Capstone and Retrospective: Turning a Team's Conventions Into a Skill Pack and Driving a Subagent Through a Real Task
Combine the week's learning into one delivery: split a team's conventions document into three skills, use subagent isolation to run a real task, use an eval to prove it's actually better than going without skills, and write it up as one portfolio entry.
Today's Goals
- Split a team's conventions document into three skills with clear, non-overlapping boundaries
- Design a set of eval cases that prove the difference in output with and without skills
- Write up this delivery as a portfolio entry that holds up under follow-up questions
The last day teaches no new concepts and instead combines six days into one complete delivery: take a real team conventions document, split it into three skills, hand a real job to a clean subagent, and use a set of assertions to prove it genuinely helps — not that it feels better, but by how many percentage points the pass rate rose. Then compress the delivery into one portfolio entry that survives follow-up questions. Once you have read the walkthrough and finished the lab, scroll back to the top and tick off the three goals.
Plain-Language Walkthrough
What you deliver is capability, not documents
Pin down the acceptance criteria first, or the last day easily becomes "wrote a few more files."
This delivery's product is not those three SKILL.md files but one conclusion that survives questioning: on a given set of tasks, the pass rate went from this to that with this package, measured under these conditions. The three SKILL.md files are only intermediate artifacts toward that conclusion.
The standard follows from that: can somebody who was not involved reproduce that number from your delivery document? If they cannot, however beautifully the files are written, the delivery is not finished.
One last use of the filing cabinet metaphor: for six days you built the cabinet, the index cards, and the distribution. What today proves is that the cabinet genuinely made new hires make fewer mistakes — not that everybody says it is handy, but that with the same batch of tasks run before and after, you counted the errors.
Interview questions
How do you prove a skill actually helps rather than just feeling better?你怎么证明一个 skill 真的有用,而不是感觉上更好?
Common in ChinaCommon overseasDeep dive#agent-skills#evaluation#methodology🔒 Unlock this chapter to see the answer keys.
How many skills should a thirty-page team convention document become, and how do you split it?一份三十页的团队规范文档要拆成几个 skill,按什么切?
Common in ChinaCommon overseasIntermediate#agent-skills#design#decomposition🔒 Unlock this chapter to see the answer keys.
What is the difference between running a task in a subagent with skills and running it in the main session?让子代理带着 skill 去执行任务,和在主会话里执行有什么区别?
Common in ChinaCommon overseasDeep dive#agent-skills#subagent#evaluation🔒 Unlock this chapter to see the answer keys.
