jev-code-review
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references the
jev-decideCLI tool, which is part of thejev-skillpackage. The instructions explicitly direct the agent to explain missing dependencies rather than performing silent installations. - [COMMAND_EXECUTION]: The skill uses the
jev-decideCLI to process code review requests. It includes instructions for validation using a--dry-runflag before making live network calls. - [DATA_EXFILTRATION]: By design, the skill sends code diffs, source snippets, and test receipts to external AI providers (OpenRouter or TypeSafe). However, this is mitigated by requiring explicit user configuration, prohibiting the inclusion of credentials in prompts, and offering a local simulation mode that avoids external network calls.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of code diffs and test receipts.
- Ingestion points: Code diffs, source files, and test command receipts are gathered into a JSON request file.
- Boundary markers: The skill uses a structured JSON schema to encapsulate data, and instructions prompt the user to review the input before submission.
- Capability inventory: The
jev-decidetool performs network operations to the selected AI provider. - Sanitization: The instructions emphasize manual review of the edited request and the use of validation flags.
Audit Metadata