chore: add a docs-style skill with RomM's house writing rules

[gantoine]

Adds .claude/skills/docs-style/SKILL.md, a companion to the vendored humanizer that holds RomM's own house style. It captures the edits made by hand on #168, plus earlier ones:

1. Don't restate context the page, heading or list intro already gives ("Email notification channels, which users add to get their notifications by email")
2. Don't use "RomM" as a filler subject ("RomM seeds two groups" → "Two groups are seeded"). Keep it where another actor is involved.
3. Don't cram clauses ("which an upgrade from a release before permission groups also uses")
4. Merge choppy sentence runs into one sentence, but keep separate points apart
5. Cross-references go in "(see [link])" parentheticals
6. No em dashes or semicolons in prose
7. Describe current behavior. History belongs in the upgrade guide.
8. No UI walkthroughs in using/, and no provider UI prose in install/
9. Keep each fact on the page that owns it (defaults, image contents, password rules)
10. Short front-matter descriptions and short warnings
11. Terms and facts: Valkey, what the slim image runs, and checking claims against the source

Each rule has before/after examples from real PRs. The skill also ends with three grep checks over the diff. The skill is model-invocable, so it loads whenever docs prose is written or reviewed, not only at PR time.

humanize-docs now applies docs-style after the humanizer. Where the two disagree, docs-style wins, for example on the passive voice from rule 2.

.gitignore now uses the same .claude rules as rommapp/romm. Skills, hooks and settings.json are tracked, and everything else under .claude/ (such as settings.local.json and worktrees/) stays local.

🤖 Generated with Claude Code