Cloudflare Platform Builder
Helps agents research and build applications on Cloudflare.
The skill directs agents to retrieve authoritative information and includes safer patterns such as parameterized SQL, secret variables, production gateway authentication, and client-side human confirmation. It lacks a unified least-privilege boundary, pre-action confirmation policy, complete data-flow disclosure, rollback guidance, and systematic sensitive-data handling; its coverage includes email, MCP, AI, databases, and deployment with consequential external effects, so points are deducted.
Decision trees, reading orders, and troubleshooting notes make the happy path understandable and acknowledge possible staleness. Nothing was executed, and no committed test suite reproduces key paths; several examples leave lifecycle async behavior, dependency versions, configuration formats, and API availability unverified, while abnormal-input handling and failure feedback are limited. The static cap and these uncertainties require deductions.
The skill covers Workers, storage, AI, networking, security, and infrastructure as code, with decision trees that support semantic invocation. The trigger “any Cloudflare development” is overly broad, non-fit scenarios and input/output contracts are not explicit, and there is no Chinese-language or mainland-China reachability guidance. Core retrieval may depend on overseas services, so points are deducted.
Front matter, retrieval-first guidance, decision trees, product indexes, reading orders, installation instructions, limitation notes, and an Apache-2.0 license provide a usable structure. There is no selected-version policy, changelog, named maintenance responsibility, update path, or systematic troubleshooting coverage; the product index also exceeds the reference material visibly supplied, so points are deducted.
The decision trees, snippets, and Agents SDK/AI Gateway patterns can directly assist with common development tasks, and the core value proposition is clear. The skill is unusually broad, many products are only indexed or delegated to external retrieval, and the supplied examples lack file-based CI or test evidence. Users must still verify versions, configuration, and API details, limiting direct effectiveness.
The material points to Cloudflare documentation, workers-types, Wrangler schema, and changelogs, and some references contain concrete configurations and code that can be audited. It provides no committed key-path tests, CI coverage for the skill content, or independent corroboration; static review cannot establish that examples run, so the static cap and evidence gaps require deductions.
- Do not treat numeric limits, API signatures, configuration fields, or dependency names in the references as verified; check current official documentation before deployment.
- For AI, MCP, email, database, gateway, deletion, or deployment actions, define permissions, sensitive-data flows, cost impact, user confirmation, and rollback procedures.
- Users in mainland China should separately verify network reachability of Cloudflare documentation, retrieval tools, and dependent services.
What does this skill do, and when should you use it?
This is a consolidated Agent Skill for Cloudflare platform development, covering Workers, Pages, KV, D1, R2, Workers AI, Vectorize, the Agents SDK, networking, security, and infrastructure as code. Decision trees and a product index help agents route a request to the relevant reference area. The skill requires fresh retrieval before citing specific numbers, API signatures, or configuration options. It is a strong fit for Cloudflare-focused development work, but it does not provide a static guarantee of current prices, limits, or versions.
Routes compute, storage, AI, networking, security, media, analytics, and IaC requests to Cloudflare product reference directories; instructs the agent to retrieve information from Cloudflare documentation, Workers types, the Wrangler configuration schema, and product changelogs; prioritizes current documentation when reference files and official docs disagree.
- A developer is choosing runtime and deployment references for a Cloudflare Workers or Pages application.
- A developer is deciding between KV, D1, R2, Queues, and Durable Objects for a data requirement.
- An AI engineer needs to distinguish between Workers AI, Vectorize, the Agents SDK, and AI Gateway.
- A platform engineer needs guidance for managing Cloudflare infrastructure with Terraform, Pulumi, or the REST API.
- A security or networking engineer needs to identify the relevant WAF, DDoS, Tunnel, or Spectrum guidance.
What are this skill's strengths and limitations?
- Covers many Cloudflare product categories with need-based decision trees.
- Explicitly prioritizes current documentation over potentially stale model knowledge.
- Identifies documentation, Workers types, Wrangler schema, and changelogs as retrieval sources.
- Follows the Agent Skills standard, with README installation paths for Claude Code, Cursor, OpenCode, Codex, and Pi.
- Depends on external documentation, packages, configuration files, and changelogs, so offline use is limited.
- Its broad coverage does not make the single SKILL.md a complete specialist guide for every product.
- The source provides no test suite, version pinning, or guarantee of current prices and limits.
- Retrieval and local reference inspection require suitable network, filesystem, and tool access.
How do you install this skill?
The repository is a collection containing 11 separate skills. Install the collection with npx skills add https://github.com/cloudflare/skills, or clone the repository and copy skills/cloudflare/ into the directory used by your agent. The README lists ~/.codex/skills/ for Codex and ~/.claude/skills/ for Claude Code. The source does not document a dedicated command for installing only this skill.
How do you use this skill?
After installing skills/cloudflare/ in an Agent Skills-compatible client, invoke it for Cloudflare development requests, for example: “Choose the right data store for my Cloudflare Workers application, and retrieve the latest documentation before confirming limits or configuration options.”