Writing for the reader
AI-generated writing can sound polished while making the reader work too hard. Models can over-explain a simple point, choose obscure language over familiar words, and turn a small request into an elaborate framework. Repeated caveats, headings, and conclusions add length without helping someone understand the subject.
The same tendency can over-engineer the work around a document: unnecessary steps, approval rituals, or deliverables that the reader never needed. A fluent answer can still be a poor use of someone’s time.
I built humanize-content around what the reader needs to understand, decide, remember, or do. That gives the model a reason to select material and a way to judge whether each paragraph belongs.
What humanize-content changes
The skill asks the agent to establish the audience, purpose, and voice before drafting or editing. It considers what can be removed before adding more explanation or structure.
Its editing checks address concrete problems:
- Cut filler, repeated arguments, stale material, and sections that don’t help the reader.
- Use clear subjects and ordinary verbs. Keep technical terms when they add precision, and explain unfamiliar ones when needed.
- Remove unnecessary frameworks, procedures, and review steps while preserving the work the task actually requires.
- Preserve facts, qualifications, links, code, and the author’s meaning. Remove or qualify claims that go beyond the supplied evidence.
Concision serves understanding. A difficult idea may need more explanation; removing a necessary condition just to shorten a sentence would make the writing worse. The skill also leaves effective prose alone.
An illustrative edit
Before:
Prior to commencing implementation, establish a comprehensive validation framework to ensure robust operational readiness. The public API must remain unchanged.
After:
Before implementation, agree on the checks the change must pass. Keep the public API unchanged.
The revision makes the action easier to follow and keeps the API constraint. It drops the unsupported promise of operational readiness.
A voice that carries across tasks
On first use, the skill asks for a voice and saves it in TONE-OF-VOICE.md at the project root. Later runs reuse it for documentation, articles, messages, UI copy, and replies. A different voice requested for one task leaves the saved default intact unless the author asks to change it.
The repository also includes skills for finding duplicated or stale code and summarizing available YouTube captions. Its README explains how to install the bundle or choose individual skills for a supported agent. Automatic use depends on the host discovering the installed skill.