expression-precision - SKILL.md Agent Skill

name: expression-precision description: Turn vague, abstract, buzzword-heavy, emotionally low-resolution, or hard-to-understand English or Chinese wording into precise, concrete, reusable language. Detect ambiguous wording, identify the missing information needed to describe the situation accurately, and iteratively ask concrete multi-select questions with a none-of-the-above path until the user confirms there is no more relevant context and the meaning is specific enough to express. Use when the user asks to clarify intent, name a feeling or state, rewrite fuzzy expression, remove "big words", prepare feedback, define requirements, sharpen strategy language, or produce [Precise Name]/[Concrete Explanation]/[Copyable Expression] or [精准命名]/[具体解释]/[可复制表达].

Expression Precision

Purpose

Transform thoughts, feelings, judgments, and abstract wording into language that another person can understand in nearly the same way and act on when appropriate. Support English and Chinese. Match the user's language; default to English only when the language is ambiguous.

Non-Negotiable Contract

Do not polish the user's sentence before understanding what it means.
First identify the exact words, references, or missing facts that make the input vague, ambiguous, or difficult to understand.
Determine which information is necessary to describe the situation accurately. Ask only for high-value missing information.
Use concrete multi-select questions to help the user recognize and recall details. Tell the user they may choose more than one option, and add a final "None of the above" / "都不是" option when presenting hypotheses. Treat that option as exclusive: choose it only when no substantive option fits. In UI implementations, pair this option with a free-text supplement field.
Treat "None of the above" / "都不是" as evidence that the current hypotheses missed the user's meaning. If the user supplies supplement text with it, treat that text as the primary factual entry point. Stop guessing within the same frame and guide the user to recall observable facts before generating new options.
Continue clarification until both conditions are true:
- The user confirms there is no more relevant context to add, or explicitly asks for direct output.
- The available information is specific enough to produce a concrete expression without pretending to know unknown facts.
Do not invent events, motives, data, or certainty. If the user requests direct output before the meaning is fully clear, state the necessary assumption briefly.
Remove "big words" and abstract labels unless they are defined by observable behavior, scene, data, threshold, boundary, risk, or requested action.

Use final output headings based on the response language:

[Precise Name]:
[Concrete Explanation]:
[Copyable Expression]:

[精准命名]：
[具体解释]：
[可复制表达]：

Working Record

Maintain a compact working interpretation during the conversation:

Intended use: self-understanding, message, feedback, document, requirement, strategy statement, or decision.
Known facts: what the user has actually said.
Competing interpretations: meanings that remain plausible.
Missing information: the smallest set of facts that would distinguish those meanings.
Readiness: whether a third party could understand the same situation and whether the final wording can avoid unresolved placeholders.

Do not expose a long internal checklist. Show only the diagnosis and questions that help the user move forward.

Operating Workflow

1. Audit The Input

Locate specific sources of imprecision:

Abstract nouns: "value", "experience", "quality", "impact", "growth", "安全感", "价值", "体验".
Big verbs: "optimize", "enhance", "empower", "align", "scale", "赋能", "优化", "闭环", "对齐".
Low-resolution feelings: "anxious", "stuck", "overwhelmed", "drained", "焦虑", "内耗", "累", "难受".
Judgments presented as facts: "unreliable", "disrespectful", "strong", "不靠谱", "不尊重", "强势".
Unclear references: "this", "that", "they", "it", "这个", "那个", "他们", "这件事".
Missing scene or evidence: who did what, when, where, how often, what was said, what changed.
Missing comparison or standard: better than what, enough by what threshold, failed which expectation.
Missing purpose or request: who will read this, what should they understand or do next, by when.
Dense or overloaded wording: one sentence that mixes several events, feelings, conclusions, or requests.

Quote short phrases from the user's input and explain why each could be interpreted in more than one way. Use plain language, not linguistic jargon.

2. Classify The Dominant Intent

Choose the closest intent so the questions target the right information:

emotion/state: name a personal feeling, need, or condition.
feedback/conflict: describe behavior, impact, boundary, and desired change.
work/requirement: define a task, product need, or acceptance standard.
strategy/value: turn abstract positioning or value language into audience, proof, and behavior change.
decision/blocker: clarify options, unknowns, constraints, tradeoffs, and next action.

3. Determine The Needed Information

Select only the dimensions that materially change the final expression:

Actor: who is involved?
Scene: when, where, or under what condition?
Observable behavior: what happened or should happen?
Evidence: exact words, example, data, or visible result?
Frequency or scale: once, repeatedly, how many people, how much?
Impact: what changed in progress, emotion, trust, cost, risk, or outcome?
Standard: compared with which expectation, threshold, or alternative?
Boundary or constraint: what must not happen or cannot be sacrificed?
Request: what should happen next, by whom, and by when?
Audience and tone: who will receive the final wording and how direct should it be?

Do not ask for every dimension. Prioritize information that separates competing interpretations or changes the copyable expression.

4. Ask One Clarification Round

Ask 1-3 short questions per round. Each question should fill one high-value gap.

Start by naming the specific fuzzy points and the information still needed.
Give 2-4 substantive options based on plausible scenes, actions, or outcomes, not synonyms or abstract labels.
Make the options multi-select by default. Say explicitly that the user may choose one or more options.
Add "None of the above" / "都不是" as the final option when the options are hypotheses about the user's meaning, situation, or desired result.
Let the user modify an option or answer outside the choices.
Add one relevant memory prompt, such as a specific occasion, exact words, frequency, before-and-after difference, or desired change.
End by inviting any other relevant detail.
Avoid generic prompts such as "Can you explain more?", "Why?", "请详细说明", or a long questionnaire.

Example:

我先定位到两个模糊点："压力大"没有说明压力来源，"想推进"没有说明希望谁做什么。
为了把这件事说具体，还需要确认压力来源和你希望发生的下一步。

1. 哪些压力来源接近你的情况？
A. 截止日期快到了，但依赖方还没给材料。
B. 目标被改了，但验收标准没有同步更新。
C. 工作量超过当前人手，需要重新排优先级。
D. 都不是。

2. 你希望对方下一步做哪些事？
A. 明确交付时间。
B. 确认验收标准。
C. 重新决定优先级。
D. 都不是。

可以多选、修改选项，或补充其他情况。如果选择"都不是"，请说一件最接近你真实感觉的具体事情。回想一下，最近有没有一次最能代表这个问题的具体场景？

5. Iterate Without Losing Context

Fold each answer into the working interpretation.
When useful, briefly restate what is now understood before asking the next question.
If the answer introduces a new ambiguity, ask about it in the next round.
Do not repeat questions the user has already answered.
If the user selects "None of the above" / "都不是", acknowledge that the options missed the point. Do not present another set of guesses immediately. If the user already supplied concrete facts, use them without asking the user to repeat them. Otherwise, ask the user to provide whichever factual entry point is easiest: a specific event, exact words, observable action, difference from their expectation, or the result that bothered or mattered to them. Use that answer to build a new set of options.
When the information appears sufficient but the user has not confirmed completion, ask a final check:

我已经可以把这件事说得比较具体了。还有没有你希望保留的细节，例如某次具体场景、对方原话、发生频率或你最希望改变的结果？如果没有，请直接说"没有了"。

6. Apply The Stop Gate

Produce the final expression only when both completion conditions in the core contract are satisfied.

Use these readiness tests:

A third party can tell what happened, what the user means, or what should happen next.
Important judgments have been translated into observable behavior or a clearly stated personal interpretation.
The expression includes the relevant scene, impact, standard, boundary, or request.
No pronoun, abstract label, or missing fact would cause two reasonable readers to understand materially different things.
The copyable expression can be used without obvious placeholder editing, unless the user truly does not know a fact.

If the user says there is no more context but the input is still insufficient, do not create false precision. Name the remaining gap, offer concrete assumptions to choose from, or produce a best-effort version that explicitly marks the uncertainty.

If the user explicitly asks for direct output, comply. State one short assumption sentence when necessary, then provide the final output.

7. Produce Precise Language

Put observable actions before judgments.
Translate feelings into trigger, unmet need, boundary, risk, or desired support.
Translate strategy words into audience, scenario, proof, and desired behavior change.
Translate requirements into current problem, target user, desired outcome, constraints, and acceptance criteria.
Fit the copyable sentence to the target audience and requested tone.
Keep the copyable expression ready to paste. Do not include method explanation inside it.
Prefer common, concrete words. Do not replace one vague phrase with a more impressive vague phrase.

Reference Loading

Load only the reference needed for the current request:

For vague feelings, needs, personal states, and business buzzwords, read references/state-translation-dictionary.md.
For identifying information gaps and generating high-quality clarification questions, read references/question-logic-chain.md.
For final wording style, bilingual headings, risk controls, and output templates, read references/style-and-output-rules.md.
For research foundations and design rationale, read references/research-basis.md only when asked to explain or improve the skill.

Quality Bar

A successful transformation must pass these checks:

A third party can understand the situation without asking what the main nouns or verbs mean.
The wording removes inflated labels unless they are explicitly defined.
The result distinguishes fact, interpretation, feeling, and request where relevant.
The request is concrete enough to accept, reject, negotiate, or execute.
The final sentence can be copied without invented details or avoidable placeholders.