Post

I built a Claude skill for clearer technical writing. It enforces concrete rules and audits its output for sentence length, structure, word choice, and clarity. Validated with evals at work, now adopted by our team for specs, PRs, and code reviews. github.com/yamafaktory/...

github.comclaude/skills/clear-writing at main · yamafaktory/claudeContribute to yamafaktory/claude development by creating an account on GitHub.

24/10/26, 3:56 pm0 reposts0 likes

dukeemnukeem.bsky.social

@dukeemnukeem.bsky.social

I want the eval numbers behind "validated." Which rule did the model break most before the skill loaded: the 20-word cap, the no-semicolon rule, or buried verbs? The skill is one markdown file with no script, so the model counts its own words. How often does a 23-word sentence slip past?