35 skills across nine pipelines.
Design, build, review, security, brand systems, frontend delivery, and more, each with a defined owner.
Fabled
A self-contained AI software-delivery team of 35 coordinated skills that design, build, review, and ship software with evidence gates at every stage.
What it is
Fabled is our software-delivery method, productized. It is a self-contained AI delivery team: 35 coordinated skills that design, build, review, and ship software, with evidence gates at every stage. It is the discipline behind every other project on this site.
The method it implements is called the Fable Method, and it fits in one sentence: observed reality outranks the model, evidence outranks confidence, and irreversible actions remain owner decisions. AI-assisted development fails when nobody checks the work. Fabled is built so the checking is not optional. Nothing passes a stage boundary without the evidence that boundary requires, and no amount of confident output substitutes for it.
In practice, that means a roster of specialists coordinated by phase leads, a gatekeeper that validates every handoff, and machine-readable contracts that define exactly what each gate demands. Deployment to production is never implied by a finished build; it always requires a separate human go.
Who it is for
Engineering organizations adopting AI-assisted delivery who want governance rather than vibes. Also, honestly, us: we run it on our own client work every day.
What it does
Nine evidence-gate boundaries.
Every stage transition is guarded by a gate with required, artifact-backed evidence.
A structured team.
Thirteen specialists, three phase leads, and a gatekeeper, with a one-writer ownership contract for every artifact.
Deterministic validation.
Gates, pipelines, and ownership live in machine-readable contracts (gates.yaml, pipelines.yaml, ownership.yaml) enforced by scripts, not prose.
A 62-check taste pre-flight.
User-facing surfaces pass a design-quality checklist before review, thirteen checks of it scanner-decidable.
Human go for anything irreversible.
Deploys and destructive actions always come back to a person.
Under the hood
A skill catalog that runs on Codex, Claude Code, and Copilot through host adapters. Validators and hooks are dependency-light Python, the runtime contract is a single manifest, and the whole catalog is documented down to an interactive atlas with 97 ready-made prompts.
Interested?
If you want AI in your delivery process without giving up engineering standards, this is the conversation we most enjoy having.