maintain-formal-verification-skill
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill reads untrusted data from git commit messages, doc comments, and changelogs to determine if verification models require updates. This creates an attack surface where instructions embedded in commit messages could influence agent behavior. Ingestion points: Step 3 in SKILL.md (Diffing covered files and reading commit messages, doc comments, and changelogs); Boundary markers: Absent (no delimiters or warnings to ignore instructions found within the ingested text); Capability inventory: File modification (updating models/harnesses) and command execution (running verifiers); Sanitization: Absent (no validation or escaping of natural language data).
- [COMMAND_EXECUTION]: The skill discovers and executes commands found within project-local configuration files to run model checkers and provers.
- [DYNAMIC_EXECUTION]: The skill identifies the verification command at runtime by reading local project metadata, which leads to the execution of commands determined by the project environment.
- [EXTERNAL_DOWNLOADS]: The THIRD_PARTY_NOTICES.md file references a repository from the Cursor project on GitHub for attribution and licensing purposes.
Audit Metadata