The best AI in the world should be available to everyone
Right now the most powerful AI setups are being built inside companies for efficiency and profits. But the purpose of technology is to serve humans—not the other way around.
PAI Packs are how we make that happen. Each pack is a battle-tested capability extracted from a production AI system, packaged so anyone can install it. Think of them like learning kung-fu in The Matrix—complete, tested capabilities you can download directly into your system.
These aren't theoretical examples or proof-of-concepts. They're the actual tools and systems running in production every day, now available to you. A full platform for magnifying yourself and your impact on the world.
The old approach was "here's my whole system—clone it and customize." That created a Jenga tower of dependencies where changing one piece broke three others.
Packs are different:
- Self-contained - Works without understanding the rest of the system
- Independently installable - Add what you need, skip what you don't
- Platform-agnostic - Works with Claude Code, OpenCode, Cursor, or custom systems
- AI-installable - Give your DA the pack directory, it handles the rest
The key insight: Give your DA the complete context it needs, and it can integrate the pack into your system autonomously.
Each pack is now a directory containing:
pack-name/
├── README.md # Pack overview, architecture, what it solves
├── INSTALL.md # Step-by-step installation instructions
├── VERIFY.md # Mandatory verification checklist
└── src/ # Actual source code files
├── hooks/ # Hook implementations (if applicable)
├── tools/ # CLI tools and utilities
├── skills/ # Skill definitions and workflows
└── config/ # Configuration files
The previous single-file approach had limitations:
- Token limits - Large packs exceeded 25k token limits
- Code simplification - AI agents would "helpfully" simplify code instead of copying verbatim
- No linting/testing - Code embedded in markdown couldn't be validated
The directory structure provides:
- Real code files - TypeScript, YAML, Handlebars files that can be linted and tested
- Clear separation - README for context, INSTALL for steps, VERIFY for validation
- Verbatim copying - AI agents copy actual files instead of extracting from markdown
Packs use wizard-style installation that leverages Claude Code's native tools:
Phase 1: System Analysis
├── Run detection commands
├── Check prerequisites
├── Find conflicts
└── Present findings to user
Phase 2: User Questions (AskUserQuestion)
├── Q1: Handle missing prerequisites
├── Q2: Resolve conflicts (if any)
├── Q3: Optional features (if any)
└── Q4: Final confirmation
Phase 3: Backup (if needed)
└── Create timestamped backup
Phase 4: Installation (TodoWrite for progress)
├── Create directories
├── Copy files from src/
├── Install dependencies
└── Pack-specific steps
Phase 5: Verification
└── Run all VERIFY.md checks
Key tools used:
- AskUserQuestion - For user decisions at each decision point
- TodoWrite - For progress tracking (user sees what's happening)
- VERIFY.md - For validation (all checks must pass)
See Tools/InstallTemplate.md for the complete INSTALL.md template.
FOR AI AGENTS INSTALLING PACKS: This section is MANDATORY. Read it before installing ANY pack.
AI agents often "helpfully" simplify pack implementations. Instead of copying the full 268-line documented specification, they create a 50-line "equivalent" version. This is WRONG.
Copy the COMPLETE files from src/. No simplification. No shortcuts. No "equivalents."
When a pack contains a 500-line TypeScript file, you copy all 500 lines. When it specifies 8 files, you copy all 8 files. When it includes detailed YAML with 100 entries, you include all 100 entries.
- Packs are production-tested - every line exists for a reason
- "Simplified" versions break integration with other packs
- Missing configuration causes silent failures
- Users shouldn't need multiple correction cycles
- NEVER reduce code complexity - If the pack has verbose implementations, keep them verbose
- NEVER skip optional sections - "Optional" means optional for the user, not for you to omit
- NEVER create "equivalent" versions - Copy the EXACT files from
src/ - NEVER summarize YAML/JSON - Include every field, every entry, every comment
- ALWAYS complete VERIFY.md - If any check fails, installation is NOT complete
After every pack installation, you MUST complete the VERIFY.md checklist:
## Installation Verification
- [ ] All directories from INSTALL.md created
- [ ] All files from `src/` copied to target locations
- [ ] Each code file contains COMPLETE implementation (diff check)
- [ ] YAML/JSON files contain ALL entries (not summarized)
- [ ] Dependencies installed (bun add, etc.)
- [ ] Verification commands from VERIFY.md run and passedIf ANY checkbox fails, you have NOT completed the installation.
PAI organizes capabilities in a clear hierarchy:
PAI System
└── Bundles (curated collections for specific goals)
└── Packs (individual capabilities)
└── Contents (code, hooks, tools, workflows, config)
- Bundles group related packs that work well together
- Packs are self-contained capabilities you can install independently
- Contents are the actual code, hooks, tools, and configuration inside each pack
| Pack | Category | Description |
|---|---|---|
| pai-hook-system | Foundation | Event-driven automation framework - the foundation for all hook-based capabilities |
| pai-core-install | Core | Skills + Identity + Architecture - the complete foundation with routing, response format, MEMORY system |
| pai-voice-system | Notifications | Voice notifications with ElevenLabs TTS and prosody enhancement for natural speech |
| pai-observability-server | Observability | Real-time multi-agent monitoring dashboard with WebSocket streaming |
| pai-statusline | Display | 4-mode responsive status line with learning signal, context usage, and trend indicators |
| Pack | Category | Description |
|---|---|---|
| pai-agents-skill | Delegation | Dynamic agent composition with unique personalities, voices, and trait combinations |
| pai-algorithm-skill | Methodology | The Algorithm implementation - ISC management, effort classification, verifiable iteration |
| pai-annualreports-skill | Research | Annual security report aggregation and threat landscape analysis |
| pai-art-skill | Creativity | Visual content generation with multi-reference image support and technical diagrams |
| pai-brightdata-skill | Scraping | Progressive URL scraping with Bright Data integration and tier escalation |
| pai-browser-skill | Automation | Debug-first browser automation with Playwright - always-on diagnostics, session auto-start |
| pai-council-skill | Analysis | Multi-agent debate system for exploring perspectives and reaching consensus |
| pai-createcli-skill | Development | Generate TypeScript CLI tools with bun runtime |
| pai-createskill-skill | Development | Create and validate PAI skills with proper structure |
| pai-firstprinciples-skill | Analysis | First principles decomposition and root cause analysis |
| pai-osint-skill | Research | Open source intelligence gathering and due diligence |
| pai-privateinvestigator-skill | Research | Ethical people-finding for reconnection and verification |
| pai-prompting-skill | Methodology | Meta-prompting system with Handlebars templates and Claude best practices |
| pai-recon-skill | Security | Security reconnaissance, bug bounty, and attack surface mapping |
| pai-redteam-skill | Security | Adversarial analysis with 32 specialized agents for stress testing ideas |
| pai-research-skill | Research | Multi-source research with parallel agent execution and Fabric patterns |
| pai-system-skill | Maintenance | System integrity checks, documentation updates, and security scanning |
| pai-telos-skill | Life OS | Deep goal capture framework - mission, goals, beliefs, strategies, learnings |
Packs have dependencies. Install in this order:
Required (install first):
1. pai-hook-system ← Foundation (no dependencies)
2. pai-core-install ← Depends on hooks, includes MEMORY system
Infrastructure (install next):
3. pai-statusline ← Depends on core-install
4. pai-voice-system ← Depends on hooks, core-install
5. pai-observability-server ← Depends on hooks
Skills (install any you need):
6+. pai-*-skill ← Most depend only on core-install
Or install the complete PAI Bundle which handles ordering automatically.
Or use a Full Release from Releases/v2.5/ which includes everything pre-configured.
Give the pack directory to your DA agent:
Install the pai-hook-system pack from PAI/Packs/pai-hook-system/.
Use PAI_DIR="~/.config/pai" and DA="MyAI".
Your DA will:
- Read
README.mdfor context - Follow
INSTALL.mdstep by step - Copy files from
src/to your system - Complete
VERIFY.mdchecklist
- Open the pack's
INSTALL.md - Follow each step, copying files from
src/to the specified locations - Complete the
VERIFY.mdchecklist to confirm success
All API keys live in ONE place: $PAI_DIR/.env
Packs that require API keys (Voice, Art, etc.) all read from a single environment file. No keys should ever be stored in pack files, configs, or code.
# Copy the example and add your keys
cp ../.env.example $PAI_DIR/.env
nano $PAI_DIR/.envSee .env.example for the complete list of supported variables.
| Category | Purpose | Examples |
|---|---|---|
| Foundation | Core infrastructure that other packs depend on | Hook System |
| Infrastructure | Systems that run automatically in the background | History System |
| Observability | Real-time monitoring and debugging tools | Observability Server |
| Routing | Intent detection and capability dispatch | Skill System |
| Notifications | User feedback and accessibility | Voice System |
| Personality | Identity, response format, principles | Identity |
| Delegation | Agent orchestration and parallel execution | Agents Skill |
| Creativity | Visual and creative content generation | Art Skill |
| Methodology | Prompt engineering and meta-prompting | Prompting Skill |
| Automation | Browser automation and web verification | Browser Skill |
See PAIPackTemplate.md for the complete pack specification. See InstallTemplate.md for the wizard-style INSTALL.md template.
Quick checklist for directory-based packs:
-
README.mdwith YAML frontmatter, problem/solution, architecture -
INSTALL.mdwith wizard-style phases (analysis, questions, install, verify) -
VERIFY.mdwith mandatory completion checklist -
src/directory with actual code files (not embedded in markdown) - 256x256 transparent icon in
icons/ - Complete, working code (no snippets or placeholders)
Every pack in this directory must be:
- Complete - Everything needed to install, no missing components
- Self-contained - Works without external documentation
- Tested - Verified working in production (extracted from Kai)
- Sanitized - No personal data, credentials, or hardcoded paths
- Documented - Clear problem statement, solution, and examples
- Verifiable - VERIFY.md checklist confirms successful installation
If you have existing single-file packs:
- Create directory with pack name (e.g.,
pai-hook-system/) - Extract frontmatter and overview sections to
README.md - Extract installation steps to
INSTALL.md - Extract verification steps to
VERIFY.md - Extract code blocks to actual files in
src/ - Update any pack references to use directory paths
- Bundles/ - Curated collections of packs
- PAIPackTemplate.md - Full pack specification
- README.md - PAI project overview
Each pack is extracted from Daniel Miessler's production Kai system - battle-tested capabilities packaged for anyone to use.