How Nullshot Turned AI into a Cloudflare Engineer

We gave AI agents real platform skills by adding 13 production-tested Cloudflare skills—so the code it generates deploys on the first try.

The Team

Share

How Nullshot Turned AI into a Cloudflare Engineer

How Nullshot Turned AI into a Cloudflare Engineer

The difference between junior and senior devs? Skills. We gave AI agents those skills.

After joining the Cloudflare Startup Program, we forked their official Skills repository and added 13 production-tested skills covering the entire platform stack. At Nullshot, your AI doesn't guess. The code it generates deploys on the first try.

What Agent Skills Actually Are

Originally, Anthropic introduced Agent Skills to make AI models good at specific things. It's pluggable expertise — folders with Markdown files containing instructions, templates, and patterns from real experience.

The AI sees short descriptions at startup. When your request matches a skill's domain, it pulls in the full instructions. This cuts hallucinations because the AI follows actual instructions instead of guessing. You can update skills anytime to stay current.

The Coverage Problem

Cloudflare released 6–7 base skills. We tried building real apps and hit walls immediately.

No dedicated skill for D1, KV, R2, Workers AI, Queues, Workflows, Vectorize, Hyperdrive, and a dozen other services we actually use. Without these, AI falls back on guesswork and won't architect proper production systems.

Our Contribution: 13 New Skills

We added new skills covering databases, AI inference, background processing, and framework deployment.

Each one captures actual usage patterns. We included the mistakes we made so others don't have to.

What Changes in Practice

Without these skills:

“Set up a Worker with D1” → code that forgets migrations, uses wrong APIs, generates invalid wrangler.toml.

With our skills:

“Detected \cloudflare-worker-base\ and \cloudflare-d1\ skills. Use them?” → deployment-ready code. Proper structure, correct configs, migrations included.

We estimate this saves 60% of context tokens and prevents hundreds of bugs you'd catch in testing. No more AI inventing functions or burning tokens “thinking out loud.”

Why This Matters Beyond Us

We open-sourced everything: `https://github.com/null-shot/cloudflare-skills`

Encode your platform knowledge once, use it everywhere. When Cloudflare ships new APIs, update the SKILL.md and every AI using it immediately knows. Way better than waiting for model retraining.

Anthropic built the mechanism, Cloudflare laid the groundwork, and we filled in the production reality. The combination turns AI from a supervised code generator into something that actually understands the platform.

Newsletter

Sign up for our newsletter

Check out our newest insights and keep up with our engaging content. Jump in now for some great perspectives!