中文 | English
CT Bridge 是一个连接中文加密社区与英文 Crypto Twitter（CT）语境的开源 Agent Skill。
它不只是逐字翻译，而是帮助用户处理两类真正困难的问题：
- Decode / 读懂：向中文读者解释英文加密内容中的术语、meme、语气、潜台词与文化背景。
- Publish / 表达：把中文加密内容翻译或校订为英语，同时保留作者本人的观点、语气与写作边界。
主要产品是可以安装到 AI Agent 中的 Skill。仓库还包含一个轻量本地 CLI，用于生成任务包和记录经过用户同意的反馈；它本身不会调用模型或上传数据。
中文和英文加密社区之间的信息差不只来自语言，也来自社区记忆、市场周期、meme、玩梗方式和默认常识。
直译经常丢掉语境；过度“润色”又可能擅自改变作者的判断。CT Bridge 希望在两者之间建立一套透明、可测试、可贡献的方法。
- 先保留作者原意，再优化英文表达。
- 区分字面含义、文化含义与市场推断。
- CT slang 只在自然且准确时使用，不把黑话当装饰。
- 不擅自增加确定性、攻击性、幽默或市场观点。
- 默认在本地处理用户内容，未经明确同意不上传反馈。
- 可以分析可迁移的写作特征，但不模仿在世作者的独特文风。
完整版本，同时支持 Decode 和 Publish。
独立的中译英版本，适合只需要翻译、双语校订或 CT-native 本地化的作者。它提供三种编辑边界：
- Faithful：忠实翻译，只修正准确性、术语、语法和明显的翻译腔。
- CT-native：在不改变原意的前提下，调整为英文加密读者熟悉的表达。
- Editorial：仅在用户明确授权时重组、压缩或改写文章。
独立的英译中版本。除了翻译文字，也会按需解释 CT slang、meme、语气、潜台词与必要背景，并明确区分事实和推断。
git clone https://github.com/Cloriskitty/ct-bridge.git
cd ct-bridge
# 安装完整 Skill
python3 scripts/install_skill.py --skills-dir "$CODEX_HOME/skills" --skill ct-bridge
# 或者只安装中译英 Skill
python3 scripts/install_skill.py --skills-dir "$CODEX_HOME/skills" --skill cn-crypto-publish
# 或者只安装英译中 + 语境解码 Skill
python3 scripts/install_skill.py --skills-dir "$CODEX_HOME/skills" --skill en-crypto-decode安装后刷新或重启 Agent，然后调用 $ct-bridge、$cn-crypto-publish、$en-crypto-decode，或直接用自然语言描述任务。
也可以把两个单向 Skill 分别打包：
python3 scripts/package_skills.pyPYTHONPATH=src python3 -m ct_bridge prompt decode examples/decode.txt
PYTHONPATH=src python3 -m ct_bridge prompt publish examples/publish-zh.txt --mode faithful
PYTHONPATH=src python3 -m ct_bridge feedback --concept 接盘 --choice "become exit liquidity" --rating usefulCLI 不调用模型，也不会通过网络发送数据。只有在用户主动指定时，它才会把反馈写入本地文件。
公开仓库包含工作流、编辑边界、基础术语、合成测试案例和扩展接口。高价值研究，例如真实文章标注、完整人物风格分析、gold eval、付费资料和用户偏好，不应进入公开仓库。
私有增强资料可以保存在独立仓库或本地目录，并通过 CT_BRIDGE_TASTE_DIR 连接。公开 Skill 在没有私有资料时仍可独立运行。详见 Private taste packs。
skill/ct-bridge/ 完整 Decode + Publish Skill
skill/cn-crypto-publish/ 独立中译英 Skill
skill/en-crypto-decode/ 独立英译中 + 语境解码 Skill
src/ct_bridge/ 本地 CLI
schemas/ 反馈数据协议
evals/ 合成行为测试
.github/ Issue 与 PR 模板
PYTHONPATH=src python3 -m unittest discover -s tests -v
python3 evals/check_cases.py项目目前处于早期阶段。术语库与评测集刻意保持精简，后续条目应有明确语境、可验证案例和人工审核。
欢迎提交术语、语境卡、测试案例和方法改进。请只使用原创、已获授权或足够短的说明性示例，不要提交私人稿件、付费课程内容或受版权保护的语料库。
本项目采用 MIT License。
CT Bridge is an open-source Agent Skill for navigating the cultural and linguistic gap between Chinese crypto communities and English-language Crypto Twitter (CT).
It supports two core workflows:
- Decode: explain English crypto terminology, memes, tone, subtext, and cultural background for Chinese readers.
- Publish: translate or review Chinese crypto writing for English-speaking audiences without silently rewriting the author's position or voice.
The main product is an installable Agent Skill. The repository also includes a small local CLI for preparing task packets and recording consent-aware feedback. The CLI does not call a model or upload data.
The gap between Chinese and English crypto communities is not merely linguistic. It also includes shared memory, market-cycle references, memes, irony, and unstated cultural assumptions.
Literal translation often loses that context, while aggressive polishing can overwrite the author's actual judgment. CT Bridge provides a transparent, testable, and community-improvable method between those extremes.
- Preserve author intent before optimizing the English.
- Separate literal meaning, cultural meaning, and market inference.
- Use CT slang precisely, not decoratively.
- Do not add conviction, aggression, humor, or market claims absent from the source.
- Keep user content local by default and never upload feedback without explicit consent.
- Analyze transferable writing traits without cloning a living writer's distinctive voice.
The complete skill, covering both Decode and Publish workflows.
A standalone Chinese-to-English skill for translation, bilingual review, and CT-native localization. It supports three editing boundaries:
- Faithful: correct accuracy, terminology, grammar, and translation artifacts while preserving structure and voice.
- CT-native: adapt established crypto phrasing and rhythm without changing the underlying meaning.
- Editorial: reorganize, compress, or rewrite only with explicit authorization.
A standalone English-to-Chinese skill. Beyond translation, it explains CT slang, memes, tone, subtext, and necessary background while keeping facts separate from inference.
git clone https://github.com/Cloriskitty/ct-bridge.git
cd ct-bridge
# Install the complete skill
python3 scripts/install_skill.py --skills-dir "$CODEX_HOME/skills" --skill ct-bridge
# Or install only the Chinese-to-English skill
python3 scripts/install_skill.py --skills-dir "$CODEX_HOME/skills" --skill cn-crypto-publish
# Or install only the English-to-Chinese decoding skill
python3 scripts/install_skill.py --skills-dir "$CODEX_HOME/skills" --skill en-crypto-decodeRefresh or restart the agent after installation, then invoke $ct-bridge, $cn-crypto-publish, $en-crypto-decode, or describe the task naturally.
Package the two directional skills as separate ZIP files:
python3 scripts/package_skills.pyPYTHONPATH=src python3 -m ct_bridge prompt decode examples/decode.txt
PYTHONPATH=src python3 -m ct_bridge prompt publish examples/publish-zh.txt --mode faithful
PYTHONPATH=src python3 -m ct_bridge feedback --concept 接盘 --choice "become exit liquidity" --rating usefulThe CLI does not call a model or transmit data over the network. It writes feedback locally only when explicitly requested.
The public repository contains the workflow, editorial boundaries, starter terminology, synthetic tests, and extension contract. High-value research such as annotated real-world drafts, complete writer studies, gold evaluations, paid material, and user preference profiles should remain outside the public repository.
Private enhancements can live in a separate repository or local directory and be selected with CT_BRIDGE_TASTE_DIR. The public skills remain fully usable without a private pack. See Private taste packs.
skill/ct-bridge/ Complete Decode + Publish skill
skill/cn-crypto-publish/ Standalone Chinese-to-English skill
skill/en-crypto-decode/ Standalone English-to-Chinese decoding skill
src/ct_bridge/ Local CLI
schemas/ Feedback data contracts
evals/ Synthetic behavioral cases
.github/ Issue and pull-request templates
PYTHONPATH=src python3 -m unittest discover -s tests -v
python3 evals/check_cases.pyThis is an early-stage, privacy-first foundation. The terminology registry and evaluation set are intentionally compact so future additions can be contextual, evidence-backed, and reviewed.
Contributions to terminology, context cards, evaluations, and methodology are welcome. Use only original, licensed, or sufficiently short illustrative examples. Do not submit private drafts, paid-course material, or copyrighted corpora.
Released under the MIT License.