A production-ready, multi-agent AI coding setup you can drop into any project. Stop configuring AI tools from scratch. Start from a battle-tested baseline that makes every coding agent smarter, safer, and more consistent from day one.
👉 Click here to generate your project from this template 👈 | Jump to Quick Start 🚀
Most developers configure their AI coding tools ad-hoc — a few lines in a CLAUDE.md, maybe a Copilot instruction file. The result: agents that hallucinate project conventions, make unsafe mutations, break existing patterns, and need constant correction.
This template encodes production-ready agentic development practices into a structured, multi-layered instruction system. It covers:
- Consistent coding standards enforced across all agents
- Security-first rules (XSS guards, escaping, forbidden APIs)
- Safe AI orchestration patterns (no hanging loops, MCP-first, strict timeouts)
- Pre-built skills and workflows for the most common engineering tasks
- Documentation drift prevention built in from the start
- Git hook integration for automated quality gates
Clone it, fill in your project details, and every AI tool working on your codebase will follow the same high-quality standards.
Whether you are a solo indie hacker using Cursor, or an enterprise team orchestrating GitHub Copilot and Google Jules, this template standardizes your AI interactions.
This template provides optimized configurations for all major AI coding tools:
| Agent | Config files | Key capability |
|---|---|---|
| Claude Code (Anthropic) | CLAUDE.md, .claude/rules/, .claude/agents/, .claude/skills/ |
Full skill system, path-scoped rules, subagents, hooks |
| GitHub Copilot (VS Code) | .github/copilot-instructions.md, .github/prompts/, .github/instructions/ |
Persistent instructions, prompt files, scoped instructions |
| Cursor | AGENTS.md |
Project-wide context via rules file |
| OpenAI Codex | AGENTS.md |
Project-wide context |
| Kilo Code | AGENTS.md, .kilocode/rules/ |
AGENTS.md context + dedicated project rules (priority 2) |
| Google Jules | AGENTS.md (read natively), README.md (env setup fallback) |
Autonomous async task execution; Jules reads AGENTS.md automatically on every task |
| Google Gemini / Antigravity | GEMINI.md, .agent/rules/, .agent/workflows/, .agent/skills/ |
Always-on rules, reusable workflow definitions, skill library |
Other tools — Windsurf, Aider, Amazon Q Developer, and Devin can leverage
AGENTS.md(widely supported) plus.agent/rules and workflows. See docs/CUSTOMIZATION-GUIDE.md for adaptation tips.
One template, every major agent, zero duplication.
.
├── CLAUDE.md ← Claude Code — always-on project instructions
├── AGENTS.md ← Copilot / Cursor / Codex / Jules — always-on
│
├── .github/
│ ├── copilot-instructions.md ← VS Code Copilot Chat — persistent instructions
│ ├── prompts/
│ │ ├── new-feature.prompt.md ← /new-feature — structured planning before coding
│ │ └── review-changes.prompt.md ← /review-changes — pre-commit rule compliance
│ ├── instructions/
│ │ ├── docs.instructions.md ← Scoped: docs/**/*.md — documentation rules
│ │ └── qa-audit.instructions.md ← Scoped: audit files — QA audit process
│ └── pull_request_template.md ← PR quality-gate checklist
│
├── .claude/
│ ├── rules/ ← Path-scoped always-on rules for Claude
│ │ ├── coding-standards.md
│ │ └── markdown-quality.md
│ ├── agents/
│ │ └── qa-reviewer.md ← Claude subagent: automated code reviewer
│ └── skills/ ← Claude skill library (16 skills)
│ ├── brainstorming/
│ ├── doc-audit/
│ ├── feature-delivery/
│ ├── frontend-design/
│ ├── hotfix/
│ ├── qa-audit/
│ ├── release-readiness/
│ ├── requesting-code-review/
│ ├── skill-creator/
│ ├── systematic-debugging/
│ ├── writing-plans/
│ └── ...
│
├── .agents/skills/ ← Universal skill library (16 skills, works with all tools)
│
├── .kilocode/
│ └── rules/
│ └── coding-standards.md ← Kilo Code project rules (priority 2, above AGENTS.md)
│
├── .agent/
│ ├── rules/ ← Gemini/Antigravity/Jules always-on rules
│ ├── skills/ ← Jules/Gemini skill library
│ ├── workflows/ ← Reusable workflow procedures
│ │ ├── doc-audit.md
│ │ ├── feature-delivery.md
│ │ ├── qa-audit.md
│ │ ├── release-readiness.md
│ │ ├── architecture-change.md
│ │ ├── hotfix.md
│ │ └── pr-analysis.md
│ └── scripts/ ← Jules API helpers, PR analysis automation
│
├── scripts/ ← Optional local adoption tools
│ ├── check-fill-markers.sh ← Scan for [FILL:] placeholder regressions (needs a baseline file)
│ ├── check-agent-context-sync.py ← Verify project-context.json is in sync across agent files
│ ├── template-health-report.py ← Generate docs/TEMPLATE-HEALTH.md metrics
│ └── check-all.sh ← Run all available checks in one command
│
├── project-context.example.json ← Optional: canonical project metadata for sync checks
│
└── docs/
├── ARCHITECTURE.md ← Why the template is structured the way it is
├── CUSTOMIZATION-GUIDE.md ← How to fill every [FILL:] marker, file by file
├── SKILLS-CATALOG.md ← Every skill: what it does, when to use it, invocation
├── TEMPLATE-HEALTH.md ← Auto-generated: placeholder counts and skill stats
├── EXAMPLE-PROFILE.md ← Reference: what a fully configured setup looks like
└── ai-coding-agents-best-practices.md ← Full reference: Claude Code, Copilot, Kilo Code, Antigravity
Click "Use this template" at the top right of this repository, or click the link below to generate it instantly:
👉 Click here to generate your project from this template 👈
git clone https://github.com/florian101010/awesome-agentic-AI-coding-template.git
cp -r awesome-agentic-AI-coding-template/. your-project/
rm -rf your-project/.gitLet the agent analyze your codebase and generate filled-in versions of every instruction file automatically.
In Claude Code — invoke the skill directly:
/setup-repo
In any other agent — bundle the repo first, then paste the prompt:
npx repomix # produces repomix-output.xmlThen open your agent (Copilot /setup-repo, Cursor, Gemini, etc.), attach the repomix output, and invoke the skill. The agent reads the codebase, infers constraints, and outputs ready-to-paste content for all 8 instruction files.
Step 1 — Find all placeholders. Every customization point is marked with [FILL: ...]:
grep -rn "\[FILL:" . --include="*.md" | grep -v READMEStep 2 — Fill in the most important files:
| File | What to provide |
|---|---|
CLAUDE.md |
Project name, key constraints, prohibited patterns, test command |
AGENTS.md |
Project name, architecture overview, file structure |
.github/copilot-instructions.md |
Always-do / never-do rules, key file table |
.claude/agents/qa-reviewer.md |
Project-specific review checklist items |
.agent/rules/coding-standards.md |
Architecture constraints, stack-specific rules |
Step 3 — Remove what you don't need. Sections like "Immutable Contract" or "Layout Constraints" are designed to be deleted for projects that don't need them.
Note: The
[FILL: e.g. "npm test"]placeholders in agent files refer to your project's test commands. This template ships no runtime code — fill in the commands that match your stack (e.g.npm test,pytest,cargo test).
Step 4 — Activate Git hooks (optional but recommended):
git config core.hooksPath .githooks
chmod +x .githooks/pre-commit .githooks/commit-msgStep 5 — Set up secrets (required for Jules AI scripts):
cp .env.example .env
# Then edit .env and add your JULES_API_KEYThe .env.example documents every variable the scripts expect. .env is git-ignored — never commit real keys.
Step 6 — Enable cross-file consistency checks (optional):
cp project-context.example.json project-context.json
# Edit project-context.json — set project_name, description, tech_stack, and commandsOnce project-context.json exists, running bash scripts/check-all.sh will verify that these values are present in all four agent instruction files (AGENTS.md, CLAUDE.md, GEMINI.md, .github/copilot-instructions.md). Useful for catching drift when one file is updated but the others are not.
Done. Your AI agents now operate with full project context.
Every skill is ready to invoke. Universal skills need no customization; project-specific ones have [FILL:] markers.
| Skill | Invoke | What it does |
|---|---|---|
| Setup Repo | /setup-repo |
AI-powered setup — analyze codebase and auto-fill all agent instruction files |
| Brainstorming | /brainstorming |
Structured ideation — explores user intent and constraints before building |
| Systematic Debugging | /systematic-debugging |
Root-cause analysis — never guess, always trace |
| QA Audit | /qa-audit |
9-phase security + robustness audit: XSS, rule compliance, race conditions |
| Doc Audit | /doc-audit |
Documentation drift detection across all instruction files and project docs |
| Feature Delivery | /feature-delivery |
End-to-end feature workflow: planning → implementation → QA → docs sync |
| Release Readiness | /release-readiness |
Pre-release quality gate with automated + manual checks |
| Hotfix | /hotfix |
Emergency fix workflow with risk assessment |
| Requesting Code Review | /requesting-code-review |
Structured review request with context |
| Writing Plans | /writing-plans |
Spec-to-plan conversion before any code is written |
| Frontend Design | /frontend-design |
Production-grade UI components — avoids generic AI aesthetics |
| Architecture Change | /architecture-change |
Controlled architecture refactor with ADR requirement |
| Skill Creator | /skill-creator |
Create new custom skills for your project |
| Find Skills | /find-skills |
Discover all available skills |
| Using Superpowers | /using-superpowers |
Skill discovery on session start |
Workflows are detailed, multi-phase procedures that agents follow step by step. Unlike skills (invoked interactively), workflows are loaded by agents working through a structured task.
| Workflow | Purpose |
|---|---|
doc-audit.md |
Comprehensive docs drift audit across all layers |
qa-audit.md |
Full-scope quality audit with 9 structured phases |
feature-delivery.md |
Feature planning → implementation → QA → docs sync |
release-readiness.md |
Pre-release quality gate |
architecture-change.md |
ADR-gated architecture refactors |
hotfix.md |
Emergency fix with risk assessment |
pr-analysis.md |
Pull request quality review |
post-merge-analysis.md |
Post-merge verification |
These rules are active for all agents on day one — no setup required.
- No hanging bash loops — agents iterate with individual tool calls
- MCP Server Tools over raw CLI calls (GitHub, databases, APIs)
- Every subprocess call has a strict timeout (
{ timeout: 10000 }) - No
eval(),document.write(), or stateful browser APIs
- Config files get a structured header comment explaining their schema
- Code comments in English, always
- UTF-8 without BOM on all files
- Markdown tables: aligned
| --- | --- |separator style - Every fenced code block includes a language specifier
- Single H1 per file, logical heading hierarchy, unique subheadings
- No duplicate subheadings within a document
- Repo-local hooks via
.githooks/ - Commit format enforced:
type(optional-scope): description - Pre-commit quality gates before every commit
All project-specific content uses a consistent placeholder convention:
[FILL: project name]
[FILL: describe your tech stack constraints here]
[FILL: e.g. "npm test" or "pytest" or "cargo test"]
This makes it trivially easy to scan for unfinished configuration:
grep -rn "\[FILL:" . --include="*.md"Inline setup annotations use a different marker so they're visually distinct:
<!-- SETUP: Replace this section with your project's file structure -->1. Multi-layered — Each AI tool gets its own config file in its native format. Claude reads CLAUDE.md and .claude/; Copilot reads .github/copilot-instructions.md; Jules reads AGENTS.md natively; Antigravity reads .agent/. One source of truth, delivered in every dialect.
2. Universal + project-specific — Universal rules (security, orchestration, markdown quality) ship ready-to-use. Project-specific rules are clearly marked with [FILL: ...] and isolated from the universal content.
3. Prevention over correction — Rules are designed to stop agents from making dangerous changes — not to catch mistakes after the fact. Security guards and forbidden patterns are enforced upfront.
4. Drift-free — Documentation drift is a leading cause of agent misbehavior. The doc-audit skill and workflow track every file that must stay in sync and surface drift proactively.
5. Skill-based — Complex multi-step tasks (audits, feature delivery, debugging) are codified as invocable skills. Skills are reusable, composable, and version-controlled alongside your code.
| Doc | What it covers |
|---|---|
| ARCHITECTURE.md | Why the template is structured the way it is — design decisions, directory map, layering principles |
| CUSTOMIZATION-GUIDE.md | How to fill every [FILL:] marker, file by file, with good/bad examples |
| SKILLS-CATALOG.md | Every skill: what it does, when to invoke it, which agents support it |
| ai-coding-agents-best-practices.md | Deep reference guide for Claude Code, Copilot, Kilo Code, and Antigravity |
Contributions are welcome. See CONTRIBUTING.md for guidelines.
Good contribution ideas:
- New universal skills (performance auditing, accessibility checking, etc.)
- Agent-specific optimizations as new tools emerge
- Improved workflow definitions
- Bug fixes and clarity improvements to existing skills
MIT — see LICENSE.
This template builds on top of the work of the teams and individuals behind these tools and projects:
AI coding agents:
| Tool | Repo / Docs |
|---|---|
| Claude Code (Anthropic) | github.com/anthropics/claude-code |
| GitHub Copilot (Microsoft) | docs.github.com/copilot |
| Google Jules | jules.google/docs |
| Google Gemini / Antigravity | github.com/google-gemini/gemini-cli |
| Cursor | cursor.com |
| OpenAI Codex | platform.openai.com/docs/codex |
| Kilo Code | github.com/Kilo-Org/kilocode |
Skills & standards:
| Project | What it contributed |
|---|---|
| obra/superpowers by Jesse Vincent (@obra) | using-superpowers, find-skills, systematic-debugging skills |
| skills.sh | frontend-design, skill-creator skills + the skills marketplace ecosystem |
| Agent Skills Open Standard | The cross-agent SKILL.md format used throughout |
Stop configuring. Start building.
Clone once. Use everywhere.
