Skill README Generation

Turn any existing skill's SKILL.md into a bilingual, human-readable pair — English README.md plus Chinese README.zh.md — ready for the internal skill catalog page.

When to use it

Document a new skill:

I just wrapped up a new skill and its SKILL.md is written. Now I want to list it on the internal catalog, so I ask this skill to hand-craft the paired README on the spot.

Add missing READMEs:

own-skills/ still holds a batch of older skills that never got a README. I want every missing one filled in with the same spec — one independent pair per directory, never merged.

Replace an old README:

Some skill's old README was hand-written way back and its shape no longer matches the current spec. I say "regenerate to the new bilingual shape and overwrite" and both files get rewritten.

Prepare for publishing:

I'm about to publish a skill to the catalog detail page. That page reads the example prompts from the README's top matter and renders them as scene tabs — no README, nothing to show.

Follow another README:

I point at a peer skill that already has a good README and say "give this one the same treatment." The skill mirrors that shape against my target's own SKILL.md.

Won't take:

Editing one paragraph of an existing README — plain editing is enough; building a SKILL.md or a whole skill from scratch → skill-create-workflow; naming a skill (display_name / display_name_zh) → skill-name-generation; picking the English directory slug → skill-domain-framing; deciding whether a piece of content deserves to become a skill → skill-content-fit; the target's SKILL.md is too thin and needs a full review → skill-improve-workflow.

What it produces

One call writes two files — English README.md and Chinese README.zh.md, matched in structure but never word-for-word translated.

• Writes: <skill dir>/README.md (English body) plus <skill dir>/README.zh.md (Chinese body)
• Top matter: a plain-language description plus no more than 3 natural prompt_examples, each covering a distinct trigger scene
• No language switcher: the H1 starts immediately after the top matter
• Four sections: opening one-liner, when to use it, what it produces, prerequisites & boundaries — each version 30-80 lines
• Plain-language pass: the Chinese version is scanned line by line, insider slang gets rewritten in plain Chinese so a non-author can still read it
• Terminal report: both output paths, line counts, prompt_examples count, covered scenes, and any "not sure about" note
• Never touches: the target's SKILL.md or any other file in the repo

Prerequisites & boundaries

Prerequisites:

Target is a skill directory containing SKILL.md with complete top matter and a description. No external dependencies — no network, no API calls, no subagent spawns.

Neighbor skills:

Action	Route to
Build a skill from scratch (SKILL.md skeleton included)	skill-create-workflow
Name a skill (display_name / display_name_zh)	skill-name-generation
Pick the English directory slug (kebab-case)	skill-domain-framing
Decide whether a piece of content deserves a skill	skill-content-fit
SKILL.md too thin — needs a full quality review	skill-improve-workflow

Won't take:

• Editing one section of an existing README — plain editing is enough
• Writing a README for a non-skill project or general docs
• CI checks on existing README top-matter compliance — that's a script's job

Fine-grained boundaries:

• Target already has README.md or README.zh.md and the user didn't say "regenerate / overwrite" → ask once, don't clobber
• Target SKILL.md too thin to yield four solid sections → stop and route to skill-improve-workflow first, rather than pad with hollow text
• Target skill has no display_name → still generate the README, and flag the routing to skill-name-generation in the report