create-verification-skill
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill is designed to generate shell scripts and markdown instructions for a new verification skill and then execute those generated commands to validate the setup. This automated execution is the core functionality intended for application verification.
- Evidence: SKILL.md Step 2 instructions the agent to "Write .cursor/skills/verify-/SKILL.md" and notes that "Any script the skill ships is executable." Step 4 then directs the agent to "Run its own instructions end to end once: launch, doctor, drive ONE mapped feature... clean up."
- Context: The instructions incorporate safety practices such as using disposable data directories (e.g.,
/tmp/) and isolated environments to minimize impact on the host system. - [INDIRECT_PROMPT_INJECTION]: The skill analyzes repository documentation and configuration files to synthesize a verification plan. This process creates a surface where instructions or patterns within the repository could influence the agent's behavior during the generation or verification phase.
- Ingestion points: Project files, READMEs, and configuration files analyzed during the repository interview (SKILL.md, Step 1).
- Boundary markers: No explicit delimiters or instructions are defined to separate untrusted repository content from the generated skill's logic.
- Capability inventory: The generated skill possesses capabilities for process launching, command execution, and file system observation (SKILL.md, Step 2 and 4).
- Sanitization: The instructions do not specify sanitization or validation of repository content before it is used to generate the new skill.
Audit Metadata