Claude Code Installation

Claude Code Installation

5 minutes to install ai-coding-ok in your project via Claude Code.

---

Prerequisites

• Claude Code installed and authenticated
• A project directory (new or existing)

---

Step 1: Install the plugin

In Claude Code, run:

/plugin install ai-coding-ok@claude-plugins-official

This installs the ai-coding-ok skill globally. You only need to do this once per machine.

Alternative (legacy git-clone):
git clone https://github.com/Mark7766/ai-coding-ok ~/.claude/skills/ai-coding-ok

---

Step 2: Initialize in your project

cd your-project
claude

In the Claude Code session, say any of these:

install ai-coding-ok

Or in Chinese:

安装 ai-coding-ok
初始化 ai-coding-ok

Claude will detect the trigger and enter Mode A (Install).

---

Step 3: Answer one question

Claude will ask:

"In one sentence, what are you building?"

Example answers:

Your project	Good answer
Personal expense tracker	"A personal expense tracker that records what I spend each day"
Team weekly report tool	"A tool for collecting weekly reports from team members, with auto-reminders every Friday"
Internal API service	"A REST API for managing customer orders, with PostgreSQL backend"

Claude will then:

1. Infer your tech stack, architecture, design principles
2. Replace all {{placeholders}} in the 16+ template files
3. Bootstrap the first task-history entry
4. Prompt you for source directory configuration (for hooks)

Source directory prompt

Claude will also ask:

"Where does your source code live? (e.g. src/, lib/, app/ — default: src/ tests/)"

This configures the Claude Code hooks to only trigger on changes to your actual source files.

---

Step 4: Review and confirm

Claude will output a summary:

✅ Installation complete

Files installed:
  AGENTS.md, CLAUDE.md, .github/copilot-instructions.md,
  .github/agent/system-prompt.md, .github/agent/coding-standards.md,
  .github/agent/workflows.md, .github/agent/prompt-templates.md,
  .github/agent/memory/project-memory.md,
  .github/agent/memory/decisions-log.md,
  .github/agent/memory/task-history.md,
  .github/project-metadata.yml, .github/PULL_REQUEST_TEMPLATE.md,
  .github/ISSUE_TEMPLATE/*, .github/workflows/*,
  .claude/settings.local.json

Key decisions:
  - Project name: your-project
  - Tech stack: Python 3.12 + FastAPI + SQLite + pytest
  - Design principle: minimalist, practical

Next: Open AGENTS.md to review. PDCA loop is active.

---

Step 5: Verify

Run the verification script:

bash scripts/verify.sh

Expected output:

✓ All required files are present.
✓ No unfilled placeholders — setup looks fully customized.

See Installation Verification for detailed exit code meanings.

---

What happens next

From now on, every coding task in this project will automatically:

1. Plan — Read 7 files (AGENTS.md + 3 agent specs + 3 memory files) before writing code
2. Do — Write code + tests in the same change
3. Check — Run tests, verify no regression
4. Act — Update task-history.md (always), and decisions-log.md / project-memory.md (when needed)

The PDCA loop is enforced by the Five-Layer Defense System.

---

Common issues

"I already have an AGENTS.md"

The installer will detect conflicts and offer three choices:

1. Overwrite (risky)
2. Copy only missing files (recommended)
3. Abort

It never silently overwrites existing files.

"settings.json conflicts with settings.local.json"

If both files exist, the installer will warn you and recommend merging into settings.local.json only. See Hooks System for details.

"I'm using Windows"

Use install.py instead of install.sh:

python install.py --claude-code

---

Next steps

• Verify your installation
• Understand the PDCA workflow
• Learn the three-tier memory system