ecc-base-template

A framework-agnostic GitHub template that ships an 18-agent development team for high-quality, high-precision collaboration with Claude Code.

日本語版 README はこちら

---

Prerequisites

This template is designed for developers who already have ECC (Everything Claude Code) installed at the user level (~/.claude/). The agents in this template reference ECC-provided rules and Skills at runtime — language-specific reviewers, framework patterns, and shared verification workflows live in your ECC install, not in this repository.

The template still runs without ECC, but agent quality is degraded: rule references resolve to nothing, Skill invocations are no-ops, and the language-specific code review path (see code-reviewer) falls back to a generic review and says so in its verdict.

New to ECC? Install it at the user level first, then return here to create your project repository from this template.

---

What you get

• 18 specialized agents covering the full product lifecycle — orchestrator, product-manager, architect, implementer, test-runner, code-reviewer, security-reviewer, performance-engineer, devops-engineer, technical-writer, and more. All ecosystem-agnostic: the agents detect your language and framework at runtime.
• Clean root directory. After forking you own the repo root — the template does not reserve docs/, scripts/, or any ADR/spec numbers.
• Document templates for ADRs and product specs at .claude/templates/, with English-first *.md and Japanese *.ja.md counterparts. Copy them wherever your project wants its decision records to live.
• Empty CI scaffold (.github/workflows/.gitkeep) so the folder is there waiting when you decide to add CI — see the CI section below.

---

Using this template

This repository is the fork payload itself: agents, document templates, skills, CI workflows, fork-facing settings, and the initializer script. "Use this template" (or git clone) copies all of it into your new repository. The template's own design history (ADRs, internal Specs) is not shipped — it lives in the upstream git history if you ever need to trace a decision.

---

Quick start

1. Create your repository

On GitHub, open b150005/ecc-base-template and click Use this template. (Use the main branch — that is the fork-clean default.)

2. Clone and open

git clone https://github.com/<your-username>/<your-repo>.git
cd <your-repo>

3. Run the initializer

.claude/init.sh

This prompts you for a project name, one-line description, and tech stack, then replaces the ## About This Project placeholder in .claude/CLAUDE.md and copies .env.example to .env. Re-running is safe.

Non-interactive form:

.claude/init.sh \
  --project-name "TaskFlow" \
  --description "Team task management API" \
  --stack "Go / Gin / PostgreSQL"

4. Start working

Open the repo in Claude Code (claude in the repo root) and give the orchestrator a real task. For example:

Design and implement a REST endpoint POST /tasks that validates input, persists to PostgreSQL, and returns the created resource. Use TDD.

The orchestrator delegates to product-manager for acceptance criteria, architect for the module boundaries, implementer for code, and the quality agents for review — you steer the hand-offs.

---

CI

This template ships no GitHub Actions workflows — .github/workflows/ contains only a .gitkeep. CI is a per-project choice; add workflows sized to your stack (build, test, lint, coverage, security scanning) when you are ready. dependabot.yml is included with github-actions updates enabled — extend it with your language ecosystem.

---

The 18-agent team

All agents are ecosystem-agnostic. They detect the project's language and framework at runtime by reading .claude/CLAUDE.md and your project's manifest files (package.json, pubspec.yaml, go.mod, Cargo.toml, etc.). The orchestrator coordinates the team; specialists are invoked by the orchestrator or directly.

Agent	Phase	Role
orchestrator	All	Analyzes issues, plans work, delegates to specialists
product-manager	Planning	Spec authoring, user stories, acceptance criteria
market-analyst	Planning	Market research, competitor analysis
monetization-strategist	Planning	Business model, pricing, revenue analysis
ui-ux-designer	Design	UI/UX design, usability review, accessibility
docs-researcher	Research	API verification, framework behavior vs. primary docs
research-critic	Research	Adversarial review of research outputs (primary-source-only)
adversarial-implementer	Build	Parallel-implementation Critic (opt-in)
architecture-critic	Design	Counter-proposal Critic for ADRs (opt-in)
architect	Design	System architecture, technology decisions, ADR creation
implementer	Build	Code implementation following architecture specs and TDD
code-reviewer	Quality	Code quality, maintainability, standards adherence
test-runner	Quality	Test execution, coverage reporting, TDD support
linter	Quality	Static analysis and code style enforcement
security-reviewer	Quality	Vulnerability detection, secret scanning, OWASP Top 10
performance-engineer	Quality	Profiling, bottleneck identification, optimization
devops-engineer	Release	CI/CD, deployment strategy, release management
technical-writer	Release	Documentation, changelog, bilingual docs

Model tiers

Each agent declares its model in frontmatter using a Claude Code alias (opus / sonnet / haiku / inherit), which resolves to the latest version in that family. The template ships a mixed fleet — the right model for the job rather than a single floor.

For the current version numbers, see the Anthropic model overview.

---

Project structure (after forking)

your-repo/
├── README.md                  ← your project's README (replace this one)
├── README.ja.md               ← optional bilingual README
├── CHANGELOG.md               ← starts at [Unreleased]; grows with your releases
├── LICENSE
├── .env.example
├── .gitignore                ← Learning Mode opt-in inversion documented inline
├── .gitattributes
├── .claude/
│   ├── CLAUDE.md              ← project instructions (edit the About section first)
│   ├── agents/                ← 18 agent definition files
│   ├── templates/             ← copy-and-fill ADR/spec/roadmap/progress templates
│   ├── skills/                ← Learning Mode, verification-layer, authoring, compliance
│   ├── meta/references/       ← reference docs the agents and Skills cite
│   ├── meta/scripts/          ← CI helper scripts
│   ├── hooks/                 ← coaching-context hook (Learning Mode)
│   ├── output-styles/         ← ecc-learn output style (opt-in)
│   ├── init.sh                ← interactive initializer
│   └── settings.json          ← Claude Code settings (Plan Mode default)
└── .github/
    ├── workflows/             ← CI workflows (most default-off; opt in per feature)
    ├── CODEOWNERS
    ├── ISSUE_TEMPLATE/
    ├── PULL_REQUEST_TEMPLATE.md
    └── dependabot.yml

You own every visible root file. The template does not reserve docs/, src/, scripts/, or any other top-level directory name.

Placing your own ADRs and specs

Copy .claude/templates/adr-template.md to wherever you want your ADRs to live. Common choices:

• adr/001-use-postgresql.md at the repo root
• adr/en/001-use-postgresql.md + adr/ja/001-use-postgresql.md for bilingual projects
• docs/adr/001-use-postgresql.md if you already have a docs/ tree

The same applies to spec-template.md. There is no forced location.

---

Contributing to the template

This section is for people working on ecc-base-template itself, not for fork users. Open issues and PRs against main. The template's own design history is recorded in the git log rather than in shipped ADR files.

---

License

MIT