AI Study Library

This repository serves as a personal knowledge base and study companion for mastering AI engineering, model interactions, and core software development foundations. It integrates evidence-based learning strategies with practical resources for maximizing the utility of Large Language Models (LLMs).

📂 Project Structure

The library is organized into the following main sections:

1. ai/docs/

Documentation and best practices for interacting with specific AI models.

• Claude: Guides on interaction, effective harnessing of long-running agents, and prompting best practices.

2. ai/prompts/

A collection of optimized prompts and meta-prompts designed to improve the performance of various models.

• Improvement: Meta-prompts specifically crafted to refine and optimize inputs for Claude, Codex, Gemini, and Grok.

3. ai/rules/

Rules written to steer coding agents, organized by the tool that loads them — the guidance is shared, only the activation metadata differs.

• Java Concurrency & static: decision rule for static in high-RPM services on Java 25 — shared state, virtual threads, DynamoDB/Redis clients, and OpenTelemetry. Available for Claude Code and for Cursor as an .mdc project rule — both scoped by globs to Java sources and build files. See ai/rules/README.md for install paths, the Cursor rule-type/frontmatter mapping, and sourced notes on the .mdc frontmatter gotchas.

4. ai/examples/

Executable reference implementations for the rules above — a rule the build does not enforce is only a suggestion.

• java/static-concurrency-enforcement/: ArchUnit test suites (v1 and v2, the latter frozen against a baseline) plus an Error Prone BugChecker that covers the synchronized blocks ArchUnit cannot see in bytecode.

5. ai/agents/

Claude Code subagents.

• product-pipeline/: 35 research-backed personas that take an idea through Discovery → PRD → Architecture → Tasks. Each stage has its own orchestrator, run in a fresh session (claude --agent <stage>-orchestrator), plus specialists, a stage critic and a shared cross-cutting pool (security, privacy/LGPD, accessibility, SRE, FinOps, analytics, traceability). Each persona states acceptance criteria and refusal criteria (blocked / rejected / out_of_scope). The folder also ships a preloaded conventions skill, default gate rubrics and install/run/lint scripts. See ai/agents/product-pipeline/README.md; the research behind it is in dev/research/2026-10-02-agentic-pipeline-personas-research.md.

6. ai/study/

Structured study plans and educational resources.

• Basics Study Plan: A comprehensive guide covering Computer Science foundations (Algorithms, Data Structures), Backend Engineering, and System Design. It incorporates neuroscientific learning principles like Active Recall and Spaced Repetition, utilizing tools like NotebookLM as a cognitive partner.
• Tools: Generators for Product Requirement Documents (PRDs) and Technical Review guidelines.

7. ai/skills/

Agent skills, each a self-contained SKILL.md folder. See ai/skills/README.md for the index, install steps, and flow diagrams.

• coding-standards: generates or updates CODING_STANDARDS.md at a repo root from the conventions the codebase actually follows, with one heading per domain. Harness-agnostic: it reads the rule files of Claude Code, Codex, Cursor, Copilot, Gemini CLI, and Windsurf, exports the standards declared in project rule files, and points at user-level rules.
• test-suite-review: reviews a codebase's unit tests, removes the ones that add volume without protecting the domain, reinforces where the domain is exposed, and ends green with at least 80% genuine coverage. It works in a git worktree, respects Sonar exclusions, and merges back with a ledger of every change.
• dream, pdf, study-plan-generator: memory consolidation, PDF to Markdown extraction, and study-plan generation.

🚀 Key Concepts

• Active Learning: The study plans encourage active engagement (teaching back, solving problems) rather than passive consumption.
• Prompt Engineering: The repository treats prompts as code, with versioning and optimization strategies for different model architectures.
• Cognitive Amplification: Using AI tools not just to generate code, but to deepen understanding of complex concepts.

🛠 Usage

1. Study: Follow the basics-study-plan.md for a rigorous CS and Engineering curriculum.
2. Optimize: Use the prompts in ai/prompts/improvement/ to refine your instructions to AI agents.
3. Reference: Consult ai/docs/ for model-specific interaction patterns and best practices.