cloudflare-deploy

The cloudflare-deploy skill comes from the curated skills collection in OpenAI’s official repository. According to the provider, it helps a Codex agent deploy, host, publish, or set up applications and infrastructure on Cloudflare. The entry is an instruction for an agent, not a Cloudflare account, not a hosting product, and not a standalone deployment API. The primary source covers Workers, Pages, and additional platform services, and provides a decision framework so a request can first be mapped to an appropriate Cloudflare product.

Purpose and product selection

The source groups common projects by their technical intent. Serverless code at the edge belongs to Workers, while a full web application with Git-based deployments belongs to Pages. Stateful coordination may call for Durable Objects, long-running multi-step jobs for Workflows, and container execution for Containers. For scheduled tasks, lightweight HTTP logic, execution-event processing, and improved latency to backend infrastructure, the skill identifies further Cloudflare services. It also separates storage and AI paths, including KV for key-value data, D1 for relational SQLite databases, R2 for objects, Queues for asynchronous messaging, Vectorize for vector search, and Workers AI, Agents SDK, and AI Gateway for different AI scenarios. This mapping is a decision aid, not an automatic architecture approval.

Preparation and operating workflow

Before a real deployment, the agent must verify the Wrangler environment’s authentication, according to the provider. An unauthenticated state should be reported clearly and resolved through the intended interactive or managed path. Credentials, tokens, and environment values must not be placed in skill text, logs, or version control. The agent then reviews the project structure, target product, existing configuration, bindings, and build output. It should distinguish the user’s requested context: a preview, a production deployment, or configuration guidance only. Timeouts need to fit builds and uploads because deployment can take several minutes.

Boundaries and security review

The skill cannot repair an application that is technically or functionally broken, and a successful deployment does not prove that it behaves correctly. A Cloudflare project can change external state, store data, enable network access, or serve content publicly. The target account, project, environment, branch, domains, bindings, and publication scope should therefore be checked before a production action. Destructive changes, deletions, permission changes, and migrations require an explicit decision and a credible recovery path. The decision trees do not replace Cloudflare documentation or architecture, privacy, and security review.

Data governance and professional context

Local inspection by an agent does not automatically make data confidential. The selected agent and its model provider may process project files and output, while Cloudflare services may process requests, logs, state, or stored objects depending on the service and configuration. Secrets belong only in the relevant protected configuration. Before upload, source maps, test data, private files, local configuration, and build artifacts should be checked for accidental disclosure. The primary source is the OpenAI repository, whose directory LICENSE.txt identifies Apache-2.0. OpenAI documentation is authoritative for the skill concept, and Cloudflare documentation is authoritative for current platform behavior. The skill is useful for structured planning and guided deployments, but it does not replace human approval, operations, monitoring, rollback, or ongoing review of provider terms. Current conditions should be checked directly with the provider; this entry states no concrete prices.