claude_skills
grahac/claude_skills
Some Claude Code skills I have created or built upon
Instalação
npx convez-ai grahac/claude_skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
Claude Code Skills
Personal productivity skills for Claude Code — tools for development, meetings, writing, and security.
Installation
Download a single skill — grab any .skill bundle from the latest release (every skill is linked individually in the table below), then:
- claude.ai (web/desktop): upload via skill settings
- Cowork: drop the file into your Cowork skills directory
- Claude Code: unzip into
~/.claude/skills/<name>/
Install the whole collection in Claude Code:
# Via the plugin marketplace
/plugin marketplace add grahac/claude_skills
# Or clone into your .claude/skills directory
git clone https://github.com/grahac/claude_skills.git ~/.claude/skills/claude_skills
Skills
Every skill below links to its downloadable .skill bundle (always the newest release). Skill names with a detailed write-up further down link to that section.
Product & Marketing
| Skill | Download |
|---|---|
| hyperslide — polished single-file HTML slide decks | ⬇ hyperslide.skill |
| website-extractor — extract a site's content + design for a rewrite | ⬇ website-extractor.skill |
| marketing-copywriter — landing pages, emails, ads, messaging | ⬇ marketing-copywriter.skill |
| innovate — the single highest-leverage addition to a plan | ⬇ innovate.skill |
| nanobanana — photo-realistic image generation via Gemini | ⬇ nanobanana.skill |
| appstore-images — device-framed App Store preview screens | ⬇ appstore-images.skill |
Strategy
| Skill | Download |
|---|---|
| indispensable-need — find the high-stakes JTBD with no substitute | ⬇ indispensable-need.skill |
Development
| Skill | Download |
|---|---|
| elixir-simplifier — refactor Elixir/Phoenix/LiveView, remove duplication | ⬇ elixir-simplifier.skill |
| elixir-deps-check — audit Elixir dependencies | ⬇ elixir-deps-check.skill |
| prompt-cache-optimizer — fix prompt-caching anti-patterns | ⬇ prompt-cache-optimizer.skill |
| changelog — Keep a Changelog-style CHANGELOG.md from git history | ⬇ changelog.skill |
| pull-main — safely sync the local branch with main | ⬇ pull-main.skill |
| pr-recap — non-technical product summary of recent PRs/commits | ⬇ pr-recap.skill |
Productivity
| Skill | Download |
|---|---|
| granola-scoop — extract Granola meeting notes from the local cache | ⬇ granola-scoop.skill |
| suggest-skills — recommend which skill fits the task at hand | ⬇ suggest-skills.skill |
Writing
| Skill | Download |
|---|---|
| voiceprint — clone your writing voice as a per-medium skill | ⬇ voiceprint.skill |
Security
| Skill | Download |
|---|---|
| security-audit-skills — audit your Claude config (skills, plugins, hooks) | ⬇ security-audit-skills.skill |
Contracts
| Skill | Download |
|---|---|
| contract-manager — multi-agent contract create/review/edit | ⬇ contract-manager.skill |
Cowork & Setup
| Skill | Download |
|---|---|
| cowork — load Cowork working principles / onboarding | ⬇ cowork.skill |
| plugin-doctor — diagnose and fix plugin/marketplace install problems | ⬇ plugin-doctor.skill |
Full file map: See SKILLS-MAP.md for the complete directory structure of every skill.
Product & Marketing
hyperslide
Generates polished, self-contained HTML slide deck presentations — no PowerPoint, no external tools.
Use when:
- Creating a presentation from a description, code, or problem statement
- Building pitch decks, architecture overviews, research presentations
- Turning any content into a visual slide deck
Key principles:
- Asks about content and visual style (URL or mood), then derives the palette and fonts
- Produces a single
.htmlfile with scroll-snap navigation, keyboard controls, and a slide counter - Uses only inline CSS/JS — no external dependencies
- Reviews the output in a browser and fixes visual issues before delivering
Invoke: /hyperslide
website-extractor
Extracts all content and design information from a website to enable a complete rewrite.
Use when:
- Rewriting or redesigning an existing website
- Creating a new version of a site based on the original
- Auditing a site's copy, structure, and brand identity before rebuilding
Key principles:
- Captures page content, brand identity (colors, fonts, tone), navigation, section-by-section copy, and CTAs
- Extracts colors from code (Tailwind classes, CSS variables), not screenshots
- Saves a structured
site-extraction.mddocument ready to pass to a rewrite agent
Invoke: /website-extractor
marketing-copywriter
Creates compelling marketing copy for landing pages, emails, ads, and product messaging.
Use when:
- Writing landing page copy
- Crafting email campaigns
- Creating ad copy
- Developing product messaging and value propositions
- Improving existing marketing copy
Key principles:
- Clarity over cleverness
- Audience-first messaging
- Specific outcomes over vague promises
- Avoids AI-sounding patterns (no "elevate", "seamless", "unlock your potential")
Invoke: /marketing-copywriter
innovate
Identifies the single most innovative, high-leverage addition to any plan or proposal.
Use when:
- Asking "what am I missing?" or "what would you add?"
- Wanting one unexpected but compelling improvement to a plan, spec, or strategy
- Looking for the boldest move before committing to an approach
Key principles:
- Picks one thing — not a list
- Must be non-obvious, concrete, and accretive (not just incrementally better)
- Presents the idea with a clear argument for why it's the highest-leverage move available
Invoke: /innovate
nanobanana
Generates professional photo-realistic images using structured JSON prompts and Google's Gemini image API.
Use when:
- Creating product shots, lifestyle images, ad creative, or hero images
- Generating consistent brand imagery with full control over lighting, camera, and composition
- Iterating on image concepts with structured prompt refinement
Key principles:
- Transforms plain descriptions into structured JSON prompts (scene, camera, lighting, style)
- Camera details and specific lighting setups produce dramatically better results than vague adjectives
- Iterates on one axis at a time — never changes everything at once
- Uses
gemini-2.5-flash-imagefor speed,gemini-3-pro-image-previewfor text and maximum quality
Invoke: /nanobanana
appstore-images
Generates polished App Store preview screens for iPhone, iPad, and Mac — composites screenshots onto colored backgrounds with realistic device mockups (dynamic island, side buttons, MacBook chrome) and bold headline typography.
Use when:
- Preparing App Store / Mac App Store listing assets
- Creating marketing screenshots with consistent device framing across a screen set
- Generating multi-device sets (iPhone + iPad + Mac) from raw simulator captures
Key principles:
- Bundles 7 font presets (montserrat default, oswald, bebas, anton, archivo, monasans, fredoka) and 20+ background colors plus custom hex/gradient support
- Renders realistic device frames matching Apple's required canvas sizes (1290×2796 iPhone, 2048×2732 iPad, 2880×1800 Mac)
- Vary the headline per screen to tell a story across a set; keep gradient and font consistent
Invoke: /appstore-images
Strategy
indispensable-need
Identifies a product's indispensable need — the high-stakes job-to-be-done that has no good substitute.
Use when:
- Stress-testing whether a product is irreplaceable or merely nice-to-have
- Mapping competitive substitutes and where they fall short
- Surfacing the functional, emotional, and social dimensions of why someone hires a product
Key principles:
- Multiple parallel agents probe functional, emotional, and social JTBD dimensions
- Combines competitive research with deep codebase reading to ground claims in what the product actually does
- Outputs a sharp statement of the indispensable need, not a feature list
Invoke: /indispensable-need
Development
elixir-simplifier
Simplifies and refines Elixir/Phoenix/LiveView code with a focus on removing duplicate code.
Use when:
- Finding and removing duplicate code across modules
- Extracting repeated patterns into shared components
- Reviewing recently written Elixir/Phoenix/LiveView code
- Refactoring existing code for clarity
- Checking if code follows Phoenix patterns
Key principles:
- DRY - Remove duplicate code (primary focus)
- KISS - Keep it simple
- LiveView over JavaScript
- Phoenix context patterns (no Repo in
_webmodules) - HEEx
{}syntax over<%= %> - No hardcoded color hex codes
- No defensive fallbacks - let it crash
Invoke: /elixir-simplifier
prompt-cache-optimizer
Audits hooks and AI/LLM API calls for prompt caching anti-patterns, then produces concrete fixes to maximize cache hit rate and reduce API costs.
Use when:
- Reviewing Claude Code hooks for cache-breaking patterns
- Auditing project AI/LLM calls for suboptimal prompt structure
- Reducing AI API costs through better caching
- Diagnosing low prompt cache hit rates
Key principles:
- Scans
~/.claude/settings.jsonand project settings for hooks - Classifies injected data as globally static / project static / session static / frequently dynamic
- Makes guesses about dynamic vs static data, then confirms with you before generating fixes
- Produces before/after code examples for every issue found
- Never changes tools mid-session; never silently assumes
Invoke: /prompt-cache-optimizer
changelog
Creates and manages CHANGELOG.md following the Keep a Changelog 1.1.0 spec.
Use when:
- Adding recent changes to the changelog
- Preparing a release and versioning the changelog
- Creating a CHANGELOG.md from scratch
Key principles:
- Reads git history to extract what changed — never invents entries
- Organizes changes into Added, Changed, Deprecated, Removed, Fixed, Security
- Supports
[Unreleased]workflow and cutting versioned releases with comparison links
Invoke: /changelog
Productivity
granola-scoop
Extracts meeting notes from Granola's local cache — no API needed.
Use when:
- Extracting recent meeting notes ("extract my granola meetings")
- Reviewing what meetings you had ("what meetings did I have this week?")
- Searching meeting content for people or topics
Output: Markdown files saved to ~/.granola-scoop/output/ with meeting metadata, your notes, and AI summaries.
Invoke: /granola-scoop
Writing
voiceprint
Clone your writing voice as an installable Claude skill, one per medium (email, LinkedIn, longform content). Analyzes your real sent emails, LinkedIn posts, or longform content — then packages your voice as a skill that auto-applies whenever Claude drafts in your name.
Source: skills/voiceprint/
Install:
Copy skills/voiceprint/ into ~/.claude/skills/voiceprint/, then /voiceprint becomes available as a slash command.
Command:
| Command | What it does |
|---|---|
/voiceprint | Build a new voiceprint OR refine an existing one — asks create-or-refine and which medium up front |
Key features:
- Corpus-first: analyzes 50 real sent emails (not a questionnaire) — or LinkedIn export / blog URLs
- Supports multiple email accounts merged into one voiceprint (personal + work)
- Quantified sentence metrics (burstiness score, avg length, typical range) as calibration targets
- Verbatim voice exemplars embedded in the output — ground truth over rules alone
- Before/after sample transformations showing generic AI vs your voice
- Output is an installable Claude skill (
myvoiceprint-<medium>) written directly to~/.claude/skills/myvoiceprint-<medium>/— active the moment it's written, no copy step - Refine mode for surgical updates to an installed voiceprint
Security
security-audit-skills
Performs security audits of your Claude Code configuration — skills, plugins, hooks, and permission allowlists.
Use when:
- Periodically reviewing your Claude setup for security issues
- After installing new plugins or skills
- Before sharing your configuration with others
Key principles:
- Nothing is silently skipped — ignored items are shown in a "Previously Reviewed" section
- Flags risky permission patterns (wildcard bash, unrestricted network access)
- Identifies third-party plugins that need review
- Checks for hooks with broad access
Invoke: /security-audit-skills
Contracts
contract-manager
Multi-agent contract management — creates, reviews, and edits contracts using three specialized Claude subagents that independently analyze, cross-review, and debate disagreements.
Use when:
- Creating a new contract (NDA, consulting/freelance engagement, or general)
- Reviewing an existing contract for risks, clarity, and business viability
- Editing an existing contract with specific changes
Key principles:
- Three independent Claude agents (The Shield, The Plain Speaker, The Deal Maker) work in parallel
- Cross-review with AGREE/ELEVATE/CHALLENGE protocol for conflict resolution
- Checklist verification against required clauses by contract type
- Produces professional
.docxoutput
Invoke: /contract-manager
Skill Folder Structure
Every skill follows a consistent structure. See SKILLS-MAP.md for the full directory listing.
skill-name/
SKILL.md # When to use, core instructions, file map
gotchas.md # Failure patterns — update when things go wrong
references/ # Detailed examples, checklists (read on demand)
assets/ # Templates, starter files, reusable formats
scripts/ # Executable scripts
Contributing
PRs welcome! Add new skills in the skills/ directory following the existing format.
License
MIT