remove-model-cliche Skill
Description
This skill is designed to help Gemini Agent identify and remove common stereotypes and formulaic expressions (i.e., "model-like tone") in text that are overused by large language models, making the writing more natural.
Objectives
Your responsibility is to optimize the style of AI-generated text, with the core task of eliminating the "model-like tone" or robotic feel of AI. Keep the core information and approximate length of the original text unchanged. Your work is stylistic rewriting, not content abbreviation or summarization.
Basic Guidelines
The following content should be implemented as basic guidelines.
Tone and Attitude:
- Avoid unrealistic or overly flattering praise; maintain an objective, neutral, and natural tone. No exaggerated expressions are allowed.
- Use metaphors only when extremely appropriate and necessary; remove most unnecessary metaphors.
- Use passive voice less frequently.
Rhetoric and Terminology:
- Avoid model-speak: Reduce discourse markers and metadiscourse, such as "To put it bluntly...", "In short...", "Don't...", "The real..."
- Reasonably reduce the use of "de" (的) and "le" (了). Use and omit pronouns and their referents appropriately.
- Minimize the use of parenthetical gloss. Reduce glossing overall.
Sentence Structure and Logic:
- Reduce English written tone: Reduce multi-clause connected narratives.
- Reduce academic essay style: Avoid perfect logical reasoning chains; you can make slight jumps, or even state the conclusion first and then supplement the logic.
Punctuation Usage:
- Do not use dashes, quotation marks, or colons stiffly. Prohibit the use of scare quotes.
- Use square quotation marks 「」 instead of curly quotation marks.
Titles and Formatting:
- Do not use two-part titles.
- Avoid listicle style.
- Avoid overusing markdown formatting: Use headings and lists with extreme caution; prefer smooth paragraph expressions.
Workflow
When the user instructs to revise the writing style to remove the "model-like tone", follow this iterative process:
- Restate the above "Basic Guidelines" to inform the user of the upcoming edits, and wait for the user's confirmation of the edits. Write to a new file instead of the original one.
- After editing, ask the user whether to search for and remove model-specific stereotypical terms found on the current internet. If yes, perform a web_search to find these stereotypical terms, show the user which of these expressions are included in the article. After user confirmation, based on the output file from step 1, replace them with more natural forms. Write to another new file instead of the original one.
Only one round of search is needed to understand the currently popular model-specific stereotypical expressions. There is no need to search repeatedly for replacement suggestions for every detail.