requesting-code-review
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The template in
code-reviewer.mdinstructs the subagent to execute shell commands (git diff) using the{BASE_SHA}and{HEAD_SHA}placeholders. Since these variables are interpolated directly into the command string, an attacker could provide SHAs containing shell metacharacters (e.g.,;,|, or backticks) to execute unauthorized commands. - [INDIRECT_PROMPT_INJECTION]: The skill processes several untrusted input fields such as
{WHAT_WAS_IMPLEMENTED}and{PLAN_OR_REQUIREMENTS}by interpolating them directly into the subagent's instructions. - Ingestion points: Multiple placeholders in
code-reviewer.mdare populated with data from the primary agent's context. - Boundary markers: None. The external data is mixed directly with the subagent's core instructions without delimiters or warnings.
- Capability inventory: The reviewer subagent is explicitly directed to use shell-based tools for code analysis.
- Sanitization: There is no evidence of validation, escaping, or filtering of the content provided to these placeholders.
Audit Metadata