Agent Skills in Claude Code and Visual Studio Code: Setup, Benefits, Risks, and a Step-by-Step Example

Image
 AI coding assistants are useful for one-off questions, but many development tasks are not one-off tasks. A team may follow the same testing procedure before every release, use a specific format for code reviews, or require certain checks whenever a database migration is created. Repeating those requirements in every prompt is inefficient. Putting all of them into a permanent instruction file is not always ideal either, because the assistant may receive irrelevant information for tasks that do not need it. This is the problem that Agent Skills are designed to solve. A skill packages reusable instructions, supporting files, examples, and optional scripts into a folder. Claude Code or GitHub Copilot in Visual Studio Code can discover that folder and load its contents when the current task matches the skill’s purpose. Both environments support the open Agent Skills format, which makes it possible to share some skills across tools. This guide explains what skills are, how they differ ...

Building a Strong Code Review Culture in the Age of AI

 

Building a Strong Code Review Culture in the Age of AI

A team's code review culture doesn't start with a tool. It doesn't start with a grand declaration from leadership. It starts with one PR, reviewed thoughtfully by someone who understands why it matters. Culture is built through accumulated small actions and habits, not through mandate or policy alone.

Write Team Guidelines, Not Commandments

When expectations are different across a team, review culture falls apart. One developer thinks [Suggestion] comments must be addressed before merge. Another thinks they're optional. One expects review within 24 hours. Another thinks several days is fine. One believes AI is authoritative. Another sees it as just another data point.

Document the minimum principles the team agrees on. Not 50 detailed rules — three to five core things that matter most.

## Code Review Guidelines

### Our Purpose
We use code reviews to improve code quality, catch problems early, and build team knowledge together.

### PR Writing Principles
- Each PR focuses on a single purpose
- Keep changed lines under 200 where possible
- Include what changed, why, and how it was tested
- Link to related issues or planning documents

### Comment Classification
- [Required]: Must fix (security, bugs, obvious errors)
- [Suggestion]: Author decides (improvements, style)
- [Question]: Asking for understanding
- [Praise]: Explicitly mention what's done well

### Review Timeline
- Start the first review within one business day
- If busy, leave a comment saying so

### Merge Criteria
- All [Required] comments resolved
- At least one human approval
- AI review [Required] items fixed
- Tests pass locally

### Using AI Review
- AI review is auxiliary, not authoritative
- Resolve AI false positives with reasoning
- [Required] human review is always needed

This is short enough to remember. Concrete enough to apply.



Create Checklists That Matter

Airline pilots use detailed pre-takeoff checklists despite thousands of hours of experience. Not because they forget things, but because attention diverts from repetitive tasks.

Code review checklists prevent "I forgot to check security this time" or "I didn't verify the business logic."

## Author Checklist
- [ ] Do variable and function names clearly explain intent?
- [ ] Are null/undefined handled? Are return types clear about nullability?
- [ ] Are there tests for failure cases, not just happy path?
- [ ] Does PR description explain why (not just what)?
- [ ] Have I checked AI review [Required] items?
- [ ] Is this PR focused on one purpose? Can it be smaller?

## Reviewer Checklist
- [ ] Does this code actually do what the requirements ask?
- [ ] Will colleagues in 6 months understand this?
- [ ] Are all connected systems and side effects handled?
- [ ] What areas did AI miss? (business logic, domain rules, context)
- [ ] Are there edge cases that need handling?
- [ ] Does this maintain consistency with existing patterns?

Checklists make abstract ideals concrete. They prevent important areas from being skipped.

How Senior Developers Shape Culture

Culture flows from top to bottom. If your team lead or senior developers ignore guidelines, guidelines become meaningless. If they follow them carefully, they become real.

What senior developers can do:

  • Share past mistakes openly. "I also hard-coded the JWT secret when I was starting out" is more powerful than any lecture. It signals that errors are part of learning, not failures of character.
  • Respond seriously to junior questions. When a junior developer asks a [Question] comment in code review, answer thoroughly and respectfully, not dismissively. This encourages more questions and real dialogue.
  • Share good review examples. "I liked this review comment in that PR because it explained the why clearly" teaches standards to the team by example.
  • Follow guidelines first. Your adherence makes them real. Your violations undermine them.
  • Publicly acknowledge mistakes. "I missed this critical issue in review last week. Next time I'll be more careful" demonstrates growth and sets psychological safety.

Measuring Culture Health

You can't manage what you don't measure. These metrics show culture direction:

  • Average PR size (lines changed) — Shrinking is good (smaller, focused changes easier to review)
  • Average time to first review — Shorter is good (review isn't a bottleneck)
  • Ratio of [Required] to [Suggestion] — Shows balance (too many Required means unclear expectations; too many Suggestion means rubber-stamping)
  • Average time to merge — Shorter means faster value delivery
  • Comment positivity — Are [Praise] comments being used? Learning is happening?

Look at trends, not absolute numbers. If PRs are getting smaller, reviews respond faster, and people are giving explicit praise, culture is improving.

Common Failure Patterns

Pattern 1: Temporary Enthusiasm

Month one: "Let's do great code reviews!" Everyone participates energetically. Month two: Schedules get busy. The pace drops. People revert to old habits.

Prevention: Make the minimum sustainable. One small principle that works even when busy beats 10 grand principles that require perfect conditions.

Pattern 2: Reviews Become a Bottleneck

More detailed reviews = more time to review = PRs queue up = pressure to "reduce reviews" and ship faster.

Prevention: Use AI to automate surface checks. Keep PR size small. Human reviewers focus on high-value issues. This actually speeds everything up.

Pattern 3: Rubber-Stamp Reviews

Reviews happen every day mechanically, but no one leaves [Required] comments. Everyone approves quickly. It's just a formality.

Prevention: Trace your last production bug back to code review. Would better review have caught it? Usually yes. When the team sees the value concretely, rubber-stamping stops. Make it personal and relevant.

The Role Evolution in the AI Age

As AI handles surface-level issues (syntax, null checks, simple patterns), human reviewers' role doesn't shrink — it elevates strategically:

  • Business logic correctness — Does this do what it's supposed to?
  • Architectural decisions — Is this maintainable long-term?
  • Production scenarios — Will this scale? Will it fail gracefully?
  • Team growth — Is this a teaching moment?
  • Strategic trade-offs — What are we optimizing for?

This is more valuable and strategic than catching syntax errors.

Measuring Success: What Better Culture Looks Like

A team with healthy review culture has these characteristics:

  • PRs are small and focused. Average PR is under 200 lines. Single purpose.
  • Reviews are fast. First review within 24 hours. No week-long waiting.
  • Comments are kind but thorough. People give reasons. People explain why.
  • Authors engage, not defend. Questions are answered. Disagreements are discussed, not argued.
  • There are [Praise] comments. Good code is acknowledged.
  • Seniors review fairly. No rubber-stamping. No rejecting good code for stylistic preference.
  • Juniors ask questions. Psychology is safe enough that they don't just copy changes silently.
  • Production incidents are rare. Good reviews catch problems before deployment.

Measure progress toward these. Not perfectly — just the direction.

The Long View

Building code review culture takes months, not weeks. There will be setbacks. Busy periods will cause backsliding. New people will need to learn the culture. But the payoff compounds.

Teams with strong review culture ship faster, not slower. They have fewer production bugs. They have higher quality code. They retain senior talent because the work is respected and learning happens continuously. They onboard juniors faster because the code is clear and review is educational.

Handling Review Bottlenecks

Teams often struggle with slow reviews. What helps:

Use AI for surface-level checks: Let AI handle null checks and simple patterns. Humans review architecture and business logic. This reduces human review time by 30-50%.

Create review SLAs: "First review within 24 hours" prevents PRs languishing for days. Even if review is incomplete, partial feedback keeps momentum going.

Smaller PRs are faster to review: A 50-line PR takes 5 minutes. A 500-line PR takes 45 minutes. Encourage authors to keep PRs focused.

Reviewer assignments: Don't rely on everyone. Assign specific reviewers based on expertise. Route frontend PRs to frontend experts. Route auth/security PRs to security-trained reviewers. This reduces review time and improves review quality.

Asynchronous reviews are okay: Not every review needs synchronous discussion. Most issues can be resolved through comments. Save video calls for complex architecture discussions.

Scaling Code Review Culture

As teams grow, maintaining review culture becomes harder. What helps at scale:

  • Explicit documentation: Written guidelines everyone can reference
  • Automation: AI tools, linters, tests run before human review
  • Clear escalation: Who decides when reviewers disagree?
  • Domain ownership: Team leads own review for their domain
  • Rotate reviewers: Don't let one person be the only reviewer of critical code
  • Regular retrospectives: Monthly: What's working? What's not? What should we change?

Start Small, Build Momentum

You don't need to transform your entire team's culture overnight. Don't try to implement 10 changes at once. Take the next PR you review. Do one thing well:

  • Distinguish [Required] from [Suggestion]
  • Leave a [Praise] comment on something well done
  • Check that business logic matches requirements
  • Read the planning document first
  • Ask about edge cases

Do one thing well. It spreads. Others notice. Others start doing it too.

In a year, look back at your reviews. They'll be markedly better. Your team will ask fewer questions about what's expected. Your code will be more reliable. Your juniors will be better trained. Not from policy written on paper, but from habit. Not from mandate from above, but from demonstrated value accumulated over many small actions.

That's how sustainable culture is built — one thoughtful review at a time, compounds into teams that ship better code.



Conclusion: Culture is Competitive Advantage

In five years, the teams that win won't be the ones with the best individual developers. They'll be the ones with the strongest code review cultures. The ones where knowledge compounds. Where mistakes become learning opportunities. Where everyone's voice is valued. Where code quality improves every quarter. Where retention is high because people feel respected and challenged.

Building that culture starts now. With this PR. With one thoughtful comment.


Comments

Popular posts from this blog

Claude Code Multi-Agent Setup for Beginners: Models, Roles, and Your First Project

The Limits and Future of Vibe Coding

How to Test and Deploy Your First Vibe-Coded App