Everyone Ships Agent Tools. Nobody Ships the Onboarding.

Look at today's GitHub Trending and you'll see a pattern:

• ayghri/i-have-adhd (+3,854⭐) — "A skill to stop your coding agent from burying the answer." It fixes agent output verbosity with a humorous name and a real pain point.
• vercel-labs/skills — "The open agent skills tool." One command: npx skills. Vercel backing, extreme simplicity.
• vastsa/PI-Desktop — "Local-first AI coding agent desktop." Electron + Rust + user-installable plugins. Developers want AI on their own machine.

Each of these repos optimizes one part of the agent experience: output quality, skill loading, or local runtime.

But there's a gap nobody is filling: the onboarding.

The Missing Layer

You found a great agent skill (i-have-adhd). You installed it with npx skills. You have a local desktop runtime (PI-Desktop).

Now what?

• How do you configure the assistant for the first time?
• How do you connect cloud AI capabilities with your desktop workflow?
• How do you go from "installed" to "actually doing useful work"?

This is the gap between "tool exists" and "tool is used." And it's where most agent ecosystems lose their users.

What We Did About It

We open-sourced the full onboarding layer: AstronClaw + Loomy assistant tutorials.

👉 https://github.com/iflytek/astronclaw-tutorial

AstronClaw is the cloud AI assistant. Loomy is the desktop AI assistant. The tutorial repo covers:

• Installation and initial setup for both cloud and desktop assistants
• Configuration walkthroughs
• First-task guides — from "I just installed it" to "I just completed my first automated task"
• Real usage patterns and playbooks

No assumptions about your prior setup. No "figure it out yourself" energy. Just step-by-step onboarding.

From "Can Use" to "Can Do Work"

Once you're past onboarding, the next question is: "Can this actually automate my work?"

That's where iflytek/astron-rpa comes in — an Agent-ready RPA suite for out-of-the-box automation.

👉 https://github.com/iflytek/astron-rpa

Think of it this way:

• astronclaw-tutorial = how to get started (cloud + desktop AI assistants)
• astron-rpa = how to get work done (automation that runs while you sleep)

The Bigger Picture

Today's Trending signals that developers are done with "another agent framework." They want:

1. Tools that respect their time (like i-have-adhd respecting your attention)
2. Tools that are trivial to install (like npx skills)
3. Tools that run locally (like PI-Desktop)
4. Tools they can actually figure out how to use ← this is where we come in

The agent ecosystem doesn't need more frameworks. It needs better onboarding. That's what we shipped.

Comments

[Jo Do]

The onboarding gap is really a success-state gap: nobody defines the user's first useful outcome, so the install instructions end at "installed." The repos that break through design the first five minutes - one command, one visible result, one obvious thing to try next.

Composability makes it worse. The user isn't onboarding to your tool; they're onboarding to the combination of your tool and the five others they installed this week, and nobody owns the combination. The closest fix I've seen is tools that ship a self-test as the onboarding: not "here's how to configure me" but "run this, and if the output looks wrong, the problem is on my side, not yours." That flips the first five minutes from setup to trust.