Claude Agent Toolkit

Advanced skills for Claude Code that enable multi-agent coordination, external service integrations, and intelligent workflow orchestration.

Built with Claude using obra's superpowers framework.

Overview

This repository contains skills that extend Claude Code's capabilities through:

Multi-agent orchestration - Coordinate multiple AI agents in parallel or sequential workflows
External integrations - Connect to Google Workspace, Perplexity AI, Folio, legacy Obsidian, and more
Task coordination - Advanced task management and synchronization patterns

Supporting Repositories:

obra/superpowers - Process skills used by local agents
obra/external-subagents - Codex subagent skill dependency
googleworkspace/cli - Google Workspace CLI used by gws-* skills
dbmcco/obsidian-mcp - Legacy Obsidian MCP server (18 tools)
dbmcco/apple-reminders-mcp - Apple Reminders MCP server (6 tools)
dbmcco/claude-tmux-orchestration - Multi-agent tmux coordination

See docs/public-skill-inventory.md for the full public skill inventory, including keynote-slides, msgvault, roborev-*, white-paper, local-only skill coverage, and external dependency links.

Prerequisites

Required

macOS - Required for Apple Reminders integration and tmux orchestration
Node.js v18 or higher - For MCP servers and build tools
tmux - For multi-agent orchestration (brew install tmux)
Claude Code CLI - Install from Claude
Git - For repository cloning and version control

Optional

Google Cloud Platform account - For Google Workspace automation
Perplexity AI account - For research integration
Folio - Default personal knowledge store; legacy Obsidian is only for explicit vault migration/debugging
Linear API key - For Linear issue tracking integration
GitHub token - For GitHub repository management

Skills

This repository is the canonical source for the shared skill definitions that codex_session_start links into ~/.codex/skills.

`planforge`

Standalone speedrift-native planning and decomposition

Uses a V2 model-mediated planner persona, schema validation, critic pass, and Workgraph materializer
Produces a lightweight spec anchor plus a seeded Workgraph task graph with tests, evals, drift gates, and fix loops
Intended to be shared and invoked directly as the planning entry point
When working in a speedrift repo, use planforge first, then hand off to speedrift for execution and drift control
Runner: skills/planforge/scripts/planforge_v2.py

`speedrift`

Workgraph-first drift-control autopilot for agentic software delivery

Orchestrates speedrift baseline + optional specdrift, datadrift, depsdrift, uxdrift, therapydrift, yagnidrift, redrift
Standard start/resume protocol for repos (driftdriver install, contracts, per-task drift checks)
Delegates planning/decomposition to planforge, then drives execution and drift control
Advisory-by-default drift handling with follow-up task redirects (no hard blocks)

`rifts` (alias)

Backward-compatible alias that routes to the speedrift workflow.

`orchestrating-tmux-claudes`

Multi-AI coordination in tmux panes (repo)

Sequential workflows (TDD → code → review → deploy → test)
Parallel dispatch (independent tasks)
Multi-model support (Claude Code, codex, ollama, haiku)
Auto-recovery for stuck agents
Smart integration (auto-commit, feature gates)

`gemini-cli-extension-porting`

Gemini CLI extension workflows for beads and Superpowers

Map skills to Gemini CLI command namespaces
GEMINI.md context and artifact guidance
Prompt workflows paired with hard guardrails (pre-commit/CI)
MCP and tool wiring patterns for extensions

`google-docs-automation`

Google Workspace automation (Docs, Sheets, Slides)

Uses the local gws-account profile wrapper around Google Workspace CLI
Common patterns: Drive, Docs, Sheets, Slides, Calendar, Tasks, and People
Email search/read is routed to msgvault, not Gmail/GWS
PAIA outbound email is routed through n8n workflows
Integration with financial modeling and reporting workflows

`keynote-slides`

Build Keynote-style single-file HTML slide decks

Brand-ready templates with multi-brand theming
Gemini-powered infographic generation (nano-banana model)
Narrative Engine with 17 storytelling frameworks
5-agent review panel for deck quality assurance
Review mode with feedback system for collaborative iteration
Content Database for entity-scoped approved-content compliance
Single portable HTML file per deck

`white-paper`

Create print-ready white papers in single HTML files

Professional 8.5x11 layout optimized for print
Strong brand adherence with entity-level theming
Research integration and visual review workflow
Gemini media generation for diagrams and charts
PDF export with proper page breaks
Collaborative document building with co-author workflow

`perplexity-research`

Perplexity API for web research

Structured query design for Claude-interpretable output
Market research, competitive intelligence, technical documentation
Citation tracking and verification
Integration with Claude analysis workflows

`obsidian-vault-intelligence`

Legacy Obsidian MCP usage patterns (repo)

Use only when explicitly working with the legacy Obsidian vault, migration, or Obsidian MCP debugging
Progressive search strategies (query_vault, intelligent_search, search_notes)
Note operations (create, update sections, append)
Cross-reference intelligence (backlinks, contextual companions)
Integration with task management and research workflows
18 available tools for comprehensive vault operations

`reminders-management`

Apple Reminders MCP integration (repo)

List organization and priority management
Time-based task tracking with due dates
Context links to Folio by default; legacy Obsidian sync only when explicitly requested
Integration with coordination workflows
6 available tools for reminder management

`linear`

Linear issue tracking integration

Create and manage Linear issues via bash scripts
GraphQL API wrapper with simple CLI interface
Project and workflow state management
Issue search and filtering
5 bash scripts for Linear workflow automation

`github`

GitHub repository and issue management

Create and manage GitHub issues and pull requests
REST API wrapper with simple CLI interface
Code search across repositories
Label and assignee management
5 bash scripts for GitHub workflow automation

`workspace-cleanup`

Intelligent workspace cleanup with multi-signal detection

Exact duplicate detection via SHA256 hashing (auto-archives older copies)
Three-layer protection: .gitignore + .cleanupignore + hard-coded patterns
Multi-signal analysis: Similarity + timestamps + import/reference detection
Tiered confidence scoring: 100% safe auto-archive, high confidence, low confidence
Two-stage safety: Central archive with 30-day review before permanent deletion
Context-aware scoping: Targeted or recursive based on user intent
Tested & validated: Handles file drift from AI-generated code variations
Reduces AI context pollution from temp files, duplicates, and obsolete versions

Additional Local Skills

This repo also publishes local skills that may be installed by Codex/OpenAI agents:

outreach-engine - Draft-only outbound outreach preparation
prospecting-partner - Evidence-backed target and relationship research
golden-voice-packets - Canonical human and agent voice packet routing
paia-program-management - Model-mediated operational program management across Workboard, PAIA Program, Linear, GitHub, Google, and Folio
paia-workboard - PAIA/Workgraph task-board operations
recover-after-reboot - tmux and repo recovery after OS restarts
roborev-review, roborev-review-branch, roborev-design-review, roborev-design-review-branch, and roborev-refine - review daemon workflows
webwright - Playwright-based browser automation for real web workflows

See docs/public-skill-inventory.md for the complete 72-skill index.

Scripts

tmux Orchestration (`scripts/tmux/`)

launch-claude-team.sh - Launch 5-pane tmux session
delegate.sh - Send tasks to specific panes
check-pane.sh - Monitor specific pane
check-all-panes.sh - Monitor all panes
recover-pane.sh - Kill and recover stuck agent

Installation (`scripts/install/`)

install-tmux-orchestration-local.sh - Local project installation

Documentation

Workflow References (`docs/`)

Reference materials for development workflows and patterns.

Installation

Clone this repository once, then expose its skills/* directories to each agent runtime you use.

hljs language-bash

git clone https://github.com/dbmcco/claude-agent-toolkit ~/claude-agent-toolkit
cd ~/claude-agent-toolkit

# Verify installation
find skills -maxdepth 2 -name SKILL.md | wc -l

Claude Code

Claude Code can read a skill bundle from ~/.claude/skills:

hljs language-bash

mkdir -p ~/.claude/skills
ln -sfn "$PWD" ~/.claude/skills/claude-agent-toolkit

Restart Claude Code, then ask it to list available skills.

Codex

Codex expects skill directories directly under ~/.codex/skills. Symlink each published skill:

hljs language-bash

mkdir -p ~/.codex/skills
for skill in "$PWD"/skills/*; do
  [ -f "$skill/SKILL.md" ] || continue
  ln -sfn "$skill" ~/.codex/skills/"$(basename "$skill")"
done

Restart Codex so it reloads the skill index.

Pi / Generic OpenAI Agents

Local OpenAI-style agents in this workspace use ~/.agents/skills:

hljs language-bash

mkdir -p ~/.agents/skills
for skill in "$PWD"/skills/*; do
  [ -f "$skill/SKILL.md" ] || continue
  ln -sfn "$skill" ~/.agents/skills/"$(basename "$skill")"
done

Restart the agent process after updating symlinks.

OpenCode

For repository-local OpenCode usage, expose skills under .opencode/skills in the target project:

hljs language-bash

mkdir -p .opencode/skills
for skill in ~/claude-agent-toolkit/skills/*; do
  [ -f "$skill/SKILL.md" ] || continue
  ln -sfn "$skill" .opencode/skills/"$(basename "$skill")"
done

If your OpenCode install uses a global config directory, use the same symlink loop with that directory instead.

Updating

hljs language-bash

cd ~/claude-agent-toolkit
git pull --rebase

Symlink installs pick up skill file changes immediately; restart the agent if it caches its skill list.

Configuration

Each skill that requires external services needs specific environment variables configured.

Google Workspace Automation

Required for skills/google-docs-automation/

Use the local GWS profile wrapper:

hljs language-bash

<workspace-root>/gws-profiles/bin/gws-account list
<workspace-root>/gws-profiles/bin/gws-account primary calendar calendarList list --params '{"maxResults":1}'

Choose the profile first: primary for Example Company, client-a for a client workspace, and personal for a personal account. If a profile auth check reports auth_method: none, stop and reauthorize that profile instead of falling back to another account.

Profiles and OAuth setup live in <workspace-root>/gws-profiles. Do not build custom OAuth flows or use service-account code for normal agent work. Use msgvault for email search/read and a draft-only workflow for outbound email.

Important: Add credentials files to .gitignore:

hljs language-bash

*.json
.env

Perplexity Research Integration

Required for skills/perplexity-research/

Get API key from Perplexity AI
Configure environment:

hljs language-bash

# In your project .env or ~/.bashrc
export PERPLEXITY_API_KEY="your_api_key_here"

Legacy Obsidian Vault Intelligence

Required for skills/obsidian-vault-intelligence/

Folio is the default personal knowledge system for current agents. Configure Obsidian MCP only for explicit legacy vault work, migration, or debugging.

Install and configure the dbmcco/obsidian-mcp server:

hljs language-bash

# Clone and build
git clone https://github.com/dbmcco/obsidian-mcp.git
cd obsidian-mcp
npm install
npm run build

Configuration File Locations:

Claude Code CLI: ~/.config/claude-code/config.json
Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json

Add to your configuration file (merge with existing mcps section if present):

hljs language-json

{
  "mcps": {
    "obsidian": {
      "command": "node",
      "args": ["/path/to/obsidian-mcp/dist/index.js"],
      "env": {
        "OBSIDIAN_VAULT_PATH": "/path/to/your/vault"
      }
    }
  }
}

See the obsidian-mcp repository for complete documentation of all 18 available tools.

Apple Reminders Management

Required for skills/reminders-management/

Install and configure the dbmcco/apple-reminders-mcp server:

hljs language-bash

# Clone and build
git clone https://github.com/dbmcco/apple-reminders-mcp.git
cd apple-reminders-mcp
npm install
npm run build

Configuration File Locations:

Claude Code CLI: ~/.config/claude-code/config.json
Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json

Add to your configuration file (merge with existing mcps section if present):

hljs language-json

{
  "mcps": {
    "reminders": {
      "command": "node",
      "args": ["/path/to/apple-reminders-mcp/dist/index.js"]
    }
  }
}

See the apple-reminders-mcp repository for complete documentation of all 6 available tools.

Linear Integration

Required for skills/linear/

Get Linear API Key
- Go to Linear Settings
- Create a personal API key
- Copy the key (starts with lin_api_)
Get Team Information
- Find your team ID in Linear settings
- Note your team key (e.g., "EX", "ENG")
- Note your organization URL slug
Configure Environment

hljs language-bash

# In your project .env or ~/.bashrc
export LINEAR_API_KEY="lin_api_xxx"
export LINEAR_TEAM_ID="xxx-xxx-xxx"
export LINEAR_TEAM_KEY="YOUR_TEAM"
export LINEAR_ORG_URL="your-org"

Usage:

hljs language-bash

# Create issue
skills/linear/scripts/linear-create-issue.sh --title "Bug fix" --description "Details"

# List issues
skills/linear/scripts/linear-list-issues.sh --state "In Progress"

# Update issue
skills/linear/scripts/linear-update-issue.sh --issue-id "ID" --state-id "state-done"

See skills/linear/SKILL.md for complete documentation.

GitHub Integration

Required for skills/github/

Create GitHub Personal Access Token
- Go to GitHub Settings > Developer settings > Personal access tokens
- Click "Generate new token (classic)"
- Select scopes: repo, workflow, admin:org (as needed)
- Generate and copy the token
Configure Environment

hljs language-bash

# In your project .env or ~/.bashrc
export GITHUB_TOKEN="<github-token>"

Usage:

hljs language-bash

# Create issue
skills/github/scripts/gh-create-issue.sh --owner org --repo repo --title "Bug"

# List issues
skills/github/scripts/gh-list-issues.sh --owner org --repo repo --state open

# Create PR
skills/github/scripts/gh-create-pr.sh --owner org --repo repo --title "Feature" --head branch

See skills/github/SKILL.md for complete documentation.

tmux Orchestration

Required for skills/orchestrating-tmux-claudes/

Scripts Included in This Repository

The tmux orchestration scripts are included in this repository under scripts/tmux/:

launch-claude-team.sh - Launch 5-pane tmux session
delegate.sh - Send tasks to specific panes
check-pane.sh - Monitor specific pane output
check-all-panes.sh - Monitor all panes
recover-pane.sh - Kill and recover stuck agents

Installation:

Install tmux:

hljs language-bash

brew install tmux

Install local scripts (if you cloned to a custom location):

hljs language-bash

# From the claude-agent-toolkit directory
./scripts/install/install-tmux-orchestration-local.sh

This installs the scripts and makes them available globally.

Role-Based Skills Pattern:

The orchestration skill enforces a coordinator-only role. When launching tmux sessions:

Coordinator Claude (you): Has all skills including orchestration
Worker Claudes (in panes): Should use role-appropriate skills only

See the skill's SKILL.md for role configuration details.

Full Implementation Available:

For the complete implementation with working scripts, see: dbmcco/claude-tmux-orchestration

This repository provides:

✅ Working scripts with automated testing
✅ Quality gates (ESLint, TypeScript, tests, coverage)
✅ Setup wizard for new projects
✅ Automated tests (13 tests passing)
✅ Documentation (installation, testing, troubleshooting)

Quick start:

hljs language-bash

git clone https://github.com/dbmcco/claude-tmux-orchestration.git
cd claude-tmux-orchestration
./setup.sh

Development

Editing Skills Locally

hljs language-bash

# Clone the repository
git clone https://github.com/dbmcco/claude-agent-toolkit
cd claude-agent-toolkit

# Edit any skill
vi skills/orchestrating-tmux-claudes/SKILL.md

# Symlink to your Claude Code config for testing
ln -s $(pwd) ~/.claude/skills/advanced

Creating New Skills

Follow the Skill Development Guide for patterns and best practices

Structure

hljs language-objectivec

claude-agent-toolkit/
├── skills/
│   ├── orchestrating-tmux-claudes/
│   │   ├── SKILL.md
│   │   └── reference.md
│   ├── google-docs-automation/
│   │   └── SKILL.md
│   ├── keynote-slides/
│   │   ├── SKILL.md
│   │   ├── assets/
│   │   ├── references/
│   │   ├── scripts/
│   │   └── skills/
│   ├── white-paper/
│   │   ├── SKILL.md
│   │   ├── assets/
│   │   ├── references/
│   │   └── scripts/
│   ├── perplexity-research/
│   │   └── SKILL.md
│   ├── obsidian-vault-intelligence/
│   │   └── SKILL.md
│   ├── reminders-management/
│   │   └── SKILL.md
│   ├── linear/
│   │   ├── SKILL.md
│   │   └── scripts/ (5 bash scripts)
│   └── github/
│       ├── SKILL.md
│       └── scripts/ (5 bash scripts)
├── scripts/
│   ├── tmux/
│   └── install/
├── docs/
├── LICENSE
├── .gitignore
└── README.md

Philosophy

Progressive disclosure - Skills load only when relevant
Focused workflows - One skill, one purpose
Quality gates - Enforce best practices
Evidence-based - Verification before claims
Systematic - Process over guessing

Contributing

Contributions welcome! Please:

Fork the repository
Create a feature branch
Follow existing skill structure and patterns
Add clear configuration documentation
Submit a pull request

See docs/skill-development-guide.md for detailed guidance.

Credits

Built with Claude by Anthropic
Built using obra's superpowers framework
Created by Project Contributors

License

MIT License - See LICENSE for details

Related Projects

obra/superpowers - Skill framework this toolkit is built on
Claude Code - AI-powered CLI development tool
dbmcco/obsidian-mcp - Legacy Obsidian MCP server for vault integration
dbmcco/apple-reminders-mcp - Apple Reminders MCP server
dbmcco/claude-tmux-orchestration - Multi-agent tmux coordination framework

claude-agent-toolkit

Claude Agent Toolkit

Overview

Prerequisites

Required

Optional

Skills

planforge

speedrift

rifts (alias)

orchestrating-tmux-claudes

gemini-cli-extension-porting

google-docs-automation

keynote-slides

white-paper

perplexity-research

obsidian-vault-intelligence

reminders-management

linear

github

workspace-cleanup

Additional Local Skills

Scripts

tmux Orchestration (scripts/tmux/)

Installation (scripts/install/)

Documentation

Workflow References (docs/)

Installation

Claude Code

Codex

Pi / Generic OpenAI Agents

OpenCode

Updating

Configuration

Google Workspace Automation

Perplexity Research Integration

Legacy Obsidian Vault Intelligence

Apple Reminders Management

Linear Integration

GitHub Integration

tmux Orchestration

Development

Editing Skills Locally

Creating New Skills

Structure

Philosophy

Contributing

Credits

License

Related Projects

Similar Packages

Repository

Original Author

Publisher

`planforge`

`speedrift`

`rifts` (alias)

`orchestrating-tmux-claudes`

`gemini-cli-extension-porting`

`google-docs-automation`

`keynote-slides`

`white-paper`

`perplexity-research`

`obsidian-vault-intelligence`

`reminders-management`

`linear`

`github`

`workspace-cleanup`

tmux Orchestration (`scripts/tmux/`)

Installation (`scripts/install/`)

Workflow References (`docs/`)