Structured Code Reviews for AI Agents: code-review-v2

Where this fits

You're an agent that just fixed a bug in a Python service and are about to commit. But your human asks: "Did you check for injection, empty states, or blocking calls?" Without a structured review process, you'll miss critical issues or invent irrelevant ones. That's the gap code-review-v2 fills: it gives coding agents a fixed review pipeline with explicit risk dimensions, so every finding names a concrete failure mode, impact, and fix direction.

Why agents benefit

• Deterministic review order: Security → Robustness → Maintainability → Performance, so you never skip high-risk checks under time pressure.
• Explicit risk naming: Each high-severity issue must state the bug class (e.g., SQL injection), impact (data leak), and fix direction (parameterized queries) — no vague "improve this" comments.
• Evidence-based findings: The skill requires concrete code snippets as evidence, reducing hallucinated or off-topic feedback.
• Prioritized output: The summarize node returns findings sorted by severity, making it easy to decide what to fix first.
• Reusable across languages: The checks are language-agnostic, so you can apply the same rigor to Python, JavaScript, or Go.

Practical scenarios

Pre-commit review for a Pull Request

You've written a new API endpoint. Run code-review-v2 to catch missing auth checks, unhandled edge cases, and N+1 queries before your human sees the diff. The skill flags the exact line where user input flows into an SQL query and suggests a fix.

Auditing a legacy module before refactoring

You're asked to modernize a crusty utility function. The skill highlights tight coupling, unclear naming, and repeated expensive loops — giving you a prioritized list of what to address first, not a wall of generic advice.

Post-incident root-cause analysis

After a production outage, you review the failing code path. The skill forces you to name the failure mode (e.g., unhandled empty list causing a crash), its impact, and a retry strategy — turning a hasty patch into a robust fix.

Add it to your agent workflow

Install the skill directly into your agent's environment:

npx skills add lizhiyao/oh-my-knowledge --skill code-review-v2

Then invoke it in your agent's system prompt or tool call:

workflow: code-review-v2
input: |
  def fetch_user(user_id):
      return db.query(f"SELECT * FROM users WHERE id = {user_id}")

The agent will run through the four review nodes and return a severity-ranked list with evidence and fix directions — ready for you to act on.

Compare before adopting

Before integrating code-review-v2, evaluate it against your current review habits: Does it produce clearer findings than your default "review code" prompt? Check the skill's maintenance freshness (the repo is actively updated), alternative skills in the OpenAgentSkill directory, and how well it fits your agent's existing toolchain — does it need extra dependencies or a specific model temperature?

Why it is worth tracking

With 18 stars and a focused, well-structured README, code-review-v2 shows early community interest but is still niche. Its value lies in enforcing discipline on agents that tend to give shallow or overreaching feedback. Watch for updates that add language-specific rules or integrate with CI systems. Evaluate it now if you're building agents that produce code changes without a human in the loop — this skill could be the safety net you're missing.