How skills actually work
Two ways to trigger anything in Claude Code — type it, or just describe what you want.
⌨️ Type a slash-command
Press / and a menu appears. Pick a skill (e.g. /code-review), optionally add arguments, and it runs in your current session.
🧠 Just say what you want
Most skills auto-trigger on intent. You never type the name — you describe the outcome and Claude loads the right skill behind the scenes.
📌 Memory & CLAUDE.md
Claude keeps a file-based memory and reads a CLAUDE.md in your repo for standing instructions — so preferences and project rules persist across sessions.
🛡️ Plan mode & permissions
Ask Claude to plan before touching files, and approve the plan first. Permission prompts gate risky actions until you allow them.
Core built-in commands
These ship with Claude Code itself — the everyday workhorses.
📄 /init
Scans your codebase and writes a CLAUDE.md documenting structure, commands and conventions so every future session starts smart.
🔍 /code-review
Reviews your current diff for real bugs plus reuse/simplification cleanups. Effort scales from low to ultra (a deep multi-agent cloud review).
🔒 /security-review
Security audit of the pending changes on your branch — injection, auth, secrets, unsafe input handling.
✅ /review
Reviews a full pull request end-to-end, not just the local diff.
🧹 /simplify
Quality-only pass over changed code — reuse, simplification, efficiency, altitude — then applies the fixes. It won’t hunt for bugs (that’s code-review’s job).
▶️ /run
Launches and drives your actual app so you can see a change working — CLI, server, browser app, whatever the project is.
🧪 /verify
Confirms a change really does what it should by running the app and observing behaviour — not just trusting the tests.
📚 /learn
Captures and manages project “learnings” — durable notes Claude reuses so the same lesson isn’t re-learned every week.
⚙️ /update-config
Configures the harness via settings.json — permissions, env vars, and hooks (automated “whenever X, do Y” behaviours the harness runs for you).
⌨️ /keybindings-help
Customise keyboard shortcuts, rebind keys, add chord bindings.
🔕 /fewer-permission-prompts
Scans your history for safe, repeated commands and builds an allowlist so Claude stops asking permission for the harmless ones.
📘 claude-api
An always-accurate reference for the Claude API / Anthropic SDK — model IDs, pricing, tool-use, caching, token counting. Auto-loads whenever you build with Claude.
Ship & release
From “it works on my machine” to a clean PR, deployed and watched. (gstack suite.)
🚀 /ship
The full release ritual: detect & merge the base branch, run tests, review the diff, bump VERSION, update CHANGELOG, commit, push, open a PR.
🌐 /land-and-deploy
Lands the change and runs your deploy flow in one move. Pair with /setup-deploy once to configure targets.
🐤 /canary
Post-deploy canary monitoring — watches the freshly shipped build for errors before you trust it.
📊 /health & /retro
/health is a code-quality dashboard; /retro runs a weekly engineering retrospective across what changed.
🐞 /investigate
Systematic debugging that drives to a real root cause instead of patching symptoms.
📐 /spec
Turns vague intent into a precise, executable spec in five phases — before a line of code is written.
Browser, QA & the web
Claude can drive a real headless browser — test flows, scrape data, dogfood your own site.
🧭 /browse
A fast headless browser for QA and dogfooding — Claude opens pages, clicks, reads, and reports what a real user would see.
🧪 /qa & /qa-only
/qa systematically tests a web app and fixes the bugs it finds; /qa-only reports without touching code.
🕷️ /scrape & /skillify
/scrape pulls structured data off any page; /skillify then freezes that flow into a permanent reusable browser-skill on disk.
🔗 /connect-chrome
Launches an AI-controlled Chromium with the sidebar extension built in, so Claude shares a live browser with you.
Design
From a blank brief to a finished, on-brand interface — with a designer’s eye on the details.
🎨 /design-consultation
Understands your product, researches the landscape, and proposes a full design system — aesthetic, type, colour, layout, spacing, motion — with font + colour previews.
👁️ /design-review
A designer’s-eye QA pass: finds spacing issues, broken hierarchy, “AI-slop” patterns and slow interactions — then fixes them.
🧱 /design-html
Generates production-quality, hand-written HTML/CSS — not framework mush.
🔀 /design-shotgun
Generates several design variants at once, opens a comparison board, collects structured feedback, and iterates.
Research, docs & developer experience
Deep answers, written documentation, and a critical look at how your tooling feels to use.
🔬 /deep-research
Fan-out web research: searches many sources, fetches them, adversarially fact-checks the claims, and writes a cited report. Ask it something specific.
📝 /document-generate
Writes missing documentation from scratch for a feature, module, or the whole project.
📦 /document-release
Post-ship docs update — keeps the manual in step with what you just shipped.
🧰 /devex-review
A live audit of your developer experience — setup friction, confusing scripts, slow loops — with fixes.
📄 /make-pdf
Turns any markdown file into a publication-quality PDF.
🔎 /seo-aeo-geo
Audits and fixes a site for classic SEO, AEO (answer-engine / featured-snippet optimisation) and GEO (getting cited by ChatGPT, Perplexity, Gemini, Claude). Runs a report → fix → verify loop.
Automation, scheduling & orchestration
The part most people miss — Claude can run on a clock, in the cloud, and fan out into a fleet of agents.
🔁 /loop
Runs a prompt or slash-command on a recurring interval inside the session — or lets Claude self-pace until a condition is met.
📆 /schedule
Creates cloud routines — cron-scheduled agents that run on their own, even when your laptop is closed. List, edit, or fire them on demand.
🕸️ Multi-agent workflows
For big jobs Claude can orchestrate dozens of subagents deterministically — fan out, verify adversarially, then synthesise. Opt in by asking for a “workflow” or using ultracode.
🧩 Hooks (via /update-config)
True “whenever X, automatically do Y” behaviour — the harness, not Claude, runs these, so they fire every time without asking.
Media & video generation
Image, video, voice, decks and marketing creative — generated and auto-published. (Higgsfield · Seedance · Veo · Nano Banana · Gamma.)
🖼️ higgsfield-generate
The core image/video engine — text-to-image, image-to-video, edits, references, and a Marketing Studio for ads with avatars and products.
🧑 higgsfield-soul-id
Trains a “Soul” model on a real face once, then reuses it for identity-faithful images and video — your digital twin across every generation.
📸 higgsfield-product-photoshoot
Brand-quality product imagery — studio shots, lifestyle scenes, hero banners, ad creative packs, virtual try-on.
🛒 higgsfield-marketplace-cards
Compliant marketplace listing images — main image, secondary shots, A+ content modules ready to upload.
🍌 nanobanana
Fast 4K image generation & editing (Gemini “Nano Banana”) with subject consistency and search grounding.
🎬 Veo
Google Veo video — generate clips, animate a still image, extend a clip, or apply a style.
📊 Gamma
AI-built presentations, documents, webpages and social posts from a prompt.
🎥 Seedance video styles (01–15)
Fifteen specialist video-prompt skills for Seedance 2.0, each tuned to a look or use-case. You never name them — describe the vibe and the right one loads.
Subagents — Claude’s teammates
Specialised agents Claude can dispatch to work in parallel, each with its own tools and focus.
🔭 Explore
A read-only search agent for sweeping many files fast — it locates code and reports the conclusion without dumping the whole repo on you.
🗺️ Plan
A software-architect agent that designs an implementation plan — steps, critical files, trade-offs — before any code is touched.
📖 claude-code-guide
An agent that answers “can Claude…/how do I…” questions about Claude Code, the Agent SDK and the Claude API.
⚡ general-purpose
The all-rounder for multi-step research and searches when you’re not sure where the answer lives.
Custom skills I’ve built
Skills aren’t just the defaults — you can author your own. These run my businesses on autopilot.
🥘 ecj-uae-brand
A locked brand system for EatCookJoy UAE — logo, colours, type, pricing and guardrails. Any branded output (deck, newsletter, social post) stays perfectly on-brand automatically.
📹 ecj-uae-ugc-reels
End-to-end branded UGC reels — image → video → captioned, voiced, branded 12-second reel saved to Downloads. Scheduled daily to the brand’s channels.
🔧 Want one for your business?
A custom skill turns a repeatable job — your reporting, your creative, your outreach — into one command (or a scheduled agent) that runs the exact same way every time.