agentleFS
mblaubergmblauberg/provenant/skills/natural-writing/SKILL.md

natural-writing

Skill1 starsChanged 6 days agoMITView on GitHub

Use to write, rewrite, tighten or diagnose emails and general, mixed, correspondence or academic/thesis prose without losing facts, citations or numbers; owns Australian English and the AI-prose mechanisms. Not for READMEs, commits or docs (engineering-writing) or Australian legal drafting (legal-writing).

15 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 12 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Don't fabricate experience, sources or detail, or add fake messiness.

    line 16
  2. Very specific:

    Use the punctuation the sentence needs, without em-dashes.

    line 20
  3. Distinctive:

    Never style-pass gates, ADR decisions or process runbooks.

    line 23

“Common” is measured: kept by at least 6 in 10 of the 12 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Natural writing

The family's writing hub, owning prose doctrine: fidelity locking, the mechanism catalogue, Australian English, claim discipline and academic prose. engineering-writing and legal-writing add a domain overlay and link here.

Hard rules

  • Facts don't move. Style passes are where they drift, so lock them (step 1) and audit against the lock (step 4).
  • Don't fabricate experience, sources or detail, or add fake messiness. Flag each gap ([FLAG: cite source]).
  • Edit for the reader, never a detector score. Formal prose isn't proof of AI authorship.
  • Use the punctuation the sentence needs, without em-dashes. Run provenant check writing when available: it flags em-dashes and lexical hints, its silence is necessary but not sufficient, and its word rules are advisory.
  • Never style-pass gates, ADR decisions or process runbooks.
  • A supplied voice sample beats these defaults.

Target

A direct claim, a plain verb, an anchor from the source (number, name, path, date, honest negative), one calibrated qualifier, and an ending on the decision, risk or next action. The default voice is plain and mildly casual (voice.md); a formal register is the exception.

Procedure

Pick the least invasive pass: light-touch (repair named defects), rewrite (default) or diagnose (findings only, no edit). If the draft already reads well, edit lightly and say so.

  1. Lock facts, numbers, names, quotes, citation keys, defined terms, modals and negations. List them first; they're the fidelity contract.
  2. Diagnose by mechanism, not phrase. Read mechanisms.md and name the two or three that cost the reader most. A swapped synonym leaves the mechanism, so the tell moves to the next word.
  3. Rewrite once, by mechanism, in voice.md's register. Restate each claim directly and add only anchors the source holds.
  4. Audit independently. Compare the locked list against the result and revert any sentence that lost or strengthened a claim.
    • High-stakes prose: a separate clean-context reviewer sees only the before and after.
    • Number-heavy text: script the comparison (claim-discipline.md).
    • Long documents: split reviewers by lens (redundancy, register) and triage their findings yourself.
  5. Report the final text first, with notes only for material flags.

References

  • voice.md: load before any write or rewrite. The default register, rules, example and formal exceptions.
  • mechanisms.md: the defects, each with a reason and a before/after pair.
  • tells-2026.md: read before judging a draft's age. What current models still do and what has faded, dated and sourced.
  • au-english.md: house style unless a repository, quote, API or form requires otherwise.
  • condense-pass.md: shortening, with a stop rule.
  • claim-discipline.md: observed, inferred, planned and pending claims; comparatives that carry their number.
  • academic-prose.md: thesis and paper register, citation-key preservation, evidence altitude. LaTeX invariants live in engineering-writing.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.