Claude Code Skills

What's enforced

SecureStartKit ships with 4 Claude Code skills in .claude/skills/. They are markdown instruction files that Claude Code reads when you type the matching slash command in your terminal. Skills run against your actual codebase, read the brand frame, respect the topical map, and write back into the project. The skill source ships with the template; you own it and can edit it.

This is the "AI-agent-optimized" claim, made concrete: the agent reads YOUR brand frame, YOUR CLAUDE.md, YOUR topical map, and produces content that fits. No subscription, no SaaS layer, no per-token charge beyond your own Claude usage.

The 4 bundled skills

/write-blog, AI blog writer with brand-frame enforcement

Researches a keyword opportunity (or accepts a topic argument), writes a full MDX post with verified primary-source citations, internal links picked from the existing corpus, GEO-compliant structure (direct-answer opening, per-H2 atomic answers, TL;DR block), and the brand-voice pass. After publish, it updates docs/topical-map.md automatically.

claude
> /write-blog
# Researches keywords, presents 5 options, writes the chosen post

> /write-blog How to add a Supabase realtime subscription
# Writes about a specific topic

/comparison-page, head-to-head and concept-vs-concept comparisons

Builds /compare/[slug] pages that target commercial-intent comparison queries. Generates the page.tsx (with FAQPage + BreadcrumbList + ItemList + Organization JSON-LD), the client component (hero, feature table, strengths, decision helper, verdict, FAQ), and registers the route in the sitemap. Handles both product-vs-product comparisons (e.g., SecureStartKit vs ShipFast) and concept-vs-concept comparisons (e.g., one-time vs subscription billing).

claude
> /comparison-page
# Asks which competitor/concept to compare against, then builds the page

/brand-frame, brand strategy definition and audit

Walks you through the Controlled Exposure Formula to define your brand's frame, anti-frame, voice, and content split. Audits existing pages against the frame and outputs brand/<project>-brand-frame.md. Other skills (/write-blog, /comparison-page) read this file to enforce voice consistency, so getting this right early pays compounding interest across every future content piece.

claude
> /brand-frame

/weekly-free-tool, build keyword-targeted free tools

Reviews the project's existing /tools/* pages to avoid duplication, finds high-value keyword opportunities, and generates a complete tool page (working interactive UI, WebApplication JSON-LD, FAQ schema, "How to use" guide, "Last updated" timestamp). Free tools are SecureStartKit's strongest TOFU asset, and this skill is how the 13 bundled tools were built.

claude
> /weekly-free-tool

How skills work mechanically

Each skill is a directory under .claude/skills/ with a SKILL.md file. When you type a slash command, Claude Code reads the corresponding skill's instructions and follows them.

.claude/skills/
├── write-blog/
│   ├── SKILL.md
│   └── reference.md
├── comparison-page/
│   ├── SKILL.md
│   └── reference.md
├── brand-frame/
│   └── SKILL.md
└── weekly-free-tool/
    └── SKILL.md

Each SKILL.md contains:

• Description and trigger conditions (when to invoke the skill)
• Step-by-step instructions for Claude to follow
• Quality checks (verification rules, anti-pattern detection)
• Output format (file paths, frontmatter, structured data)
• Honest constraints (what the skill does NOT do, where it asks the user instead of auto-deciding)

Skills also load shared resources like brand/<project>-brand-frame.md, CLAUDE.md, and docs/topical-map.md to keep their output aligned with the project's strategy.

Creating custom skills

Add a directory under .claude/skills/ with a SKILL.md:

---
name: my-skill
description: What this skill does and when to use it.
allowed_tools:
  - Read
  - Write
  - Edit
  - Glob
  - Grep
---

# My Skill

Instructions for Claude to follow when this skill is invoked.

Skills are project-local (versioned in your git repo, not your Claude account), so they ship with the template and any teammate who clones the repo gets the same skills.

Requirements

• Claude Code CLI installed
• Any Claude plan that supports Claude Code
• Skills are included in the template source code; no additional setup needed

Why this matters

Most "AI-ready" SaaS templates ship a one-line README mention of Cursor compatibility. SecureStartKit ships the actual instruction files that AI agents follow, written for THIS codebase, with THIS brand frame, against THIS topical map. The skills are the difference between an agent guessing what your project wants and an agent reading what your project actually documents.

The same pattern is documented externally for AI coding agents in general at public/llms.txt.