requesting-code-review

Warn

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • Indirect Prompt Injection (MEDIUM): The skill is designed to ingest external, potentially untrusted content such as placeholders and git diffs.
  • Ingestion points: Variables such as {WHAT_WAS_IMPLEMENTED}, {PLAN_OR_REQUIREMENTS}, and {DESCRIPTION} in SKILL.md, along with the source code diffs between BASE_SHA and HEAD_SHA.
  • Boundary markers: Absent. No delimiters are used to separate untrusted content from instructions.
  • Capability inventory: The skill directs the agent to 'Act on feedback' and 'Fix Critical issues immediately.' This creates a risk where malicious instructions embedded in code comments or PR descriptions could influence the agent's subsequent write/execute actions.
  • Sanitization: No sanitization or validation of input variables is described.
  • Command Execution (LOW): The skill utilizes standard git commands (rev-parse, log, awk) to manage commit history. While these execute subprocesses, they are restricted to local repository metadata and are standard for the described purpose.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 16, 2026, 01:00 PM