@mikefluff/skills

skills

A small, opinionated collection of Claude Code skills for editing prose without producing text that reads like LLM output. Russian-first, English-capable.

Seventeen skills, one base linter + twelve wrappers + three linters + one meta-skill. Plain markdown, MIT-licensed, no required external deps.

Install

# Curl (5 seconds, no deps)
curl -fsSL https://raw.githubusercontent.com/Mikefluff/skills/main/install.sh | bash

# npm
npm install -g @mikefluff/skills && skills install

# Homebrew (after tap)
brew tap mikefluff/tap https://github.com/Mikefluff/homebrew-tap
brew install mikefluff/tap/skills && skills install

# Docker (for CI — no install into ~/.claude needed)
docker run --rm -v "$PWD:/work" ghcr.io/mikefluff/skills lint /work/draft.md

Skills appear after Claude Code session restart. Discovery is automatic via name: and description: in each skill's frontmatter — no ~/.claude/settings.json edits required.

For all install options + troubleshooting, see docs/INSTALL.md.

First time? Start here

→ User Guide — pick your scenario, walk through it end-to-end.

→ Skill Index — all 17 skills indexed by layer, domain, and language.

→ Composing recipes — 14 named workflows showing how to chain skills.

Quick scenario picker:

You want to …	Walkthrough
Write a viral social-media post	RU · EN
Edit a fiction chapter	fiction-chapter
Draft a long-form essay	non-fiction
Verify a multilingual translation	translation-parity
Auto-lint every commit	pre-commit-hook
Audit a chapter against your story bible	canon-check-audit
Insert a Pelevin-vector digression	digression-insertion
Run a read-only quality gate	style-check-gate
Rewrite text in a different register	USER-GUIDE
Draft a cold outreach email	USER-GUIDE
Generate an AI image prompt (Midjourney/DALL-E/Flux)	USER-GUIDE
Generate an AI video prompt (Kling/Veo/Sora/Runway)	USER-GUIDE
Write UX microcopy (errors, empty states, tooltips, buttons)	USER-GUIDE
Write release notes / changelogs	USER-GUIDE
Write an RFC / ADR / design doc	USER-GUIDE
Write marketing copy (landing / SEO / ads)	USER-GUIDE

If something looks wrong: FAQ · Troubleshooting.

What's in the box

Skill	Layer	Languages	Purpose
writer	base	ru/en	Base clean-prose editor — antinyeyroslop (28 categories), typography, structural synthetics, RU calques. Invoked by all other prose skills.
viral-text	wrapper	ru/en	Write viral social media content — hooks, numbered points, micro-conclusion with NLP question, CTA. 41 viral content rules + platform adaptation.
prose-edit	wrapper	ru	Fiction rewrite layer — Pelevin/Manson voice vector, 10-item style drift checklist, no meta-refs / anglicisms in narrator voice, long artistic rewrite (no comma-stitching), ToV pattern, 5-trigger structural-synthesis detector, Postirony depth-pass.
essay-write	wrapper	ru	Non-fiction layer — long subordinate sentences (Manson style), source-backed claims, philosophy through humor, biography through scenes, plain-Russian for complex content.
style-check	linter	ru/en	Read-only pre-commit lint that stacks writer + prose-edit + essay-write rules. Routes by configurable path patterns (fiction vs non-fic), BLOCKING/WARNING/INFO severity, exit-code semantics for git hook.
translation-sync	linter	ru/en/pt-br	Read-only pre-commit parity checker for trilingual book translations (RU↔EN↔PT-BR) — typography per language, terminology canon, anchor-quote drift, names/patronymics/diminutives, cultural realia footnotes, no-smoothing of numbers/brands/dates. BLOCKING/WARNING/INFO severity with exit-code semantics for git hook.
canon-check	linter	ru/en	Story-bible consistency auditor for any book series. Greps entities (characters / artifacts / locations) in changed chapters, cross-references the project's story-bible document, flags BLOCKING contradictions / WARNING gaps / INFO new details. Read-only — trust the text, not memory.
pelevin-digression	wrapper	ru	Write a Pelevin-style digression for a fiction or non-fiction passage — 12 structural techniques + 5 banned constructions. Wraps prose-edit (fiction) or essay-write (non-fic). Invoked by request, not auto-applied.
skills-update	meta	en/ru	User-invocable update check + apply for this collection. Compares local install marker with latest GitHub release, shows CHANGELOG diff, asks for confirmation, runs install.sh --update.
tone-shifter	wrapper	en/ru	Rewrite a passage in a different register (formal↔casual, business↔academic, technical↔friendly, plain-explainer) without changing meaning. 6 registers + named transformation deltas. Wraps writer as final cleanup.
cold-email	wrapper	en/ru	Write or rewrite cold outreach emails (first-touch, follow-up, intro request, re-engage). 5-block structure, ≤120-word budget, banned ceremony patterns, anti-template subject lines. Wraps writer as final cleanup.
image-prompt	wrapper	en/ru	Write prompts for AI image generators (Midjourney, DALL-E, Flux, Nano Banana, Stable Diffusion). 6-part formula (subject + setting + style + lighting + camera + texture), per-model deltas, negative prompts.
video-prompt	wrapper	en/ru	Write prompts for AI video generators (Kling, Veo, Sora, Runway, Pika, Hailuo, Luma). CHARACTER FIRST law, beat structure (Beat 1/2/3), exact camera vocabulary, pacing modes (narrative/action/comedy/documentary/timelapse).
microcopy	wrapper	en/ru	Write UX microcopy — error messages, empty states, tooltips, button labels, helper text, modals, 404/500 pages, onboarding. Plain language, action-oriented, never blame user, length budgets per element type. Wraps writer for final cleanup.
release-notes	wrapper	en/ru	Write user-facing release notes + changelogs. Keep-a-Changelog format, sections Added/Changed/Fixed/Deprecated/Removed/Security. Per-audience tone (user/dev/ops). Anti-marketing-fluff bans. Wraps writer.
rfc-writer	wrapper	en/ru	Write engineer-facing design docs — RFCs, ADRs, Tech Specs, Design Docs. Structure: context/problem/proposal/alternatives/consequences/decision. RFC 2119 (MUST/SHOULD/MAY). Review checklist for spotting weak alternatives sections.
landing-copy	wrapper	en/ru	Write marketing copy — landing page sections (hero/features/pricing/FAQ), SEO meta (title+description+OG+Twitter), ad copy (Google/Facebook/LinkedIn/X). Julian Shapiro hero formula, char limits per platform. Wraps writer.

Skills compose: wrappers call writer internally; linters reference the same rule files but don't mutate. See docs/COMPOSING.md for the dependency graph.

Updates

Three ways, increasing in eagerness:

• On demand: invoke /skills-update inside Claude Code.
• Ambient status-line banner (opt-in): bash scripts/install-hook.sh. Shows · skills v1.0.1→1.2.0 +1 skill when an update exists.
• CLI: bash install.sh --check / bash install.sh --update [--prune].

The banner / /skills-update never updates without explicit user confirmation.

Common install flags

# install a subset
curl -fsSL .../install.sh | bash -s -- --skills writer,viral-text

# install to a custom prefix
curl -fsSL .../install.sh | bash -s -- --prefix /tmp/skills

# pin a specific version
curl -fsSL .../install.sh | bash -s -- --version 1.0.1

# re-install (overwrite existing skills)
bash install.sh --update

# check what's installed vs what's available
bash install.sh --check

# uninstall everything
bash install.sh --uninstall

Full installer help: bash install.sh --help.

Repo layout

skills/
├── README.md                # this file
├── VERSION                  # semver, single source of truth
├── CHANGELOG.md             # Keep-a-Changelog
├── skills.json              # machine-readable manifest used by installer
├── install.sh               # pure-bash installer, curl-pipeable
├── Makefile                 # local dev convenience
├── CONTRIBUTING.md          # how to add a skill / report a bug / propose new one
├── docs/
│   ├── USER-GUIDE.md        # ← start here as a user
│   ├── walkthroughs/        # detailed per-scenario flows
│   ├── FAQ.md
│   ├── TROUBLESHOOTING.md
│   ├── COMPOSING.md         # dependency graph + composition patterns
│   ├── VERSIONING.md        # semver policy + release flow
│   └── LINTER-COVERAGE.md   # auto-generated regex coverage
├── scripts/
│   ├── validate.sh          # frontmatter + cross-link + description-quality check
│   ├── check-docs-consistency.sh  # skills.json ↔ README ↔ USER-GUIDE ↔ walkthroughs
│   ├── gen-skills-table.py  # regenerate the README skills table from skills.json
│   ├── smoke.sh             # validate + writer linter regression + fixture snapshots
│   ├── coverage.py          # regenerate docs/LINTER-COVERAGE.md
│   ├── bump.sh              # bump VERSION + promote [Unreleased] CHANGELOG section
│   ├── new-skill.sh         # bootstrap a new skill folder
│   ├── decide-bump.sh       # parse conventional commits since last tag
│   ├── lint-description.py  # frontmatter description quality (advisory)
│   └── install-hook.sh      # idempotent status-line banner installer
├── hooks/
│   └── skills-update-banner.js
├── tests/                   # fixture snapshots for writer/scripts/lint.py
├── .github/                 # workflows + issue/PR templates + SECURITY.md
└── <skill-name>/            # the 17 skills, one folder each

Local development

make help                       # list all targets
make install                    # install from this checkout to ~/.claude/skills/
make smoke                      # validate + linter regression + fixture snapshots
make check-docs                 # docs-consistency gate
make gen-readme                 # regenerate skills table
make new-skill NAME=foo-bar DESC="..."

Releases are automatic — push a conventional-commit message (feat:, fix:, feat!:, etc.) and .github/workflows/release.yml bumps + tags + publishes. See docs/VERSIONING.md.

Want to contribute? Read CONTRIBUTING.md. PRs welcome.

License

MIT — see LICENSE.