It does not assume one tool can plan, build, review, and approve by itself. Its value is the repo contract: file ownership, startup files, branch rules, credential boundaries, and proof steps live in the repo instead of in chat state.

If you are new here, use Ask the wiki in the header. It explains wiki terms and repo concepts such as GitHub branches, previews, pipelines, Vercel, and launch readiness from the approved GoingNinja sources.

1. Start Here

Read these in order if you are new to the system:

Supporting context:

2. What The System Is

The system is split into three repos:

Repo Purpose
goingninja-os bootstrap script, starter template, invariants, renderers; GoingNinja OS is the orchestration and skeleton layer, not a device operating system
goingninja-manual explanation, source citations, proof status, audit criteria
goingninja-project-<slug> one generated MVP repo with runtime, state, and deployments

3. How New Products Start

The normal GoingNinja path is:

  1. confirm the provider and domain prerequisites
  2. confirm the product brief
  3. start the active product repo from goingninja-os
  4. derive the concrete requirement set
  5. derive the architecture requirements from that requirement set
  6. pass the readiness gate
  7. continue implementation inside that generated project repo
  8. keep platform-wide rules and bootstrap logic in the global repos

The full planning-to-launch sequence is documented in Planning To Launch.

The important boundary is simple:

  • goingninja-os and goingninja-manual define the system
  • goingninja-project-<slug> is where one concrete product is built

If a pre-platform product already exists, it is an exception case and should be treated as reference only. It is not the main product-genesis model.

4. What It Fits

Good fit:

  • authenticated operator consoles
  • internal tools with durable server state
  • review queues and approval workflows
  • gated portals or manual surfaces that should not start from an empty repo

Not a good default fit:

  • native mobile apps
  • purely static marketing sites
  • highly custom frontends with little workflow logic
  • products that do not need state, access boundaries, or review-before-merge

See Stack And Scope for the default stack and setup assumptions.

5. What Breaks Without It

Failure mode How GoingNinja addresses it
state lives in chat instead of files STATE.json, DELTA.json, and RISKS.json hold machine truth; Markdown files are rendered views
one tool builds, reviews, or approves its own work the builder lane builds; Claude Code and Gemini CLI sit on the review side by contract today and by CI later; the human remains merge owner
repo instructions drift across tools AGENTS.md is the shared contract; CLAUDE.md and GEMINI.md stay thin adapters
review is implicit, late, or left to the builder a pre-merge check on required JSON files and keys runs on every pull_request; rendered Markdown state is maintained by CI on main after merge
each new MVP reconstructs auth, CI, state, and provider wiring from scratch goingninja-os/scripts/bootstrap-project.sh generates the initial file contract, CI workflows, and provider wiring in one resumable run

6. Where Proof Lives

The authoritative proof surface lives in Workflow And Proof.

The current product-case proof lives in Product Proof.

Today that proof surface shows:

  • local deterministic bootstrap is reproducible
  • generated repos contain the deterministic file contract
  • JSON is machine truth and Markdown is rendered view
  • post-merge rendering is encoded into the generated workflow
  • open proof items remain explicit instead of being hidden in prose

If you need the provider and token prerequisites before that proof path starts, read Before You Start.

Notes

  1. Source. Repository family and the generated project contract documented in Repo Structure.