You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The command library behind CogniSwitch's website — a ~50-page Next.js site run by one person with Claude Code, Gemini CLI, a headless CMS (Prismic), and about $20/month in tooling. No designer, agency, or page builder.
Each command covers one stage of the same pipeline a marketing team runs — brief, review, build, QA, ship. Work that would normally need a strategist, designer, QA lead, or DevOps engineer runs as a named, repeatable step instead.
Describe intent instead of hand-writing prompts. Say what you want; the command builds the structured prompt and you refine it. (/prompt, /idea, /content)
One command per stage. Chain them rather than asking for everything in a single shot. (/dev → /review-local → /deploy)
Explicit handoffs. Each transition records what was done, what to check, and what comes next. (/end-session, /review-handoff, /sandbox)
The pipeline
The stages map onto a standard marketing workflow:
Skills are different from commands. Commands are things you invoke (/deploy). Skills are knowledge that Claude loads automatically when relevant — writing guides, visual treatment patterns, design system rules.
Copy individual commands or skills into your project's .claude/ folder:
# A command
cp commands/deploy.md your-project/.claude/commands/deploy.md
# A skill
cp -r skills/copyreview/ your-project/.claude/skills/copyreview/
Most of these are CogniSwitch-specific (Prismic, our design system, our editorial pipeline). Yours will look different. What transfers is the approach: document your process, give each stage a name, and let the AI execute the parts you've already figured out.