| name | project-onboard |
| description | Fast project onboarding for AI agents entering this SO-101 robotics project. Use this skill at the START of every new conversation, before doing any work. Also use when: the user says "onboard", "catch me up", "what's the status", "where are we", "專案狀況", "目前進度", or any question about overall project state. This skill ensures you never waste time re-discovering what's already documented. If you're unsure whether to use this skill, use it — the cost of reading a few files is far less than the cost of giving outdated advice.
|
Project Onboard
This is a robotics learning project building toward a dual-arm mobile robot.
The docs are well-structured — read them in the right order and you'll have
full context in under 30 seconds of tool calls.
Onboarding Sequence
Read these files in this exact order. Each one is short and gives you a
different layer of understanding. Parallelize the reads where marked.
Step 1 — Big Picture (read in parallel)
docs/README.md → Current Focus (what's happening NOW), hardware list, nav links
docs/goals.md → Vision, 8-phase milestones, success criteria, reproducibility requirements
docs/roadmap.md → Phase table with priorities (P0-P3), status, dependencies, Mermaid graph
These three files tell you: what, why, and when.
Step 2 — Architecture
docs/architecture/overview.md → Module diagram, data flow, which modules are implemented vs planned
This tells you: how the pieces fit together, and what exists today vs what's future.
Step 3 — Current Phase (read only what's active)
Check roadmap.md for which phases have status in-progress. Then read only
those phase files:
docs/phases/01-hardware-setup.md (if Phase 01 is active)
docs/phases/02-teleoperation.md (if Phase 02 is active)
...etc
Each phase file has 10 sections: 目標, 前置條件, 輸入/輸出, 步驟, 驗證方式,
關鍵連結, 成功標準, 產出物, 踩坑紀錄, 風險/Blockers.
Do NOT read all 8 phase files. Only read active ones + the one immediately
before (for context on what was already done) and the one immediately after
(to understand what's coming next).
Step 4 — Latest Journal (if exists)
docs/journal/_index.md → Current Phase pointer + recent entries
If there are journal entries, read the most recent 1-2 for ground-level context
(what happened last week, current blockers, discoveries).
Step 5 — References (only on demand)
Do NOT read reference files during onboarding. They're for when the user asks
about specific topics:
docs/references/hardware.md → BOM, datasheets, wiring
docs/references/algorithms.md → ACT, Diffusion, VLA papers
docs/references/software.md → LeRobot, uv, accelerate, tools
docs/references/simulation.md → Isaac Sim, MuJoCo
docs/references/community-projects.md → Hackathon projects, XLeRobot
After Onboarding
Once you've read Steps 1-4, respond with a brief status summary:
目前狀態:Phase [N] — [name] ([status])
Current Focus:[from README.md]
下一步:[next phase]
Blockers:[any, or 無]
Keep it to 3-4 lines. The user already knows their project — they just need
confirmation that you're up to speed.
Key Project Facts (stable, rarely change)
- Hardware: 2x SO-101 arms (Feetech STS3215), Intel RealSense D435, Jetson Orin Nano 8GB, 5x RTX 8000 training workstation
- Framework: HuggingFace LeRobot (Python, installed via
uv from source)
- Ultimate goal: XLeRobot dual-arm mobile robot
- Language: 繁體中文為主,專有名詞保留英文
- Owner: 盧柏宇 (roy4222), experienced with Python/ROS2/Jetson/ML from PawAI project (Unitree Go2)