33 engineering skills for Claude Code, optimized for solo developers building AI systems and multi-service applications.
Install: /plugin install curated-skills in Claude Code (or add this repo as a plugin).
| Skill | Category | Trigger | Source |
|---|---|---|---|
| spec-driven-development | engineering-process | "spec", "design doc" | addyosmani |
| planning-and-task-breakdown | engineering-process | "plan", "break down" | addyosmani |
| incremental-implementation | engineering-process | "incremental", "step by step" | addyosmani |
| shipping-and-launch | engineering-process | "ship", "launch", "deploy" | addyosmani + gstack |
| debugging-and-error-recovery | engineering-process | "debug", "what's wrong" | addyosmani + karpathy |
| performance-optimization | engineering-process | "optimize", "slow", "performance" | addyosmani |
| investigate | engineering-process | "debug", "bug", "root cause" | gstack (rewritten) |
| code-review-and-quality | code-quality | "review", "code review" | addyosmani |
| code-simplification | code-quality | "simplify", "clean up" | addyosmani |
| security-and-hardening | code-quality | "security", "hardening" | addyosmani + affaan-m |
| api-and-interface-design | code-quality | "API design", "interface" | addyosmani |
| pr-review | code-quality | "review PR", "code review" | gstack (rewritten) |
| test-driven-development | testing | "TDD", "tests first" | addyosmani |
| browser-testing-with-devtools | testing | "test UI", "browser test" | addyosmani + anthropics |
| agent-eval | testing | "eval", "test agent" | affaan-m (rewritten) |
| qa | testing | "QA", "test", "find bugs" | gstack (rewritten) |
| ci-cd-and-automation | infrastructure | "CI/CD", "automation" | addyosmani |
| git-workflow-and-versioning | infrastructure | "git", "version" | addyosmani |
| docker-patterns | infrastructure | "Docker", "containerize" | affaan-m (rewritten) |
| fastapi-patterns | infrastructure | "FastAPI", "async Python" | affaan-m |
| frontend-ui-engineering | frontend | "UI", "React", "frontend" | addyosmani |
| design-system | frontend | "design system", "components" | affaan-m (rewritten) |
| context-engineering | ai-and-agents | "context", "prompts" | addyosmani |
| mcp-builder | ai-and-agents | "MCP", "tools" | anthropics |
| autonomous-loops | ai-and-agents | "loops", "agent retry" | affaan-m (rewritten) |
| deep-research | ai-and-agents | "research", "understand" | affaan-m (rewritten) |
| llm-behavior-guidelines | ai-and-agents | "LLM behavior", "agent mistakes" | karpathy (synthesized) |
| cost-aware-llm-pipeline | ai-and-agents | "cost", "budget", "LLM pricing" | affaan-m |
| foundation-models-on-device | ai-and-agents | "iOS", "on-device", "Apple" | affaan-m |
| pytorch-patterns | ai-and-agents | "PyTorch", "ML" | affaan-m |
| documentation-and-adrs | documentation | "docs", "ADR", "document" | addyosmani |
| source-driven-development | documentation | "read source", "understand code" | addyosmani |
| api-finder | (root) | "find API", "search APIs" | existing |
Workflows for planning, implementing, shipping, and debugging. Core patterns: spec-driven development, incremental implementation, root cause investigation.
Code review, security hardening, simplification, and API design. Focus: catching bugs, security issues, and scope drift before merge.
TDD, browser testing, agent evaluation, and QA. Focus: systematic testing, regression prevention, and behavior validation.
CI/CD, git workflows, Docker containerization, and FastAPI patterns. Focus: reproducible builds, multi-architecture support, automation.
UI engineering and design systems. Focus: component libraries, design tokens, consistency at scale.
Context engineering, MCP tools, autonomous loops, agent evaluation, and behavior guidelines. Focus: building reliable, safe, and efficient agent systems.
ADRs, documentation patterns, and source-driven development. Focus: capturing decisions and knowledge for future maintainers.
All skills follow the Agent Skills specification with a consistent 6-section format:
- Overview — what the skill does and why it matters
- When to Use — triggering conditions and when NOT to use
- Core Workflow — step-by-step process
- Common Rationalizations — anti-patterns table (the most valuable section)
- Red Flags — warning signs the skill is being misapplied
- Verification — checklist before declaring done
The Anti-Rationalizations section is the most important—read it first to understand common pitfalls and how to avoid them.
- addyosmani/agent-skills (MIT) — 14 skills, verbatim copies. Production-grade engineering patterns.
- garrytan/gstack — 3 skills (investigate, pr-review, qa), conceptually extracted and rewritten for standalone use.
- affaan-m/everything-claude-code — 7 skills (4 verbatim copies: fastapi-patterns, cost-aware-llm-pipeline, foundation-models-on-device, pytorch-patterns; 3 rewritten: agent-eval, docker-patterns, autonomous-loops, deep-research, design-system).
- anthropics/skills — 1 skill (mcp-builder), with added Anti-Rationalizations and Verification sections.
- multica-ai/andrej-karpathy-skills — 1 skill synthesized (llm-behavior-guidelines), converted from flat guidelines to 6-section format.
This repository prioritizes:
- Quality over quantity — 33 curated skills, not 229 unfiltered ones
- Portability — no platform dependencies (no gstack binaries, no infrastructure lock-in)
- Comprehensiveness — covers all major domains (process, quality, testing, infra, frontend, AI/agents, docs)
- Deduplication — no duplicate skills; where multiple repos have similar content, the best version is included
- Solo developer focus — skills relevant to building multi-service AI systems (Athena context: Mac Mini + Raspberry Pi deployment patterns)
This repository is curated and not accepting external PRs. To propose a new skill:
- Fork one of the source repos above
- Submit changes to the original repository
- Cross-link your skill to this collection if it's a good fit
Skills are sourced from:
- addyosmani/agent-skills — MIT License
- garrytan/gstack — License varies (research); concepts extracted under fair use for education/reference
- affaan-m/everything-claude-code — License varies (Apache 2.0 primary)
- anthropics/skills — Apache 2.0
- multica-ai/andrej-karpathy-skills — MIT License
Synthesized and rewritten skills in this repository are released under MIT License. See individual skill files for original source and license.
-
Install the plugin:
# In Claude Code /plugin install curated-skills -
Trigger a skill by phrase:
"I need to debug this issue" → invokes investigate "Let's write tests first" → invokes test-driven-development "Ship this to production" → invokes shipping-and-launch -
Read anti-rationalizations first: Every skill has a "Common Rationalizations" section. This is where the value is—it lists common mistakes and how to avoid them.
Found an error? Want to suggest a skill? Open an issue at the original source repositories: