Ptlimited

Writing Craft — Testing skill for Claude Code

Testing community

A blind tested skill for better creative writing.

How to install Writing Craft

This entry records only its repository, not the path inside it, so there is no exact command to give. Open Ptlimited/writing-craft and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Writing Craft does

A blind tested skill for better creative writing. Nine signals three AI judges named after scoring 50 samples blind, then measured in a second blind test. Two files, works in Claude Code, Codex, Gemini or any chat.

Alternatives in Testing

Agent Skills Spec — A skill is a folder of instructions, scripts, and resources that agents can discover and load dynamically to p 1.9k ★
Web Design — A Claude Code SKILL for designing beautiful, consistent web pages — spec first, code second 686 ★
Craft Analyze — Post-cycle analysis — QA, UX, Creative, and Style audits using MCP browser tools 52 ★

README

Most writing skills are someone's opinion about good writing.
These nine came out of a blind test of 50 samples, scored by 3 AI judges who had no idea who wrote what. Then they were measured in a second blind test.

Works in Claude Code, Codex, Gemini, or pasted into any chat.
The skill itself is two files: SKILL.md and check.py.

Install · The nine rules · Does it work · Where they came from

▶ Watch the blind test on YouTube

Five models were given the same five everyday writing jobs, twice each. 50 outputs. Filenames scrambled, punctuation stripped, the answer key locked until every score was in. Three judges then scored all fifty blind and were each asked one open question, in their own words, with no vocabulary supplied: **what separated the high scorers from the low ones?**

Nine signals came back. All three judges named the first four independently, without seeing each other's answers.

The skill turns those nine signals into rules the model writes to. **Load it, ask for the copy you wanted anyway, and the draft comes back better.** The model writes to the nine, checks its own work against the three that are measurable, fixes what missed, and only then hands it to you.

You don't run anything. You don't paste a draft in for marking. You ask for a post and get a better post.

Does it actually work

The same blind test was run on the skill itself. Sonnet 5 with it and without it, judged in one sitting by GPT-5.6 Sol, same five prompts, twice each.