teach-me
Teach me my unknowns — an explainer for a field's vocabulary and mental model, so vague requests become precise. Use on "teach me", "make me an explainer". Not a plan investigation.
- License 许可
- MIT
- 最近更新
- 2026-11-13
- Stars
- ★ 2
- Forks
- 0
01 前言 Front Matter
- name
- teach-me
- description
- Teach me my unknowns — an explainer for a field's vocabulary and mental model, so vague requests become precise. Use on "teach me", "make me an explainer". Not a plan investigation.
02 资源文件 1
- 参考文档agents/openai.yaml268 B
03 SKILL.md 正文
Teach Me — Domain Vocabulary Explainer (Teach Me My Unknowns)
This skill serves what the user does not know and you do, and surfaces the user's gaps.
When you and the user both seem to understand the request, do it without this method: if its result carries weight (it ships, is sent or is hard to undo), first read back in 2–3 lines the request as you understand it, the assumptions you are making and what outcome counts as done, as the whole reply, and do it once the user confirms; a self-evident request (fix a typo) needs no read-back. When the read-back exposes a mismatch, route by who does not know: the user → the teach-me skill; you → the interview skill; neither of you → the blindspot skill. A misread weighty request costs the work built on it; a read-back costs one exchange.
Vague requests ("make it better") usually stem from missing vocabulary, not missing taste. This skill fills the vocabulary gap.
Iron Rules
- Teach first; the work starts from the precise request. The user needs the words before the request can be precise.
- Teach only vocabulary needed for decisions, not an encyclopedia — start from the axes the user must choose on in this task.
Procedure
-
Parse the input.
a. Parse the domain, goal and current level from the text the user gave with the request or after the skill name.
b. With no request text, take the domain of the work under discussion; if the domain is still unclear, ask once, briefly.
-
Open with the mental model: the domain's pipeline in 3–5 ordered stages (color grading: ingest → correct → grade → match). The pipeline is there so the user knows what comes before what — correction first, then the creative look.
-
Pick 3–7 decision axes the user will decide on in this task (e.g. color grading: exposure / white balance / contrast curve / saturation vs naturalness / look).
-
Build a vocabulary ladder: per axis, everyday word → expert term, numbering each term. Give each term:
- 1 example sentence of "what you can request with this term"
- one source (a doc, standard or reference text)
Above the ladder, say in one line which sources you opened in this session and which you cite from memory, so a source cited from memory does not read as one you opened. When you can name no source for a term, say so where its source would go rather than supplying one: an invented citation reads as evidence. Mark any term whose usage you are not sure of as unverified rather than defining it confidently: a precise term can make a wrong premise sound right.
-
Show what each decision axis changes.
a. Show a before/after comparison per decision axis — same subject at two contrasting settings on that axis. Build it by domain:
Domain Comparison Visual Numbered Markdown before/after descriptions of the same subject, naming the visible differences. If a real photo is essential, ask the user for a file and refer to its visible regions. Any other domain (e.g. text, code, process) Comparison examples. b. Add 2–3 named presets (e.g. flat / corporate clean / cinematic teal-orange), with a description of each preset's overall result and its settings.
-
Give what good looks like: 4–6 judging criteria stated in the new vocabulary (e.g. "skin tones stay believable", "blacks are rich but not crushed"). Keep this step in a short explainer too. These criteria buy later revision rounds: judging criteria transfer far better than production vocabulary, because recognising a bad result is cheaper than knowing which axis to move.
-
Close with the payoff.
a. End the reply by asking which numbered terms, comparisons and presets to include in the request and which settings to change; the draft waits for the answers.
b. From the answers, write the precise-request draft: the user's original request rewritten in the new vocabulary with the picks applied. Mark every detail the user did not give (their taste, their numbers, their situation) as an assumption to confirm, so the draft does not decide for them. The draft is the precise request the work starts from.
Output
Reaction: number the terms, before/after comparisons, and presets; the user's picks assemble into the precise-request draft (step 7b); a missing pick is undecided.
Related
- Blind spots in a specific plan, system or situation: use the blindspot skill — this skill covers blind spots in a field's vocabulary. Running both is fine: when the user cannot yet describe the task, run the teach-me skill first; otherwise run the blindspot skill first and hand its ③ list (where the user does not know how to judge) to the teach-me skill.
- If requirements still diverge after gaining vocabulary (revisions stop converging), lock decisions with the interview skill rather than teaching more vocabulary.
04 同仓库更多 Skills
unknowns-loop
Run the know-your-unknowns workflow end to end, ending with an outcome check. Use on "unknowns loop", "run the operating loop", "know your unknowns". A bare "loop" is NOT this.
tweakable-plan
The tweakable plan — decisions most likely to change come first, with alternatives, for any work. Use on "plan this", "make a plan", "tweakable plan".
teach-me
Teach me my unknowns — an explainer for a field's vocabulary and mental model, so vague requests become precise. Use on "teach me", "make me an explainer". Not a plan investigation.
reference
Point at a reference — read an example, show your reading for correction before building on it; runs even when the adaptation looks simple. Use on "use this as a reference", "make it like this".
quiz
Comprehension quiz — explain the work, then quiz you so you can defend it at handover; runs even when a quiz is easy to write. Use on "quiz me", "test my understanding", or before merging.
prototypes
Prototypes — several versions of anything to react to, each on a different philosophy. Use on "divergent prototypes", "design options", "versions to react to", "know it when I see it".