pw-learnai-core

A modular, open-source library of ideas, frameworks, and interactive tools for applied AI strategy and AI-assisted software development. Built by Protocol Wealth LLC. Licensed MIT-0 OR Apache-2.0. No certificate. No paywall. No marketing funnel.

The repository is named pw-learnai-core; the learning library and its private frontend package retain the pw-learnai name.

What this is

A library of self-contained modules you can read, fork, remix, or drop into an AI notebook (NotebookLM, Claude Projects, ChatGPT custom GPTs) for your own learning. Each module is a three-file directory:

• A module.md file — the ideas in prose, with falsifiable claims and worked examples
• An exercises.md file — applied work that produces a reviewable artifact
• A references.md file — sources, reviewed-date notes, and further reading

Interactive .jsx components live separately under components/interactive/ so the module directories stay stable for bundling.

Pick the modules you want. Ignore the rest. Nothing here assumes you take them in order.

What this is not

• Not a course. There is no enrollment, no grading, no certificate.
• Not a training program with a fixed path. The modules are deliberately independent.
• Not a product. Nothing here sells Protocol Wealth services.
• Not a complete map of the field. It is a library of the ideas we have found useful and wish we had encountered earlier.

Start here

If you only know ChatGPT or Claude prompts, start with 00 - Getting Started as an AI Operator.

The first routing decision is level, not tool:

Level	Use this when	Next step
Beginner	You know ChatGPT or Claude prompts, but repositories and diffs are new	Create GitHub, clone one repo, install one desktop or IDE agent, and use the Setup Path Builder
Intermediate	You can work in a repo and need repeatable agent practice	Add AGENTS.md, CLI-agent prompts, and a documented verification loop
Advanced	You are connecting runtimes, MCP, memory, remote access, cloud, or regulated workflow patterns	Use Module 16 to map the system, study the OSS labs, and deploy only after the local loop is stable

The full beginner path is deliberately practical:

1. Get a GitHub account, enable 2FA, and learn to read a diff.
2. Install one desktop or IDE agent such as Codex app, Claude Code, or Antigravity.
3. Add one CLI agent such as Codex CLI or Claude Code CLI.
4. Put durable repo guidance in AGENTS.md. Check that your coding agent loads it; do not maintain a second, divergent instruction file.
5. Record meaningful changes in CHANGELOG.md. Outstanding work lives in GitHub issues.
6. Learn public data sources with read-only examples before writing harvesters or deploying services.
7. Use Module 15 to map untrusted content, secrets, tools, and egress.
8. Use Module 16 to separate the runtime, control plane, capabilities, memory, recovery, remote path, and accountable humans before adding autonomy.

The live site now starts with a visual first-hour walkthrough before sending anyone into raw markdown. The first file pair is simple:

• README.md or SETUP-NOTES.md for instructions, notes, and agent-readable context.
• first-page.html or index.html for a visible browser artifact with no backend required.

For AI notebooks, upload notebooklm/starter-bundle.md. It combines the beginner path, prompt engineering, agentic coding, agent instructions, evaluation design, public-data source discipline, and governed agent-system architecture.

Project state

• CHANGELOG.md records shipped changes.
• GitHub issues are the authoritative tracker for outstanding and planned work. Re-query them; no checked-in file mirrors them.
• The estate design record, live and planned, is at pwos.app/build.
• The agent-systems entryway shipped in PR #55, closing #54.

Module index

Each module is self-contained. Start with 00 if you are new to repository-based agent work, or jump to the track that matches your problem.

Getting started

#	Module	Core question
00	Getting Started as an AI Operator	How do you move from prompt literacy to GitHub, coding agents, repo state files, public data, and safe first deployment?

Strategy and judgment

#	Module	Core question
01	Reading Disruption	How do you tell real disruption from incumbent-friendly change?
02	The AI Advantage Matrix	Where does AI create durable advantage vs a race to the floor?
03	Where AI Destroys Value	What are the failure modes that look like wins in the short run?
05	Platform Economics	When is an opportunity actually a platform, and when is it a product with an API?
09	AI Judgment Under Uncertainty	How do you make AI decisions when the capability curve keeps moving?

Organization and delivery

#	Module	Core question
06	Organizational Design for Continuous Change	How do you structure for exploration without starving the core?
07	Disciplined Experimentation	What makes an experiment actually diagnostic vs theater?
08	Stakeholder Buy-in Through Evidence	How do you convert skeptics by sequence rather than persuasion?

Practice

#	Module	Core question
04	Building Decision-Support Artifacts	How do you build a tool that actually gets used?
10	Prompt Engineering for Operators	What separates a durable prompt from a lucky one?
11	Evaluation Design for AI Systems	How do you know your AI system is getting better, not just different?
12	AI-Assisted Coding in Practice	How do you use Codex CLI, Claude Code, and coding agents without creating maintenance debt?
13	Designing Agent Instructions	What makes an AGENTS.md one a coding agent actually follows, not decoration?
15	Security & Secrets Hygiene	How do you keep untrusted content, secrets, and excessive agency from collapsing the trust boundary?
16	Building Agent Systems	How do you reuse a mature agent loop while separating intent, tools, permissions, memory, recovery, remote access, and human accountability?
17	Models, Effort and Cost	Which model and effort should each stage run at, and what does a completed task actually cost?

Working with data

#	Module	Core question
14	Working with Public Data	How do you use Data.gov, archival APIs, and OAI-PMH without confusing metadata with production-ready data?

Protocol Wealth OSS Labs

The labs connect the learning library to Protocol Wealth's open-source starting points.

Lab	Source repo	Core question
PWCLI Core Lab	Protocol-Wealth/pwcli-core	How does a runtime-neutral control plane govern intent, redaction, approvals, and receipts around the Claude Agent SDK or another runtime?
Shard Core Lab	Protocol-Wealth/shard-core	Where do authenticated encryption, threshold recovery, and human ceremony belong—and where do they not belong—in an agent system?
Nexus Core Lab	Protocol-Wealth/nexus-core	How do MCP tools expose public, read-only financial analysis, planning tools, and on-chain analytics without carrying client identity?
PWOS Core Lab	Protocol-Wealth/pwos-core	How do PII boundaries, audit trails, confirmation gates, tool tiers, and compliance primitives become structural controls?
PWPlan Core Lab	Protocol-Wealth/pwplan-core	How does a thin planning UI consume a 34-tool engine contract while rejecting named direct-identifier keys without claiming de-identification?
System-of-Systems Lab	Public repositories + live/public boundaries	How do the learning, control, runtime, capability, planning, governance, recovery, private-surface, and human layers fit without being confused for one product?

Public material treats durable semantic memory as a provider-neutral boundary and does not publish private implementation details or require a private component.

How to use this library

As a reader

Clone the repo or read the files on GitHub. Each module.md is complete on its own.

git clone https://github.com/Protocol-Wealth/pw-learnai-core.git
cd pw-learnai-core

With NotebookLM (or Claude Projects, or a custom GPT)

The notebooklm/ directory contains pre-assembled bundles — concatenated markdown files grouped by theme — that you upload directly as sources.

# Upload any file from notebooklm/ as a NotebookLM source
ls notebooklm/

Recommended first upload:

notebooklm/starter-bundle.md

Or roll your own bundle:

# Combine the strategy modules into one source file
cat modules/01-*/module.md modules/02-*/module.md modules/03-*/module.md > my-strategy-bundle.md

With Codex CLI or Claude Code

Module 12 is written around practical terminal workflows for Codex CLI and Claude Code. Module 16 continues into Agent SDK adapters, MCP, memory, remote access, system boundaries, and human decision rights. This repo also includes lightweight agent guidance files:

• AGENTS.md as the shared project guidance for coding agents

Useful starting prompts:

codex "Read AGENTS.md and summarize how to update a module safely."
claude --permission-mode plan

Useful OSS lab prompts:

codex "Read labs/protocol-wealth-oss/README.md and propose the smallest safe lab enhancement."
claude --permission-mode plan "Read the PWOS Core lab and propose a client-only confirmation-gate simulator improvement."

For concrete Python and TypeScript Claude Agent SDK reference adapters, read the PWCLI Core Lab and the implementation shipped in pwcli-core PR #7. The bounded examples are read-only, use API-key or documented cloud-provider authentication, and do not extract, copy, or proxy Claude subscription credentials.

PR hygiene across the OSS surface

Audit open PRs across pw-learnai-core, pwcli-core, shard-core, nexus-core, pwos-core, and pwplan-core:

pnpm pr:audit

Merge only PRs classified as clean, non-draft, check-passing, and not review-blocked:

pnpm pr:mergeable

With the interactive components

Use them in your browser — no install required:

→ https://protocol-wealth.github.io/pw-learnai-core/

The interactive tools (including the setup path builder, Agent Systems Architect, MCP planner, PII guard simulator, confirmation gate simulator, planning contract validator, CLI coding playbook, prompt evaluator, and agent-instructions auditor) run client-side. No login. No telemetry. No external API calls. Each session evaporates when you close the tab — fill in, copy or screenshot if useful, leave.

Or run them locally:

pnpm install
pnpm dev

Or copy any .jsx file in components/interactive/ into your own project. They are deliberately written as single-file components with no private dependencies.

As a remix target

Fork the repo. Edit the modules to fit your industry. Submit a PR if you want your work upstreamed, or keep it in your fork — that's what the permissive license is for.

Structure

pw-learnai-core/
├── modules/            # Self-contained learning units
│   └── NN-name/
│       ├── module.md           # The content
│       ├── exercises.md        # Applied work
│       └── references.md       # Sources and further reading
├── labs/               # Applied OSS labs connected to Protocol Wealth repos
├── prompts/            # Standalone copyable prompts linked from modules/tools
├── components/
│   ├── interactive/    # Single-file React tools (diagnostics, calculators)
│   └── ui/             # Shared primitives (buttons, cards)
├── notebooklm/         # Pre-assembled bundles for AI notebooks
├── assets/             # Images, diagrams, SVGs
└── CONTRIBUTING.md     # How to submit a module or fix

Contributing

Contributions welcome. The bar is specific:

• Falsifiable claims over fluent ones. A claim that could describe the opposite outcome does not count.
• Artifacts over essays. If you submit a new module, include at least one exercise producing something the reader could show another human.
• Honesty about limits. State what the module does not cover and where its reasoning breaks down.

See CONTRIBUTING.md for the submission template and review criteria.

License

Licensed under either MIT-0 or Apache-2.0, at your option.

• MIT No Attribution: LICENSE-MIT-0
• Apache License 2.0: LICENSE-APACHE

SPDX-License-Identifier: MIT-0 OR Apache-2.0

Use commercially, fork it, rename it. The only ask: if the module helped you, send a PR with what you learned.

Who built this and why

Protocol Wealth LLC is an SEC-registered RIA that does its own engineering. We built this library because the AI-strategy content available when we started was either expensive and generic (big-name university certificates) or free and shallow (blog posts and LinkedIn slop). We wrote what we wished we had. Keeping it open-source is how we make sure the next person does not have to re-derive it.

Maintainer: Nicholas Rygiel — Managing Partner / CTO / CISO, Protocol Wealth LLC.

Acknowledgments

This library draws on work by Clayton Christensen, Rita McGrath, Geoffrey Parker, Marshall Van Alstyne, Charles O'Reilly, Michael Tushman, and a generation of practitioners who wrote honestly about what worked and what did not. Where a specific idea traces to a specific source, it is cited in the module's references.md.