claude-dev-setup（Claude DEV 设置）

Claude Dev Setup

Complete package for coding agents: Claude Code CLI integration, session management, and APEX cognitive framework.

What This Skill Does

Installs and configures:

1. Session Tracking — memory/claude-code-sessions.md for background task management
2. AGENTS.md Section — Workflow documentation for all coding agents
3. APEX Stack Reference — How to add cognitive framework to project CLAUDE.md files
4. OAuth Token Handling — Automatic token extraction and usage

Prerequisites

• Node.js 18+
• Claude Pro/Max subscription (for OAuth token)
• OpenClaw agent workspace

OAuth Token (CRITICAL)

Where the Token Lives

Claude Code stores its OAuth token in:

~/.claude/.credentials.json

Token Format

sk-ant-oat01-xxxxx...

How to Get the Token

Option 1: Extract from credentials file (RECOMMENDED)

# Extract the accessToken from credentials
cat ~/.claude/.credentials.json | grep -o '"accessToken":"[^"]*"' | cut -d'"' -f4

Option 2: Authenticate via browser (first-time setup)

claude setup-token
# Opens browser for OAuth flow
# Paste authorization code when prompted

ALWAYS Use the Token When Calling Claude Code

Every Claude Code CLI call MUST include the token:

# ✅ CORRECT - Token set explicitly
export CLAUDE_CODE_OAUTH_TOKEN="sk-ant-oat01-xxxxx..."
claude --print --dangerously-skip-permissions "your task"

# ❌ WRONG - Token not set, will fail in non-interactive mode
claude --print "your task"

Automatic Token Extraction

Add to your shell profile for automatic token loading:

# Add to ~/.bashrc or ~/.zshrc
export CLAUDE_CODE_OAUTH_TOKEN=$(cat ~/.claude/.credentials.json 2>/dev/null | grep -o '"accessToken":"[^"]*"' | cut -d'"' -f4)

For Agent Executors

When using the exec tool, always extract and set the token:

exec({
  command: "export CLAUDE_CODE_OAUTH_TOKEN=$(cat ~/.claude/.credentials.json | grep -o '\"accessToken\":\"[^\"]*\"' | cut -d'\"' -f4) && claude --print --dangerously-skip-permissions 'Task description'",
  background: true,
  yieldMs: 10000
})

Or pre-extract and use:

// First, extract token
exec({
  command: "cat ~/.claude/.credentials.json | grep -o '\"accessToken\":\"[^\"]*\"' | cut -d'\"' -f4"
})

// Then use in all subsequent calls
exec({
  command: "CLAUDE_CODE_OAUTH_TOKEN=sk-ant-oat01-xxx claude --print --dangerously-skip-permissions 'Task'"
})

Architecture

Agent Workspace
├── AGENTS.md                          # + Session management section
├── memory/
│   └── claude-code-sessions.md         # NEW: Session registry
└── skills/
    ├── claude-code-cli-openclaw/       # CLI integration
    ├── apex-stack-claude-code/         # Cognitive framework
    └── claude-dev-setup/               # THIS SKILL

System Level
└── ~/.claude/.credentials.json         # OAuth token (DO NOT COMMIT)

Session Management

Registry Format

Sessions are tracked in memory/claude-code-sessions.md:

| Session ID | Label | Task | Started | Status |
|------------|-------|------|---------|--------|
| tender-nexus | build-auth | Build auth module | 2026-03-24 08:50 UTC | running |

Starting a Background Task (CORRECT WAY)

// Step 1: Extract token from credentials
// Step 2: Use in Claude Code call
exec({
  command: "CLAUDE_CODE_OAUTH_TOKEN=$(cat ~/.claude/.credentials.json | grep -o '\"accessToken\":\"[^\"]*\"' | cut -d'\"' -f4) claude --print --dangerously-skip-permissions 'Task description'",
  background: true,
  yieldMs: 10000
})

After starting, log the session:

1. Note sessionId from response
2. Append to memory/claude-code-sessions.md
3. Report to user: "Started X (session: label)"

Checking Status

When user asks "what's the status?":

process({ action: "log", sessionId: "session-id" })

Label Naming Convention

• build-feature-X — Building a new feature
• refactor-Y-module — Refactoring code
• fix-bug-Z — Bug fix
• test-coverage-A — Adding tests
• cleanup-legacy-B — Removing old code

Quick Tasks (< 30 seconds)

For simple tasks, use direct exec without tracking:

exec({
  command: "CLAUDE_CODE_OAUTH_TOKEN=$(cat ~/.claude/.credentials.json | grep -o '\"accessToken\":\"[^\"]*\"' | cut -d'\"' -f4) claude --print --dangerously-skip-permissions 'Quick fix'",
  timeout: 60
})

AGENTS.md Integration

After installing this skill, your AGENTS.md gets a new section:

## 🤖 Claude Code CLI Sessions

You have access to Claude Code CLI for coding tasks.

### OAuth Token (Required)

**Token location:** `~/.claude/.credentials.json`

**ALWAYS extract and use the token:**

CLAUDE_CODE_OAUTH_TOKEN=$(cat ~/.claude/.credentials.json | grep -o '"accessToken":"[^"]*"' | cut -d'"' -f4) claude --print --dangerously-skip-permissions 'task'

### Session Tracking

All background sessions tracked in `memory/claude-code-sessions.md`.

**Before starting:** Check for running sessions
**After starting:** Log session ID + label
**On completion:** Update status column

### When User Asks Status

1. Read sessions file
2. Find session by label
3. Poll for output
4. Summarize progress

APEX Stack for Projects

The cognitive framework (apex-stack-claude-code) should be added to project CLAUDE.md files, not agent memory.

Adding to a Project

cd /path/to/your/project

# If CLAUDE.md doesn't exist:
cat ~/.openclaw/workspace-YOURS/skills/apex-stack-claude-code/SKILL.md > CLAUDE.md

# If CLAUDE.md exists, append:
cat ~/.openclaw/workspace-YOURS/skills/apex-stack-claude-code/SKILL.md >> CLAUDE.md

What APEX Stack Does

Project CLAUDE.md Structure

# Project: [Name]

## Overview
[1-2 sentences]

## Tech Stack
- Language: ...
- Framework: ...

[APEX Stack content appended here]

Complete Workflow

User gives task
    ↓
Agent reads memory/claude-code-sessions.md (check for conflicts)
    ↓
Agent extracts token: cat ~/.claude/.credentials.json | grep accessToken
    ↓
Agent starts: exec({ command: "CLAUDE_CODE_OAUTH_TOKEN=xxx claude ...", background: true })
    ↓
Agent logs session to claude-code-sessions.md
    ↓
Agent reports: "Started X (session: build-feature-X)"
    ↓
User: "What's the status?"
    ↓
Agent reads sessions file → polls process → summarizes

Installation Checklist

Run this after installing the skill:

• [ ] Claude Code CLI installed: which claude
• [ ] Authenticated: check ~/.claude/.credentials.json exists
• [ ] Token extractable: cat ~/.claude/.credentials.json | grep accessToken
• [ ] Session file created: memory/claude-code-sessions.md
• [ ] AGENTS.md updated with session section
• [ ] APEX Stack added to project CLAUDE.md (if applicable)

Security Notes

• OAuth token stored in ~/.claude/.credentials.json (valid 1 year)
• NEVER commit CLAUDE_CODE_OAUTH_TOKEN to git
• ALWAYS extract token from credentials file, never hardcode
• Add .claude/ to .gitignore
• Sessions are local to each agent workspace
• MEMORIA explicitly forbids storing credentials

Troubleshooting

"Authentication failed" or "No token"

# Check if credentials exist
ls -la ~/.claude/.credentials.json

# If missing, run setup
claude setup-token

"command not found: claude"

# Install Claude Code CLI
npm install -g @anthropic-ai/claude-code

# Verify installation
which claude
claude --version

Token works in terminal but not in exec

The token must be exported in the same shell session:

# ✅ CORRECT - Same session
CLAUDE_CODE_OAUTH_TOKEN=xxx claude --print "task"

# ❌ WRONG - Separate commands
export CLAUDE_CODE_OAUTH_TOKEN=xxx
claude --print "task"  # Token not available

Token Efficiency

Savings: 80-90% reduction in token usage.

Related Skills

Install these alongside for full capability:

• self-improving — Learn from corrections
• claude-code-cli-openclaw — CLI integration details
• apex-stack-claude-code — Cognitive framework

Files in This Skill

claude-dev-setup/
├── SKILL.md              # This file
├── sessions-template.md  # Template for claude-code-sessions.md
├── agents-section.md     # AGENTS.md section template
└── setup.sh              # Installation script

Publishing

To publish to ClawHub:

clawhub publish ./skills/claude-dev-setup \
  --slug claude-dev-setup \
  --name "Claude Dev Setup" \
  --version 1.1.0 \
  --changelog "Added explicit OAuth token extraction guide"

Feedback

• If useful: clawhub star claude-dev-setup
• Issues: Open issue on ClawHub
• Stay updated: clawhub sync