English | 简体中文

Learning OS

Type: AI Harness Mode: Local-First Sources: BYOS License: MIT

Learning OS is a local-first study workspace you open in Codex. You bring your own sources, Codex routes the work, and the repo helps you keep durable study files instead of losing everything in chat.

Prefer a packaged snapshot over git clone? See Releases.

Start Here

1. Download the repo snapshot (source.zip) from Releases or GitHub.
2. Unzip it.
3. Open the folder as a project in Codex.
4. If you want the shortest first success path, open docs/demo-flow.md.
5. If you want to use your own materials later, follow docs/public-setup.md and docs/bring-your-own-sources.md.

You can understand the repo and try the included demo immediately. You do not need to set up a private library before the public surface makes sense.

What This Repo Does

• turns one local source into reusable study state
• keeps the next step, open questions, and distinctions in files instead of only in chat
• supports deep reading, synthesis, thesis-style reading, and research intake
• stays local-first and BYOS, so the public repo can remain clean and reusable

What You Get

When a pass goes well, the useful output is not only the answer in chat. You also keep a small packet that survives the session:

• project.md
• session-log.md
• open-questions.md
• distinctions.md

How It Works

One source in, one useful packet out

flowchart LR
    A["Your local source"] --> B["Open Learning OS in Codex"]
    B --> C["Choose the right workflow"]
    C --> D["Run one bounded pass"]
    D --> E["Write back durable files"]
    E --> F["Resume next time without starting over"]

Loading

What lives where

flowchart TB
    A["Public repo<br/>README / docs / tools / templates / examples"] --> C["Codex"]
    B["Your local sources<br/>books / papers / notes"] --> C
    C --> D["Your local packet<br/>project.md / session-log.md / open-questions.md / distinctions.md"]

Loading

Try It In 5 Minutes

If you want to see whether the repo is real before reading deeper architecture, do this:

input   -> samples/open/demo-source.md
guide   -> docs/demo-flow.md
output  -> examples/research-intake-packet/{project.md,session-log.md,open-questions.md,distinctions.md}

1. Open docs/demo-flow.md.
2. Compare samples/open/demo-source.md with examples/research-intake-packet.
3. If you want to verify the public surface, run:

pwsh -NoProfile -File ./tools/Test-All.ps1 -RepoOnly

On Windows, the convenience wrapper still works:

.\tools\Test-All.cmd -RepoOnly

Expected result:

project.md        -> workflow_mode: research / paper workflow
session-log.md    -> what changed and what to do next
open-questions.md -> what still needs evidence or clarification
distinctions.md   -> important differences worth keeping

That is the core public proof surface of the repo: one open sample goes in, and a reusable packet comes out.

What Makes It Different

Surface	Prompt-only chat	Learning OS
Output	one answer in chat history	durable packet files
Continuity	manual recap next session	packet state survives the pass
Routing	one generic conversation	explicit workflow choice
Validation	little visible proof	repo-safe checks and boundary-sensitive validation
Sources	context pasted ad hoc	explicit local source boundary

What It Supports

• Single-book deep reading One primary source, slow mechanism-first reading, cumulative write-back.
• Multi-book synthesis Several sources routed into one structured program without flattening them into one book.
• Thesis / non-textbook reading Argument-heavy reading that needs a different protocol from textbook-style study.
• Research / paper workflow Papers, reports, and captured articles that need intake and classification before deeper study.

Next Reads

• Run it with Codex Start with docs/run-with-codex.md.
• AI entry Start with AI_CONTEXT.md, then ai-context.json, task-router.json, and writeback-map.json.
• Architecture Read docs/architecture.md, docs/ai-harness.md, and docs/agent-architecture.md.
• Contribute Use ROADMAP.md, CONTRIBUTING.md, and Issues.

Portable CLI Path

pwsh -NoProfile -File ./tools/Test-All.ps1 -RepoOnly
$SOURCE_ROOT = "/absolute/path/to/your/files" # or C:\path\to\your\files on Windows
pwsh -NoProfile -File ./tools/Import-LocalSources.ps1 -SourceRoot $SOURCE_ROOT
pwsh -NoProfile -File ./tools/Test-PublicSetup.ps1

Windows .cmd wrappers are included, but the documented primary path uses pwsh so the repo does not read as Windows-only.

Repo Map

• system.md Public identity and operating principles.
• AI_CONTEXT.md The shortest AI-facing entrypoint.
• docs/demo-flow.md The shortest public walkthrough.
• docs/run-with-codex.md How to operate the repo as a Codex-driven harness.
• docs/architecture.md The simple public architecture view.
• agent/README.md Minimal public-safe agent layer.
• templates/project-template Minimal reusable project skeleton for durable write-back.
• examples Worked public-safe packets you can compare against.
• tools Validation and local-source import helpers.

Public Boundary

This public repository does not include third-party books, papers, PDFs, slides, or proprietary study materials.

You are expected to bring your own lawfully obtained sources.

That boundary is deliberate:

• it keeps the repo safer to publish and easier to share
• it keeps the harness reusable beyond one private library
• it lets the same operating model work across finance, economics, philosophy, policy, ML, and other reading-heavy domains

License

This repository is licensed under the MIT License. See LICENSE.

Third-party books, papers, PDFs, slides, and other source materials are not included and are not covered by this repository license.