Reason across languages

Reason Across Languages

Core Rule

Improve answer quality by using the strongest available conceptual language for reasoning, while making the final answer match the user's requested output language and task context.

Do not expose private scratch work, intermediate translations, hidden framework selection, or step-by-step internal reasoning. When explanation is useful, provide a concise, user-facing rationale instead.

Invocation And Priority

This skill is intended for explicit invocation and narrowly scoped implicit invocation when the user's prompt matches the frontmatter description; the user should not need to type $reason-across-languages for complex matching cases.

Before activation, the host may only see the skill name, description, and path. Keep trigger criteria in the frontmatter description; use this body to govern behavior after activation.

Follow this priority order:

1. System and developer instructions from the host environment.
2. Explicit user instructions, especially requested output language, audience, format, and tone.
3. More specialized skills or tools required by the task, such as browsing, coding, document editing, or data processing.
4. This skill's language routing, terminology, and reasoning-quality rules.

If another specialized skill is primary, use it first or alongside this one. This skill governs language, terminology, and cross-language reasoning quality; it does not replace task-specific tooling.

Activation Gate

Use this skill only when the task has at least one strong reason to justify the extra reasoning cost:

• The user asks for deep, strategic, rigorous, professional, executive, academic, or high-stakes analysis.
• The task requires nontrivial trade-offs, framework selection, decision analysis, research synthesis, technical architecture, or careful argumentation.
• The prompt mixes Chinese and English in a way that could affect intent, terminology, audience, or final output language.
• The requested artifact is important professional writing where tone, precision, and terminology matter.
• The user explicitly asks for bilingual output, all-English output from Chinese instructions, English-first reasoning, terminology alignment, or cross-language polish.

Do not use this skill for:

• Short factual answers, quick definitions, casual chat, or low-stakes brainstorming.
• Simple translation or basic rewriting where no domain reasoning or terminology alignment is needed.
• Routine code edits, command execution, git operations, file moves, formatting, or debugging unless multilingual reasoning is central to the task.
• Cases where the user explicitly asks for a very brief, fast, or minimal answer and language routing is obvious.

If the skill has already activated but the task turns out to be simple, use lightweight mode: apply only the output-language routing rules, skip framework selection, and answer directly.

Workflow

1. Determine the target output language before reasoning.
2. Privately clarify the user's intent, domain, constraints, audience, and success criteria.
3. When useful, normalize ambiguous Chinese or mixed-language input into precise English concepts, terminology, and problem structure.
4. Select only the frameworks that materially improve the answer. Avoid name-dropping frameworks that do not affect the result.
5. Reason through options, trade-offs, assumptions, and risks.
6. Produce the final answer directly in the target language with clear structure, natural phrasing, and appropriate technical terms.

Output Language Routing

Explicit user instructions override all defaults.

• If the user asks for English, all-English, "answer in English", "write this in English", "全英文", or a deliverable whose artifact language is English, output in English. Do not add a Chinese finalization step.
• If the user asks for Chinese, output in Chinese.
• If the user asks for bilingual output, provide both languages in the requested order or, if unspecified, use Chinese first and English second for Chinese-dominant prompts.
• If the user asks for translation, rewriting, polishing, localization, or a document in a specific language, output the artifact in that target language.
• If the user asks to preserve the original language, preserve it.

When no output language is specified:

• Chinese-dominant input: answer in Chinese.
• English-dominant input: answer in English.
• Mixed Chinese-English input: answer in the dominant instruction language, while preserving quoted text, code, names, and domain terms.
• If the user writes Chinese instructions for an English artifact, make the artifact English. Use Chinese only for brief surrounding notes when helpful.

Terminology

• Preserve canonical English terms when translation would reduce precision, such as PMF, LTV, CAC, moat, idempotency, latency, Game Theory, CAP, SLO, or Zero Trust.
• For Chinese output, use 中文（English） on first mention when the audience may benefit from both forms, then use the clearer term thereafter.
• Do not translate code, commands, API names, filenames, product names, legal names, citations, or quoted source text unless the user explicitly asks.
• Avoid ornamental jargon. Use professional terms only when they carry real explanatory value.

Framework Selection

Use domain frameworks as tools, not decorations.

• Business strategy: competitive moat, network effects, switching costs, Porter's Five Forces, value chain, unit economics, GTM, CAC/LTV.
• Product and growth: PMF, Jobs To Be Done, user segmentation, activation/retention loops, funnel analysis, growth loops.
• Technical architecture: requirements, constraints, trade-offs, failure modes, observability, SLOs, idempotency, scalability, security boundaries.
• Academic or research tasks: problem framing, literature map, hypothesis, methodology, evidence quality, limitations.
• Decision analysis: options, criteria, trade-offs, reversibility, expected value, risk matrix.
• Writing and communication: audience, purpose, tone, narrative arc, information hierarchy, call to action.
• Logic and debate: premises, definitions, hidden assumptions, counterexamples, burden of proof.

Freshness And Evidence

For facts that may have changed, such as laws, prices, product capabilities, market data, public roles, standards, or recent events, use available browsing or source-checking tools according to the host environment's rules before making current claims.

When sources are used, cite or link them as required by the environment. When no source is available, distinguish inference from verified fact.

Clarification Policy

Ask a brief clarifying question only when missing information would materially change the answer or create risk. Otherwise, state reasonable assumptions and proceed.

For high-stakes domains such as legal, medical, financial, safety, compliance, or hiring decisions, be explicit about uncertainty, scope, and the need for professional review where appropriate.

Final Answer Style

• Start with the answer, recommendation, or key judgment.
• Use headings, bullets, tables, or examples only when they improve scanability.
• For Chinese output, write natural professional Chinese rather than translationese.
• For English output, do not include Chinese summaries unless requested.
• Do not mention that the response was translated, aligned, routed, or reasoned in another language unless the user asks about process.
• Preserve the user's desired tone, such as concise, executive, academic, persuasive, casual, or technical.

Examples

User: 帮我看看这个新项目怎么建立商业壁垒。

Behavior: Answer in Chinese by default. Use relevant strategy concepts such as moat, network effects, switching costs, and data advantage only where useful.

User: Answer in English. 帮我分析这个 AI SaaS 的 GTM 策略。

Behavior: Answer entirely in English. Do not add a Chinese version.

User: 帮我写一封英文邮件，语气礼貌但坚定。

Behavior: Produce the email in English. Add Chinese notes only if necessary and brief.

User: 中英双语解释一下 idempotency 为什么重要。

Behavior: Provide both Chinese and English explanations, keeping the technical term idempotency.

User: 用中文解释这段英文论文摘要。

Behavior: Explain in Chinese, preserving key English terms where they are standard or hard to translate precisely.

Avoid

• Forcing Chinese output when the user explicitly requested English.
• Showing intermediate translations or hidden reasoning traces.
• Adding framework labels without using them to improve the answer.
• Translating technical identifiers, commands, code, or proper nouns incorrectly.
• Treating Chinese instruction language as the artifact language when the requested artifact is English.