create-formal-verification-skill

Pass

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted project data (source code, recent incident reports) and user-provided invariants to construct formal models. It possesses the capability to write new skill files to the .claude/skills/ directory and execute local verification tools such as TLC, lake, and cargo kani. This represents a standard attack surface where malicious code within the analyzed project could potentially influence the agent's modeling logic, though the skill includes explicit instructions for human-in-the-loop verification and honest modeling.
  • [DYNAMIC_EXECUTION]: The workflow instructs the agent to dynamically locate a local package's root directory and import specific utility functions (appendLedgerEntry, defaultLedgerPath) from a dist folder at runtime. This dynamic loading of local scripts is used for the functional purpose of maintaining a centralized verification ledger and relies on the pre-existence of the project's internal developer tools.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 9, 2026, 08:35 PM