', 'doctor', and 'drive' commands. This 'generate-then-execute' pattern lacks a human-in-the-loop review before initial execution.
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it 'interviews' the repo's codebase to determine its execution logic.
  • Ingestion points: The agent reads repo-level documentation, package scripts, and Makefiles to determine how to drive the application (SKILL.md, Section 1).
  • Boundary markers: There are no instructions to treat repository content as untrusted or to ignore instructions embedded within the codebase files.
  • Capability inventory: The agent is empowered to execute shell commands for application startup, terminal interaction (via tmux/PTY), and cleanup.
  • Sanitization: The skill does not define validation or sanitization rules for the commands it derives from the repository's 'dev commands' or 'existing harnesses'.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 15, 2026, 08:46 PM
Security Audit — agent-trust-hub — create-verification-skill