Best Agent Skills for AI Code Review: 8 Evaluated Skills For Dev Workflows

Yesterday I asked an agent to review a PR.

It replied: “Looks good. Consider adding comments.” Cool. That’s definitely worth the API cost. The real issue is it doesn’t know our conventions, security posture, or what “risky” looks like for our codebase - so it treats a 500-line refactor like a one-line typo.

Agent skills solve this by packaging expert knowledge into reusable instructions that agents load when relevant.

For code review, this means skills that know how to analyze diffs systematically, check for security anti-patterns, verify test coverage, and produce actionable feedback, based on your ways of working and project or team policies, rather than generic surface-level comments.

We looked through the skills ecosystem on the Tessl Registry and GitHub, and we found 3 distinct categories of code review skills that you might want to consider looking at for your projects:

The Reviewers: Skills that generate review content (comments, risk notes, suggested diffs). These are the workhorses: multi-agent reviewers, security analyzers, and checklist-based review generators.

The Workflow: Skills that decide and coordinate (routing, sequencing, policies). They don’t write review comments; they decide when to run which reviewer, how to combine results, and what to do next (block, request changes, escalate).

The Plumbing: Skills that move data between systems (GitHub/GitLab/Jira/Slack). Fetching diffs, mapping files to owners, posting inline comments, updating statuses - pure integrations and utilities

Here are the skills we’ll look at in this post:

• Reviewers: getsentry/skills (code-review), secondsky/claude-skills (code-review), langgenius/dify (frontend-code-review), bkircher/skills (gh-code-review)
• Workflow: obra/superpowers (requesting-code-review, and receiving-code-review)
• Plumbing: YPares/agent-skills (github-pr-workflow), alchemiststudiosDOTai/coderabbit-fix-flow-plugin (coderabbit-fix-flow),

How does Tessl evaluate each skill?

Skill evaluations are scored via registry reviews that grade quality, validity, implementation, and activation.In Tessl, a skill’s evaluation in the registry is broken into multiple scores:

• Review Score (0–100%): overall quality (structure/clarity, spec conformance, workflow integration). This score is made of the calculated from implementation + activation + validation score
• Validation Score: pass/fail checks like required frontmatter, trigger hints, step-by-step workflow, license/metadata.
• Implementation Score: quality of code examples (tested, secure, error handling).
• Activation Score: how discoverable the skill is (clear triggers, keywords, scope).

We’ll assess below the plugins, skills, tiles based on the results from Tessl evals as well as skill popularity, and show you how to add them into your project and start using them.

You can manually download skill files from GitHub, but Tessl’s package manager gives you versioning (pin to specific releases), evaluations (tests), and a manifest that tracks all skills and context for your projectlike npm, Maven, or pip, but for skills and context.

Getting Started with Tessl

Before we jump into the skills, let's set up Tessl.

# Install Tessl
npm i -g @tessl/cli

# Login to Tessl (you'll create a free account)
tessl login

That’s it! Let’s now jump in and look at the skills that perform the actual code reviews

The Reviewers: Skills That Generate Feedback

The following SKILLs scored among the highest in the “analyze code and produce actionable review feedback” category.

1. code-review by Sentry

Sentry's code-review skill packages how one of the industry's most engineering-focused companies approaches review internally

• Review Score: 86% (quality checks on the skill definition—implementation + activation + validation)
• Validation Score: 12/16 (manifest/metadata checks)
• Implementation Score: 88% (structure, clarity, actionability)
• Activation Score: 82% (how well it’s described/discoverable)
• View the full evaluation here.

tessl i github:getsentry/skills --skill code-review

2. code-review by secondsky

code-review from secondsky/claude-skills provides general-purpose code review with emphasis on best practices, maintainability standards, and overall code quality. Clean, straightforward approach to review fundamentals.

• Review Eval: 88%
• Validation Score: 13/16
• Implementation Score: 100%
• Activation Score: 67%
• View the full evaluation here.

tessl i github:secondsky/claude-skills --skill code-review

3. frontend-code-review by langgenius

frontend-code-review from langgenius/dify specializes in frontend codebases: React/Vue component review, state management patterns, styling approaches, accessibility compliance, and performance optimization. From the team behind Dify, an open-source LLM app platform.

• Review Eval: 76%
  Validation Score: 14/16
• Implementation Score: 85%
• Activation Score: 50%
• View the full evaluation here.

tessl i github:langgenius/dify --skill frontend-code-review 

4. gh-code-review by bkircher

gh-code-review from bkircher/skills is a straightforward, checklist-based PR reviewer using the GitHub CLI. It covers correctness, security, performance, concurrency, API contracts, and dependencies a solid general-purpose option.

• Review Eval: 83%
• Validation Score: 12/16
• Implementation Score: 92%
• Activation Score: 67%
• View the full evaluation here.

tessl i github:bkircher/skills --skill gh-code-review

The Workflow: Analyze, Fix, Ship

These skills don't generate feedback themselves, they orchestrate the review process and help you handle feedback effectively.

Superpowers: A Complete Development Methodology

obra/superpowers has exploded to 43,000+ GitHub stars, but it’s often mislabeled as a “code review tool.” It’s closer to an agentic SDLC toolkit: it lets you compose end-to-end workflows like triage an issue → generate a plan → open a PR → run checks → request review → apply feedback → ship, with code review being just one step.

1. requesting-code-review

requesting-code-review teaches agents when to dispatch code-reviewer sub-agents at task boundaries with proper SHA tracking. It's about triggering reviews at the right moments in your workflow.

• Review Eval: 71%
• Validation Score: 13/16
• Implementation Score: 100%
• Activation Score: 15%
• View the full evaluation here.

tessl i github:obra/superpowers --skill requesting-code-review

2. Receiving-code-review

receiving-code-review is perhaps more interesting as it teaches agents how to handle feedback. Verify claims before implementing fixes. Push back with technical reasoning when appropriate. Don't blindly accept every suggestion.

• Review Eval: 79%
• Validation Score: 13/16
• Implementation Score: 100%
• Activation Score: 40%
• View the full evaluation here.

tessl i github:obra/superpowers --skill receiving-code-review

The Plumbing: PR Utilities and Integrations

Lower-level tools that work with PRs and external review services.

1. github-pr-workflow

github-pr-workflow from YPares/agent-skills handles the mechanics of working with PRs: fetching diffs, posting inline comments, checking CI status, handling approvals. The plumbing that review skills build on.

• Review Eval: 88%
• Validation Score: 13/16
• Implementation Score: 88%
• Activation Score: N/A
• View the full evaluation here.

tessl i github:YPares/agent-skills --skill github-pr-workflow

2. coderabbit-fix-flow

coderabbit-fix-flow from alchemiststudiosDOTai processes feedback from CodeRabbit systematically save comments, analyze with MCP tools, implement fixes. Useful if your team uses CodeRabbit for automated reviews.

• Review Eval: 66%
• Validation Score: 13/16
• Implementation Score: 50%
• Activation Score: 75%
• View the full evaluation here.

tessl i github:alchemiststudiosDOTai/coderabbit-fix-flow-plugin --skill coderabbit-fix-flow

How to Choose the Right Agent Skills?

The same skill won't be right for everyone.

I’d suggest you use recommendations from colleagues and friends who have had success with code-review skills using a similar technology stack. Evals tell you whether a skill is reliable at its task (correctness, consistency, regressions).

Popularity tells you whether it’s adopted - a weak but useful signal that people find it valuable in practice.

Below is a ranking of the skills we’ve discussed in this post, sorted by the number of GitHub stars per repository. Note that each repo can host multiple skills or projects, but it also gives you a sense of the community's size.

The agent skills ecosystem is evolving rapidly. What started as simple prompt templates has grown into sophisticated multi-agent systems with confidence scoring, security analysis, and intelligent routing. The best part? They compose together, letting you build review workflows tailored to your team's needs.

What code review skills are your team using? Join our community, and let us know.