The Engine, Open Source, MIT

An open-source AI writing pipeline that turns build sessions into publishable content. 4 files, 2 loops, 19 commands, MIT licensed. Clone it, run it, own it.


Here Is What That Looks Like

A two-hour debugging session becomes a post. You never open a writing tool.

I ship consistently. Before the engine, I did not. Every Tuesday I would finish a build session, look at the work, think ‘I should post about this,’ and not post. Not because I could not write. Because writing meant switching from builder to creator. That switch never happened.

Session log

2 minutes of notes

Spent 2 hours fixing context drift in a LangGraph workflow. Root cause: too many hidden state transitions between nodes. Fix: explicit state schema with required fields per transition.

Generated post

The output

Most AI agents don’t fail because of prompting. They fail because nobody knows what state they’re in. Every hidden transition is a drift point. Every drift point is a silent bug. Make the state explicit.

Same session. Different frame. Extracted, not written. The engine pulls what is already there.


How It Works

Two state machines run independently. LLM handles creative steps. Scripts handle mechanical steps. The two never overlap.

01
Wiki loop IDLE → INGEST → ANALYZE → RECONCILE → INDEX → VALIDATE → COMPLETE. Ingests raw notes, extracts entities, reconciles into wiki pages, links them.
02
Content loop IDLE → SESSION → STRATEGY → COMPILE → DRAFT → GATE → QUEUE → PUBLISH. Classifies by archetype, runs the 8-step Compiler, drafts to platform-native format, gates every draft.
03
The quality gate 16 mechanical gates (char count, hook, audience, frontmatter, word repetition) plus 14 creative gates. Composite score must hit 0.85. Failing drafts do not enter the queue.

What You Get

Clone the repo and you get the engine I run daily. 19 commands, 8 templates, 30 gates, two loops, one architecture.

19

Slash commands

/extract, /post, /publish, /state, /health, /config, /optimize, /prune, and 11 more. Each is a markdown file your agent reads at invocation.

8

Templates

Blog, LinkedIn, X, concept, entity, summary, comparison, session-log. Each with the right frontmatter and structure for its platform.

30

Quality gates

16 mechanical (code-enforced) + 14 creative (LLM-judged). Each gate is a hard floor the pipeline cannot cross. Composite score must hit 0.85.

2

State machines

Wiki loop for compound knowledge. Content loop for publishing. Same Act→Observe→Evaluate→Update→Repeat pattern. Different scope, same architecture.

14

Setup questions

Paste SETUP.md into any LLM agent. The agent auto-detects your platform, installs all commands, runs a 14-question ICP/voice/brand setup. ~15 minutes.

Zero dependencies

Python 3, bash, git. Everything else is optional. No npm. No Docker. No SaaS subscription. The system lives in your repo.


Setup: Clone + Paste

Four steps. One paste. ~15 minutes. The engine is wired to your workflow.

git clone https://github.com/ShayanSpiel/SpielOS my-os
cd my-os
01
Clone the repo git clone https://github.com/ShayanSpiel/SpielOS my-os && cd my-os
02
Open SETUP.md in any LLM agent Cursor, Claude Code, opencode, Continue, ChatGPT. Copy the prompt block, paste it.
03
Agent handles install + setup Detects platform, installs commands, runs 14-question setup. Writes all config files. ~15 minutes.
04
Run your first commands /state for status, /extract my-notes.md to ingest, /post “my first session” to draft, /health to check.

The Two-Layer System

Open source. Or done-for-you. Your call.

Layer 1

Open Source · Free

The repo, methodology, pipeline. Clone, paste, customize. This is what you get from the GitHub repo. The engine I run daily. Clone it, paste the setup prompt into your agent, answer 14 questions, and you have a working content pipeline in an afternoon.

Layer 2

DFY Install · $990

Full pipeline installed inside your workflow, positioned to your voice, 30 days review. Positioning strategy, offer design, bio rewrite, agents and templates in your voice, workflow design, 3 sample pillars ready to publish, 30 days post-publish review.

14 days

Install

Full pipeline in your workflow. You own every file. No subscription. No ‘we changed the pricing’ email in 2 years.

30 days

Money-back guarantee

If after 30 days you have run 5 sessions through the engine and you do not have 5 standalone-tested drafts, full refund. You keep the system.

3

Slots per month

Hard cap. The 14-day install is hands-on. 3 founders per month. The rest wait or run the open-source version themselves.


Why This Exists

I built this for myself first. The repo was extracted from my working vault after months of iteration. The templates are the ones I use every day. There is no theory here. Every file has been tested in production.

The insight is simple: you should never ‘go create content.’ Your work should already contain it. The engine extracts what is already there.

If you ship and do not post

This is the system that fixes it. The work happens. The capture happens. The drafts happen. You just review and publish.

If you do not ship

This system will not save you. There is no content to extract. Build first. Then run the engine.

If you want both

DFY install gets you the system AND the strategy. You get the methodology installed, the templates positioned to your voice, 3 sample pillars ready, and 30 days of review.