v1.9.4

Release Notes - v1.9.4

Release Date: May 11, 2026
Version: 1.9.4

Overview

Version 1.9.4 focus on architectural refactoring and standardizing the skill structure. The primary change is the transition from a deprecated workflow-based directory structure to a root-level SKILL.md utilizing YAML frontmatter, ensuring better compatibility with modern AI agent environments.

Key Changes

🏗️ Architectural Refactoring

• Root-level SKILL.md: Migrated all core project logic to a single, high-fidelity SKILL.md in the root directory.
• YAML Frontmatter Integration: Standardized versioning and metadata using YAML frontmatter within SKILL.md, enabling programmatic parsing by AI tools.
• Cleanup: Automated the removal of the deprecated workflows/ directory via setup scripts.

📜 Documentation & Templates

• README Updates: Revised documentation to explain the new SKILL.md usage, updated paths, and new installation instructions.
• .rulesrc.yaml Template: Updated the configuration template to reflect Superpowers and Karpathy methodologies (v1.9.4+).
• CHANGELOG: Fully documented the transition from v1.9.3 to v1.9.4.

🛠️ Setup & Automation Improvements

• Improved Version Extraction:
  • setup.ps1: Updated PowerShell regex logic to accurately parse the version: field from YAML frontmatter.
  • setup.sh: Updated Bash/Awk logic to match the new frontmatter standard, fixing a bug where older versions looked for markdown tables.
• Enhanced Prerequisite Checks: Improved error handling and connectivity checks during the installation process.

How to Update

To update to version 1.9.4, run the appropriate setup script for your platform:

Windows (PowerShell):

.\setup.ps1 -Update

Unix (Bash):

./setup.sh --update

---

For more details, visit the official repository.

v1.9.3

🚀 v1.9.3: The Agentic Alignment Update

This major release transforms the AI Project Rules Generator into a professional-grade framework for AI Agent orchestration. By integrating the methodologies of Andrej Karpathy and the Superpowers Framework, v1.9.3 enforces strict discipline, extreme context efficiency, and systematic reliability.

🌟 Major Highlights

🧠 Agentic Alignment Integration

We've embedded world-class AI engineering principles directly into the generator's core logic:

• Andrej Karpathy's Methodology: Enforces "Simplicity First," "Surgical Changes," and "Think Before Coding" to prevent AI over-engineering and context drift.
• Superpowers Framework: Implements "Verification Before Completion," "Subagent-Driven Development," and "TDD for Documentation."

⚡ Deterministic JIT & Deep Context Savings

To solve the "Context Bloat" problem in long-running AI sessions:

• Pointer-based Retrieval: The generator now produces Pointers to detailed rules instead of dumping full Markdown files. Agents are instructed to read_file specific skills only when contextually triggered.
• Trigger-First Metadata: The indexing engine now prioritizes specific "Use when" conditions over generic summaries, significantly reducing LLM hallucination.

⚙️ Enterprise-Grade Performance & Audit

• Parallel Skill Indexing: Refactored indexer.py with ThreadPoolExecutor for high-performance processing of large skill catalogs.
• Systematic Debugging: Audit logs (.agent/logs/) now capture full Stack Traces and Execution Durations (ms), enabling professional root-cause investigation for every AI action.
• Lean Memory Management: Refactored persistent state to exclude transient session data, keeping the AI's "long-term memory" focused on stable project architecture.

🛠 What's New

Added

• Subagent Orchestration: Rules for AI to autonomously dispatch tasks to specialized subagents or invoke interactive MCP tools (e.g., Python REPL, Chrome DevTools).
• Exact Word Boundary Matching: Upgraded tag discovery algorithm using regex \b patterns to prevent false-positive skill matches.
• Verification Before Completion: New mandatory section in .cursorrules forcing AI agents to run and verify tests before claiming success.

Changed

• Config-Driven Orchestration: Added new methodology toggles in .rulesrc.yaml for granular control over Agent discipline.
• Refined Confidence Gate: Enhanced heuristic scoring (80% threshold) to prevent the AI from generating rules without a clear project understanding.
• Acknowledgments: Full attribution added for ComposioHQ, Tech Leads Club, and OthmanAdi skill sources.

---

Full Changelog: v1.9.2...v1.9.3

“We practice what we preach. This repository itself was developed and maintained using the very same v1.9.3 rules it generates.” 🦾

v1.9.2

AI Project Rules Generator v1.9.2 🚀

We are thrilled to announce v1.9.2, a massive upgrade focused on making the generator truly agentic, environment-aware, and context-efficient.

This release introduces Flexible Agentic Execution (Zero-Install) and Tiered Skill Orchestration, ensuring your AI assistants not only write better code but also think, plan, and reason like senior engineers—all without requiring any local Python setup for quick-start users.

🌟 Major Highlights

1. Flexible Agentic Execution (Zero-Install Fallback)

The workflow now dynamically adapts to your local environment using a Dual-Mode Strategy:

• Mode A (Enhanced): If you clone the full repository, the AI utilizes our Python helper scripts (indexer.py, wizard.py) for maximum speed and efficiency.
• Mode B (Autonomous): If you are a Quick Start user with just the markdown file, the AI falls back to its native file-reading capabilities. No Python installation required. The workflow is now 100% AI-self-sufficient.

2. Tiered Skill Orchestration (Double Search)

Previously, a single limit forced AI agents to choose between technical skills (like React) and functional skills (like Planning or Security). We've fixed this "blind spot" with a new double-search architecture:

• agentic_match_limit (Default: 3): Guarantees the inclusion of foundational reasoning skills (Planning, System Debugging, Memory).
• skill_match_limit (Default: 5): Dedicated capacity for technical and domain-specific implementation skills.
• Result: Your generated rules now ensure the AI knows "How to think" alongside "How to code".

3. Smart Intent Discovery (Custom Specification Support)

We've removed rigid filename requirements (like prompt.txt). The AI now uses Pattern-Based Semantic Matching to autonomously discover your custom project specifications (e.g., README.md, spec.md, architecture.md, instructions.txt). If found, these documents are treated as the Primary Source of Truth for high-fidelity architectural intent.

4. Full Design Token Extraction

Upgraded extract_design_tokens() from a runtime stub to a fully functional regex-based parser. The AI now actively extracts brand colors, font families, and spacing units directly from tailwind.config.* files and CSS custom properties, eliminating visual hallucinations.

5. Anti-Overload Rule

To protect context window limits during manual skill discovery, we've introduced strict constraints: AI agents must now list directories first, filter by intent, and read a maximum of 3-5 relevant files, rather than attempting to read entire skill repositories at once.

---

🛠️ Changelog

• Added: Tiered Skill Orchestration with agentic_match_limit configuration.
• Added: Flexible Agentic Execution (Dual-Mode Scripted vs. Autonomous fallback).
• Added: Full Design Token Extraction logic in project_rules_runtime.py.
• Added: Anti-Overload Rule for manual skill discovery.
• Changed: Replaced hardcoded prompt.txt references with "Smart Intent Discovery".
• Fixed: Encoding/BOM issues that caused grep false negatives during verification.
• Testing: Added robust regression coverage for design token parsing and semantic versioning detection in setup scripts (setup.sh / setup.ps1).

---

Upgrade Instructions:
If you are using the Quick Start method, your AI assistant will automatically use the new v1.9.2 workflow the next time you prompt it. If you cloned the repository, simply run git pull to fetch the latest scripts and templates.