Skip to content

JeremyJC67/super-translator

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🌐 Super Translator

A translator that explains in your language, shows you several, and learns you over time.

A mother-language-anchored, multilingual, self-evolving translation & learning skill for Claude Code — built for non-native English speakers who want to understand deeply and write professionally.

License: MIT Claude Code skill PRs welcome Stars

Super Translator demo

Most translators give you a translation. Super Translator helps you understand: every explanation is grounded in your mother language, the same meaning is shown across several languages so you grasp nuance by comparison, and it quietly gets to know you the more you use it — adapting to your weak spots and, when needed, proposing changes to its own behaviour with your approval.

✨ Why it's different

Existing tools (Nuance AI Translator, Learnables, Talkpal, Preply, Duolingo…) are strong but share three gaps:

Gap in existing apps 🌐 Super Translator
📱 Closed SaaS / mobile app Open, local, lives in your terminal & agent workflow — composable with your other skills
🔒 Personalization is a black box Your learner profile is a plain, version-controlled file you own
🔁 Mostly L1 ↔ L2 only Mother-language anchor + multilingual triangulation to expose nuance
🪨 The tool never changes for you Two-level self-evolution: silent profile adaptation + approval-gated, auditable self-rewrite

📑 Contents

🔧 How it works

flowchart LR
    A[Your sentence<br/>any language] --> B{Load context<br/>config + profile}
    B --> C[Detect direction<br/>& intent]
    C --> D[Multilingual<br/>anchor block]
    D --> E[Relevant lenses<br/>register · rewrite · vocab]
    E --> F[📤 Card]
    F --> G[(profile.md<br/>Level-1 evolution)]
    G -. divergence large .-> H[Propose self-rewrite<br/>Level-2 · you approve]
Loading
  1. Load context — reads your config.md (anchor + targets) and profile.md (what it knows about you).
  2. Detect direction & intent — comprehend vs produce; both directions first-class.
  3. Multilingual anchor block — meaning grounded in your mother language, triangulated across languages.
  4. Apply only the lenses this input needs — register, professional rewrite, key vocab & grammar.
  5. Level-1 evolution — append what it learned about you to profile.md.
  6. Level-2 evolution — when the skill itself no longer fits, it proposes edits you approve and commit.
📄 Example output (click to expand)

Input: 他这个人说话总是绕弯子,让人摸不着头脑。

🌐 Super Translator · anchor: 中文 · intent: produce · lenses: multilingual, register, rewrite, vocab

Multilingual anchor block
  中文(原意)   他说话老是兜圈子、不直说,让人搞不懂他想表达什么。
  English (nat.) He always beats around the bush, leaving people unable to follow him.
  English (i+1)  He never says things directly, so he is hard to understand.
  Nuance   "绕弯子" → beat around the bush(地道成语);比直译 talk in circles 更自然。

Professional Rewrite · 语域阶梯
  casual    He always talks in circles — you never know what he is getting at.
  neutral   He has a habit of beating around the bush, which makes him hard to follow.
  formal    He tends to be indirect and evasive, leaving his meaning unclear.
  回译核对  他习惯绕弯子,让人难以抓住重点 —— 与原意一致 ✓

— learning note logged to profile.md —

📦 Install

Requires Claude Code.

git clone https://github.com/JeremyJC67/super-translator.git
cd super-translator
./install.sh          # symlinks this repo into ~/.claude/skills/super-translator

Then, in Claude Code:

/super-translator 他这个人说话总是绕弯子,让人摸不着头脑。

…or just send a sentence to translate — the skill triggers automatically.

⚙️ Configure

Edit config.md:

  • Anchor (mother language) — default zh. All explanations ground here.
  • Targets — default en; add a triangulation language anytime (e.g. [ja] for Japanese) to compare nuance across three languages.
  • Preferences — adaptive depth, register ladder, back-translation check.

🧬 Self-evolution (two levels)

A markdown skill can't safely rewrite itself at call time — so evolution is explicit and auditable. See evolution.md.

Level Trigger What happens Safety
1 · Adapt to you Every interaction Appends your errors / learned items / preferences to profile.md; output adapts next time Additive only, plain file you own
2 · Rewrite the skill Divergence is large Proposes concrete edits to SKILL.md / lenses with evidence Approval-gated + git-versioned

This split — fast personal adaptation vs slow auditable structural change — is what keeps it both responsive and under your control.

🗂 What's inside

SKILL.md      orchestrator (the skill entrypoint)
config.md     your settings: anchor, targets, preferences
profile.md    self-evolving learner memory (yours, version-controlled)
evolution.md  the two-level self-evolution protocol
lenses/       composable understanding modules — extensible
  multilingual.md   anchor + multilingual triangulation (always on)
  register.md       formality / tone / connotation
  rewrite.md        professional rewrite + register ladder
  grammar-vocab.md  key vocab, collocations, grammar
docs/         research landscape & design notes
demo/         the demo GIF + its vhs script

🧩 Contributing: add your own lens

A lens is one focused understanding module. Adding one is the easiest way to extend Super Translator:

  1. Create lenses/your-lens.md describing what to produce and when to apply it (copy an existing lens as a template).
  2. Reference it from SKILL.md step 3.
  3. Open a PR. Issues and ideas welcome.

To regenerate the demo GIF after changes: vhs demo/demo.tape (needs vhs).

🔬 Research notes

This is a personal tool first — and a candidate testbed for self-evolving, measurable skills second. The honest competitive scan and the "is there a paper here?" analysis live in docs/research-landscape.md.

📜 License

MIT — see LICENSE.

About

Mother-language-anchored, multilingual, self-evolving translation & learning skill for Claude Code

Topics

Resources

License

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages