readabilitylisted

Check whether a draft is actually readable by locating where a reader falls off — weak paragraph junctions, back-references with no antecedent nearby, and terms used before they are explained — then dispatch a fresh-context second reader to catch contradictions and repetition that no script can see. Use this whenever the user asks whether something is readable, clear, hard to follow, confusing, well-structured or "does this flow"; whenever they ask for a readability check or score, a Flesch or grade-level reading, or a second opinion; and as the last review pass before publishing a blog post, README, ADR, runbook or long technical explanation. Also trigger on "where does this lose people", "is this too dense", "cohesion", "flow check". Reports locations and never a grade level — if the user explicitly wants a Flesch or Gunning Fog number, use this skill and explain why it is not computed. For AI texture, voice and register use clear-and-human instead; for a CV use cv-and-human.
kevin-burns/claude-skills · ★ 8 · AI & Automation · score 78

Install: claude install-skill kevin-burns/claude-skills

# Readability Two things a draft can be wrong about, and only one of them is visible to a script. **A script can find where the argument jumps.** Two paragraphs that share no vocabulary, a paragraph opening on "This" with the referent stranded in the previous one, a term used forty lines before it is explained. Those have locations, and `cohesion_report.py` names them. **Only a reader can find a contradiction.** On 2026-09-01 a second reader found this author's draft asserting that a German edition "sidesteps the whole thing" four paragraphs after describing a German tripwire built for that same problem. Vale, Harper and every formula missed it, because none of them read for meaning. That is why this skill is a script **and** a reader, and why the reader is not optional. **Run the reader again after a revision, rather than treating the last pass as a clearance.** On 2026-09-02 a second reader on that same draft — by then through `check.sh`, a full register pass and the earlier reader — found four more things, including a contradiction and a numeric ambiguity in the one footnote whose job was to make the evidence believable. Findings do not converge on a fixed list. One reader is a sample, not an audit. **No grade level, ever.** Not because grade levels are unfashionable, but because Redish (2000) reports that whether the formulas are valid for technical material read by adults is unknown, and because the grade-level criterion means 50% of children at that grade answered