gs-create-formal-verification-skill

Pass

Audited by Gen Agent Trust Hub on Nov 20, 2026

Risk Level: SAFEDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill is designed to generate a new, project-local skill (containing instructions and potentially scripts) and then immediately execute that generated content to ensure it correctly identifies breaks in the code. This 'self-testing' phase involves executing logic created at runtime.
  • [COMMAND_EXECUTION]: To perform verification, the skill executes various shell commands, including Git operations ('git status --porcelain') and the specific verification tools (e.g., TLA+ model checkers) selected based on the codebase's needs.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local codebase to determine its verification targets. This creates an attack surface where malicious comments or code could attempt to influence the logic or configuration of the generated verification skill.
  • Ingestion points: The skill reads source files across the repository to 'Choose the target from the code'.
  • Boundary markers: No specific delimiters are used to isolate codebase content from the agent's instructions.
  • Capability inventory: The skill possesses the ability to write files (creating the local skill) and execute shell commands (running the verification tools).
  • Sanitization: No specific sanitization or escaping of ingested code content is mentioned before it is used to describe the verification results.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 20, 2026, 07:02 PM