Welcome
testing-os is the centralized dogfood evidence system for the Dogfood Lab and mcp-tool-shop-org GitHub orgs — it accepts dispatched submissions from repos under dogfood-lab/* and mcp-tool-shop-org/* (per the README threat model). It proves, with auditable evidence, that each governed repo was actually exercised in a dogfood-worthy way.
What This System Does
Section titled “What This System Does”- Source repos define scenarios and run dogfood workflows
- Central verifier validates schema, provenance, and policy compliance
- Accepted records are persisted with full audit trail
- Generated indexes make dogfood status queryable across the org
Key Concepts
Section titled “Key Concepts”| Concept | What It Means |
|---|---|
| Record | A structured JSON document proving a dogfood run happened |
| Scenario | A YAML definition of what constitutes real exercise |
| Policy | Per-repo rules governing enforcement and freshness |
| Surface | The product type being dogfooded (CLI, desktop, web, etc.) |
| Finding | An evidence-bound lesson extracted from dogfood runs |
| Pattern | A repeated lesson cluster backed by 2+ accepted findings |
| Doctrine | A hardened portfolio rule earned from repeated patterns |
Current Coverage
Section titled “Current Coverage”14 governed repos across 8 product surfaces (cli, web, mcp-server, desktop, plugin, npm-package, api, library); every latest accepted record is verified pass. 13 of the 14 are past the freshness window — only this repo’s own self-dogfood is current — and indexes/stale.json names them; indexes/latest-by-repo.json is the live verdict map. (Measured 2026-09-06 from those two files; an earlier version of this line counted a dogfood-labs legacy entry that the map no longer carries.)
Getting Started
Section titled “Getting Started”Start here → Beginner’s Guide walks you through testing-os end-to-end on your first repo. The other pages below are reference material — open them as you need them.
- Beginner’s Guide – new to testing-os? start here
- Architecture – how the system works
- Contracts – the seven defining contracts
- Operating Guide – day-to-day operations
- Integration – how other systems consume dogfood status
- Intelligence Layer – how evidence becomes reusable portfolio memory
- State Machines – the four distinct status vocabularies (record classification, finding review, wave-finding classification, agent_run lifecycle) and which layer each one operates at
- Recovery — The Three R’s – revalidate, rewind, redrive — the lawful recovery contract
- swarm history – deep-audit the wave_state_events transition chain
- swarm CLI reference – per-verb quick reference for all 32
swarmverbs, with cross-links to the deep-dive pages - Error Code Reference – structured error codes surfaced by testing-os CLIs
