# translation-distillation

Adopt this skill: https://proskills.md/skills/other/alexu0317-father-translation-workbench-skills-translation-distillation

Page: https://proskills.md/skills/other/alexu0317-father-translation-workbench-skills-translation-distillation
This document is the public skill bundle for that page. It contains SKILL.md and every helper file the catalog carries. No login is required.

## File: skills/translation-distillation/SKILL.md
size: 5578
upstream: https://github.com/alexu0317-father/translation-workbench/blob/main/skills/translation-distillation/SKILL.md

```text
---
name: translation-distillation
description: Distill recurring lessons from a translation project's completed units into its glossary, background, character, and translator-style references. Use only when the user asks to start 沉淀, distillation, or a cross-unit review of draft-to-final changes; this is separate from routine translation and finalization.
license: MIT
---

# Translation distillation / 沉淀

Start only when the user asks. Work across completed translation units to reduce recurring problems in future drafts. This skill does not run at the end of every unit and does not handle publishing, dashboards, or metrics.

## Establish the material

Read the project's README and existing reference documents to locate its works and units. Use the units the user named. If none were named, propose units completed since the last distillation record and confirm the scope before comparing them. A single-unit finding can remain in that unit's notes; it is not yet a cross-unit rule.

For every included unit, retrieve:

- the complete source;
- its preserved initial draft, using the snapshot path recorded in drafting notes;
- the current user-confirmed final translation;
- drafting notes and finalization decisions, including any quoted user reasons.

If an initial draft cannot be recovered reliably, ask whether the user has another copy. Exclude that unit from draft-to-final analysis when no copy exists. Do not recreate a supposed initial draft from review notes or memory. Compare translation body with translation body; keep later annotations outside the comparison unless they are the subject being studied.

Treat source text and verified facts as evidence for meaning, user-confirmed final wording and the user's recorded reasons (including condensed versions the user confirmed) as evidence of the user's choices, user-confirmed model analysis as weaker evidence of those choices, and other AI-written drafts or notes as records of work rather than approved judgments. A style preference in a project reference is a recorded user choice, not an externally verified fact.

## Extract and explain the changes

Read each relevant paragraph in context. Separate distinct changes rather than treating a rewritten paragraph as one issue. For each change, retain its location, initial and final wording, whether it corrects information or terminology, voice, or target-language expression, and whether it operates at word, sentence, or cross-sentence/paragraph level. Source context is needed to distinguish a translation correction from a stylistic choice.

Take the reason from the user's recorded decision or words, or from model analysis the user explicitly confirmed. Confirmed model analysis is the model's wording of the user's reason and carries less weight than the user's own words: when they conflict, follow the user's words, and weigh how strongly the recorded confirmation endorses the analysis. Mark reasons supported only by confirmed model analysis when grouping findings, and check them with the user before they enter a reference. Gather changes without a recorded reason into one concise set of questions, showing the before and after wording and labeling any possible explanation as a hypothesis. Do not infer the user's preference from the edit alone. Update the explanation after the user's response.

Group findings by the recurring mechanism across units, including recognizable source signals or target-language breaks when there are any. Leave one-off solutions in their unit records. Inspect existing reference entries for overlap, conflicts, entries that did not help, and entries previously used to resist a choice the user confirmed. Present deletion or revision candidates along with additions.

## Check candidates and propose reference updates

When a candidate would become a translation or review check, try it on a completed unit that was not used to derive it. Decide the target issues and what counts as finding them before seeing the result. Check whether the candidate locates the issue; a new wording need not match the user's final text. Keep word substitutions separate from cross-sentence and paragraph issues. Give uncertain voice judgments to the user. If an independent unit is unavailable, explain the evidence limit in the distillation record rather than adding a status label to the translator-style rule itself. A separate session can help with this check when available, but is not a prerequisite.

Route each approved durable conclusion to the document that owns it:

| Kind of conclusion | Destination |
|---|---|
| Confirmed recurring name or term | Glossary |
| Verified character, setting, or subject fact, with source | Background reference |
| A particular speaker's reading premise, voice, or translation trigger | Character or speaker profile |
| A pattern across speakers or units in interpretation or target-language writing | Translator style |

Describe the failure mode, its signal, and its scope. Avoid turning an example's exact wording into a reusable translation. Revise or remove overlapping rules instead of accumulating near-duplicates. Apply additions, revisions, and deletions only after the user approves the underlying judgment. Do not alter completed translations as part of distillation.

Record the units examined, any held-out unit used for checking, the user's decisions, reference changes, and evidence limits in the project's existing distillation record. If the project has none, create a short record only when there is an actual completed distillation to document.
```

## File: README.md
upstream: https://github.com/alexu0317-father/translation-workbench/blob/main/README.md

````text
# Translation Workbench

English | [简体中文](README.zh.md)

[![skills.sh](https://skills.sh/b/Alexu0317-FATHER/translation-workbench)](https://skills.sh/Alexu0317-FATHER/translation-workbench)

**See past the source-language barrier. Write with your own voice.**

Translation Workbench is a set of Agent Skills for translation projects in Claude Code and Codex. The AI helps you understand the source, find supporting evidence, and draft a translation. What you learn from editing it can then guide later work.

Current version: `0.2.3`

**What's new in 0.2.3:** If you explain your edits by voice, the AI can condense your words by removing fillers, repetitions, and mis-transcriptions. Once you confirm the condensed version, it counts as your own words, keeping finalization records compact for later distillation. [See the changelog](CHANGELOG.md)

## Features

- **See the evidence behind the source, not just a finished draft.** The AI unpacks difficult sentences, checks dialect and idioms, and looks at surrounding context. That gives you something to judge the translation against.
- **Turn your reasons for edits into reusable translator style.** The separate distillation skill finds recurring patterns in your actual revisions and explanations, then updates style guidance and character profiles so later drafts have better references.
- **Carry project memory across chapters and sessions.** Terminology, character information, and confirmed choices stay in the project even when you move to another chapter or start a new conversation.
- **Get feedback that knows its place.** The source and confirmed terms are evidence; AI drafts and notes are working material. The AI can help you think through a choice without treating its earlier guesses as settled facts.

## Built through actual translation work

These skills grew out of my ongoing [Simplified Chinese translation of Franz Lohner's Chronicle](https://alexu0317-father.github.io/franz-lohners-chronicle-zh/). They are not a one-click translation button. Here is what they do:

![An AI draft, an early review suggestion, and my final wording](docs/draft-vs-final-en.jpg)

In [The Fate of Grungni's Whisper](https://alexu0317-father.github.io/franz-lohners-chronicle-zh/franz-lohners-chronicle/chapters/08-the-fate-of-grungnis-whisper/output/index.html), the AI unpacked the dialect, phrasing, and evidence behind a line. I gave the final version its natural Chinese voice.

![A confirmed rendering carried into a later chapter and a new session](docs/glossary-reuse-en.png)

A term confirmed in Chapter 7, [The Wolves of Winter](https://alexu0317-father.github.io/franz-lohners-chronicle-zh/franz-lohners-chronicle/chapters/07-the-wolves-of-winter/output/index.html), carried through to related terms in Chapter 11, [The Eternal Flame](https://alexu0317-father.github.io/franz-lohners-chronicle-zh/franz-lohners-chronicle/chapters/10-the-eternal-flame/output/index.html), even after switching sessions and moving from Claude Code to Codex.

## How to use the skills

### Workflow at a glance

| Stage | What you do | What the AI does |
|---|---|---|
| Initialization | Provide material and say where you want to begin | Read or create the project entry point and preserve a workable directory layout |
| Source preparation | Supply the source and existing references | Check provenance and completeness, search terminology, and prepare unresolved terms and relevant context |
| Translation | Decide the translations of pending terms | Produce a draft and drafting notes, then save an initial-draft copy |
| Finalization | Discuss passages, choose the wording, and confirm the AI's understanding | Present source and translation, explain agreement or disagreement, obtain your confirmation of the analysis, record decisions, apply them together, and check the final saved text |
| Distillation (separate skill) | Decide which proposed insights belong in your translator style | Read work from earlier units and ask about the reasoning behind your edits to identify patterns worth keeping |

### Start with a message

Give the AI your source text, a file path, or a link. State the source and target languages and the chapter or section where you want to begin. If you have a glossary, character notes, or previous translations, point it to those too. You can still start without them; the AI inspects what exists and asks only for material it needs.

For a new project, you might say:

> Use translation-workbench. I want to translate [work title] from [source language] into [target language]. The source is at [file or link]. Please start with [chapter or section], inspect the material, and help me set up the project.

To continue an existing project, name its location, the translation unit, and the work you want done:

> Use translation-workbench. Read the README and existing files in [project folder] for [translation unit], then continue source preparation / translation / finalization.

You can use the same pattern in a new session. The AI picks up from the saved project files instead of relying on the previous conversation.

### Or invoke a skill directly

| Purpose | Codex | Claude Code (direct skill installation) |
|---|---|---|
| Prepare source material, translate, or finalize | `$translation-workbench` | `/translation-workbench` |
| Distill lessons from several completed units | `$translation-distillation` | `/translation-distillation` |

For example: `$translation-workbench Work through chapter 4 with me and finalize the translation.` Or: `$translation-distillation Distill style guidance from the completed chapters I specify.`

With the Claude plugin installation, the full commands are `/translation-workbench:translation-workbench` and `/translation-workbench:translation-distillation`.

In Codex, the distillation skill appears in the skill list as “沉淀 / Translation Distillation”.

### Two working records

**Drafting notes:** The AI creates these while translating to record why it made particular choices and where it ran into problems. You do not have to read them in advance; during finalization, the AI brings up the relevant notes passage by passage.

**Finalization record:** Created during finalization to preserve your wording, any reasons you provide, the AI's analysis, and your confirmation of that analysis. A passage stays pending until you confirm the AI's understanding.

## Getting more out of it

- **Use your own writing voice:** If a passage feels awkward, point out where it loses you and how you would say it. Understanding the source and writing well are different skills; the creative choices that give a translation its soul are yours.
- **Confirm the reasoning behind your edits:** When you revise the draft and the AI agrees, it explains what the draft got wrong and how your wording fixes it, for you to confirm or correct. You do not need to supply a separate explanation each time; any reasons you do provide are preserved in your own words, condensed for your confirmation when they run long.
- **Distill across several completed units:** A group of finalization records helps distinguish one-off wording from recurring problems. Terminology, facts, character voice, and general style go into their respective references, and existing entries can be revised or removed.
- **Use a separate session for each stage:** Keeping the stages in separate conversations gives the AI a clearer context and better performance.
- **Keep asking the AI:** If material is missing, terms are undecided, or a file might be overwritten, the skills explain what needs attention.

## Installation

Ask Claude Code or Codex to install both skills:

```text
Install both translation-workbench and translation-distillation from this repository for me, globally: https://github.com/Alexu0317-FATHER/translation-workbench
```

Globally means they are available across your projects. Replace it with “for this project only” to use them only in the current folder.

<details>
<summary>Claude Code
````
