Concept: setup-prompt onboarding — agent-driven local install via a public skill file

[iret77]

Concept — not yet scheduled. Captures a research session's findings so this can be picked up.

Goal

A copy-paste onboarding path for non-technical evaluators, alongside Docker (Path c) and the native installer (byte5ai/omadia-ui#40, Path a): the user pastes a small setup prompt into their AI desktop app (Claude / Codex), the agent follows a public skill file that installs and starts omadia locally, and afterwards the user just opens the browser/app. This is Path b of #337.

Precedent (and its caveat)

byte5 has already shipped this pattern for kemia and palaia (setup prompt → ClawHub skill / public SKILL.md, idempotent install-or-update + doctor --fix). Caveat: those ran through developer tools (OpenClaw, Claude Code CLI). The actual target group here uses the consumer desktop apps, not the CLIs. Verified separately that the Claude desktop app can execute locally on the user's machine, so the path is viable for laypeople in principle — but the desktop-app route is new for us and unproven for omadia's heavier stack.

Mechanic

• Small setup prompt, copy-pasted from the website / README → points the agent at a public skill file → agent installs.
• Idempotent / self-healing, modeled on palaia (install-or-update even if present, doctor-style verification).

Dependencies (this is the key insight)

• b is the shell over an install path — it's only as robust as what it installs. The robust version installs the prebuilt artifact from Path a (byte5ai/omadia-ui#40): the skill file shrinks to "download the app, start it, connect your subscription." Without a, b has to orchestrate the heavy from-source / Docker install (fragile on arbitrary lay machines). → b benefits strongly from a; a→b is a stack, not an either/or.
• Subscription-backed agents via official CLIs (Claude / Codex / Gemini) using an MCP tool bridge #309 (subscription via the official CLI) makes b key-free, and fits naturally: the installing agent already runs on the subscription. Not a show-stopper (~1–2 days for Shape 2; provider seam already landed).

Advantage over Path a (the benchmark b must beat)

	a (installer)	b (setup prompt)
user download	150–250 MB, per-OS	none
byte5 maintenance	3 OS builds + native-deps + signing + auto-update	one versioned skill file
credential	API key in /setup (or bundle the CLI)	uses the subscription the user already has
prerequisite	double-click only	user must have a Claude/Codex desktop app

If a and b end up roughly even per scenario, both can ship (different funnels).

Risks / open

• Robustness of the agent-driven install on arbitrary lay machines (the real unknown).
• Permission-confirmation friction inside the consumer desktop apps (UX tuning, not a blocker).
• The skill file must stay in sync with omadia releases (versioning + doctor, like palaia).
• Next step: a hands-on dry-run of the non-Docker install — it measures robustness and produces the first draft of the skill file.

Related

• #337 — onboarding hub (Path a/b/c)
• #309 / #280 — subscription engine (key-free)
• byte5ai/omadia-ui#40 — Path a (the prebuilt artifact b should wrap)

Status

Concept. Hands-on validation pending.