FLME Wiki

• Agents & How We Work — who the named FLME agents are, what they own, and the shared conventions they all follow. Moved here intact from the original single-file wiki; unchanged in content.
• Demos — real, narrated walkthroughs of the build pipeline actually running, for a non-engineering audience. New.
• Engineering Reference — real technical reference material, one subsystem at a time; the UI Pipeline (markup → build → deployment) is the first section built. New.
• Repo Atlas — every FLME repo (there are about 15), what it actually is, its status, and what it's authoritative for. New.
• Skills Map — where FLME’s skills live and which copy is the real one. 197 SKILL.md files exist and most are copies of each other, so this maps originals against copies rather than counting. Generated; a skill in an undeclared location fails the build. New.
• Open Questions — a short list of real, currently-unresolved ambiguities worth knowing about before you assume something. New.
• Usage — the practical guide to starting/resuming an agent session with scripts/start-agent.sh. Unchanged.

The wiki's scope was designed to eventually cover more than what's built so far — deliberately staged rather than all at once, so each page is real and verified rather than a thin placeholder. Not yet started:

• How the Pipeline Works — the end-to-end requirements → UI codegen → backend codegen → deployment → release narrative. Engineering Reference's UI Pipeline section covers the UI-codegen-to-deployment slice of this; the requirements and backend-codegen ends still aren't written up.
• Engineering Reference — UI Codegen & the IR (built 2026-09-06, as "UI Pipeline: markup to deployment"); Requirements Framework, Backend Architecture & Codegen, Privacy & Compliance, and Tooling remain not built.
• Release, Deployment & Environments — gates and the release process (the source docs for this haven't been surveyed in depth yet).
• Clients & Deployments — Browser, Pet Shop, and the other client/reference repos, with real status.
• Company & Product, Brand & Marketing — thin sections; several inputs (the company's own name, some site copy) are still undecided.

Each of these will be built from real source documents already surveyed this session — see agents/sage/STATE.md for the full research digest if you want the detail before the pages themselves exist.

• IR — the intermediate representation ux-template generates client UIs from; sits between a ReqDoc spec and generated code.
• Gate — an automated or human check a release must pass at a given pipeline step (e.g. a step-9 Integrate gate).
• Fence rule — ReqDoc's dependency rule: Core → Domain → Client, one direction only. A Core entity never depends on a Domain or Client one.
• Core / Domain / Client — ReqDoc's three layers of requirements content: Core (shared framework), Domain (e.g. ECommerce, Medical — a vertical overlay on Core), Client (one real deployment's specific instantiation).
• Elephant / Goldfish — the persistent-identity vs. memoryless-companion session roles every FLME agent can use. See Agents & How We Work.
• "Specstrum" — two different things, don't conflate them: (1) the resolved product/brand name FLME's public site now presents itself under, with six purchased domains redirecting to flme.ai; and (2) a separate folder inside ReqDoc/Requirements/Specstrum/ holding its own parallel requirements-framework content. Whether/how those two relate hasn't been confirmed — see Open Questions.