code-review
Warn
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill mandates the execution of
yarn cli installandyarn test:viteston untrusted code from Pull Requests and branches to verify logic. While it uses temporary git worktrees to isolate the file system state, it does not isolate the execution environment. Maliciously craftedpackage.jsonlifecycle scripts or test files in a PR could execute arbitrary code on the host machine. - [COMMAND_EXECUTION]: The skill relies on shell-based tools including
git,gh(GitHub CLI), andyarnto interact with the repository and verify changes. The instructions specifically direct the agent to 'double run' (base vs head) verification scripts, which involves executing code derived from external PR OIDs. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from Pull Request diffs and metadata, creating a vulnerability surface where a malicious PR could attempt to influence agent behavior.
- Ingestion points:
gh pr view,gh pr diff,git diff, and reading the contents of changed or untracked files. - Boundary markers: Absent. The skill lacks instructions for using delimiters or escape sequences to isolate untrusted code diffs from the agent's core instructions.
- Capability inventory: Access to shell commands (
git,gh) and high-risk execution capabilities viayarn(installation and testing). - Sanitization: Absent. There is no explicit sanitization or validation of the code content before it is passed to the agent or executed via verification tools.
Audit Metadata