requesting-code-review
Pass
Audited by Gen Agent Trust Hub on Feb 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- Prompt Injection (LOW): The template in
code-reviewer.mdis vulnerable to Indirect Prompt Injection. It ingests potentially untrusted data through the{DESCRIPTION}and{PLAN_REFERENCE}placeholders without utilizing boundary markers or sanitization. * Ingestion points:{DESCRIPTION}and{PLAN_REFERENCE}incode-reviewer.md. * Boundary markers: Absent; there are no delimiters separating instructions from external data. * Capability inventory: Subprocess execution ofgit diffandgit log. * Sanitization: None detected. - Command Execution (LOW): The skill performs shell command execution using the
gitutility. Placeholders such as{BASE_SHA}and{HEAD_SHA}are interpolated directly into shell command strings. If these parameters are sourced from untrusted metadata (e.g., in a public pull request scenario), it creates a surface for command injection.
Audit Metadata