Official AI Agent Skills Are Here — Install Stripe and Cloudflare Best Practices in One File

You integrate Stripe. You read the docs. You miss edge cases. Claude Code helps — but it’s working from training data, not from Stripe’s internal engineering playbook. The recommendations are reasonable but generic. You know there’s a gap between “correct” and “what Stripe’s own team would do.”

That gap has a fix now. Companies are shipping SKILL.md files — single Markdown documents that encode their engineering team’s best practices directly into your Claude Code context. Stripe has one. Cloudflare has nine. Anthropic has seventeen. And they’re all free.

What Are Official Skills?

A skill is a Markdown file with YAML frontmatter that Claude Code discovers automatically. Drop it in .claude/skills/ and it becomes part of every session’s context. The format is simple:

---
name: stripe-best-practices
description: Guides Stripe integration decisions — API selection, Connect setup, billing, webhooks.
---

## Integration routing

| Building...                  | Recommended API          |
| One-time payments            | Checkout Sessions        |
| Subscriptions                | Billing APIs + Checkout  |
| Connect platform             | Accounts v2              |
...

Claude Code reads this before answering your questions. Instead of guessing which Stripe API to use for subscriptions, it knows. Instead of generating a wrangler.toml config, it knows Cloudflare moved to wrangler.jsonc and that newer features are JSON-only.

The key insight: these aren’t tutorials. They’re decision trees. The Stripe skill routes you to the right API before you write a line of code. The Cloudflare Workers skill flags anti-patterns (like await response.text() on unbounded data) that would pass code review but fail at scale.

The Ecosystem

Official skills are not centralized. Each company maintains their own repo:

Repo	Owner	What’s Inside
anthropics/skills	Anthropic	17 skills: frontend-design, mcp-builder, skill-creator, pdf, docx, webapp-testing
cloudflare/skills	Cloudflare	9 skills: wrangler, workers-best-practices, web-perf, durable-objects, agents-sdk
stripe/ai	Stripe	3 skills: stripe-best-practices, upgrade-stripe, stripe-projects
VoltAgent/awesome-agent-skills	Community	Link directory to 1,060+ skills across the ecosystem

All use the same SKILL.md format. All update independently when the vendor’s best practices change.

The Three That Matter Most

If you’re building anything on Cloudflare or accepting payments through Stripe, these three skills give you the highest return for the least effort.

1. cloudflare/wrangler

What it covers: Every Wrangler CLI command, every config field, every binding shape — D1, KV, R2, Vectorize, Workers AI, Queues, Workflows, Containers, Pages.

Why it matters: Wrangler moves fast. Config fields change. New binding shapes appear. The skill tells Claude Code to prefer retrieval over pre-training for any Wrangler task — and gives it the exact URLs to fetch from. No more generating deprecated TOML configs or wrong CLI flags.

Key behavior it installs:

• Prefer wrangler.jsonc over TOML (newer features are JSON-only)
• Run wrangler types after config changes to generate TypeScript bindings
• Use environments (env.staging, env.production) for deployment stages
• Local dev defaults to local storage — use remote: true for bindings that need real resources

2. cloudflare/workers-best-practices

What it covers: Production patterns and anti-patterns for Cloudflare Workers code.

Why it matters: Workers have a 128 MB memory limit. They share isolates across requests. The patterns that work in a Node.js server can fail silently in a Worker. This skill encodes the rules that the Cloudflare team uses internally.

Anti-patterns it catches:

• await response.text() on unbounded data (memory exhaustion)
• Module-level mutable variables (cross-request data leaks)
• Math.random() for security tokens (not cryptographically secure)
• Destructuring ctx (breaks this binding, throws “Illegal invocation”)
• Hand-written Env interfaces (drift from actual wrangler config)
• ctx.passThroughOnException() as error handling (hides bugs)

3. stripe/stripe-best-practices

What it covers: API selection, Connect platform setup, billing, Treasury, and integration surfaces.

Why it matters: Stripe has multiple ways to accept payments. Checkout Sessions, PaymentIntents, the Payment Element — the right choice depends on your use case, and picking wrong means a rewrite later. This skill routes Claude to the right API before it writes code.

Decision routing it installs:

Building…	Recommended API
One-time payments	Checkout Sessions
Custom payment form	Checkout Sessions + Payment Element
Saving a payment method	Setup Intents
Marketplace / Connect	Accounts v2 (/v2/core/accounts)
Subscriptions	Billing APIs + Checkout Sessions

It also pins the latest Stripe API version (2026-03-25.dahlia as of this writing) so Claude generates code against the current API, not a training-data snapshot.

How to Install

Three commands, five minutes. No package manager, no build step.

Step 1: Clone the vendor repos

mkdir -p resources
cd resources
git clone https://github.com/cloudflare/skills cloudflare-skills
git clone https://github.com/stripe/ai stripe-ai

Step 2: Symlink the skills

mkdir -p .claude/skills
cd .claude/skills
ln -s ../../resources/cloudflare-skills/skills/wrangler/SKILL.md cloudflare-wrangler.md
ln -s ../../resources/cloudflare-skills/skills/workers-best-practices/SKILL.md cloudflare-workers.md
ln -s ../../resources/stripe-ai/skills/stripe-best-practices/SKILL.md stripe-best-practices.md

Step 3: Restart Claude Code

Skills are loaded at session start. Start a new session from the directory where .claude/skills/ lives. That’s it.

Updating Later

cd resources/cloudflare-skills && git pull
cd ../stripe-ai && git pull

Symlinks pick up changes automatically. Restart Claude Code to load the updated content.

Why Symlinks Instead of Copies

You could copy SKILL.md files directly into .claude/skills/. But symlinks give you one thing copies don’t: git pull updates propagate instantly. No sync scripts, no version tracking, no forgetting to update. The vendor pushes a change, you pull, you restart Claude Code. Done.

The tradeoff: if you delete or move the cloned repo, the symlinks break. Keep your vendor repos in a stable location (we use a gitignored resources/ directory) and this is a non-issue.

What Not to Install

Not every skill is relevant to every stack. Install what you actually use:

• durable-objects — Only if you’re building stateful Workers
• agents-sdk — Cloudflare’s agent runtime, not Claude’s. Monitor, don’t install.
• upgrade-stripe — Useful during SDK version bumps. Install when you need it, remove after.
• anthropics/frontend-design — Good for general frontend work. If you have your own design system standards, yours are more specific.

The VoltAgent/awesome-agent-skills directory lists 1,060+ skills from the community. Most are links, not installable files. Browse for discovery, but install from the vendor repos directly.

The Pattern

The shift here isn’t about any single skill. It’s about where best practices live.

Before: best practices lived in documentation you read once and forgot. They lived in blog posts that went stale. They lived in the heads of senior engineers who reviewed your PRs.

Now: best practices live in files that load into your AI coding assistant’s context on every session. The vendor updates them. You pull. The knowledge is always current, always present, always applied.

Three files. Fifteen minutes. Permanently better code.