gepetto
Pass
Audited by Gen Agent Trust Hub on Aug 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill recommends the installation of external CLI tools (Gemini and Codex) from well-known technology providers (Google and OpenAI) to facilitate independent plan reviews.
- [COMMAND_EXECUTION]: The workflow involves executing bash commands to run the
geminiandcodexCLI tools. These commands are used to process the generated implementation plan content for external review. - [DATA_EXFILTRATION]: Implementation plans and codebase research findings are transmitted to external LLM services via CLI tools. This is a core documented feature of the 'External Review' step and targets established AI service providers.
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from the local codebase and web research to synthesize implementation plans. These plans are then provided to subagents and external reviewers. While this creates a potential surface for indirect injection where malicious codebase comments could influence the planning process, the risk is mitigated by the planning-focused context and lack of automated execution of the resulting plans. (Severity: LOW)
Audit Metadata