update-claude-md
Update CLAUDE.md
Philosophy: Less Is More
CLAUDE.md is not documentation. It's a small, surgical file that steers agents away from mistakes they'd otherwise make. Every line you add goes into context for every single token the agent generates — biasing its behavior, costing time, and competing for attention with the actual task.
A recent study benchmarking context files across multiple LLMs found that developer-written CLAUDE.md files only marginally improved task completion (+4% average), while auto-generated ones actually hurt performance (-3% average). Both increased cost by over 20%. The takeaway isn't that CLAUDE.md is useless — it's that most CLAUDE.md files contain the wrong things.
The core principle: If an agent can discover it by reading the codebase, it doesn't belong in CLAUDE.md. These models are RL-trained to explore codebases aggressively — they read package.json, grep for patterns, trace imports, check configs. They're genuinely good at this. Telling them what they'd find anyway just wastes context and creates staleness risk.
The Litmus Test
More from scalecommerce/skills
landing-page-guide
Build high-converting landing pages with proven conversion psychology and layout patterns (9% → 20%+ conversion rates). Use this skill whenever the user wants to create, build, or optimize a landing page, lead generation page, squeeze page, opt-in page, or any single-page conversion-focused experience. Also trigger when the user asks about landing page copywriting (headlines, CTAs, benefit statements), conversion rate optimization for a landing page, A/B testing landing page elements, landing page layout or structure advice, or improving an existing page's conversion rate. Covers the full workflow from strategy to code (Nuxt4 + NuxtUI) to deployment, including AI-generated visuals via OpenRouter. Even if the user doesn't say "landing page" explicitly — if they're describing a page whose primary goal is capturing leads, driving signups, or getting a single conversion action, this skill applies.
8code-review
Analyze codebases for quality issues, security vulnerabilities, duplicate code, inconsistencies, and bad patterns. Use this skill whenever the user asks to review code quality, find code smells, detect duplicates, check for anti-patterns, audit a codebase, clean up code, find inconsistencies, run a security audit, check for vulnerabilities, or do any kind of static analysis or code health check. Also trigger when the user says things like 'review my code', 'find problems in my codebase', 'what should I refactor', 'is this code any good', 'check for bad patterns', 'find duplicate code', 'code audit', 'tech debt review', 'security review', 'check for vulnerabilities', 'is this code secure', or 'find security issues'. This skill should be used even for casual requests like 'take a look at my project and tell me what's wrong' or 'anything I should fix in here'. Works across all languages and frameworks.
7frontend-design
Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, artifacts, posters, or applications (examples include websites, landing pages, dashboards, React components, HTML/CSS layouts, or when styling/beautifying any web UI). Generates creative, polished code and UI design that avoids generic AI aesthetics.
6nano-banana
Generate images using the OpenRouter API with the Nano Banana model (Gemini 3.1 Flash Image). Use this skill whenever the user asks to generate, create, or make an image, picture, illustration, photo, or visual using OpenRouter or Nano Banana. Also trigger when the user wants AI image generation and has an OpenRouter API key available, even if they don't mention "Nano Banana" by name — any request like "make me a picture of...", "generate an image of...", or "create a visual of..." should use this skill. Also trigger for image editing requests like "remove the background", "change the color", "add X to this image".
5