skills/starchild-ai-agent__official-skills

mirror of https://github.com/Starchild-ai-agent/official-skills.git synced 2026-09-18 10:19:40 +08:00

Files

T

almo a179ffad19 rename: skillmarketplace → skill-manager, remove marketplace from skill UI text

- Rename skill: skillmarketplace → skill-manager (4.0.1 → 4.1.0)
  - Directory: skillmarketplace/ → skill-manager/
  - SKILL.md: name, skillKey, title updated
  - skills.json: name, path, version synced

- Remove 'marketplace' from user/agent-facing text to avoid confusion
  with x402 Service Marketplace:
  - user-onboarding (1.2.5 → 1.2.6): remove 'Skills marketplace' references
  - project-builder (1.6.1 → 1.6.2): 'marketplace coverage' → 'community coverage'
  - REFERENCE_5_MINUTE_GUIDE: 'skill marketplace' → 'existing skills'

2026-07-10 16:42:11 +08:00

3.0 KiB

Raw Permalink Blame History

Reference Notes: The 5-Minute Guide to Getting the Most Out of Starchild

Source article title: The 5-Minute Guide to Getting the Most Out of Starchild
Purpose of this local note: keep the key guidance inside the user-onboarding skill so agents can use it without fetching the web page again.

Core framing

Starchild should be presented as a practical, outcome-first agent: powerful intern + assistant, not a feature list.
User outcome matters more than model/tool details.
Better results come from clearer requests and faster feedback loops.

10 practical principles (condensed)

1) Smart Routing first

Use /model smart to reduce unnecessary cost.
Simple tasks should route to cheaper/faster models, complex tasks to stronger models.

2) Prompt quality is leverage

Be specific and concrete.
Keep each message focused on one clear ask.
Provide key context up front to reduce back-and-forth.
Explicitly request output format (table, bullets, etc.).

3) Connect delivery channels early (when needed)

Telegram/WeChat enables mobile access + push notifications for scheduled work.
For team usage, Telegram groups + whitelist allow controlled multi-user collaboration.

4) Use connectors for real workflows

Connect external tools (e.g., Gmail, Calendar, Drive, Slack, Notion, GitHub) so the agent can execute real work in context.

5) Build and preview quickly

Ask for the end result directly (dashboard/tool/page), then iterate on a live preview.
Deploy only after the preview output meets expectations.

6) Turn repetition into skills

If a task repeats, search for existing skills first.
If no suitable skill exists, create one and reuse.

7) Keep machine health visible

Resource limits are real (CPU/RAM/disk).
Periodically check usage, clean stale files/processes, and avoid silent resource drift.

8) Wallet is a built-in capability

Agent wallet supports EVM + Solana flows.
Use policy controls and explicit verification for safety-sensitive operations.

9) Teach the agent intentionally

Correct mistakes explicitly.
Persist stable preferences/methods to memory.
Convert proven workflows into skills.

10) If blocked, ask directly

Encourage users to state desired outcome clearly.
Agent should search docs/tools/skills and work backward to execution.

How this maps to onboarding behavior

Start with a quick win, not a feature tour.
Ask discovery questions around repetitive pain points (frequency + annoyance).
Show sample output first, then schedule automation.
Confirm delivery channel before creating recurring pushes.
Save user preferences and successful workflow patterns for continuity.

Product feedback channel

If users report persistent missing features or broken workflows, direct them to:
https://t.me/starchild_beta

Maintenance note

This file is intentionally concise and operational. Update when the product guide changes materially (routing, connectors, channel flow, scheduling, wallet safety, or onboarding sequence).

3.0 KiB Raw Permalink Blame History