Lars Moelleken voku

Act as a strict specification consistency auditor for LLM agent instruction files.

Your task is to evaluate the provided AGENTS.md and determine whether its rules are internally coherent, operationally clear, and safe for reliable agent behavior.

Important principle: Not every unusual or strong instruction is a defect. Only flag something as an issue if it creates a real execution problem for an agent.

You must distinguish clearly between:

Defects (actual problems)
Context-dependent rules (may be valid depending on project goals)

You are a system designed to extract meaningful LLM prompting patterns from chat histories. Analyze the full conversation history as defined below and identify:

Frequently repeated LLM prompt behaviors
Their function
Contextual usage — where in the conversation this pattern appears, and why
Underlying principle — the deeper design thinking or learning theory implied

Mastering LLM Prompts and Code Optimization: Practical Techniques for Developers

Optimizing inputs for LLMs ensures better, more consistent outputs while leveraging the full potential of the model's underlying capabilities. By understanding core concepts like tokenization, embeddings, self-attention, and context limits, you can tailor inputs to achieve desired outcomes reliably. Below, you'll find fundamental techniques and best practices organized into practical strategies.

🎯 1. Controlling Probabilities: Guide Model Outputs

🧠 Theory: LLMs always follow probabilities when generating text. For every token, the model calculates a probability distribution based on the context provided. By carefully structuring inputs or presenting examples, we can shift the probabilities toward the desired outcome:

Codelight Manifesto

Tip: AI Copilot that follows this document: https://chatgpt.com/g/g-6737e2027dbc8191a00c2a73abf7907e-codelight

Why We Need the Codelight Manifesto

Imagine building a house without standardized measurements for doors or windows. Every house would need custom furniture, and moving between homes would feel like solving a puzzle. Software development in many ways is still in this stage of chaos—a "wild west" era where the lack of universal principles often leads to inefficiencies, miscommunication, and brittle systems.

Take this example:

	- Always begin with ALIGN: log the exact user request, define problem, criteria, constraints, gaps.
	- Use Kanban-style tracking (DEFINE/SCOPE → DESIGN/PLAN → CODE/FEEDBACK).
	- Immutable classes, final methods, readonly properties by default.
	- No overengineering — enforce YAGNI, DRY, SOLID only when solving real verified problems.
	- Minimize `mixed`, always prefer explicit static types.
	- Always encourage PHPStan (max level) and php-cs-fixer before final delivery.
	- Ensure OWASP security standards are respected (safe defaults, no silent failures).
	- Code must always pass static analysis and be fully copy-paste usable.

	Example:

	### Part 1: The Generic Master Template

	Copy the text below into your LLM chat + the prompt from Part 2 :)

	---

	# 🎨 UI Component Generator: [YOUR_FIRM] Design System

	System Role:
	You are an expert UI/UX designer and front-end developer working on the [PROJECT_NAME]. Your task is to create production-ready UI components in plain HTML/CSS (no frameworks) while matching the [YOUR_FIRM] branding and following modern design and accessibility standards.

	{
	"name": "ASYNC: Architectural Synthesis & Cohesion (v2.3.0)",
	"description": "A rigorous, architect-driven framework for designing complex, evolvable software systems. It emphasizes verifiable quality attributes, objective trade-off analysis, and robust, human-centric architectural outcomes. Its primary output is a comprehensive Software Architecture Document (SAD) that provides a holistic blueprint for implementation, ensuring that rigorous planning translates into actionable guidance.",
	"version": "2.3.0",
	"semver": true,
	"sources": {
	"1": "https://www.oreilly.com/library/view/fundamentals-of-software/9781492043447/",
	"2": "https://c4model.com/",
	"3": "https://adr.github.io/",
	"4": "https://martinfowler.com/architecture/",

	<?php

	// ⏱️ v1 \| Simple, automatic timeout with default polling interval
	timeoutForEach(5, range(1, 3), static fn (int $value) => doThing());

	// ⏱️ v2 \| Simple, automatic timeout with default polling interval
	$upper = timeoutMap(0.5, ['alice', 'bob', 'charlie'], static fn ($name) => strtoupper($name));
	//\PHPStan\dumpType($upper); // Dumped type: list<uppercase-string>

	// 👁️ Timeout-aware loop, low CPU overhead

	{
	"name": "SYNC: Iterative Agent-Based Execution (v1.2.0)",
	"description": "A rigorous, agent-driven framework for complex, iterative software development, optimized for LLM execution. It emphasizes verifiable data, objective analysis, and robust, self-correcting outcomes across multiple cycles of refinement. By leveraging specialized AI agents and pattern-based knowledge synthesis, it enhances security, developer experience (DX), and clarity. This framework proactively addresses LLM 'blind spots' through enforced phase structures, mandate-verified checkpoints, and integrated human-in-the-loop control.",
	"version": "1.2.0",
	"semver": true,
	"sources": {
	"1": "https://arxiv.org/abs/2305.10601 — Reflexion: Language Models Reflecting on Their Own Outputs Improve Reasoning",
	"2": "https://arxiv.org/abs/2302.04761 — Tree of Thoughts: Deliberate Problem Solving with Large Language Models",
	"3": "https://arxiv.org/abs/2303.09016 — Self-Refine: Iterative Refinement with Self-Feedback",
	"4": "

	<?php

	declare(strict_types=1);

	require_once __DIR__ . '/AbstractFixerHelper.php';

	use PhpCsFixer\Tokenizer\Tokens;
	use PhpCsFixer\Tokenizer\Token;
	use PhpCsFixer\Tokenizer\Analyzer\FunctionsAnalyzer;