requesting-code-review

Pass

Audited by Gen Agent Trust Hub on Mar 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8). The subagent prompt in code-reviewer.md interpolates untrusted content including code diffs and task descriptions without sanitization or boundary markers. This allows malicious instructions embedded in reviewed files to potentially influence the subagent's assessment.\n
  • Ingestion points: code-reviewer.md processes external data from {DESCRIPTION}, {PLAN_REFERENCE}, and Git command outputs.\n
  • Boundary markers: Absent. No specific delimiters or warnings are used to isolate user-provided code from the subagent's system instructions.\n
  • Capability inventory: The skill invokes shell commands such as git diff to analyze code.\n
  • Sanitization: No validation or escaping is applied to the data before interpolation into the prompt.\n- [COMMAND_EXECUTION]: The skill constructs and executes shell commands (git rev-parse, git log, git diff) using variable interpolation in SKILL.md and code-reviewer.md. This is required for its functionality but introduces a potential injection risk if the commit SHAs or other variables are manipulated.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 14, 2026, 11:30 AM