Description

将输入法或语音输入系统已经识别出的口语文本，整理成适合即时通讯发送的自然书面语，并在用户明确指定目标语言时执行翻译。用于语音输入结束后的文本润色、消息发送前整理、口语转消息文案、聊天文本优化等场景。输入应为文本而不是音频；当用户已经有 asr 结果、想让句子更顺、更干净、更适合发送时使用本技能。默认使用 doub...

README (SKILL.md)

输入法语音消息书面化

Name: ime_message_skill
Author: hei-maom

核心原则

将已经识别出的文本改成适合发送消息的自然书面语。
默认保留原意、语气和立场，只清理口语噪声。
默认不翻译；只有用户明确指定目标语言或明确要求翻译时才翻译。
输出应短、自然、可直接发送；不要改成公文、纪要、通知或长篇说明。
不要凭空补充事实，不要把犹豫语气改成确定结论。
默认模型为 doubao-seed-2-0-lite-260215。

输入要求

优先接收以下任一文本输入：

输入法语音识别结束后得到的一整段文本
输入框中已经插入的口语识别文本
用户粘贴的一小段口语化聊天文本

不要把本技能当作音频转写技能使用。若用户提供的是音频文件、音频 URL、base64 音频或要求直接做 ASR，请不要使用本技能的主流程。

默认处理流程

获取待处理文本。
判断用户是否要求翻译。
生成改写要求：
- 未要求翻译时，使用默认消息书面化指令。
- 要求翻译时，在默认消息书面化基础上补充目标语言要求。
运行 scripts/rewrite_client.py 调用模型接口。
从返回结果中提取最终文本作为输出。
默认只返回最终文本；除非用户要求，否则不要展示接口原始 JSON。

默认改写指令

默认使用下面这条改写指令：

将口语输入整理成适合即时通讯发送的自然书面语，保留原意，去掉语气词和重复，不要过度正式。

如果用户提出更具体风格要求，可在此基础上追加：

更正式：改成较正式但仍适合聊天发送的表达。
更口语：保留轻松自然的聊天语气，但去掉明显口头禅和重复。
更简洁：精简表达，保持关键信息完整。
翻译：整理成自然书面语后翻译成 \x3C目标语言>。

模型与鉴权

默认服务基地址：https://models.audiozen.cn
默认模型：doubao-seed-2-0-lite-260215
api token 不写入技能包；使用者自行通过环境变量填写。
默认脚本按 OpenAI 兼容 Chat Completions 方式组装请求。
如果你的网关路径或字段与默认兼容格式不同，请修改 scripts/rewrite_client.py 中的请求构造，或通过环境变量覆盖 endpoint。

输出要求

只输出最终可发送文本。
优先输出 1 段短文本；不要拆成分析说明。
保留原有意图，例如请求、解释、迟到说明、确认、婉拒、提醒。
对金额、时间、日期、数字保持自然规范写法。
如果原文极短且已经足够自然，可以只做极小修改。

不要这样做

不要改成会议纪要。
不要改成总结报告。
不要补充没有说过的信息。
不要把“可能、应该、想、先、再看看”改成绝对结论。
不要把聊天消息改得过于官腔。

示例

示例 1：迟到说明

输入： 嗯我这边可能晚十分钟到你们先开始吧

输出： 我这边可能会晚十分钟到，你们先开始吧。

示例 2：请求查看文件

输入： 我一会把那个文件发你你先帮我看一下

输出： 我一会儿把文件发给你，你先帮我看一下。

示例 3：带翻译要求

输入： 我今天可能来不及了明天给你 要求：翻译成英文

输出： I might not make it today. I'll send it to you tomorrow.

接口与接入说明

模型接入细节见 references/api_cn.md
产品接入建议见 references/integration_cn.md
脚本入口见 scripts/rewrite_client.py

Usage Guidance

This skill is internally consistent in function — it rewrites text by calling a remote model — but it will send user-provided message text to an external service (default https://models.audiozen.cn) and requires an API token (IME_MODEL_API_KEY) even though the registry metadata doesn't declare it. Before installing: 1) Verify you trust the endpoint and the operator of models.audiozen.cn; decide whether sending potentially sensitive chat content to that service is acceptable. 2) Provide the API token via environment variables (do not hard-code it in the package). 3) Review scripts/rewrite_client.py (it will include the full raw model response if run with --json and returns response.model_dump() in JSON) to ensure no unexpected logging or storage. 4) If privacy is a concern, consider running a locally hosted/enterprise model or an allowed endpoint, and set IME_MODEL_BASE_URL accordingly. 5) Note the metadata mismatch (required env var not declared) — make sure deployment automation sets IME_MODEL_API_KEY or the skill will fail.

Capability Analysis

Type: OpenClaw Skill Name: ime-message-skill Version: 1.0.1 The skill is designed to rewrite spoken-style text into natural written language for instant messaging. It uses a standard Python script (scripts/rewrite_client.py) and the OpenAI SDK to communicate with a specified model endpoint (https://models.audiozen.cn/v1). The code and instructions in SKILL.md are transparent, follow best practices by requiring API keys via environment variables rather than hardcoding them, and lack any indicators of malicious intent, data exfiltration, or unauthorized execution.

Capability Assessment

ℹ Purpose & Capability

Name/description match the implementation: the skill rewrites already-recognized spoken text into sendable IM messages and uses a remote LLM for that. However, the package registry metadata lists no required env vars while the SKILL.md and scripts require IME_MODEL_API_KEY (and allow IME_MODEL_BASE_URL/IME_MODEL_NAME). This metadata omission is an inconsistency that could mislead integrators.

✓ Instruction Scope

SKILL.md and scripts confine behavior to taking input text, optionally adding a translation instruction, calling the configured model endpoint, and returning the rewritten text. There are no instructions to read unrelated files, shell-history, or other system secrets. The skill explicitly warns not to accept audio/ASR work itself.

✓ Install Mechanism

No install specification (instruction-only plus a small helper script) — no archives or remote installers. The only runtime dependency is the openai Python package used by scripts/rewrite_client.py, which is noted in the script.

ℹ Credentials

The script requires one API credential (IME_MODEL_API_KEY) plus optional IME_MODEL_BASE_URL/IME_MODEL_NAME/TEMPERATURE/TIMEOUT — reasonable for a remote-model integration. The registry metadata, however, does not declare any required env vars or primary credential, which is a mismatch and may cause integrators to miss that they must supply a token. The default base URL (models.audiozen.cn) is a third-party endpoint; confirm you trust it before sending user content.

✓ Persistence & Privilege

The skill does not request permanent/always-on status, does not modify other skills or global agent config, and does not persist credentials in the package. There is no installation step that writes to system paths.

Version History

v1.0.1

No functional or content changes; only the default model name in the description updated for accuracy. - 默认模型由 doubao-seed-2-0-pro-260215 正确标注为 doubao-seed-2-0-lite-260215。

v1.0.0

ime-message-rewriter v1.0.0 初始发布 - 支持将输入法或语音识别得到的口语文本整理为适合即时通讯发送的自然书面语 - 默认不翻译，只有在用户明确指定目标语言时才会翻译 - 保留原意、语气和立场，仅去除口语噪声、重复和语气词 - 输入需为已识别文本，不支持直接音频输入或ASR - 仅输出最终可发送的简洁短文本，不生成分析说明 - 支持通过api token调用模型服务，配置灵活

Metadata

Slug ime-message-skill

Version 1.0.1

License MIT-0

All-time Installs 0

Active Installs 0

Total Versions 2

Frequently Asked Questions

What is ime_message_skill?

将输入法或语音输入系统已经识别出的口语文本，整理成适合即时通讯发送的自然书面语，并在用户明确指定目标语言时执行翻译。用于语音输入结束后的文本润色、消息发送前整理、口语转消息文案、聊天文本优化等场景。输入应为文本而不是音频；当用户已经有 asr 结果、想让句子更顺、更干净、更适合发送时使用本技能。默认使用 doub... It is an AI Agent Skill for Claude Code / OpenClaw, with 250 downloads so far.

How do I install ime_message_skill?

Run "/install ime-message-skill" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is ime_message_skill free?

Yes, ime_message_skill is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does ime_message_skill support?

ime_message_skill is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created ime_message_skill?

It is built and maintained by HeiMaoM (@hei-maom); the current version is v1.0.1.

More Skills

ime_message_skill