Natural writing

Natural writing

Rewrite general prose in a natural, direct voice. Preserve meaning, facts, citations, numbers and intent while removing filler, canned framing, synthetic rhythm and chatbot residue.

Hub role

This skill is the writing family's hub: it owns the shared prose doctrine, the anti-AI taxonomy, condense pass, claim-discipline/evidence schema, voice matching, and the always-on Australian English default. academic-writing, engineering-writing and legal-writing link here instead of cloning it, keeping only their own domain specialization and linking back for the rest. Load au-english.md as the default house style for any prose without a project convention that overrides it.

Boundary

Use this skill for general, mixed or unclassified prose. Defer technical docs and READMEs to engineering-writing, scholarly work to academic-writing, and Australian legal drafting to legal-writing.

Do not fabricate experience, invent sources, add fake messiness or optimise for AI-detector scores. Formal prose is not proof of AI authorship.

Modes

• rewrite (default): clean the full draft.
• light-touch: preserve most wording and repair only clear defects.
• match-voice: follow a supplied writing sample.
• precision-preserving: protect high-stakes or citation-heavy wording.
• full-rewrite: reshape low-stakes prose when a normal pass is insufficient.
• diagnose: identify the strongest problems before rewriting.

Choose the least invasive mode that solves the problem. Combine any mode with match-voice when the user supplies a sample.

Workflow

1. Lock facts, logic, stance, citations, numbers, names, quotes and required terms.
2. Diagnose the real defect: weak information, assistant residue, generic or inflated language, repeated templates, unclear sentences or a voice mismatch.
3. Rewrite structure, emphasis and sentence rhythm rather than swapping synonyms. Use specifics already present in the source; never invent texture.
4. Match confidence to evidence. Narrow or flag an unsupported claim instead of strengthening it.
5. Repair remaining quality and voice mismatches, then compare the result against the locked invariants and any writing sample.

Load patterns.md for the sign catalogue, research, voice-matching rubric, genre guidance or a stubborn second pass. Load anti-ai-taxonomy.md for the tiered sweep, condense-pass.md for shortening without losing meaning, and claim-discipline.md for wording observed, inferred, limitation and pending claims.

Bright-line rules

• Preserve facts, meaning, citations, named entities and quoted terms.
• Delete chatbot framing and tool residue instead of paraphrasing them.
• Prefer concrete nouns and verbs; state each supported point once.
• Repair repeated templates without manufacturing quirks or forced variation.
• Keep one honest qualifier for weak evidence; do not hedge strong evidence.
• If the draft already reads naturally, edit lightly and say so.

Output

For a rewrite, return the final text first and add notes only for material flags or choices. For a diagnosis, lead with the strongest quality and voice risks. Keep inline flags short, such as [FLAG: cite source].