PAI v4.0.3 — Community PR Patch

4 community-contributed fixes — no new features, no breaking changes.

What Changed

Inference & Parsing

• #800 Inference.ts JSON parsing only matched objects {} — now handles arrays [] too, with validation via JSON.parse

Documentation & Portability

• #836 CONTEXT_ROUTING.md had 29 dead references to files removed in v4.0 — consolidated to 4 README pointers
• #817 WorldThreatModelHarness hardcoded ~/.claude/ path — now uses $PAI_DIR for portability

Installer

• #846 Upgrading from v2.5/v3.0 stranded user context at skills/PAI/USER/ — installer now migrates files to PAI/USER/ and creates symlinks for backwards compatibility

Installation

git clone https://github.com/danielmiessler/Personal_AI_Infrastructure.git
cd Personal_AI_Infrastructure/Releases/v4.0.3
cp -r .claude ~/ && cd ~/.claude && bash install.sh

See the Releases directory for all versions and upgrade instructions.

Contributors

Thank you to @rikitikitavi2012-debug, @Spirotot, and @catchingknives for these contributions!

PAI v4.0.1 — Upgrade Path & Preferences

PAI v4.0.1 — Upgrade Path & Preferences

Patch release addressing community feedback on upgrade difficulties, missing preferences, and statusline bugs.

What Changed

1. Upgrade Path Documentation

The main README now has separate Fresh Install and Upgrading from a Previous Version sections with explicit backup, merge, and post-upgrade checklist steps.

2. Temperature Unit Preference

New preferences.temperatureUnit field in settings.json — choose Fahrenheit or Celsius. The installer prompts for it during setup, and the statusline honors the setting.

3. Statusline Bug Fixes

Community-reported fixes from PRs #762, #780, and #806:

• Hardcoded timezone removed — reads principal.timezone from settings.json instead of hardcoded America/Los_Angeles
• Broken context fallback removed — /clear no longer produces stale context percentages
• Startup estimate removed — the self-calibrating estimate that inflated fresh session context % is gone
• f-string syntax fix — nested escaped double quotes inside Python f-strings caused parse errors on some shells

4. FAQ Fixes

• Removed stale Python reference — PAI v4.0 is TypeScript and Bash, not Python
• Improved "What if I break something?" answer with backup/merge/preserve guidance

Files Changed (from v4.0.0)

File	Change
settings.json	Added preferences.temperatureUnit field
statusline-command.sh	Temp unit pref, dynamic timezone, context/startup/f-string fixes
PAI-Install/engine/types.ts	Added temperatureUnit to collected data and PAIConfig
PAI-Install/engine/config-gen.ts	Outputs preferences section
PAI-Install/engine/actions.ts	Temp unit prompt in identity step, preserves preferences on upgrade

Quick Manual Upgrade (from v4.0.0)

If you just want the fixes without re-running the installer:

1. Add to your ~/.claude/settings.json:

"preferences": {
  "temperatureUnit": "celsius"
}

2. Copy the updated statusline-command.sh from this release over your existing one.
3. Delete your weather cache: rm -f ~/.claude/MEMORY/STATE/weather-cache.json

Full Release Notes →

PAI v4.0.0 — Lean and Mean

PAI v4.0.0 — Lean and Mean

38 flat skill directories compressed into 12 hierarchical categories. Dead systems removed. Context footprint cut in half.

Highlights

1. Skill Reorganization — 38 flat directories compressed into 12 hierarchical categories with clear ownership boundaries.
2. Algorithm v3.5.0 — Major reasoning upgrade with improved phase execution and ISC tracking.
3. CLAUDE.md Template System — BuildCLAUDE.ts + SessionStart hook generates CLAUDE.md from template using settings.json variables.
4. Security Sanitization — 33+ files cleaned of hardcoded paths, API keys, and personal data.
5. Context Reduction — ~19% context at startup (down from ~38% in v3.0).

Stats

• 63 Skills (up from 38)
• 21 Hooks (up from 20)
• 180 Workflows (up from 162)
• Algorithm v3.5.0 (up from v1.4.0)
• 13 Skill Categories

Breaking Changes

• Skill directories restructured — flat skills/SkillName/ → hierarchical skills/Category/SkillName/
• CLAUDE.md is now generated from template — edit CLAUDE.md.template instead
• Several deprecated systems removed (old voice server, legacy migration tools)

Full Release Notes →

PAI v3.0.0 — The Algorithm Matures

PAI v3.0.0 — The Algorithm Matures

Constraint extraction, build drift prevention, persistent PRDs, and parallel loop execution

Seven Major Features

1. Algorithm v1.4.0 — From prototype to production. Constraint extraction, self-interrogation, build drift prevention, verification rehearsal, loop mode with parallel workers, plan mode integration.
2. Full Installer System — Electron GUI wizard + CLI fallback with auto-discovery of API keys and template-based settings generation.
3. 10 New Skills — IterativeDepth, Science, Remotion, WorldThreatModelHarness, WriteStory, Evals, USMetrics, ExtractWisdom, Cloudflare, Sales.
4. Agent Teams / Swarm — Coordinated multi-agent execution with shared task lists, parallel workers, and message passing.
5. Persistent PRDs — Requirements documents that survive across sessions with full lifecycle tracking.
6. Voice Personality System — Configurable personality traits that shape vocal expression.
7. Inline Verification Methods — Every ISC criterion tagged with explicit verification approach (CLI, Test, Static, Browser, Grep, Read, Custom).

Stats

• 38 Skills (up from 28)
• 20 Hooks (up from 17)
• 162 Workflows
• 1,229 Total Files
• Algorithm v1.4.0 (up from v0.2.25)

Breaking Changes

• INSTALL.ts replaced by PAI-Install/ directory (full installer system)
• Algorithm v0.2.x → v1.4.0 (complete reasoning rewrite)
• Voice server includes Qwen3 local TTS alongside ElevenLabs
• settings.json structure updated (installer handles generation)

Full Release Notes →

---

The version where the Algorithm stopped being a prototype and started being systematic.

PAI v2.5.0 — Think Deeper, Execute Faster

PAI v2.5.0 — Think Deeper, Execute Faster

The Algorithm learns to think about how it thinks—and do more work in parallel

---

What Changed

v2.4 introduced The Algorithm—a 7-phase approach to problem-solving with verifiable criteria. It worked, but it had blind spots: it picked capabilities based on habit rather than analysis, it skipped thinking tools without justification, and it ran independent tasks one at a time when they could have run simultaneously.

v2.5 fixes all three.

The Problem v2.5 Solves

v2.4's Algorithm:

Pick familiar tools → Execute sequentially → Hope the right capabilities were chosen

v2.5's Algorithm:

Hook suggests capabilities → Reverse-engineer true intent → Validate suggestions against ISC →
Justify every tool inclusion/exclusion → Execute independent work in parallel → Verify

The difference is metacognition. The Algorithm now thinks about how it thinks before it acts.

---

Three Major Features

1. Two-Pass Capability Selection

Previously, capability selection happened implicitly—the system defaulted to familiar patterns (usually Engineer + Research) regardless of what the task actually needed.

Now selection happens in two explicit passes:

Pass 1 — Hook Hints (before the Algorithm starts)

The FormatReminder hook runs AI inference on the raw prompt and suggests capabilities, skills, and thinking tools. These are draft suggestions—a head start, not a decision.

Pass 2 — THINK Validation (after OBSERVE completes)

With the full context of reverse-engineering and ISC criteria, the THINK phase validates every suggestion. Skills get added, removed, or confirmed. Pass 2 is authoritative.

🔍 SKILL CHECK (validate hook hints against ISC):
│ Hook suggested:   CreateSkill:UpdateSkill
│ ISC requires:     Architecture design (not just skill update)
│ Final skills:     CreateSkill:UpdateSkill + Architect consultation

Why two passes? The hook gives a head start from the raw prompt. But OBSERVE changes the picture. A request that looks like a simple skill update might actually require architectural decisions, or might rest on questionable assumptions that need First Principles analysis. Pass 2 catches what Pass 1 cannot see.

---

2. Thinking Tools with Justify-Exclusion

v2.4 had thinking tools (Council, RedTeam, FirstPrinciples, etc.) but they were rarely used because the default was to skip them. v2.5 inverts this: thinking tools are opt-OUT, not opt-IN.

For every request, the Algorithm must evaluate each thinking tool and justify why it is NOT being used:

🔍 THINKING TOOLS ASSESSMENT (justify exclusion):
│ Council:          EXCLUDE — single clear approach, no alternatives to debate
│ RedTeam:          INCLUDE — proposal could fail in non-obvious ways
│ FirstPrinciples:  INCLUDE — requirement rests on unexamined assumption
│ Science:          EXCLUDE — not iterative/experimental
│ BeCreative:       EXCLUDE — clear requirements, no divergence needed
│ Prompting:        EXCLUDE — not generating prompts

The burden of proof is on exclusion. "Too simple" and "already know the answer" are not valid reasons—simple tasks can have hidden assumptions, and confidence without verification is exactly the failure mode this catches.

The Six Thinking Tools

Tool	What It Does	Include When
Council	Multi-agent debate with 3-7 agents	Multiple valid approaches exist
RedTeam	Adversarial analysis with 32 agents	Claims need stress-testing
FirstPrinciples	Deconstruct, challenge, reconstruct	Assumptions need examining
Science	Hypothesis-test-analyze cycles	Iterative experimentation needed
BeCreative	Extended thinking, 5 diverse options	Creative divergence needed
Prompting	Meta-prompting with templates	Prompt optimization needed

---

3. Parallel-by-Default Execution

When the Algorithm identifies independent tasks during BUILD/EXECUTE, they now launch simultaneously as concurrent agents in a single message. Serial execution of independent tasks is a failure mode.

BEFORE (v2.4):          AFTER (v2.5):
Task A → wait           Task A ─┐
Task B → wait           Task B ─┼─ all at once
Task C → wait           Task C ─┘
Task D → done           Task D → done

The rule is simple: If tasks don't depend on each other's output, they run at the same time. Period.

This applies automatically to:

• Multiple file edits with no cross-dependencies
• Multiple research queries on different topics
• Multiple audits or scans of independent systems
• Multiple creation tasks with no shared state

Fan-out is the default pattern for 3+ independent workstreams.

---

Composition Patterns

Capabilities now combine using 7 named patterns, making orchestration explicit and visible:

Pattern	Shape	Example	When
Pipeline	A → B → C	Explore → Architect → Engineer	Sequential domain handoff
TDD Loop	A ↔ B	Engineer ↔ QA	Build-verify cycle until ISC passes
Fan-out	→ [A, B, C]	3 Researchers in parallel	Multiple perspectives needed
Fan-in	[A, B, C] → D	Researchers → Synthesis	Merging parallel results
Gate	A → check → B or retry	Engineer → QA → Deploy or fix	Quality gate before progression
Escalation	A(haiku) → A(sonnet) → A(opus)	Model upgrade on failure	Complexity exceeded model tier
Specialist	Single A	Pentester for security review	One domain, deep expertise

The full Capability Selection block in THINK phase now looks like:

🎯 CAPABILITY SELECTION:
│ Skills:     CreateSkill:UpdateSkill
│ Thinking:   FirstPrinciples, RedTeam
│ Primary:    Architect  — system design decision (ISC #1)
│ Support:    Engineer   — implementation (ISC #2-4)
│ Verify:     QATester   — browser verification (ISC #5)
│ Pattern:    Pipeline
│ Sequence:   Architect → Engineer → QATester
│ Rationale:  Architecture must be decided before implementation begins

---

What's New in v2.5

Major Features

Feature	Description
Two-Pass Capability Selection	Hook hints validated against ISC in THINK phase
Thinking Tools Assessment	6 tools evaluated with justify-exclusion principle
Parallel-by-Default Execution	Independent tasks run concurrently, not sequentially
7 Composition Patterns	Named patterns for combining capabilities
Mandatory AskUserQuestion	All user questions use structured tool, not inline text
28 Skills	Refined from 29—CORE renamed to PAI
17 Hooks	2 new hooks for deeper system awareness
356 Workflows	Expanded automation coverage

Algorithm Upgrades (v0.2.23 → v0.2.25)

Version	Change
v0.2.23	Two-Pass Capability Selection, Skill Check in THINK, FormatReminder hook enrichment
v0.2.24	Thinking Tools Assessment, justify-exclusion principle, mandatory AskUserQuestion
v0.2.25	Parallel-by-default execution, fan-out as default pattern

Structural Changes

• CORE → PAI — The core skill renamed from CORE to PAI for clarity
• Consolidated SKILL.md — Single authoritative source for all PAI components with dynamic context loading
• INSTALL.ts — Wizard renamed from PAIInstallWizard.ts to INSTALL.ts
• INSTALL.md — New installation documentation added
• MEMORY structure — Updated directory organization for learning capture
• Settings template — Cleaner defaults with contextFiles updates

---

Full Release Contents

.claude/
├── INSTALL.ts              # Interactive setup wizard
├── INSTALL.md              # Installation documentation
├── settings.json           # Template configuration
├── statusline-command.sh   # 4-mode responsive status line
├── statusline-debug.sh     # Status line debugging
│
├── hooks/                  # 17 event hooks
│   ├── FormatReminder.hook.ts      # AI-powered depth classification
│   ├── ExplicitRatingCapture.hook.ts
│   ├── ImplicitSentimentCapture.hook.ts
│   ├── RelationshipMemory.hook.ts
│   ├── SoulEvolution.hook.ts
│   └── ... (12 more)
│
├── skills/                 # 28 production skills
│   ├── PAI/                # The Algorithm and system core
│   ├── Agents/             # Agent personalities and spawning
│   ├── Art/                # Visual content creation
│   ├── Browser/            # Playwright automation
│   ├── Research/           # Multi-model parallel research
│   ├── Fabric/             # 235 prompt patterns
│   └── ... (22 more)
│
├── agents/                 # 12 named agent definitions
├── lib/                    # Shared utilities
├── MEMORY/                 # Learning capture system
├── Observability/          # Real-time monitoring dashboard
└── VoiceServer/            # Voice notification server

---

Quick Start

# 1. Clone the repo (if you haven't already)
git clone https://github.com/danielmiessler/PAI.git
...

PAI v2.4.0 — The Algorithm

PAI v2.4.0 — The Algorithm

Our first attempt at a general problem solver built into PAI to pursue Euphoric Surprise.

The Algorithm

v2.4 introduces a structured, 7-phase approach to problem-solving. It's an experiment in making AI work more verifiable.

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

Ideal State Criteria (ISC)

The core mechanism is exactly 8-word, binary-testable conditions that define success. If you can define testable criteria upfront, you can actually verify whether something worked.

What's New

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
Voice Notifications	ElevenLabs integration for spoken alerts
Observability Dashboard	Real-time system monitoring
AllowList Enforcement	Security control for releases

Installation

git clone https://github.com/danielmiessler/PAI.git
cd PAI/Releases/v2.4

[ -d ~/.claude ] && mv ~/.claude ~/.claude-backup-$(date +%Y%m%d)
cp -r .claude ~/

cd ~/.claude && bun run PAIInstallWizard.ts

Named Agents

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

Full Documentation

See Releases/v2.4/README.md for complete documentation.

PAI v2.3.0 - Full Releases Return

PAI v2.3.0 - Full Releases Return

The most significant release since v2.0. PAI v2.3 brings back complete, functional releases while maintaining the modular Packs system. Now you can install PAI two ways: grab individual packs for specific capabilities, or download a complete .claude/ directory and start immediately.

---

🎯 Highlights

Full Releases Are Back

The Releases/v2.3/.claude/ directory contains a complete, functional PAI installation. Copy it to ~/.claude/ and you're running. No assembly required.

This addresses the most common feedback since v2.0: "I love the modular approach, but I just want the whole thing." Now you have both options.

Euphoric Surprise

PAI v2.3 introduces Euphoric Surprise as the north star—the singular pursuit that drives every interaction. Not just task completion, but results so thorough and thoughtful that you're genuinely surprised and delighted. This philosophy is now embedded throughout the documentation and skill design.

Customization as Principle #1

The PAI Founding Principles have been updated. Customization for Your Goals is now Principle #1—PAI exists to help you accomplish your goals in life. It democratizes access to personalized agentic infrastructure that knows your context, preferences, and history.

---

📦 What's in v2.3

Complete Release (Releases/v2.3/.claude/)

Component	Count	Description
Skills	22	Full skill modules with workflows
Named Agents	11	Agents with personalities (Artist, Architect, Engineer, etc.)
Hooks	15	Lifecycle event handlers
Fabric Patterns	100+	Reusable prompt patterns

Skills Included:

• Core: CORE, System, THEALGORITHM
• Research: Research, OSINT, PrivateInvestigator, AnnualReports
• Creative: Art, Prompting, Council, RedTeam
• Development: Browser, CreateCLI, CreateSkill, Agents
• Infrastructure: PAIUpgrade, BrightData, Telos, FirstPrinciples, Recon

Named Agents:
Artist, QATester, Architect, Designer, GeminiResearcher, GrokResearcher, CodexResearcher, Pentester, Intern, Engineer, ClaudeResearcher

Updated Packs (23 total)

All packs updated to v2.3 standards:

• Standardized INSTALL.md with AI wizard format
• Icon references in all READMEs
• Consistent versioning

---

⚡ Installation

Option 1: Full Installation (Recommended for New Users)

# Download v2.3 release
# Copy .claude/ to your home directory
cp -r Releases/v2.3/.claude ~/.claude

# Run installation wizard
cd ~/.claude && bun run install.ts

# Start using PAI
claude

Option 2: Individual Packs (Existing Users)

Browse Packs/ and install specific capabilities. Each pack is self-contained with complete instructions.

---

🔄 Changes Since v2.1.0

Major

• Full releases restored - Complete .claude/ directory in Releases/v2.3/
• 23 packs updated for v2.3 release
• Euphoric Surprise added as north star
• Customization elevated to Principle #1
• Visual README redesign with technical diagrams

Infrastructure

• Retired pai-history-system, migrated to MEMORY in core
• Security system upgrade to directory-based architecture
• Voice integration improvements in pai-core-install
• All pack INSTALL.md standardized to AI wizard format

Documentation

• Comprehensive README overhaul
• ELI5 section for newcomers ("Wait, Hold On—Assume I'm New to All This")
• Updated typing animation and badges
• Logo updated to v7

---

🚀 Getting Started

1. Watch the walkthrough: PAI Overview Video
2. Read the philosophy: The Real Internet of Things
3. Understand the vision: Personal AI Maturity Model

---

📊 Stats

• 34 commits since v2.1.0
• 22 skills in full release
• 11 named agents with personalities
• 23 packs available
• 100+ fabric patterns included

---

PAI - Magnifying human capabilities through personalized AI infrastructure.

Everyone deserves AI that gets better at helping them over time. This is open-source. This is free. This is for everyone.

PAI v2.1.0 - Major Core Upgrade & New Upgrades Skill

PAI v2.1.0 - Major Core Upgrade & New Upgrades Skill

This release includes significant updates to the core infrastructure and introduces a new skill for tracking system upgrades.

Major Updates

pai-core-install v1.2.0 (Major Upgrade)

The core pack received its biggest update yet:

• NEW: MEMORY/ skeleton - 11-directory structure for session history, learnings, decisions, and state management
• NEW: settings.json.template - Complete hook configuration template with all events pre-configured
• USER/ directory (15 templates) - Personal configuration for identity, contacts, preferences
• SYSTEM/ directory (17 templates) - System architecture documentation for skills, hooks, memory, delegation
• Wizard-style installation - The installer now walks you through setup step-by-step with clear phases

This update transforms pai-core-install from a basic foundation into a complete infrastructure framework.

pai-browser-skill v1.2.0

Enhanced browser automation with debug-first philosophy:

• Always-on diagnostics and console capture
• Session auto-start capability
• 99% token savings through file-based MCP pattern

NEW: pai-upgrades-skill v1.0.0

A new skill for tracking and managing PAI system upgrades:

• Monitors for new features and capabilities
• Tracks upgrade opportunities
• Maintains upgrade history
• Helps plan and prioritize system improvements

Documentation Updates

• Updated pack count badge (8 → 10 packs)
• Fixed bundle references (Kai/ → Official/)
• Updated installation order with new skill
• Refreshed version numbers across all documentation

Install

# Fresh install
cd PAI/Bundles/Official && bun run install.ts

# Upgrade existing installation
cd PAI/Bundles/Official && bun run install.ts --update

What's Next

• More skills coming to the pack ecosystem
• Enhanced wizard-style installation across all packs
• Community contributions welcome!

---

Full Changelog: v2.0.0...v2.1.0

v2.0.0 - PAI Packs System Launch

PAI v2.0.0 - PAI Packs System Launch

Major Architecture Shift

• Transitioned from "mirrored system" approach to modular PAI Packs
• Packs are self-contained, AI-installable capability bundles
• Platform-agnostic design: works with Claude Code, OpenCode, Gemini Code, GPT-Codex, or custom systems

Available Packs (8)

• kai-core-install - Foundation pack with skill routing, identity, and architecture tracking
• kai-hook-system - Event-driven hook architecture
• kai-history-system - Automatic context-tracking
• kai-voice-system - Voice notifications with ElevenLabs TTS
• kai-art-skill - Visual content creation
• kai-prompting-skill - Meta-prompting system
• kai-agents-skill - Dynamic agent composition
• kai-observability-server - Real-time agent monitoring

Available Bundles (1)

• kai-core-bundle - Complete Kai infrastructure (4 foundational packs)

Why the Change?

• v1.x tried to mirror the entire Kai system - too fragile, too many interdependencies
• v2.0 extracts battle-tested features as independent, installable modules
• Each pack is like learning kung-fu in The Matrix - a complete capability download

Documentation

• Full pack template specification
• Complete pack system documentation
• Updated README with 14 Founding Principles and installation guide

v0.9.0 - Platform Agnostic Release

PAI v0.9.0 - Platform Agnostic Release

This release focuses on making PAI fully portable and fork-friendly. Your AI, your identity, your system.

Highlights

Observability Dashboard

• Complete real-time agent monitoring at .claude/Observability/
• WebSocket streaming of all agent activity
• Live pulse charts, event timelines, and swim lanes
• Multiple themes (Tokyo Night, Nord, Catppuccin, etc.)
• Security obfuscation for sensitive data
• Run with: ~/.claude/Observability/manage.sh start

Genericized Agent Identity

• All agent references now use process.env.DA || 'main'
• No more hardcoded names - your DA name flows through the entire system
• Observability dashboard shows your configured identity
• Color system supports any DA name with intelligent defaults

Platform-Agnostic Configuration

• Added _envDocs to settings.json explaining all environment variables
• Clear separation: settings.json for identity/paths, .env for API keys
• DA (Digital Assistant name) - your AI's identity
• PAI_DIR - root directory for all configuration
• TIME_ZONE - configurable timezone for timestamps

Statusline Enhancements

• Terminal color compatibility mode for different terminal emulators
• Dynamic Claude Code version display
• Cleaner greeting format

Skill System Improvements

• Canonical TitleCase file naming throughout
• Standardized skill-workflow-notification script for dashboard detection
• All paths use ${PAI_DIR}/ for location-agnostic installation

Breaking Changes

None - this release is backwards compatible.

Upgrade Instructions

cd ~/.claude  # or your PAI_DIR
git pull origin main

If using a fresh install, follow the Quick Start in README.md.

---

Full Changelog: v0.7.0...v0.9.0