A reusable Agent Skill for producing more natural, voice-aware writing without relying on deliberate mistakes or artificial randomness.
It focuses on voice matching, genre fidelity, sentence rhythm, specificity, and level-appropriate language.
- matches a demonstrated writing level and voice
- reduces repetitive, template-driven prose
- varies sentence and paragraph rhythm naturally
- prioritizes concrete details over generic filler
- preserves meaning, stance, tone, and genre
- supports school, university, professional, and casual writing
Clone the repository into a lowercase skill directory:
git clone https://github.com/lmkn-dev/Human-Writing.git human-writingThen place or reference the human-writing directory from your agent's skills location.
The skill follows the Agent Skills SKILL.md format and includes the required discovery metadata in YAML frontmatter.
Use the skill when generating or rewriting prose that should sound natural and match a specific writer, proficiency level, tone, or format.
Conceptually:
user prompt
+
human-writing/SKILL.md
↓
model
↓
output
No runtime dependencies are required. The skill is instruction-only.
The skill can be validated with the Agent Skills reference tooling:
skills-ref validate ./human-writingThe evaluation/ directory contains a small paired exploratory evaluation across five writing tasks.
The current dataset records:
- baseline and skill-assisted outputs
- one external classifier score per output
It does not include independent human ratings. The classifier results therefore should not be interpreted as proof of authorship or as evidence that naturalness, voice match, or genre fidelity improved.
See evaluation/REPORT.md for the recorded results and limitations.
├── README.md
├── SKILL.md
├── LICENSE
└── evaluation/
├── REPORT.md
├── prompts.jsonl
└── results.csv
MIT. See LICENSE.