create-verification-skill
Warn
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill is designed to identify and execute arbitrary shell commands found in the repository, such as build scripts, dev commands, and test harnesses (e.g., 'package scripts, Makefile, README quickstart').
- [DYNAMIC_EXECUTION]: The skill follows a 'generate-and-execute' pattern where it writes a new skill file (SKILL.md) and then immediately executes its instructions end-to-end to verify functionality.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from the repository codebase to determine its driving logic, creating a surface where a malicious repository could influence the agent's behavior.
- Ingestion points: Codebase files including READMEs, Makefiles, and package scripts (Step 1: 'Interview the repo').
- Boundary markers: None present in the instructions to the agent when reading files.
- Capability inventory: The agent uses filesystem operations (mkdir, ln) and executes commands discovered during the interview.
- Sanitization: No explicit sanitization or validation of repository-derived commands is performed before the 'Prove the generated skill' execution step.
Audit Metadata