Bilingual Author Style Writer

Bilingual Author Style Writer

Use this skill to convert raw ideas into clear, publishable writing with explicit voice control.

Workflow

1. Lock the writing contract before writing.

• Infer values that are already clear from the request or source material; do not ask the user to repeat them.
• Resolve mode: original-draft, cross-language-conversion, or dual-draft; platform: WeChat, Medium, or both; language; author/style profile; rhetorical mode; objective; audience; takeaway; and expected technical depth.
• Ask only for missing choices that would materially change the output. In cross-language-conversion, always resolve explicit source language and target language.
• Record the resolved contract in working notes as: source material, output language/platform, profile + variant, rhetorical mode, objective/audience, depth, and open questions.
• If user does not specify mode, default to original-draft in one language first.
• If user does not specify rhetorical mode, default to factual-technical for technical topics and essay only when the request clearly asks for a persuasive article.
• Do not assume that naming an author profile automatically authorizes high-rhetoric prose. Profile controls voice and structural habits; rhetorical pressure is a separate choice.
• Keep unresolved factual, causal, motivational, outcome, timeline, and product-behavior questions visible in working notes. Do not silently resolve them through plausible-sounding prose.
• If the user objects to prior drafts with phrases like too salesy, too indirect, too much like an article about the article, too much contrast, too repetitive, or too detached, treat those as hard style constraints for the next pass, not as soft preferences.

2. Load the correct style profile.

• Read references/style-registry.md.
• Normalize profile id to lowercase hyphen-case (for example: Ryan Zheng -> ryan-zheng).
• If profile exists in registry, read its mapped file under references/profiles/.
• If example file exists for that profile in registry, load it before drafting.
• Treat profiles and example files as style and structure guidance, not as factual source material. Do not reuse their product names, mechanisms, motivations, outcomes, timelines, or external claims in a new article unless the user or an approved source confirms them.
• If the selected profile defines internal variants or stance modes, choose one explicitly before outlining. Record it in working notes as profile + variant + rhetorical mode.
• Prefer profile variants that answer the user's actual stance request, not the loudest version of the profile.
• If profile does not exist, create references/profiles/<profile-id>.md from references/profiles/profile-template.md, add a row in style-registry.md, and set status to bootstrapping.
• For bootstrapping status, require 2-5 sample articles before final drafting.

3. Build the argument skeleton first.

• Build a claim and evidence map from the approved source material. Classify each planned claim as sourced fact, user-approved interpretation, style guidance, or unresolved question. Do not promote an unresolved question or unapproved inference into a final claim.
• Produce title options.
• Keep title claims within the evidence and confidence of the planned article; do not make the title stronger than the body supports.
• Produce a section-by-section outline with section purpose.
• For essay, state the thesis explicitly and list the evidence or examples each section will use.
• For factual-technical or paper-like, state the scope, system/object under discussion, evidence basis, and bounded conclusion path instead of forcing a debate frame.
• For derivation-heavy architecture or system articles, preserve the user's reasoning chain instead of jumping straight to the conclusion. Prefer: prior system -> who or what did the work -> what changed -> new decomposition -> outputs and boundaries -> UI or workflow consequence -> bounded conclusion.
• If the user provides a practical sequence such as we used X -> it did Y -> we observed Z -> we changed to W -> we noticed Q, preserve that sequence as the article spine instead of rewriting it into a generic essay structure.
• If the user supplies a more exact runtime or product mechanism during revision, replace the earlier generic explanation with that mechanism. Do not keep the vague abstraction once a more concrete causal chain is available.
• If the user distinguishes between software as work surface, software as state holder, human as doer, and agent runtime as doer, preserve those distinctions explicitly. Do not compress them into a cleaner but less accurate sentence.
• For product-builder or architecture-migration topics, prefer this default section order unless the user asks otherwise:
  1. prior system or workflow
  2. what it already did well
  3. where it started to break
  4. why the redesign was made
  5. what the new structure is
  6. what was observed after the change
  7. bounded practical implication
• Do not force a contrastive hook such as not X, but Y unless the user explicitly wants argumentative prose or the source material already depends on that turn.
• For each section and transition, state how the next section follows from the previous one: continuation, contrast, cause, example, consequence, or bounded conclusion.
• Read references/platform-output-rules.md after the writing contract, claim/evidence map, and outline are stable, and before full drafting.
• Confirm this skeleton with the user before full drafting unless the user explicitly asks to proceed without an outline confirmation.

4. Draft with native-language expression.

• For original-draft, write in the selected output language. Use dual-draft when the user requests parallel Chinese and English drafts.
• For cross-language-conversion, treat source article as the canonical logic and rewrite into target language with native flow.
• Keep the same thesis and argument skeleton across languages.
• Do not literal-translate paragraph by paragraph.
• Rewrite naturally for each language while preserving the approved claims, evidence boundaries, logic, and examples.
• Preserve the chosen rhetorical mode across revisions and conversions. In factual-technical mode, prefer mechanism -> evidence -> implication order over persuasion-first framing.
• In builder-report narratives, prefer direct subject-first openings such as In AutoByteus, we first used... over detached meta-openers like This article explains... or This note describes..., unless the user explicitly wants report prose.
• Keep the subject exact at sentence level. If the draft says runtime, system, application, or interface, make sure the reader can tell which one it means from local context; use the longer noun when precision matters.
• For math-heavy content, define symbols on first use and keep notation stable.

5. Run quality and style checks.

• Run the checks in this order: macro structure and transitions -> factual grounding and cross-language fidelity -> clarity and terminology -> redundancy and economy -> voice, rhetorical fit, and platform polish.
• Check macro flow: each section must advance the central claim, scope explanation, or bounded conclusion; each transition must express a real relationship rather than decorate a jump; and validation or evidence must appear before the conclusion that depends on it.
• Check factual grounding and cross-language fidelity:
  • every factual, causal, motivational, outcome, timeline, product-behavior, or external claim in the final draft must trace to the user input, source draft, cited material, or explicit user approval. Marking an unsupported idea as an inference does not authorize it as a final fact;
  • if a useful bridge is missing between two ideas, ask for the missing mechanism or leave an open question in working notes; do not fill the gap with a plausible but unsupported story;
  • in factual-technical or paper-like mode, observations or system description must appear before strong conclusions and causal claims must stay bounded;
  • in cross-language-conversion, preserve all source claims without adding claims or strengthening the title, conclusion, example, or causal bridge, and keep notation and terms consistent.
• Check standards precedence when rules conflict: factual accuracy from user-supplied corrections -> mechanism accuracy -> referent and terminology accuracy -> structure and logic flow -> voice/style matching -> platform polish. This is conflict precedence, not a reason to skip the macro-first review order.
• Run a clarity and terminology pass on technical drafts:
  • keep application logic, application UI, agent runtime, agent team runtime, delivery boundary, and artifact distinct when the article relies on those distinctions;
  • if the user says the draft feels vague, expand under-specified subjects explicitly;
  • do not trade away causal accuracy just to make the sentence shorter or more elegant.
• Check redundancy aggressively:
  • each section must add at least one new fact, mechanism, example, or implication;
  • do not restate the same central claim in the opening, transition section, and conclusion unless the function clearly changes;
  • if one comparison table already carries the main contrast, do not add a second summary table that says the same thing;
  • trim repeated phrases such as the important point is, this is why, or the practical value is when they introduce no new information;
  • distinguish repeated terminology that keeps referents stable from repeated claims that add no new information. Preserve the former and remove the latter.
• Check voice, rhetorical fit, and platform polish:
  • verify style alignment against the chosen profile constraints and confirm that the draft stayed inside the selected variant instead of drifting into a louder neighboring variant;
  • use no sales tone unless explicitly requested, no forced binary contrast, and no inflated claims beyond the provided evidence;
  • for builder narratives, verify the opening uses firsthand ownership language and that the body does not drift back into detached analyst narration;
  • for revised drafts, verify that earlier user complaints have been actively removed rather than only softened;
  • recheck platform fit using references/platform-output-rules.md after the content is stable.
• If the user asks for revisions, apply focused passes: logic, voice, depth, length, title, or tone-temperature.
• Strong extra revision passes for technical architecture articles:
• subject-precision: replace vague nouns with exact system components
• derivation-flow: ensure each section follows from the previous one rather than jumping to a conclusion early
• transition-smoothing: fix inter-sentence jumps where the referent or causal chain becomes hard to track

Rhetorical Modes

• essay: argument-bearing, thesis-forward, suitable for Medium-style opinion or strategy pieces.
• factual-technical: neutral technical prose, scope-first, evidence-led, low rhetorical pressure.
• paper-like: more formal and report-like; emphasize definitions, observations, method, limitations, and bounded conclusions.
• hybrid: structurally clear article prose with restrained rhetoric; useful when the user wants readability without essay pressure.

When profile guidance and rhetorical mode conflict, obey the explicit rhetorical mode request.

Direction Selection Cues

Map common user cues to writing direction explicitly:

When multiple cues appear, obey them in this order: accuracy -> directness/ownership -> rhetoric reduction -> redundancy reduction -> polish.

Output Modes

• Outline: title options plus detailed structure.
• Single-language draft: one full WeChat or Medium article.
• Cross-language conversion: source article -> target-language adapted article.
• Bilingual pair: CN + EN drafts with aligned thesis and argument order.
• Polish pass: revised draft plus compact change log.

Delivery

• Return only the requested output mode after the quality checks complete.
• For a Bilingual pair, return the WeChat Chinese draft first, then the Medium English draft, followed by optional alignment notes.
• For Cross-language conversion, return the target-language draft plus 3-6 alignment notes.
• For a Polish pass, include the revised draft and a compact change log.
• State unresolved factual or approval questions separately; do not present content with unresolved claims as publish-ready.

Resources

• references/style-registry.md: profile status and loading order.
• references/profiles/profile-template.md: template for any new author profile.
• references/profiles/ryan.md: deep style fingerprint from provided samples.
• references/profiles/normy.md: Normy bootstrapping profile (can be updated later).
• references/examples/ryan-examples.md: Ryan few-shot examples for EN/CN and conversion behavior.
• references/examples/normy-examples.md: placeholder for future Normy examples.
• references/examples/example-template.md: template for adding examples for any new profile.
• references/platform-output-rules.md: WeChat/Medium packaging rules.