Six passes over a draft.
One of them is mine.

Each skill is a directory under skills/ holding a SKILL.md with YAML front matter: a name, a description that tells the agent when to reach for it, and the instructions that follow. Two skills carry a references/ file they read (the humanizer rubric, and the Microsoft term substitutions). Four carry evals/ cases. Nothing is compiled, and there is no service to run.

That shape is the whole design. The description is the trigger surface, so the skill loads when a draft is in front of the agent, and the body is the procedure it follows once loaded.

humanize

Edits a draft into Shane’s voice, then scores it against the vendored humanizer rubric. The voice pass, not a general clarity edit.

ms-style-pass

Applies Microsoft Writing Style Guide term preferences, bias-free language, and heading conventions. Runs after humanize.

plain-language

Applies ISO 24495-1:2023. Restructures a document for a reader’s task, which is a different job from editing its sentences.

ste100

Applies ASD-STE100 Simplified Technical English to procedures, descriptions, and safety text. Deliberately austere.

writing-review

Diagnoses a draft and ranks the findings under keep, revise, ask-author, and cut. Never produces a replacement draft.

writing-interview

Interviews the author before drafting and writes no prose until the answers arrive. Written for Shane, and not portable as it stands.

Choosing a pass covers when to reach for which one, and which of them need a voice profile before they do anything useful.

The source README states the rule plainly: Profiles/voice.md is authoritative for voice across every skill here, and where a skill’s defaults and that profile disagree, the profile wins. That file is not in the repository. It is a private note, and it is not published on this site.

So the reusable part is the procedure: the structural edits, the precedence rule that protects an author’s earned choices, the pass order, the review vocabulary that refuses to rewrite. The profile is the material. Bring your own, and humanize, ms-style-pass, and writing-review become yours. Run them without one and they will still edit, but the voice they converge on is Shane’s, not yours.

Two skills sit outside that rule on purpose. plain-language and ste100 apply a published standard to text that often is not the author’s personal voice at all, and both say so in their own instructions.

Version 0.5.0, dated 2026-10-19, MIT licensed, distributed through the slogsdon-claude-code-config marketplace. The changelog is public and worth reading before you rely on a pass: 0.3.0 removed an enrichment step that was inventing hedges, 0.4.0 added two register rules taken from hand edits, and 0.5.0 added the scoring gate.

Two diagnostics ship alongside the skills in scripts/, vendored byte-identical from addyosmani/clarity (MIT). They report cadence, hedge and booster density, anchor density, and stock-lexis hits. There is no composite score, by design. The upstream project found its own stylometric composite came out inverted, so a single number here would invite optimising toward the machine register.

This site publishes no voice measurements, no baseline figures, and no detector claim in either direction. Nothing here asserts that a pass makes text pass a detector, and the skills do not promise it.