mrgediao/shuorenhua
WritingThis page was translated automatically. Technical details may differ from the original. View the original
Speak plainly | Chinese first, go to AI-style rewriting skills: Preserve facts, divide scenarios, and after editing, post directly. Chinese-first rewrite skill for Codex / Claude Code / Cursor / ChatGPT — removes AI tone, preserves facts.
Overview
Speak Human is a rewriting skill for Chinese text. It is mainly used to clean up template clichés, commercial packaging, engineer posturing, translation style, and unfounded qualitative assertions in AI-generated content. It is suitable for developers, maintainers, and writers who routinely write READMEs, release notes, work synchronization, issue replies, forum discussions, and technical lengthy articles. The core logic is: all numeric indicators, version numbers, command parameters, file paths, causal conditions, and responsibility attribution from the original text are retained, with only unnecessary rendering and empty groundwork removed. Supports integration with Claude Code, Codex, Cursor, ChatGPT, and various custom agents. - : Use Speech-Speaking GPT (ChatGPT Plus / Pro) and paste text to rewrite. - : Execute the following command in the conversation to install the plugin: After installation, enter "Remove this section of the AI flavor" to call it. For manual installation and update instructions, see install/claude-code.md. - : After cloning the repository, execute once: - : - If you only want to troubleshoot the issue and do not want to modify the main text directly, you can add "Click annotation mode to mark only, not rewrite" in the command. For more access methods (Cursor, Windsurf, OpenClaw, etc.), see installation methods. 1. : Retain specific metrics and do not crudely generalize "p95 latency reduced from 480ms to 160ms" as "significant optimization." 2. : Do not arbitrarily rewrite "demonstrating cloud-native potential" as "has adopted cloud-native architecture." 3. : Conditions, scope, negation, tense, and intensity are facts and should not be omitted along with tone. 4. If the original text only contains abstract qualitative content (such as "improving efficiency"), do not arbitrarily expand it to "save development time." 5. Ensure that all factual points in the original text correspond after rewriting, and every conclusion after revision has a source in the original text. - : This optimization achieved significant performance results, effectively improving interface response issues. P95 latency dropped from 480ms to 160ms, fully demonstrating the team's ability to continuously optimize. - : This optimization significantly reduced interface latency. (Core metrics for p95, 480ms, and 160ms were lost) - : This optimization reduced interface p95 latency from 480ms to 160ms. Fixed execution process: Identify text scenarios → Lock factual fragments (numbers/commands/responsible parties) → Assess problem strength → Prioritize sentence patterns and paragraph structure (phrase lists only serve as a safety net) → Bidirectional fidelity replay → Residual auditing. - structural: Allows merging, deleting, and rearranging structures across sentences, suitable for short texts or manuscripts that explicitly require major revisions.
Overview (machine translated)
Speaking plainly is a revision skill packed into tools like Claude Code, Codex, and ChatGPT. Give it your finished text, and it will remove clichés, repeated setup, and stiff sentences, and return a draft you can use directly. Numbers, conditions, promises, and your tone remain the same; the text that was already fine remains unchanged.
Once the command is installed, Claude Code, Codex, and Cursor can all be used. Other methods can be found in installation:
Can’t see the image clearly? This is the text version, feel free to copy and try
Original text:
Hello everyone, now let me give you a detailed update on this week’s progress. This week, we completed the adjustments to the retry strategy. The retry strategy has been adjusted. Notably, after the adjustment, the number of duplicate requests per day dropped from 24 to 7. If no new errors appear before Friday, we will merge this change into the main branch. That’s all for this week’s progress. Thank you all for your attention!
After revision:
This week, we adjusted our retry strategy, reducing repeat requests from 24 times per day to 7 times. If no new errors appear before Friday, we will merge this change into the main branch.
The request was “remove this cliché, keep the information and my tone,” and was loaded with GPT-6.1-sol v2.6.0 and output, without manual changes. The same passage was given to Claude Opus 5.5, but only “after adjustment,” was added.
Installation
Claude Code
Codex, Cursor, and other tools that support skills
After running, select which tools to install. Add --global to the user directory, and all projects can use it.
ChatGPT: Directly speak human speech GPT, no installation required.
If you want to manually copy files, install them in OpenClaw, or just paste a prompt, see the installation instructions.
How to use it
After installation, you don’t need to remember commands; just paste the original text in the dialogue and say a request, and the tool will automatically activate it; If you want to ensure it works, you can specify “speak human language” in the request.
By default, only the revised full text is submitted, no rating or explanation attached. If the original text is fine, return as is. Mainly in Chinese, but English can also be edited.
Change what you keep
| Will clean | Will keep |
|---|---|
| No content greetings, announcements, or acknowledgments | Number, date, version number |
| “Notably” prompts like this | Conditions, commitments, and proportions: “if” “after approval” “possibly” |
| “Completed the adjustment work on …” These roundabout expressions | Quotes, terminology, code blocks |
| Repeating the same thing twice | Your judgment and tone are basically even about promotional reviews |
It does not determine whether the article was written by AI, nor does it guarantee it will pass AI detection.
Control the scope of changes
| Requirements | Scope of changes |
|---|---|
Freely adjust / structural |
Can be deleted, merged, or rearranged, retaining valid information and the author’s intent |
Keep structure / bounded |
Only clean within sentences; Delete entire sentences only as suggestions, no merging or rearranging |
Only change in place / in-place |
No deleting sentences, no merging sentences, no rearranging sentences; When preservation of sentence count is required, sentence separation is not done |
Your request takes priority.
This page was translated automatically. Technical details may differ from the original.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add mrgediao/shuorenhua