Our first attempt at a general problem solver built into PAI to pursue Euphoric Surprise
v2.4 introduces The Algorithm—a structured, 7-phase approach to problem-solving. It's an experiment in making AI work more verifiable.
Most AI assistants work like this:
User asks → AI responds → Hope it's right
The Algorithm works like this:
User asks → Understand true intent → Build verifiable criteria →
Execute systematically → Verify against criteria → Learn from result
The idea is that if you can define testable criteria upfront, you can actually verify whether something worked.
| Phase | Name | Purpose |
|---|---|---|
| 1 | OBSERVE | Gather information, understand current state |
| 2 | THINK | Analyze intent, desired outcome, failure modes |
| 3 | PLAN | Build Ideal State Criteria (ISC) - testable success conditions |
| 4 | BUILD | Construct the solution components |
| 5 | EXECUTE | Take action, track progress against criteria |
| 6 | VERIFY | Confirm all criteria met with evidence |
| 7 | LEARN | Capture insights, define next steps |
The core mechanism is Ideal State Criteria—exactly 8-word, binary-testable conditions that define success:
🎯 ISC TRACKER ════════════════════════════════════════════════════════════════
│ # │ Criterion (exactly 8 words) │ Status │ Evidence │
├───┼────────────────────────────────────┼─────────────────┼────────────────┤
│ 1 │ All unit tests pass after fix applied │ ✅ VERIFIED │ pytest: 47/47 │
│ 2 │ No regressions in existing functionality │ ✅ VERIFIED │ CI green │
│ 3 │ Error message matches user expectation │ ✅ VERIFIED │ Screenshot │
└───┴────────────────────────────────────┴─────────────────┴────────────────┘
SCORE: 3/3 verified │ ANTI: 0 triggered │ RESULT: COMPLETE
Why 8 words? It forces precision. You can't be vague in 8 words. Each criterion should be testable in under 2 seconds.
The Algorithm also tracks what must NOT happen:
⚠️ ANTI-CRITERIA CHECK
│ ! │ No credentials exposed in output │ ✅ AVOIDED │
│ ! │ No breaking changes to public API │ ✅ AVOIDED │
This prevents the common AI failure mode of solving one problem while creating another.
The Algorithm has access to various capabilities it can use during each phase:
| Capability | Purpose | When to Use |
|---|---|---|
| Skills | Pre-built expertise for specific domains | Domain knowledge needed |
| Agents | Specialized personalities for focused work | Parallel work, delegation |
| Research | Multi-model research with synthesis | Information gathering |
| Red Team | Adversarial analysis, stress testing | Validating ideas |
| First Principles | Fundamental analysis without assumptions | Complex problems |
| Be Creative | Extended thinking, expanded creativity | Ideation, novel solutions |
| Browser | Visual verification, screenshots | UI testing |
| Plan Mode | Deep analysis before implementation | Major work |
Before each phase, the system considers which capabilities might help. The goal is to avoid defaulting to "just do it directly" when a skill or agent would do better.
| Feature | Description |
|---|---|
| The Algorithm | 7-phase problem-solving with ISC tracking |
| 29 Skills | Up from 20—expanded domain coverage |
| 15 Hooks | Refined hook architecture |
| 331 Workflows | Comprehensive automation library |
| Improved Wizard | Better interactive setup with voice support |
| Cross-Platform | Enhanced bash/zsh detection and support |
| Voice Notifications | ElevenLabs integration for spoken alerts |
| Observability Dashboard | Real-time system monitoring |
- Faster hook execution — Streamlined hook architecture
- Reduced context overhead — Smarter context loading
- Parallel agent execution — Multiple agents work simultaneously
- Incremental file operations — Only process what changed
- Interactive wizard with readline prompts (not just piped stdin)
- Voice ID selection — Choose from Adam, Rachel, or Antoni voices
- Auto shell detection — Works with bash and zsh automatically
- Projects directory configuration — Set your workspace path
- Custom catchphrase — Personalize your DA's greeting
- Cleaner settings.json — Template with sensible defaults
- Sanitized paths — No hardcoded
/Users/daniel/leaks - Better hooks organization — Clear lib/ separation
- Allowlist enforcement — Only approved content in releases
.claude/
├── PAIInstallWizard.ts # Interactive setup wizard
├── settings.json # Template configuration
├── statusline-command.sh # 4-mode responsive status line
├── statusline-debug.sh # Status line debugging
│
├── hooks/ # 15 event hooks
│ ├── LoadContext.hook.ts
│ ├── ExplicitRatingCapture.hook.ts
│ ├── ImplicitSentimentCapture.hook.ts
│ └── ... (12 more)
│
├── skills/ # 29 production skills
│ ├── CORE/ # Algorithm and system core
│ ├── Agents/ # Agent personalities
│ ├── Art/ # Visual content creation
│ ├── Browser/ # Browser automation
│ ├── Research/ # Multi-model research
│ └── ... (24 more)
│
├── agents/ # Named agent definitions
├── lib/ # Shared utilities
├── MEMORY/ # Learning capture system
├── Observability/ # Real-time dashboard
└── VoiceServer/ # Voice notification server
# 1. Clone the repo (if you haven't already)
git clone https://github.com/danielmiessler/PAI.git
cd PAI/Releases/v2.4
# 2. Backup existing installation (if any)
[ -d ~/.claude ] && mv ~/.claude ~/.claude-backup-$(date +%Y%m%d)
# 3. Copy the complete release
cp -r .claude ~/
# 4. Run the configuration wizard
cd ~/.claude && bun run PAIInstallWizard.ts
# 5. Restart Claude Code to activate hooks| Prompt | Purpose |
|---|---|
| Your name | Personalization throughout the system |
| Projects directory | Where your work lives (default: ~/Projects) |
| AI name | What to call your assistant (default: Kai) |
| Startup catchphrase | What your DA says on startup |
| ElevenLabs API key | Optional voice notifications |
The wizard auto-detects your shell and configures the appropriate file:
- zsh →
~/.zshrc - bash →
~/.bashrc
| Skill | Purpose |
|---|---|
| CORE | The Algorithm, steering rules, identity |
| Agents | Agent personalities and spawning |
| PAIUpgrade | System improvement extraction |
| Skill | Purpose |
|---|---|
| Research | Multi-model parallel research |
| OSINT | Open source intelligence gathering |
| PrivateInvestigator | Ethical people-finding |
| AnnualReports | Security report aggregation |
| SECUpdates | Security news monitoring |
| Skill | Purpose |
|---|---|
| Art | Visual content, diagrams, icons |
| BeCreative | Extended thinking mode |
| FirstPrinciples | Fundamental analysis |
| RedTeam | Adversarial validation |
| Council | Multi-agent debate |
| Prompting | Meta-prompt generation |
| Skill | Purpose |
|---|---|
| Browser | Playwright automation |
| CreateCLI | TypeScript CLI generation |
| CreateSkill | Skill structure creation |
| Evals | Agent evaluation framework |
| Documents | Document processing |
| Skill | Purpose |
|---|---|
| Telos | Life goals and projects |
| Recon | Security reconnaissance |
| WebAssessment | Web security testing |
| PromptInjection | LLM security testing |
| BrightData | Progressive URL scraping |
| Apify | Social media scraping |
| Fabric | 240+ prompt patterns |
| Aphorisms | Quote management |
| VoiceServer | Voice notification control |
Specialized personalities for focused work:
| Agent | Specialty |
|---|---|
| Algorithm | ISC tracking, verification work |
| Architect | System design, distributed systems |
| Engineer | TDD, implementation patterns |
| Artist | Visual content, prompt engineering |
| Designer | UX/UI, accessibility |
| QATester | Browser automation, verification |
| Pentester | Security testing |
| Intern | High-agency generalist |
| GeminiResearcher | Multi-perspective research |
| GrokResearcher | Contrarian analysis |
| CodexResearcher | Technical archaeology |
| ClaudeResearcher | Academic synthesis |
The target we're aiming for is Euphoric Surprise—results thorough enough that you're genuinely delighted, not just satisfied.
The Algorithm tries to make this measurable:
- Ideal State — What would make you say "wow"?
- Criteria — Break that into testable conditions
- Verification — Prove each condition is met
- Learning — Capture what worked for next time
Whether it actually achieves this consistently is still being figured out.
v2.4 is a significant upgrade. Recommended approach:
# 1. Backup your current installation
mv ~/.claude ~/.claude-v2.3-backup
# 2. Install v2.4
cp -r .claude ~/
# 3. Run the wizard
cd ~/.claude && bun run PAIInstallWizard.ts
# 4. Migrate personal content
# Copy USER/ and MEMORY/ from backup if desiredBreaking changes:
install.ts→PAIInstallWizard.ts- Hook architecture refined (check custom hooks)
- Settings structure updated (wizard handles migration)
- GitHub: github.com/danielmiessler/PAI
- Video: PAI Overview
- Philosophy: The Real Internet of Things
PAI v2.4.0 — The Algorithm
An experiment in verifiable problem-solving.