requesting-code-review
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from git repositories and planning documents through multiple placeholders in the
code-reviewer.mdtemplate without using boundary markers or sanitization techniques. - Ingestion points: The placeholders
{WHAT_WAS_IMPLEMENTED},{PLAN_OR_REQUIREMENTS},{DESCRIPTION}, and{PLAN_REFERENCE}incode-reviewer.mdingest external content. - Boundary markers: Absent. The template lacks delimiters (such as XML tags or explicit 'ignore instructions' warnings) to prevent the subagent from following malicious instructions embedded in the injected text.
- Capability inventory: The subagent has the capability to execute shell commands (e.g.,
git diff) and dispatch further sub-tasks based on the review. - Sanitization: Absent. The instructions do not define validation or escaping for the content interpolated into the prompt template.
- [COMMAND_EXECUTION]: The
code-reviewer.mdtemplate instructs the subagent to execute git commands using interpolated variables for commit SHAs. If these SHAs are manipulated by an attacker (e.g., via a compromised plan document), it could lead to command injection. - Evidence: The bash blocks containing
git diff --stat {BASE_SHA}..{HEAD_SHA}andgit diff {BASE_SHA}..{HEAD_SHA}incode-reviewer.mdrely on runtime interpolation of variables that may originate from untrusted sources.
Audit Metadata