user-onboarding

An onboarding playbook for first-time or uncertain users of the Starchild agent, a personal assistant positioned as an intern with a full computer: anything achievable on a computer is in scope, from browsing, reading, and writing to coding, connecting services, operating files, deploying pages, and pushing results back to the user. The skill is meant to fire on cold starts and vague openers such as "what can you do", "how do I start", "help", or a generic hello, and explicitly not when the user already has a clear, specific request.

Its guiding philosophy is "show, don't tell": rather than reciting a feature list, the agent does something concrete in the very first exchange so the user feels the capability instead of reading about it. The flow begins by checking memory (USER.md Context and MEMORY.md) before composing anything, applying a hard language gate to reply in the user's preferred language from the first sentence. If memory holds useful signal, the agent opens with continuity rather than a cold introduction and never surfaces the memory machinery by saying things like "according to my memory". Otherwise it uses a default three-beat opener that anchors who the agent is, mentions the $5 free credit as a fact, proactively offers to bind push channels like Telegram and WeChat, and asks one open pull-question seeded with concrete verbs.

Discovery focuses on the verb-plus-frequency pair where automation lives, mapping user pain (weekly reports, inbox overload, tracking a topic or KOL, research, prototypes, database inspection, news digests, price alerts) to a real underlying need. Crypto and trading are treated as a deep specialty to lean into only when the user signals interest, never forced. For any recurring request the skill prescribes a quick-win pattern: show a real sample now, take one round of edits, then confirm cadence and channel.