Cloudflare has launched a CI software development kit that allows developers to write continuous integration pipelines using TypeScript instead of YAML configuration files, executing every stage as a durable Cloudflare Workflow. The package, called @cloudflare/ci, was published under Apache 2.0 license in early August 2026, according to a report by InfoQ. The SDK represents a shift from declarative configuration to programmable pipelines, trading portability for resumable execution that can survive failures without starting over.
The toolkit runs on Cloudflare Workers runtime rather than Node.js and publishes TypeScript source code for Workers-aware bundlers like Wrangler. The current version pulls repositories from Cloudflare Artifacts, which remains in private beta. A functioning pipeline requires bindings to Workflows, Sandboxes, Containers and Durable Objects, plus R2 storage when caching is turned on. Commands execute inside isolated Sandbox containers, which provide the filesystem snapshots that caching relies on. Pipeline stages run independently and therefore execute concurrently unless developers specify dependencies—wrapping separate checks in a Promise.all() ensures that lint, test, typecheck and build operations all complete before deployment begins.
The report describes two mechanisms designed to address common CI frustrations. Dependency caching saves the outcome of an install stage as a Sandbox filesystem snapshot in an R2 bucket, letting subsequent stages reuse it rather than reinstalling packages. A new events field in the Wrangler configuration then launches a Workflow directly when a cf.artifacts.repo.pushed event fires, replacing earlier subscription, queue and consumer wiring. "In essence, a CI/CD pipeline is just a Workflow," the announcement states. The repository highlights two production constraints: runner commands operate inside retryable Workflow steps, meaning any command with external side effects must be idempotent or a retry will duplicate it, and the SDK returns raw command output in CiRunnerResult.logs without redacting secrets.
The architectural advantage stems from Workflows' checkpointed execution model: each stage maps to a Workflow step that inherits state-preserving retries, so a failed step can restart with context intact rather than forcing the entire pipeline to rerun from scratch. Cloudflare published self-healing as a standalone example that wraps the pipeline in error handling and calls an application-owned agent when a runner fails—the agent runs a Think harness over Workers AI, using Moonshot's Kimi coding model to propose a patch, commit it to a branch and leave the original run marked as failed until an engineer merges the fix. The report notes that Cloudflare isn't the first to migrate pipelines into a general-purpose language: Dagger offers SDKs for eight languages and runs pipelines in containers on any OCI-compatible system, caching operations by content address so the same pipeline executes on a laptop or CI server. Cloudflare makes the opposite exchange, binding execution to Workflows and Sandboxes in return for resumable state.
Cloudflare says it plans to add triggers from any version control system, deployment and preview primitives, and monorepo support. For teams already using Cloudflare's stack, the SDK eliminates glue code and delivers step-level observability. For everyone else, the transferable concepts are durable retries per step and a repair agent that stops short of merging. The SDK positions CI as platform infrastructure rather than a standalone service—teams gain programmability and fault tolerance, but commit to the Cloudflare ecosystem to get it. Declarative configuration remains simpler to inspect, diff and govern with policy, while a general-purpose language purchases expressiveness at the expense of that transparency.