Agentic Coding Template

A GitHub repository template for building software with AI coding agents.

Plan projects, implement features, review code independently, and verify changes with a repeatable workflow. You choose which agents and models do the work, and you keep the important decisions.

It works with Claude Code and Codex, and with any programming language or framework.

Get started View on GitHub Explore the workflow

How it works

  1. 1. Plan

    Clarify the requirements, define the architecture, and approve a roadmap.

    You approve the plan

  2. 2. Build

    An AI agent implements a feature in its own Git branch and worktree.

  3. 3. Verify and review

    Run the checks, review the change independently, and address the findings.

  4. 4. Deliver

    Open a pull request with the recorded verification and review evidence.

    An authorized maintainer merges

Each feature goes through build, review, and delivery. The project plan can be updated when requirements change.

Agents handle much of the development work. You stay in control of requirements, important approvals, and how changes are merged.

Explore the detailed workflow

What's inside the template

The template gives you a ready-to-use repository structure for working with AI coding agents.

It includes instructions for agents, configurable roles and models, scripts for planning and development, independent code reviews, and automated verification.

Everything runs from your own repository, without a separate orchestration platform.

Explore the project structure · Get started

agentic-coding-template/

  • AGENTS.mdShared instructions for all agents
  • .agents/Agent configuration, roles, and policies
    • agents.confWhich agent and model has each role
    • prompts/The roles: planning, implementation, review
    • policies/Permissions and safeguards
    • schemas/The format of review records
    • template.confWhich files a project receives from the template
  • scripts/Workflow automation
    • start-planning.shProject planning
    • start-feature.shFeature implementation
    • review-feature.shIndependent review
    • run-feature.shUnattended feature execution
    • verify.shThe project's checks, with one command
    • report-runs.shWhat your runs took: time, tokens, rounds
    • analyze-runs.shProposals drawn from several runs
    • doctor.shChecks your local setup
  • docs/Requirements, architecture, roadmap
    • PROJECT_REQUIREMENTS.mdWritten and approved during planning
    • architecture.mdWritten during planning
    • roadmap.mdThe features, in order
    • decisions/Records of important technical decisions
    • development.mdEvery command and option
  • .github/CI and GitHub configuration
    • workflows/ci.ymlRuns the project's checks on every pull request
    • ISSUE_TEMPLATE/The form of a feature issue
  • tests/Tests of the workflow scripts

A simplified view of the repository; select a folder to open it. The requirements, the architecture, and the roadmap are written during planning.

Why use it

Clear requirements

A round of questions and your approval come before any plan, so agents work toward a goal you agreed to.

Independent reviews

A separate, read-only agent reviews every change, which gives you a second perspective on the code.

Meaningful verification

One command runs your project's checks and records what was checked. If the code changes, the checks run again.

Traceable changes

Planning, GitHub issues, review decisions, and pull requests stay connected, so you can see why a change was made.

How the workflow flags risky changes, and what you configure in GitHub to enforce approvals, is in Configuration and security.

Get started

Create a repository from the template, configure your coding agents, and check your setup with ./scripts/doctor.sh. Then you are ready to plan your first project.

Create a repository Setup guide Explore the workflow

What the template does not try to do

Explore further

The workflow

Every step from requirements to a reviewed pull request, with its script and its approvals.

Configuration and security

Agent roles and models, what an agent may do, and what GitHub has to enforce.

Research and design choices

Why we built it, the frameworks and practices that influenced it, and the ideas we left out.

History

How the workflow developed and which changes made it more reliable.

Possible improvements

Features we are considering, such as better agent isolation and smarter orchestration.

Worked example

A sample feature from planning through implementation, review, and merge.