create-verification-skill
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions (Step 4) direct the agent to execute shell commands (Launch, Doctor, Drive, Cleanup) that it discovers within the target repository's build files and documentation.\n- [DYNAMIC_EXECUTION]: The skill performs runtime generation and immediate execution of a new verification skill (
.agents/skills/verify-<app>/SKILL.md) to prove its functionality, which involves executing dynamically created instructions.\n- [INDIRECT_PROMPT_INJECTION]: The skill facilitates indirect prompt injection by ingesting untrusted codebase data to generate executable instructions.\n- [INDIRECT_PROMPT_INJECTION]: Ingestion points: Reads repo codebase, READMEs, and build scripts (SKILL.md Section 1: Interview the repo).\n- [INDIRECT_PROMPT_INJECTION]: Boundary markers: Delimiters and 'ignore embedded instructions' warnings are absent.\n- [INDIRECT_PROMPT_INJECTION]: Capability inventory: Shell command execution (Launch, Drive, Cleanup commands) (SKILL.md Section 4: Prove the generated skill).\n- [INDIRECT_PROMPT_INJECTION]: Sanitization: No validation or sanitization of the commands discovered in the repo is performed before execution.
Audit Metadata