Usage Guide
Learn how to use AI Instruction Kits v2.0 with its skill-based architecture and best practices.
📖 Basic Usage
In v2.0, simply referencing CLAUDE.md is all you need. ROOT_INSTRUCTION automatically selects the optimal skill for your task. After setup, the workflow is extremely simple.
Setup
# Integrate into your project
bash scripts/setup-project.sh
Daily Usage
# That's all you need! ROOT_INSTRUCTION auto-selects the right skill
claude "Refer to CLAUDE.md and implement user authentication"
# Same entry point regardless of task type
claude "Refer to CLAUDE.md and investigate performance bottlenecks"
claude "Refer to CLAUDE.md and design a RESTful API"
How it works: CLAUDE.md → ROOT_INSTRUCTION.md (Skill Orchestrator) → auto-selects optimal skill from .claude/skills/ → executes task
🎯 Skill-Based Workflow
The core of v2.0 is four core skills placed in .claude/skills/. ROOT_INSTRUCTION automatically suggests the appropriate skill based on the task at hand.
checkpoint-manager (Task Progress Management)
Automatically suggests checking pending tasks at the start of a conversation.
# Start a task
scripts/checkpoint.sh start "New feature development" 5
# → Task ID: TASK-123456-abc is issued
# Report progress
scripts/checkpoint.sh progress TASK-123456-abc 2 5 "Design complete" "Start implementation"
# Complete task
scripts/checkpoint.sh complete TASK-123456-abc "5 features implemented, 20 tests created"
Auto-suggestion timing: Suggests checking pending tasks at conversation start
worktree-manager (Git Worktree Management)
Suggests creating a worktree for complex tasks or multi-file changes.
# Create worktree
scripts/worktree-manager.sh create TASK-123456-abc "feature-auth"
# → .gitworktrees/ai-TASK-123456-abc-feature-auth/ is created
# Move to working directory
cd .gitworktrees/ai-TASK-123456-abc-feature-auth/
# After completing work
scripts/worktree-manager.sh complete TASK-123456-abc
Auto-suggestion timing: Suggests worktree creation for complex tasks
auto-build (Automatic Build & Test)
Automatically detects project type and runs the appropriate build command.
# Auto-detects project type and builds
# package.json → npm run build
# Cargo.toml → cargo build
# go.mod → go build
# etc.
Auto-suggestion timing: Suggests build/test execution after code changes
commit-safe (Safe Commits)
Creates clean commits without AI signatures.
# Clean commit with AI signatures automatically removed
scripts/commit.sh "feat: Add user authentication"
Auto-suggestion timing: Suggests file-specific commits after changes
Basic Workflow
1. Check pending → 2. Start task → 3. Create worktree (optional) → 4. Work → 5. Commit → 6. Complete
🛒 Marketplace Skills
Install additional skills from claude-skills-marketplace.
Install via Claude Code /plugin Command
# Step 1: Register the marketplace (one-time only)
/plugin marketplace add dobachi/claude-skills-marketplace
# Step 2: Install skills
/plugin install code-reviewer@dobachi-skills
/plugin install data-analyst@dobachi-skills
Run /plugin alone to open the plugin manager UI for interactive skill browsing, installation, and management.
Installation Scopes
| Scope | Description | Saved to |
|---|---|---|
| User | Available across all your projects | ~/.claude/settings.json |
| Project | Available to team members | .claude/settings.json |
| Local | Only you, this repo only | .claude/settings.local.json |
Manual Installation
If not using the /plugin command, you can place skill files directly:
# Copy a skill directory from the marketplace
cp -r path/to/code-reviewer .claude/skills/code-reviewer
Creating Custom Skills
If you need custom skills, use the skill-creator skill from the marketplace. Simply place skill files in .claude/skills/ to make them available.
📊 Checkpoint Management
scripts/checkpoint.sh is a script that records and manages task progress in detail.
Key Commands
| Command | Purpose | Example |
|---|---|---|
start |
Start a new task | scripts/checkpoint.sh start "API development" 5 |
progress |
Report progress | scripts/checkpoint.sh progress TASK-xxx 2 5 "Design complete" "Start implementation" |
complete |
Complete a task | scripts/checkpoint.sh complete TASK-xxx "3 endpoints implemented" |
pending |
List pending tasks | scripts/checkpoint.sh pending |
summary |
Show task details | scripts/checkpoint.sh summary TASK-xxx |
error |
Report an error | scripts/checkpoint.sh error TASK-xxx "Dependency error" |
Tracking Skill Usage
# Record skill usage start
scripts/checkpoint.sh instruction-start ".claude/skills/auto-build" "API development" TASK-xxx
# Record skill usage completion
scripts/checkpoint.sh instruction-complete ".claude/skills/auto-build" "3 endpoints implemented" TASK-xxx
Visualizing Progress
# Check pending tasks
scripts/checkpoint.sh pending
# View detailed task history
scripts/checkpoint.sh summary TASK-xxx
# Show help
scripts/checkpoint.sh help
🌲 Git Worktree Workflow
For complex tasks or changes spanning multiple files, working in a Git worktree is recommended.
Recommended Flow
# 1. Start task
scripts/checkpoint.sh start "Auth feature development" 5
# → TASK-123456-abc
# 2. Create worktree
scripts/worktree-manager.sh create TASK-123456-abc "feature-auth"
# 3. Move to worktree and work
cd .gitworktrees/ai-TASK-123456-abc-feature-auth/
# 4. Work and commit
scripts/commit.sh "feat: Add authentication feature"
# 5. Complete task and clean up worktree
scripts/checkpoint.sh complete TASK-123456-abc "Auth feature implemented"
scripts/worktree-manager.sh complete TASK-123456-abc
Worktree Management Commands
# List worktrees
scripts/worktree-manager.sh list
# Switch to a specific worktree
scripts/worktree-manager.sh switch TASK-xxx
# Complete and clean up
scripts/worktree-manager.sh complete TASK-xxx
# Bulk remove unused worktrees
scripts/worktree-manager.sh clean
⚙️ Customizing PROJECT.md
Centralize project-specific settings in PROJECT.md.
Basic Configuration Example
## Project-Specific Instructions
### Coding Standards
- ESLint config: Follow .eslintrc.js
- Naming convention: camelCase
- Comments: JSDoc format
### Test Requirements
- Coverage: 80% minimum
- E2E tests: Using Cypress
### Build Settings
- Build command: npm run build
- Lint command: npm run lint
- Test command: npm run test
### Instruction Priority
1. PROJECT.md (highest priority)
2. Task-specific instructions
3. Skill auto-selection
🤖 Codex CLI / Gemini CLI
AI Instruction Kits supports AI CLI tools beyond Claude Code.
Codex CLI
Custom prompts are placed in .codex/prompts/. The filename becomes the /command-name you can invoke.
# Available commands
/build # Detect project type and assist with build
/checkpoint # Guide through checkpoint.sh subcommands
/commit-safe # Safe commits without AI signatures
/commit-and-report # Commit, push, and report to Issues
/reload-instructions # Reload instructions
Gemini CLI
TOML command definitions are placed in .gemini/commands/. The same command set as Codex CLI is available.
🎯 Best Practices
1. Skill Selection Tips
- Let ROOT_INSTRUCTION handle it by default (auto-selection is optimal)
- Reference
.claude/skills/files directly when you need a specific skill - Search the marketplace for missing skills
2. Managing Customizations
- Centralize project-specific settings in PROJECT.md
- Track changes with version control
- Share with team members
3. Feedback Loop
- Accumulate work history with the checkpoint feature
- Evaluate skill effectiveness and customize as needed
- Consider creating new skills or contributing to the marketplace
🔍 Troubleshooting
Q: Skills are not being auto-selected?
A: Verify that skill files are placed in the .claude/skills/ directory.
# Check for installed skills
ls .claude/skills/
# Re-run setup
bash scripts/setup-project.sh
Q: Checkpoint log not found?
A: checkpoint.log is created in the project root. It is auto-generated when you start your first task.
# Start a new task to create the log
scripts/checkpoint.sh start "Test task" 1
Q: Worktree creation fails?
A: Make sure you are running from the Git repository root directory.
# Navigate to repository root
cd $(git rev-parse --show-toplevel)
# Create worktree
scripts/worktree-manager.sh create TASK-xxx "description"
Q: Unsure about instruction priority?
A: Instructions are applied in the following priority order:
1. PROJECT.md (highest priority)
2. Task-specific instructions
3. ROOT_INSTRUCTION skill selection
4. Skill default behavior
📚 Learn More
- Features - Detailed feature explanations
- Quick Start - Get started in 5 minutes
- GitHub - Source code
- Skills Marketplace - Community-built skills
💡 Tip
In v2.0, simply saying "Refer to CLAUDE.md" is enough for ROOT_INSTRUCTION to auto-select the optimal skill. Start simple, then add skills from the marketplace as needed.