foundations-formal-methods

Pass

Audited by Gen Agent Trust Hub on Oct 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The scripts/test_check_finite_model.py file uses subprocess.run to execute the local check_finite_model.py script for unit testing. This implementation is restricted to testing the command-line interface of the skill's own utility and does not provide a path for arbitrary command injection.
  • [EXTERNAL_DOWNLOADS]: The data/sources.json file references numerous external repositories, documentation, and academic papers related to formal verification tools (e.g., Z3, TLA+, Lean, Dafny). All URLs target well-known academic domains or official project repositories on established platforms.
  • [INDIRECT_PROMPT_INJECTION]: The skill includes an ingestion surface where the agent processes state machine definitions in JSON format using scripts/check_finite_model.py.
  • Ingestion points: The agent is instructed to run the checker on JSON files (e.g., model.json) which could contain untrusted graph data.
  • Boundary markers: The script enforces a strict JSON schema and utilizes a custom parser hook to reject non-standard numeric constants or duplicate keys.
  • Capability inventory: The analysis script is restricted to graph traversal and result reporting; it contains no network, file-system write, or shell execution capabilities.
  • Sanitization: The script performs extensive validation of the input, ensuring all transition endpoints are valid members of the defined state set and using safe JSON loading practices.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 17, 2026, 08:31 AM