v4.0

PAI v4.0.0 — Lean and Mean

63 skills, 180 workflows, 21 hooks — and 81% of the context window is free for your actual work.

---

Context is Everything

The primary point of v4.0 is becoming lean and mean, and that primary means massive reduction in context usage.

Using CC's system itself has become heavier recently, and we wanted to look at how we could really reduce PAI's default context load.

So we basically re-architected PAI to be much more of a dynamic loading system. Super lean, and then expanding as needed.

This also means using NATIVE mode a lot more, for things that don't require the algorithm. So it's a lot more intelligent about that.

Tons of other updates as well, all part of this efficiency theme.

---

What All Changed

v3.0 made the Algorithm mechanically rigorous. But the system around it had grown unwieldy — 38 flat skill directories competing for context, dead build systems nobody used, stale version numbers scattered across dozens of files, and a Components/ directory that hadn't been functional in months.

v4.0 cuts the fat. The same capabilities, dramatically less overhead.

The Problem v4.0 Solves

v3.0's system:

38 flat skill directories → each loaded its own SKILL.md description into context
Components/ directory with dead InstantiatePAI.ts → never called
DocRebuild.hook.ts → referenced dead RebuildSkill.ts handler
CLAUDE.md → static file, manually updated, easy to drift
Version refs → v3.0 hardcoded in 14+ files across the installer

v4.0's system:

12 hierarchical categories → sub-skills nested under parent SKILL.md
Dead systems removed entirely → zero orphaned hooks or handlers
CLAUDE.md generated from template → BuildCLAUDE.ts resolves variables automatically
Algorithm at PAI/Algorithm/ with LATEST pointer → single source of truth
Every version ref → 4.0.0, verified by grep

---

Six Major Changes

1. Skill Hierarchy — 38 Directories → 12 Categories

The single biggest change. Every skill was reorganized into a hierarchical category system:

Category	Sub-skills	What moved here
Thinking	FirstPrinciples, IterativeDepth, BeCreative, Council, RedTeam, WorldThreatModel, Science	7 standalone dirs → 1 category
Security	WebAssessment, PromptInjection, SECUpdates, AnnualReports	4 dirs → 1 category
Utilities	CreateCLI, CreateSkill, Evals, Documents, Parser, Fabric, Cloudflare, Browser, Prompting, Aphorisms	10 dirs → 1 category
Media	Art, Remotion	2 dirs → 1 category
Investigation	OSINT, PrivateInvestigator, Recon	3 dirs → 1 category
Scraping	Apify, BrightData	2 dirs → 1 category
Research	Research, ExtractWisdom	2 dirs → 1 category
Telos	Telos	Preserved
Agents	Agents	Preserved
USMetrics	USMetrics	Preserved
ContentAnalysis	ContentAnalysis	New — extracted from Research

Why this matters: Each skill category loads ONE SKILL.md that routes to sub-skills. Instead of 38 competing descriptions in the context window, there are 12. The routing table in CLAUDE.md is half the size. The model spends less context on skill discovery and more on actual work.

2. Dead System Removal

Three entire systems were removed because they hadn't been functional in months:

Removed	What it was	Why it was dead
Components/ directory	Dynamic component loading system	InstantiatePAI.ts was retired, nothing called it
DocRebuild.hook.ts	Stop hook that rebuilt CLAUDE.md	Called RebuildSkill.ts which referenced dead Components system
RebuildSkill.ts handler	Handler for DocRebuild	Imported from Components that no longer existed

Zero functionality lost. These systems had been silently failing for months.

3. CLAUDE.md Template System

CLAUDE.md is no longer a static file. It's generated from a template:

CLAUDE.md.template          ← Source of truth (with variables)
    ↓
PAI/Tools/BuildCLAUDE.ts   ← Resolves variables from settings.json + LATEST
    ↓
CLAUDE.md                   ← Generated output (loaded by Claude Code)

Variables resolved:

• {DAIDENTITY.NAME} → Your DA's name (from settings.json → daidentity.name)
• {PRINCIPAL.NAME} → Your name (from settings.json → principal.name)
• {{PAI_VERSION}} → Current PAI version
• {{ALGO_VERSION}} → Algorithm version (from PAI/Algorithm/LATEST)
• {{ALGO_PATH}} → Path to algorithm file

SessionStart hook runs BuildCLAUDE.ts automatically — keeps CLAUDE.md fresh for the next session without manual intervention.

4. Algorithm v3.5.0

The Algorithm jumped from v1.4.0 (shipped with v3.0) to v3.5.0. Now lives at PAI/Algorithm/v3.5.0.md with a LATEST file pointing to the current version. No more digging through a Components directory to find it.

5. Installer Updates

Every version reference across the entire installer was audited and updated:

File	Was	Now
install.sh banner	v3.0	v4.0
cli/display.ts	Algo v1.4.0, Installer v3.0, "Alpha"	Algo v3.5.0, Installer v4.0, "Lean and Mean"
engine/state.ts	version: "3.0"	version: "4.0.0"
engine/config-gen.ts	version: "3.0"	version: "4.0.0"
engine/validate.ts	paiVersion: "3.0"	paiVersion: "4.0.0"
engine/actions.ts	LATEST writes v1.4.0	LATEST writes v3.5.0
electron/package.json	version: "3.0.0"	version: "4.0.0"
public/index.html	v3.0	v4.0
7 documentation files	"PAI 3.0 Alpha"	"PAI 4.0"
Banner.ts	paiVersion = "3.0"	paiVersion = "4.0.0"
merger.ts	paiVersion = '3.0'	paiVersion = '4.0.0'

Electron fix: Pre-built node_modules removed from release. Electron installs fresh on the target machine, avoiding the macOS quarantine crash that plagued v3.0.

6. Security & Sanitization

Comprehensive audit with multiple agent sweeps:

• Personal branding removed — "Unsupervised Learning" / "UL" references cleaned from 33+ files
• Identity templated — Pronunciation, timezone, voice clone ID, DA personality all use template variables
• Voice clone protected — Personal voice clone ID replaced with {YOUR_VOICE_CLONE_ID} (premade voices kept as useful defaults)
• Pronunciations cleaned — Removed DA-revealing phonetics from VoiceServer
• Code comments sanitized — No personal usernames in example paths
• Pre-commit hook updated — New v4.0-aware bash hook replaces broken validate-protected.ts reference

---

By The Numbers

Metric	v3.0	v4.0	Change
Context at startup	~38%	~19%	-50% context footprint
Skill directories	38	13 categories	-66% top-level dirs
SKILL.md files	41	63	+22 (sub-skill routing)
Workflows	162	180	+18
Hooks	20	21	+1
Agent templates	13	14	+1
Total files	1,227	1,151	-76 files
Dead systems	3	0	Cleaned
Algorithm version	v1.4.0	v3.5.0	+21 minor versions
Stale version refs	14+	0	All updated

---

Installation

# Clone the repo
git clone https://github.com/danielmiessler/Personal_AI_Infrastructure.git
cd Personal_AI_Infrastructure/Releases/v4.0

# Copy the release and run the installer
cp -r .claude ~/ && cd ~/.claude && bash install.sh

The installer handles prerequisites, configuration, and setup. Run source ~/.zshrc && pai after installation.

---

Upgrading from v3.0

If you're running v3.0:

1. Back up your ~/.claude/ directory
2. Copy v4.0 over your existing installation: cp -r .claude ~/
3. Run BuildCLAUDE to generate your CLAUDE.md: bun ~/.claude/PAI/Tools/BuildCLAUDE.ts
4. Verify your settings.json merged correctly — your identity, voice, and preferences should persist

The template system will detect your existing settings.json values and generate CLAUDE.md accordingly.

---

What's Next

v4.0 is the foundation for speed. With the context footprint cut and dead systems cleared, future releases can focus on:

• Smarter routing — Sub-skill routing with confidence scoring
• Local model support — Run skills with local models for privacy and cost
• Granular model routing — Route tasks to different models by complexity
• Remote access — Mobile and web interfaces to your PAI