<div align="center">

🧠 OpenSkills

49 production-grade AI engineering skills for Markdown-aware agents.

github.com/CODE-SAURABH/OpenSkills

</div>

---

openskills is an open-source collection of 49 production-grade AI agent skills. The skills are plain Markdown and can be used with any agent that lets you load instructions.

Each skill is a single SKILL.md file that gives your AI agent opinionated, expert-level instructions for a specific engineering role — from system design and backend dev to security, QA, and retrospectives. Your agent must be configured to load the file or rule; copying a file alone does not guarantee that every product will activate it.

Start with one skill, confirm that your agent loads it, then add the rest you need.

---

Table of Contents

• Why OpenSkills?
• Quick Install
• The Sprint Flow
• Skills
• Agent Compatibility
• How Skills Work
• Contributing
• License

---

Why OpenSkills?

Most AI coding assistants are generalists. They write code, but they don't ship like a senior engineer. They don't ask the hard questions before building, they don't catch security holes, they don't enforce architecture decisions, and they don't know when to stop.

OpenSkills fixes this. Each skill turns your agent into a domain expert with:

• ✅ Explicit trigger conditions — the agent knows when to activate
• ✅ Numbered execution workflows — step-by-step, not a pile of tips
• ✅ Output templates — structured, consistent deliverables every time
• ✅ Definition of Done checklists — verifiable completion criteria
• ✅ Edge cases and anti-patterns — the things juniors miss

---

Quick Install

Claude Code

Clone this repository, then copy the individual skill folder you want into Claude Code's skills directory. Do not copy the entire repository: each installed skill should contain its own SKILL.md file.

git clone https://github.com/CODE-SAURABH/OpenSkills.git openskills
cd openskills

For example, install the code-reviewer skill:

mkdir -p ~/.claude/skills
cp -r code-reviewer ~/.claude/skills/code-reviewer

To install another skill, substitute its directory name:

cp -r <skill-name> ~/.claude/skills/<skill-name>

Start a new Claude Code session after copying the skill. It will then be available for use.

Codex

Codex discovers skills in .agents/skills/ within a repository and in ~/.agents/skills/ for your personal, cross-project skills. Copy a selected OpenSkills folder to one of those locations.

# Install for every project you use with Codex
mkdir -p ~/.agents/skills
cp -r code-reviewer ~/.agents/skills/code-reviewer

# Or, from a target project's root, install for that project only
mkdir -p .agents/skills
cp -r /path/to/openskills/code-reviewer .agents/skills/code-reviewer

Codex detects skill changes automatically. Mention the skill explicitly with $code-reviewer when you want to ensure it is used, or let Codex choose it when the task matches the skill description. See the Codex skills documentation for details.

Antigravity

Antigravity uses the same workspace skill layout: <workspace-root>/.agents/skills/<skill-name>/SKILL.md. From the root of the workspace you opened in Antigravity, copy the skill you want:

mkdir -p .agents/skills
cp -r /path/to/openskills/code-reviewer .agents/skills/code-reviewer

Start a new conversation after adding the skill. For personal skills shared across Antigravity workspaces, use ~/.gemini/config/skills/<skill-name>/ instead. See the Antigravity skills documentation.

Custom agents

Use the same folder structure when your agent supports the Agent Skills format:

<agent-config-root>/skills/
└── code-reviewer/
    └── SKILL.md

Configure the agent to list the available SKILL.md files at session start and load the selected file when its description matches the task. If your agent does not support skill discovery, read the chosen SKILL.md and pass its contents as the agent's system instructions or task context.

Other agents

For Cursor, Windsurf, GitHub Copilot, Kiro, Codex, and other agents, use that product's documented instruction mechanism and add only the skills that fit your workflow. A safe universal approach is to provide the contents of <skill-name>/SKILL.md as agent instructions or context. Do not concatenate all 49 skills into one instruction file: they can conflict and consume unnecessary context.

---

The Sprint Flow

OpenSkills maps to a complete engineering sprint. Use the right skill at each phase and hand off deliberately.

| Phase | Skill(s) | What Happens | |---|---|---| | 💡 Think | office-hours · plan-ceo-review · autoplan | Stress-test the idea. CEO validates scope. AutoPlan chains all reviews automatically. | | 📋 Plan | spec-author · system-design · product-engineer · plan-eng-review · plan-design-review · doc-coauthoring | Turn validated ideas into unambiguous specs and architecture blueprints. | | 🏗️ Build | backend-dev · frontend-design · api-design · database-architect · ai-engineer · data-engineer · mcp-builder · devops-platform · security-engineer · performance-optimizer · prompt-engineer · design-explorer · brand-guidelines · theme-factory · pptx · pdf · xlsx · safety-guardrails | Full-stack implementation across every layer. | | 🔍 Review | code-reviewer · devex-engineer · research-agent | Catch bugs, security holes, and regressions before they ship. | | 🧪 Test | test-writer · qa-engineer · debugger · benchmark · webapp-testing | Write tests, validate behaviour, find what's hiding. | | 🚀 Ship | technical-writer · commit-message · release-engineer · land-and-deploy · sre-canary · internal-comms | Document, commit, release, monitor, and communicate. | | 🔄 Reflect | retro-engineer · agent-memory · skill-creator | Learn from what happened. Improve the system itself. | | 🌐 Deploy | forward-deployment-engineer | Customer-site deployment and field technical delivery. |

---

Skills

All 49 skills, ready to use.

| Skill | Role | Description | Phase | |---|---|---|---| | java-development | Java Engineer | Production Java services, APIs, persistence, concurrency, resilience, and JVM-safe delivery. | Build | | python-development | Python Engineer | Production Python applications, services, typing, asyncio, testing, and reliable I/O. | Build | | react-development | React Engineer | Accessible production React interfaces, state, data flows, rendering, and testing. | Build | | typescript-development | TypeScript Engineer | Type-safe TypeScript applications, Node.js services, runtime validation, and packaging. | Build | | agent-memory | Memory Manager | Persists learnings across sessions, runs browser-based QA with real Chromium, and coordinates multi-agent workflows. | Reflect | | ai-engineer | AI/ML Engineer | LLM integration, RAG pipelines, embedding stores, fine-tuning, and AI evaluation frameworks. | Build | | api-design | API Designer | Design production-grade REST, GraphQL, gRPC, and WebSocket APIs with versioning, error contracts, and OpenAPI schemas. | Build | | autoplan | Review Orchestrator | Chains CEO → Design → Eng → DX reviews automatically. Auto-decides resolvable questions; surfaces only human taste-gates. | Think | | backend-dev | Backend Engineer | Production-grade server-side code across any language or framework — auth, business logic, data access, and more. | Build | | benchmark | Benchmark Auditor | Baseline and compare page load times, Core Web Vitals, bundle sizes, and API response times. | Test | | brand-guidelines | Brand Compliance | Apply brand voice, tone, colors, typography, and visual style consistently across any artifact. | Build | | code-reviewer | Code Reviewer | 6-pass code review for bugs, security, performance, and maintainability before shipping. | Review | | commit-message | Commit Author | Conventional Commit messages and PR descriptions from staged changes or diffs. | Ship | | data-engineer | Data Engineer | ETL/ELT pipelines, streaming architectures, data quality gates, and analytical data modelling. | Build | | database-architect | Data Modeller | Schema design, indexing strategy, migration plans, and query optimisation across SQL and NoSQL. | Build | | debugger | Debugger | Systematic root-cause debugging. Iron law: no fixes without investigation first. 4-phase workflow. | Review | | design-explorer | Design Strategist | Generative design exploration — multiple directions before committing, with clear trade-off framing. | Build | | devex-engineer | DX Engineer | Developer experience audit — TTHW measurement, friction tracing, competitor benchmarking. | Review | | devops-platform | DevOps Engineer | CI/CD pipelines, containerisation, IaC, cloud architecture, and deployment automation. | Build | | doc-coauthoring | Doc Co-Author | Structured co-authoring for PRDs, design docs, RFCs, and decision docs through three guided stages. | Plan | | forward-deployment-engineer | Field Engineer | Enterprise deployment support — integration reviews, customer onboarding, and field technical guidance. | Deploy | | frontend-design | Frontend Designer | Distinctive, intentional UI — typography, layout, and design that doesn't look like a template. | Build | | internal-comms | Comms Writer | Incident reports, post-mortems, launch announcements, 3P updates, and leadership briefings. | Ship | | land-and-deploy | Deploy Orchestrator | Merges the PR, monitors CI/CD, waits for deploy, verifies production health, and confirms the feature is live. | Ship | | mcp-builder | MCP Server Builder | Build production-grade MCP servers that connect AI agents to external APIs, databases, and services. | Build | | office-hours | Idea Validator | YC-style pre-code interrogation. Stress-tests assumptions, produces a design doc, prevents building the wrong thing. | Think | | pdf | PDF Engineer | Read, extract, create, merge, split, and manipulate PDF files — including OCR and form filling. | Build | | performance-optimizer | Performance Engineer | Profiling, bottleneck identification, query optimisation, caching strategy, and load test design. | Build | | plan-ceo-review | Strategic Reviewer | Pressure-tests scope, value, and prioritisation decisions before any architectu