requesting-code-review

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through placeholders in the code-reviewer template.
  • Ingestion points: The {WHAT_WAS_IMPLEMENTED}, {PLAN_OR_REQUIREMENTS}, and {DESCRIPTION} placeholders in code-reviewer.md ingest content from local project files or agent-generated descriptions.
  • Boundary markers: Absent. The template does not use specific delimiters (like XML tags or triple quotes with instructions) to separate untrusted data from the system instructions.
  • Capability inventory: The subagent executes shell commands (git diff, git log) and its output is used to determine if a task is complete.
  • Sanitization: Absent. The skill does not provide instructions to escape or validate the input content before interpolation.
  • [COMMAND_EXECUTION]: The skill involves the execution of local git commands to retrieve and compare code changes.
  • Evidence: Both SKILL.md and code-reviewer.md include shell commands such as git rev-parse, git log, and git diff. These are standard tools for identifying changes between commits and are used here for their intended purpose without requesting elevated privileges.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 02:24 AM