Agent Deep Toolkit

A portable, cross-agent meta-cognitive toolkit for development workflows.

This directory is designed to become the root of a standalone repo that packages the full workflow suite for:

• Windsurf – .windsurf/workflows-style deep workflows.
• Claude Code – SKILL-based slash commands using the Agent Skills standard.
• Cursor – .cursor/commands-style slash commands.
• OpenCode – .opencode/commands-style commands with native OpenCode frontmatter.

The goal is to keep procedures conceptually identical across agents while keeping their implementations independent so you can install Windsurf, Claude Code, or Cursor tooling separately.

Quick Start (5 Minutes)

1️⃣ Install

# Clone repository
git clone https://github.com/AIntelligentTech/agent-deep-toolkit.git
cd agent-deep-toolkit

# Interactive wizard (recommended for first-time users)
./install.sh --wizard

# Or direct install for all detected agents
./install.sh --agent all --level user

2️⃣ Choose Your Agent

The installer automatically detects:

• Claude Code → Skills installed to ~/.claude/skills/
• Windsurf Cascade → Workflows installed to ~/.windsurf/workflows/
• Cursor → Commands installed to ~/.cursor/commands/
• OpenCode → Commands installed to ~/.opencode/commands/

3️⃣ Try Your First Skill

Claude Code: Type in chat

/think "How should I architect a real-time notification system?"

Windsurf: Type in chat (auto-suggests contextually)

/architect

Cursor: Command Palette (Cmd+K or Ctrl+K)

think <Enter>

OpenCode: Type in chat

/think "notification system design"

4️⃣ Explore 50 Skills

• Type /index (Claude/OpenCode) or /help to see all skills
• Read EXAMPLES.md for practical recipes
• Review ARCHITECTURE.md to understand the system

How It Works

┌──────────────────────────────────────────┐
│   Single CACE Source (skills/*.md)       │  ← One skill definition
│   Frontmatter + XML-tagged instructions  │
└──────────────┬───────────────────────────┘
               │
               ▼
       ┌───────────────┐
       │ Build Pipeline │  ← Convert to agent formats
       └───────┬────────┘
               │
    ┌──────────┼──────────┬──────────┐
    ▼          ▼          ▼          ▼
 Windsurf   Claude     Cursor    OpenCode   ← Agent-specific outputs
 Workflows  Skills     Commands  Commands

Key Benefits:

• Portable: One skill, four agent formats
• Deep Work: Strategic thinking, not just code completion
• Battle-Tested: 50 skills refined over real-world projects
• Open Source: MIT licensed, community-driven

Workflow Suite (v2.0)

The toolkit provides 48 specialized workflows organized into 7 categories:

🧠 Core Reasoning (5)

Command	Aliases	Purpose
/think	/reason, /analyze, /ponder	Deep multi-perspective analysis
/decide	/consider, /choose, /weigh	Structured decision-making with frameworks
/search	/find, /lookup, /locate, /research	Local + web search combined
/investigate	/dig, /probe, /examine	Root-cause analysis and hypothesis testing
/explore	/understand, /learn, /discover	Build mental models of code/concepts

🏗️ Architecture & Design (9)

Command	Aliases	Purpose
/architect	/design-system, /structure, /blueprint	System architecture design (C4, DDD)
/design	-	Product and interaction design (Double Diamond)
/ux	-	UX evaluation and improvement
/spec	-	Write specifications and ADRs
/document	/docs, /write-docs, /docstring	Documentation discovery and creation
/tokens	/design-tokens	Extract and systematize design tokens
/svg	-	Generate and validate SVGs
/onboard	/onboarding, /getting-started	Developer onboarding materials
/api	/api-design, /endpoint	API design, contracts, versioning

💻 Implementation (11)

Command	Aliases	Purpose
/code	/implement, /build, /develop	High-quality code implementation
/debug	/fix, /troubleshoot, /diagnose	Systematic debugging
/refactor	/restructure, /reorganize	Safe refactoring and design improvement
/prune	/purge, /cleanup, /sweep	Aggressive dead code and legacy removal
/test	/verify, /validate, /check	Test strategy and implementation
/optimize	/perf, /performance, /speedup	Performance improvement
/polish	-	Final refinement to world-class quality
/review	/code-review, /cr	Structured code review
/dependency	/deps, /dependencies	Dependency management and security
/benchmark	/perf-test, /measure	Performance benchmarking
/simplify	/reduce-complexity, /declutter	Complexity reduction

⚙️ Operations (9)

Command	Aliases	Purpose
/bash	/shell, /cli, /terminal	CLI automation and scripting
/git	/github, /vcs	Git workflows and best practices
/data	-	Data modeling and governance
/infra	/infrastructure, /devops, /platform	Infrastructure design
/observe	/observability, /monitor, /metrics	Logs, metrics, traces
/incident	/retrospective, /postmortem, /outage	Incident response + retros
/inventory	/catalog, /enumerate	Systematic enumeration
/migrate	/migration, /upgrade	Safe data/schema migrations
/deploy	/release, /rollout	Deployment strategy

📋 Governance (2)

Command	Aliases	Purpose
/compliance	/ethics, /regulation, /legal	Ethics + regulatory alignment
/threat	/threat-model, /security	Security threat modeling

🔄 Change Management (6)

Command	Aliases	Purpose
/integrate	/integration, /integrate-plan	Holistic integration planning
/impact	-	Change impact analysis
/propagate	-	Safe change rollout
/brainstorm	/ideas, /alternatives, /ideate	Idea generation and evaluation
/followup	/next, /nextsteps	Next steps planning
/experiment	/test-hypothesis, /ab-test	Experiment design and execution

🔧 Meta (7)

Command	Aliases	Purpose
/loop	/ralph-loop, /autonomous, /self-directed	Autonomous execution until completion
/iterate	/cycle, /increment, /chunk	Break tasks into verifiable iterations
/relentless	/try-hard, /dont-stop, /ultrathink	Multiply effort and depth for high-stakes work
/index	/help, /workflows, /skills	Navigator for all workflows
/audit	-	Comprehensive system review
/plan	/roadmap, /schedule	Implementation planning
/estimate	/scope, /sizing	Effort estimation

Repository layout

agent-deep-toolkit/
  README.md          # This file (agent-agnostic overview)
  LICENSE            # MIT license for the toolkit
  .gitignore         # Ignore rules for this repo
  install.sh         # Unified installer for Windsurf, Claude Code, Cursor, and OpenCode

  skills/            # Canonical skill definitions (source of truth)
    think/SKILL.md
    decide/SKILL.md
    search/SKILL.md
    ... (45 total skills)

  templates/
    windsurf/
      workflow-template.md      # Single Windsurf workflow template
    claude-code/
      skill-template.md         # Single Claude skill template
    cursor/
      command-template.md       # Single Cursor command template
    open-code/
      command-template.md       # Single OpenCode command template

  outputs/
    windsurf/workflows/         # Generated Windsurf workflows
    claude/skills/              # Generated Claude Code skills
    cursor/commands/            # Generated Cursor commands
    opencode/commands/          # Generated OpenCode commands

Each skill directory is structured according to the Agent Skills standard:

• SKILL.md – YAML frontmatter (name, description, command, aliases, and policy fields) + concise instructions.

Installing

Interactive Wizard (Recommended)

For first-time users, use the interactive wizard:

./install.sh --wizard

The wizard will guide you through:

• Prerequisite validation (Node.js, fzf)
• Multi-select agent selection
• Installation level choice (user or project)
• Preview of what will be installed
• Safe dry-run mode by default

Command-Line Installation

For automation or CLI-only environments:

# Install for all agents at user level
./install.sh --agent all --level user

# Install specific agent (windsurf, claude, cursor, opencode)
./install.sh --agent claude --level project

# Force overwrite existing installation
./install.sh --agent all --level user --force

# Dry-run preview (no filesystem changes)
./install.sh --agent all --level user --dry-run

Command naming

All tools use canonical command names without the legacy deep- prefix:

• /think instead of /deep-think
• /debug instead of /deep-debug
• etc.

Commands also support aliases that trigger the same workflow:

• /fix triggers /debug
• /perf triggers /optimize
• /docs triggers /document

Per-agent independence

• Windsurf installation only touches Windsurf workflow locations.
• Claude Code installation only touches .claude/skills locations.
• Cursor installation only touches .cursor/commands locations.
• OpenCode installation only touches .opencode/commands locations.

You can install for just one agent without affecting others.

Requirements

• A POSIX-like environment (macOS or Linux) with bash available.
• Access to your home directory for user-level installs.
• Optionally, a project directory for project-level installs.

Versioning

Agent Deep Toolkit follows Semantic Versioning. The canonical version is stored in the top-level VERSION file.

Uninstalling

# Uninstall everything
./install.sh --agent all --level user --uninstall --yes

# Detect existing installations
./install.sh --agent all --level user --detect-only

Contributing and support

See:

• CONTRIBUTING.md for contribution guidelines.
• CODE_OF_CONDUCT.md for community standards.
• SECURITY.md for reporting security issues.

License

MIT. See LICENSE for details.