On this page: How to read this guide · Get started · Concepts · How-to · Reference
The README.md pitches the framework. This is the book. It walks a developer from a first
green PR all the way to writing a custom runtime adapter, with every concept, mission, and
schema documented against the code on main. If you only have ten minutes, read the
Quickstart. If you want to understand the thing, read tier by tier.
Twenty chapters, four tiers. The tiers are an order, not a menu: each builds on the last.
Tier 1 GET STARTED ── read in order, ~30 min ──► a green PR in your own repo
│
Tier 2 CONCEPTS ── read for the "why", ~90 min ──► explain it to a colleague
│
Tier 3 HOW-TO ── consult per-chapter when you need to do a thing
│
Tier 4 REFERENCE ── look-up material, never sequential
- New here? Read Tier 1 top to bottom. By the end of chapter 3 you have run
doc-syncend-to-end and seen the run-archive it produced. - Want the model in your head? Tier 2. Chapter 04 is the mental model; 06 and 07 are the dense ones (the engine, then the 4-layer substrate).
- Need to do one specific thing? Jump straight into Tier 3. Each how-to chapter stands alone and links back to the concepts it assumes.
- Looking up a field, a flag, or a term? Tier 4. The Glossary defines every framework-specific word and links to the chapter that explains it.
Two honest limitations are surfaced where they live, not hidden:
The trace stream emits exactly one event in production today,
T-FINAL, written byscripts/lib/fleet_run.py(before the manifest, per the engine doctrine). The schema covers its 11-value trace primitive enum; per-transition emission is rolling out. See Trace schema.
Headless campaign mode (
run-campaign.sh) is not yet validated end-to-end. The interactive path (your agent's chat,/goal) is the supported flow today. See Safety and secrets.
- Quickstart, your first PR in 10 minutes
- Installation, full setup across all 4 runtimes
- Your first mission, running doc-sync end-to-end
- Mental model, what a "run" actually is
- Missions vs campaigns, when to chain
- The engine, primitives, ledger, frozen DAG
- The substrate, 4-layer verification
- Roles and blindness, why review is structural