Best Skills for Workflow and Productivity

The AI writes decent code. But it does not plan before coding. It does not verify before claiming success. It writes commit messages that say “update stuff.” It cannot run a browser, audit a website, or create a PR that follows your repo’s template. The code is fine; everything around the code is amateur.

Productivity skills fix this. Every skill listed below is real, published on skills.sh, and used by thousands of developers. Install counts are included so you know what the community actually trusts.

What You’ll Walk Away With

• Real skills with verified install counts for planning, debugging, review, and git workflows
• Installation commands that work immediately with npx skills add
• Prompts that leverage each skill for common workflows
• A recommended stack that covers the biggest productivity gaps

Planning and Execution Skills

obra/superpowers — The Top Productivity Skill Repo

With over 78,000 total installs, obra/superpowers is the single most popular productivity skill collection on skills.sh. It covers the full development lifecycle from brainstorming through verification.

npx skills add obra/superpowers

This installs all skills from the repo. Here are the individual skills and their install counts:

Brainstorming (12.7K installs)

A Socratic questioning protocol for complex requests. Instead of jumping straight to code, the AI asks clarifying questions, explores edge cases, and validates assumptions before proposing a solution.

Writing Plans (6.2K installs)

Generates detailed implementation plans broken into bite-sized tasks. Each task is scoped to a single concern with clear acceptance criteria, making it easy to review the plan before any code gets written.

Executing Plans (5.5K installs)

Executes plans in controlled batches with review checkpoints. Instead of implementing an entire plan in one pass, the AI works through tasks in small groups and pauses for your feedback between batches.

Subagent-Driven Development (4.8K installs)

Delegates plan execution to subagents. Each subagent handles a scoped task, and the orchestrating agent reviews and integrates the results. This is how you scale AI-assisted development across a large feature.

Dispatching Parallel Agents (4.2K installs)

Runs multiple agents concurrently on independent tasks. When your plan has tasks with no dependencies between them, this skill parallelizes the work instead of running sequentially.

Tip

Copy-paste prompt for plan-then-execute workflow:

“I need to add OAuth2 login with Google and GitHub providers. Use your brainstorming skill first to ask me clarifying questions about scope and requirements. Once we agree on the approach, use writing-plans to create a detailed implementation plan. Do not write any code until I approve the plan.”

Verification Before Completion (4.5K installs)

Part of obra/superpowers. This skill forces the AI to run verification steps (tests, type checks, linting) before claiming a task is complete. Without it, the AI says “Done!” while the build is broken.

What it prevents:

• Claiming success when tests fail
• Missing type errors that only show up at build time
• Forgetting to run the linter after refactoring
• Skipping manual verification steps defined in your project

Create Plan — openai/skills

npx skills add openai/skills

OpenAI’s official skill collection (2,900 installs, 38 skills) includes create-plan, which generates concise implementation plans for coding tasks. Lighter weight than obra’s planning skills, useful when you want a quick outline rather than a full task breakdown.

Debugging Skills

Systematic Debugging (7.2K installs)

Part of obra/superpowers. A root cause analysis methodology that stops the AI from guessing. Instead of trying random fixes, the AI follows a structured process: reproduce the issue, form hypotheses, test each one, and confirm the root cause before proposing a fix.

Tip

Copy-paste prompt for structured debugging:

“The checkout flow fails silently when a user has both a gift card and a discount code applied. Use your systematic-debugging skill: reproduce the issue, list possible root causes, test each hypothesis, and only propose a fix after confirming the root cause. Do not guess.”

Test-Driven Development (6.2K installs)

Part of obra/superpowers. Comprehensive TDD patterns that teach the AI to write tests first, then implement code to pass them. Covers unit, integration, and edge case test patterns.

What it teaches:

• Write the failing test before the implementation
• Red-green-refactor cycle discipline
• Edge case identification and test coverage
• When to use mocks versus real dependencies

Code Review Skills

Requesting Code Review (5K installs)

Part of obra/superpowers. Dispatches a code-reviewer subagent that examines your changes with fresh context. The subagent categorizes findings as critical, suggestion, or nitpick and provides specific fix recommendations.

npx skills add obra/superpowers

Tip

Copy-paste prompt for AI code review:

“Review the changes on this branch compared to main. Use the requesting-code-review skill to dispatch a reviewer subagent. Categorize each finding as critical, suggestion, or nitpick. For critical findings, explain the risk and provide a fix. Ignore pure style issues.”

Code Reviewer — google-gemini/gemini-cli (1.6K installs)

Google’s code review automation skill supporting TypeScript, JavaScript, Python, Go, Swift, and Kotlin. Focused on language-specific idioms and best practices.

npx skills add google-gemini/gemini-cli

What it covers:

• Language-specific anti-patterns and idiomatic improvements
• Performance concerns per language runtime
• Security patterns relevant to each language ecosystem
• Constructive feedback with specific suggestions

Git Workflow Skills

Using Git Worktrees (4.5K installs)

Part of obra/superpowers. Teaches the AI to isolate feature work using git worktrees instead of stashing or switching branches. Each feature gets its own working directory, so you can have multiple features in progress without conflicts.

What it teaches:

• Creating worktrees for feature branches
• Managing multiple worktrees without confusion
• Cleaning up worktrees after merge
• When worktrees are better than branch switching

Finishing a Development Branch (3.8K installs)

Part of obra/superpowers. A complete integration workflow for merging feature branches: rebase on main, run the full test suite, resolve conflicts, update documentation, and create the merge commit with a proper message.

Yeet — openai/skills

Part of openai/skills (2,900 installs total). Stages, commits, pushes, and opens a PR in one flow. When you just need to ship the current changes without ceremony, this skill handles the entire git-to-PR pipeline in a single command.

npx skills add openai/skills

PR Creator — google-gemini/gemini-cli (481 installs)

Creates pull requests following your repo’s PR template. Reads the template from .github/PULL_REQUEST_TEMPLATE.md, fills in each section based on the actual changes, and submits the PR.

npx skills add google-gemini/gemini-cli

Tip

Copy-paste prompt for PR creation:

“Create a pull request for the changes on this branch. Follow the repo’s PR template. Include a summary of what changed, why it changed, how to test it, and any risks for reviewers. Link to the relevant ticket if one is mentioned in the branch name.”

GitHub and CI Skills

Address PR Comments — openai/skills

Part of openai/skills. Reads GitHub PR review comments and issue comments, then addresses them: fixes code issues, responds to questions, and marks conversations as resolved.

Fix CI — openai/skills

Part of openai/skills. Debugs and fixes failing GitHub Actions workflows. Reads the CI logs, identifies the failure, and proposes or applies a fix. Saves you from manually parsing through hundreds of lines of CI output.

Sentry — openai/skills

Part of openai/skills. Inspects Sentry issues and pulls error summaries. When a production error fires, this skill fetches the stack trace, context, and frequency data so the AI can start debugging with full information.

Linear — openai/skills

Part of openai/skills. Manages Linear issues, projects, and team workflows. Creates issues, updates status, assigns team members, and links issues to PRs.

Documentation Skills

Docs Writer — google-gemini/gemini-cli (686 installs)

Writing and reviewing documentation. Teaches the AI to write docs that developers actually read: clear structure, practical examples, and appropriate detail.

npx skills add google-gemini/gemini-cli

What it teaches:

• Lead with the “why,” then the “how”
• Use concrete examples, not abstract descriptions
• Write for scanners: headings, bullet points, code blocks
• Keep paragraphs short (3-4 sentences maximum)
• Include working code examples that can be copied directly

Tip

Copy-paste prompt for documentation generation:

“Generate documentation for the authentication module in src/lib/auth/. Read the code first, then use the docs-writer skill: start with what the module does and why someone would use it, show the main use cases with code examples, and end with configuration options and troubleshooting.”

Browser and Web Skills

browser-use/browser-use (24.4K installs)

Browser automation, testing, form filling, and screenshots. This skill gives the AI the ability to interact with actual web pages: navigate, click, fill forms, take screenshots, and verify UI state.

npx skills add browser-use/browser-use

Use cases:

• E2E testing without writing Playwright scripts manually
• Filling out forms to test validation logic
• Taking screenshots for documentation or bug reports
• Verifying deployed UI matches expectations

Playwright — openai/skills

Part of openai/skills. E2E testing patterns using Playwright. More structured than browser-use for teams that want repeatable, committed test suites rather than ad-hoc browser interaction.

squirrelscan/skills — Website Auditing (14.8K installs)

Audits websites for SEO, performance, accessibility, and security using over 230 rules. Run it against your staging URL before every release to catch issues the AI would otherwise miss.

npx skills add squirrelscan/skills

firecrawl/cli — Web Scraping (1.9K installs)

Web scraping, crawling, URL discovery, and structured data extraction. When the AI needs to fetch content from external sites, convert pages to structured data, or discover all URLs on a domain.

npx skills add firecrawl/cli

Design and Implementation Skills

Figma Design Implementation (755 installs)

Part of figma/mcp-server-guide (1,100 installs total). Translates Figma designs to code with 1:1 fidelity. Reads the Figma file, extracts layout, spacing, colors, and typography, then generates component code that matches the design.

npx skills add figma/mcp-server-guide

Create Design System Rules (184 installs)

Also part of figma/mcp-server-guide. Generates design system rules from your Figma components so the AI consistently uses the right tokens, spacing, and patterns across all generated code.

Skill Discovery

find-skills — vercel-labs/skills (154K installs)

The most-installed skill on the entire platform. Helps the AI discover and install other skills. When you describe a workflow need, this skill searches the registry and recommends relevant skills.

npx skills add vercel-labs/skills

Building Your Skill Stack

Recommended Starter Stack

For most teams, start with these three installs to cover planning, review, git workflow, and verification:

npx skills add obra/superpowers
npx skills add openai/skills
npx skills add google-gemini/gemini-cli

This gives you 50+ skills covering the full development lifecycle. The obra/superpowers collection alone (78K installs) handles planning, debugging, TDD, code review, git workflows, and verification.

Adding Project-Specific Conventions

After installing community skills, create a project-specific skill that captures your team’s unique conventions:

Cursor

Create .cursor/rules/team-conventions.md covering:

• Branch naming: type/TEAM-123-description
• PR reviewers: backend changes need @backend-team, frontend needs @frontend-team
• Deploy process: merge to main triggers staging, manual promotion to production
• Communication: post in #releases after each production deploy

Claude Code

Create .claude/skills/team-conventions/instructions.md covering:

• Branch naming: type/TEAM-123-description
• PR reviewers: backend changes need @backend-team, frontend needs @frontend-team
• Deploy process: merge to main triggers staging, manual promotion to production
• Communication: post in #releases after each production deploy

Codex

Add to .codex/instructions.md:

• Branch naming: type/TEAM-123-description
• PR reviewers: backend changes need @backend-team, frontend needs @frontend-team
• Deploy process: merge to main triggers staging, manual promotion to production
• Communication: post in #releases after each production deploy

Tip

Copy-paste prompt for convention extraction:

“Analyze the last 50 commit messages, the last 20 PR descriptions, and the README files in this repository. Identify recurring patterns in how the team communicates: commit message format, PR structure, documentation style, and naming conventions. Generate a team conventions skill file that codifies these patterns.”

When This Breaks

Skills from obra/superpowers make the AI too methodical. The brainstorming and planning skills are thorough by design. If you need a quick fix, be explicit: “Skip planning. Just fix the null check on line 42.” The AI will respect the override.

Verification skill blocks completion on flaky tests. The verification-before-completion skill runs your test suite before claiming success. If you have flaky tests, the AI will keep retrying. Either fix the flaky tests or tell the AI: “The payment integration test is flaky, ignore failures in that test file.”

Too many skills slow down context loading. Installing obra/superpowers, openai/skills, and google-gemini/gemini-cli adds significant context. If responses feel slow, check which skills are active and disable ones you do not need for the current task.

Parallel agent skills require careful scoping. The dispatching-parallel-agents and subagent-driven-development skills work best when tasks have clear boundaries. If tasks have hidden dependencies, agents will produce conflicting changes. Review your plan for task independence before dispatching.

Browser-use skill needs a running browser. The browser-use skill requires a browser instance. If you are running in a headless CI environment, ensure you have a headless browser configured. The skill will fail silently without one.

find-skills returns too many results. The vercel-labs find-skills skill searches broadly. Be specific in your query: “Find skills for Playwright E2E testing” works better than “Find testing skills.”

What’s Next

Building Custom Skills Create and publish skills tailored to your team's workflows.

Skills Best Practices Optimize your skill setup for maximum team impact.