Skip to content

Using GuardKit with Claude Code Web

Overview

GuardKit is fully compatible with Claude Code Web (claude.ai/code), providing a complete AI-assisted development workflow with quality gates directly in your browser. This guide covers installation, persistence, and best practices for using GuardKit across multiple repositories.

Table of Contents


How It Works

Architecture

GuardKit uses Claude Code's slash command system and agent framework:

/root/.agentecflow/          # Global installation (persistent storage)
├── commands/                # Command definitions (.md files)
│   ├── task-create.md      → /task-create
│   ├── task-work.md        → /task-work
│   ├── task-complete.md    → /task-complete
│   └── ...
├── agents/                  # AI agent definitions (.md files)
│   ├── task-manager.md
│   ├── code-reviewer.md
│   ├── test-verifier.md
│   └── ...
└── templates/               # Stack templates (react, python, .NET, etc.)

/root/.claude/               # Claude Code configuration
├── commands -> /root/.agentecflow/commands  # Symlink
└── agents -> /root/.agentecflow/agents      # Symlink

Command Discovery Process

  1. On startup, Claude Code scans ~/.claude/commands/ for .md files
  2. Each .md file becomes a slash command (filename determines command name)
  3. Commands are registered and available for autocomplete
  4. Agents are loaded from ~/.claude/agents/ for specialized tasks

Why This Works in Web

  • Persistent file system: /root/ and /home/user/ persist across sessions
  • No installation required: Slash commands are markdown files, not executables
  • Browser-based: All operations use Claude Code's built-in tools (Bash, Read, Write, Edit)
  • Full functionality: Access to all GuardKit features without limitations

Installation

One-Time Setup

Run the installation script from the GuardKit repository:

# 1. Navigate to guardkit repository
cd /home/user/guardkit

# 2. Run installer (creates global installation)
./installer/scripts/install.sh

What Gets Installed

Global Components (~/.agentecflow/): - Commands: 9 slash commands (task-create, task-work, etc.) - Agents: 55 AI agents (14 global + 41 stack-specific) - Templates: 5 high-quality templates (react-typescript, fastapi-python, nextjs-fullstack, react-fastapi-monorepo, default) - Libraries: Python modules for complexity evaluation, plan rendering - Documentation: Methodology instructions and references

Claude Code Integration (~/.claude/): - Symlinks to commands and agents directories - Enables automatic command discovery - Compatible with Conductor.build for parallel development

Shell Integration (~/.bashrc): - PATH configuration for CLI commands (guardkit, guardkit-init) - Environment variables (AGENTECFLOW_HOME)

Installation Output

✅ Installation Complete!

Installed Components:
- 9 slash commands available
- 55 AI agents (14 global + 41 stack-specific)
- 4 high-quality project templates
- Python dependencies (Jinja2, python-frontmatter)

Commands Available:
- /task-create
- /task-work
- /task-complete
- /task-status
- /task-refine
- /debug
- /figma-to-react
- /zeplin-to-maui
- /mcp-zeplin

Installation size: ~4.0 MB

Verification

Verify installation success:

# Check installation directory
ls -la ~/.agentecflow/
# Should show: agents, bin, commands, templates, etc.

# Check Claude Code symlinks
ls -la ~/.claude/
# Should show: commands -> /root/.agentecflow/commands
#              agents -> /root/.agentecflow/agents

# List available commands
ls -1 ~/.claude/commands/*.md
# Should show all 9 command files

# List available agents
ls -1 ~/.claude/agents/*.md | head -10
# Should show agent definitions

Persistence Across Sessions

How Persistence Works

File System Persistence: - Claude Code Web provides persistent storage in /root/ and /home/user/ - Installation creates files that survive session restarts - No re-installation needed for future sessions

Automatic Command Loading:

Session Start → Claude Code scans ~/.claude/commands/
             → Registers slash commands
             → Loads agent definitions
             → Commands immediately available

What Persists

Persists Across Sessions: - All installed commands and agents - Project files and code - Task states and history - Implementation plans - Configuration settings

Does NOT Persist: - Running processes (background tasks) - Environment variables (set via export in shell) - Temporary files in /tmp/

Testing Persistence

In Current Session:

# Verify installation
ls ~/.claude/commands/task-work.md
# Output: /root/.claude/commands/task-work.md

In Future Sessions:

# Commands automatically available
/task-status
# Should work immediately without re-installation

Re-installation

Not Required! Installation is one-time only.

When to Re-install: - Upgrading to a new GuardKit version - Fixing corrupted installation - Adding new templates or agents


Multi-Repository Usage

Global Availability

Commands are available in all repositories:

# GuardKit repository
cd /home/user/guardkit
/task-status    # ✅ Works

# RequireKit repository
cd /home/user/requirekit
/task-status    # ✅ Works

# Any other project
cd /home/user/my-project
/task-status    # ✅ Works

Command Scope

Command Available Globally? Notes
/task-create ✅ Yes Creates tasks in current directory's tasks/ folder
/task-status ✅ Yes Shows tasks from current directory
/task-work ✅ Yes Must run from project root (see below)
/task-complete ✅ Yes Works on tasks in current directory
/task-refine ✅ Yes Works on tasks in current directory
/debug ✅ Yes Troubleshooting for current context

Repository-Specific Usage

GuardKit Development:

cd /home/user/guardkit
/task-status                  # View guardkit tasks
/task-work TASK-002          # Implement guardkit feature
/task-complete TASK-002      # Complete guardkit task

RequireKit Development:

cd /home/user/requirekit
/task-status                  # View requirekit tasks
/task-work REQ-TASK-001      # Implement requirekit feature
/task-complete REQ-TASK-001  # Complete requirekit task

Other Projects:

cd /home/user/my-web-app
guardkit init react-typescript  # Initialize with React template
/task-create "Add user auth"  # Create task in this project
/task-work TASK-001          # Implement in this project


Working Directory Requirements

Critical: Run from Project Root

The /task-work command MUST be run from your project's root directory where source code should be created.

Why This Matters

/task-work uses the current working directory to:

  1. Detect technology stack
  2. Looks for .csproj → .NET project
  3. Looks for package.json → Node.js/TypeScript project
  4. Looks for requirements.txt → Python project

  5. Create source files

  6. Implements code in correct directories
  7. Follows project structure conventions

  8. Run tests and builds

  9. Executes stack-specific test commands
  10. Validates compilation and coverage

  11. Generate artifacts

  12. Creates implementation plans in .claude/task-plans/
  13. Updates task states in tasks/ folders

Correct Usage Examples

Correct - Working on GuardKit Tasks:

# 1. Navigate to guardkit project
cd /home/user/guardkit
pwd  # Verify: /home/user/guardkit

# 2. Confirm you see project files
ls   # Should show: installer/, tasks/, README.md, etc.

# 3. Work on task
/task-work TASK-001
# Creates files in /home/user/guardkit/
# Detects: Python/Shell project

Correct - Working on RequireKit Tasks:

# 1. Navigate to requirekit project
cd /home/user/requirekit
pwd  # Verify: /home/user/requirekit

# 2. Confirm you see project files
ls   # Should show: requirements.txt, src/, tests/, etc.

# 3. Work on task
/task-work TASK-042
# Creates files in /home/user/requirekit/
# Detects: Python project with pytest

Incorrect Usage Examples

Wrong - Files Go to Wrong Location:

# In guardkit directory, working on requirekit task
cd /home/user/guardkit
/task-work REQ-TASK-001
# Problem: Creates files in guardkit/ instead of requirekit/
# Problem: Detects wrong tech stack

Wrong - Wrong Tech Stack Detected:

# In wrong directory
cd /home/user/
/task-work TASK-001
# Problem: Can't detect technology stack
# Problem: Doesn't know where to create files

Directory Validation

Before running /task-work, verify location:

# Check current directory
pwd

# List project files (should see your project's main files)
ls -la

# For .NET projects, should see:
ls *.csproj 2>/dev/null

# For Node.js projects, should see:
ls package.json 2>/dev/null

# For Python projects, should see:
ls requirements.txt 2>/dev/null || ls setup.py 2>/dev/null

Quick Validation Script

# Verify you're in a project root
if [ -f "*.csproj" ] || [ -f "package.json" ] || [ -f "requirements.txt" ]; then
    echo "✅ In project root - safe to run /task-work"
else
    echo "❌ Not in project root - navigate to project first!"
fi

Integration with RequireKit

Automatic Detection

GuardKit automatically detects if RequireKit is installed and enhances workflow:

GuardKit Only: - Task workflow with quality gates - Implementation planning - Architectural review - Test enforcement

GuardKit + RequireKit: - All the above PLUS: - Loads EARS requirements automatically - Includes Gherkin scenarios for BDD - Epic/feature context and hierarchy - Requirements-based acceptance criteria

No Configuration Required

Detection is automatic - just install both packages:

# Install GuardKit (already done)
cd /home/user/guardkit
./installer/scripts/install.sh

# Install RequireKit (optional)
cd /home/user/requirekit
./installer/scripts/install.sh

# Both commands now available everywhere
# Integration happens automatically

Usage Pattern

Create requirement in RequireKit:

cd /home/user/requirekit
/req-create "User authentication system"
# Creates: REQ-001

Create linked task in your project:

cd /home/user/my-web-app
/task-create "Implement user auth" requirements:[REQ-001]
# Creates: TASK-001 linked to REQ-001

Work on task (from project directory):

cd /home/user/my-web-app
/task-work TASK-001
# Automatically loads REQ-001 context
# Implements in /home/user/my-web-app/

Benefits of Integration

Feature GuardKit Only GuardKit + RequireKit
Task workflow
Quality gates
EARS requirements
Gherkin scenarios
Epic/feature hierarchy
Requirements traceability
BDD workflow

Available Commands

Task Management Commands

Command Description Example
/task-create Create a new task /task-create "Add login feature"
/task-work Execute task workflow (Phases 2-5.5) /task-work TASK-001
/task-complete Mark task as complete and archive /task-complete TASK-001
/task-status View task status and summary /task-status TASK-001
/task-refine Lightweight improvements without re-work /task-refine TASK-001
/task-review Analysis and decision workflows /task-review TASK-001 --mode=architectural --depth=standard

Design Integration Commands

Command Description Example
/figma-to-react Convert Figma designs to React components /figma-to-react <file-key>
/zeplin-to-maui Convert Zeplin designs to .NET MAUI /zeplin-to-maui <project-id> <screen-id>
/mcp-zeplin Zeplin MCP integration utilities /mcp-zeplin

Utility Commands

Command Description Example
/debug Troubleshoot issues and check configuration /debug

Task Work Options

# Standard mode (default)
/task-work TASK-001

# TDD mode (test-driven development)
/task-work TASK-001 --mode=tdd

# Design-first workflow (complex tasks)
/task-work TASK-001 --design-only
# [Review and approve plan]
/task-work TASK-001 --implement-only

Command Flags

Flag Description Use Case
--mode=standard Implementation + tests together (default) Straightforward features
--mode=tdd Test-Driven Development (Red → Green → Refactor) Complex business logic
--design-only Planning phase only (Phases 2-2.8) Complex tasks requiring review
--implement-only Implementation phase only (Phases 3-5) After design approval

Troubleshooting

Commands Not Found

Problem: Slash commands don't appear in autocomplete

Solutions:

# 1. Verify installation
ls ~/.claude/commands/
# Should show: task-create.md, task-work.md, etc.

# 2. Check symlinks
readlink ~/.claude/commands
# Should show: /root/.agentecflow/commands

# 3. Re-create symlinks if broken
mkdir -p ~/.claude
ln -sf /root/.agentecflow/commands ~/.claude/commands
ln -sf /root/.agentecflow/agents ~/.claude/agents

# 4. Verify Claude Code can read files
cat ~/.claude/commands/task-work.md | head -20
# Should show command definition

Wrong Directory Errors

Problem: /task-work creates files in wrong location

Solution:

# Always verify location before running /task-work
pwd
ls -la

# Navigate to correct project root
cd /home/user/your-project

# Verify you see project files
ls package.json  # or *.csproj, requirements.txt, etc.

# Now run task-work
/task-work TASK-001

Tech Stack Misdetected

Problem: GuardKit detects wrong technology stack

Cause: Running from wrong directory

Solution:

# Ensure you're in project root with correct markers
cd /home/user/your-project

# For .NET projects, verify:
ls *.csproj

# For Node.js projects, verify:
ls package.json

# For Python projects, verify:
ls requirements.txt

Tasks Not Found

Problem: /task-status shows no tasks

Possible Causes: 1. No tasks/ directory in current location 2. Tasks in different repository 3. Running from wrong directory

Solution:

# Check if tasks/ directory exists
ls -la tasks/

# Verify task files exist
ls -la tasks/backlog/
ls -la tasks/in_progress/

# Navigate to correct repository
cd /home/user/guardkit  # or requirekit, or your project
/task-status

Installation Issues

Problem: Installation script fails

Solution:

# 1. Check prerequisites
python3 --version  # Should be 3.7+
node --version     # Optional but recommended

# 2. Check permissions
ls -la installer/scripts/install.sh
# Should be executable (-rwxr-xr-x)

# 3. Re-run with explicit bash
bash installer/scripts/install.sh

# 4. Check installation log
tail -50 ~/.agentecflow/installation.log  # If exists

Commands Work But No Output

Problem: Commands execute but produce no visible results

Possible Causes: 1. Command is running but needs time to complete 2. Error occurred but wasn't displayed 3. Output is being written elsewhere

Solution:

# 1. Check command output explicitly
/task-status 2>&1

# 2. Verify task files
ls -la tasks/backlog/

# 3. Check for error logs
ls -la ~/.agentecflow/logs/  # If exists

# 4. Run debug command
/debug


Quick Reference

Installation Checklist

  • Navigate to guardkit repository
  • Run ./installer/scripts/install.sh
  • Verify symlinks: ls -la ~/.claude/
  • Test command: /task-status
  • Verify 9 commands available
  • Verify 55 agents installed

Pre-Flight Checklist (Before /task-work)

  • Navigate to project root: cd /home/user/your-project
  • Verify location: pwd
  • Check project files exist: ls
  • Confirm task exists: ls tasks/backlog/TASK-XXX*
  • Run command: /task-work TASK-XXX

Common Workflows

Simple Task:

cd /home/user/your-project
/task-create "Feature name"
/task-work TASK-001
/task-complete TASK-001

Complex Task with Design Review:

cd /home/user/your-project
/task-create "Complex refactoring" priority:high
/task-work TASK-002 --design-only
# [Review plan in .claude/task-plans/]
/task-work TASK-002 --implement-only
/task-complete TASK-002

TDD Workflow:

cd /home/user/your-project
/task-create "Calculate tax rates"
/task-work TASK-003 --mode=tdd
/task-complete TASK-003

Directory Structure Reference

/root/.agentecflow/              # Global installation
├── agents/                      # 55 AI agents
├── commands/                    # 9 slash commands
├── templates/                   # 5 high-quality templates
└── bin/                        # CLI commands (guardkit, gk)

/root/.claude/                   # Claude Code config
├── commands -> ../agentecflow/commands
└── agents -> ../agentecflow/agents

/home/user/guardkit/           # GuardKit repository
├── tasks/
│   ├── backlog/
│   ├── in_progress/
│   └── completed/
├── .claude/
│   └── task-plans/
└── installer/

/home/user/requirekit/           # RequireKit repository (optional)
├── requirements/
├── tasks/
└── .claude/

/home/user/your-project/         # Your project
├── tasks/                       # Created by guardkit init
├── .claude/                     # Created by guardkit init
├── src/                        # Your source code
└── tests/                      # Your tests

Key Concepts

Concept Description
Global Installation Commands available in all repositories
Working Directory Must be project root for /task-work
Persistence Installation survives session restarts
Auto-Detection RequireKit integration automatic
Symlinks Claude Code reads from ~/.claude/
Stack Templates 4 high-quality reference templates + custom via /template-create

Command Hierarchy

All Commands Available Globally
Task Commands Need Project Context
/task-work Needs Project Root Directory
Creates Files in Current Directory
Detects Tech Stack from Current Directory

Getting Help

Documentation: - GuardKit Workflow - Complete workflow guide - Quick Reference - Command cheat sheet - Creating Local Templates - Custom templates - MAUI Template Selection - .NET MAUI guidance

Commands:

/debug                    # Troubleshooting and diagnostics
/task-status             # View all tasks
/task-status TASK-001    # View specific task

Verification:

# Check installation
ls ~/.agentecflow/
ls ~/.claude/commands/

# Test commands
/task-status

# Verify agents
ls ~/.claude/agents/


Summary

GuardKit works seamlessly with Claude Code Web:

One-time installation - Persists across all sessions ✅ Global commands - Available in all repositories ✅ Multi-repository support - Use with guardkit, requirekit, and your projects ✅ Full functionality - No limitations compared to desktop ✅ Automatic integration - RequireKit detection when both installed ✅ Quality gates - 80% coverage, 100% test pass, architectural review

Remember: - Install once, use everywhere - Always run /task-work from project root directory - Commands persist across Claude Code Web sessions - No re-installation needed

Ready to start:

cd /home/user/your-project
/task-create "Your feature"
/task-work TASK-001
/task-complete TASK-001

Happy building! 🚀