wrangler-skilld

name	wrangler-skilld
description	ALWAYS use when writing code importing "wrangler". Consult for debugging, best practices, or modifying wrangler, workers-sdk, workers sdk.
metadata	{"version":"4.107.0","generated_by":"Anthropic · Haiku 4.5","generated_at":"2026-07-02T00:00:00.000Z"}

cloudflare/workers-sdk wrangler@4.107.0

Tags: wrangler@2.2.4: 2.2.4, legacy: 3.114.17, latest: 4.107.0

References: package.json • README • Docs • Issues • Discussions • Releases

Search

Use skilld search "query" -p wrangler instead of grepping .skilld/ directories. Run skilld search --guide -p wrangler for full syntax, filters, and operators.

API Changes

This section documents version-specific API changes in wrangler v4.x — prioritizing recent minor releases where APIs changed in ways that older training data may not reflect.

Detailed Changes

• NEW: createTestHarness() — v4.99.0 adds integration testing for Workers. Runs Workers in a local preview environment using production build output. Supports multiple workers, request triggering, scheduled events, test server reset, and structured log capture via getLogs() and debug(). source

• NEW: defineWorker() and defineWranglerConfig() from wrangler/experimental-config — v4.100.0 introduces experimental TypeScript config authoring. When --x-new-config flag is passed, wrangler dev, wrangler build, wrangler deploy, wrangler versions upload, and wrangler versions deploy load from cloudflare.config.ts (required, worker runtime config via defineWorker) and optional wrangler.config.ts (tooling config via defineWranglerConfig). Per-environment config via ctx.mode branching. This is experimental and may change. source

• BREAKING: web_search binding renamed to websearch — v4.98.0. Wrangler config key, binding-type string, and miniflare option key all move from web_search/webSearch to websearch. Update "web_search": { "binding": "WEBSEARCH" } to "websearch": { "binding": "WEBSEARCH" }. Runtime type WebSearch on env.WEBSEARCH unchanged. source

• NEW: agent_memory bindings — v4.96.0 allows Workers to connect to Cloudflare Agent Memory service for storing/retrieving agent conversation state. Remote-only binding. Configure with binding name and namespace; wrangler auto-provisions namespace on deploy. Adds agent-memory:write OAuth scope to login. Supports type generation via wrangler types. source

• NEW: wrangler agent-memory namespace commands — v4.96.0 adds create, list, get, delete subcommands for managing Agent Memory namespaces. Supports --json output. source

• NEW: wrangler websearch search command — v4.96.0 CLI for running ad-hoc Web Search queries. Syntax: wrangler websearch search "query" with optional --limit <N> (default 10, max 20) and --json for raw response. source

• BREAKING: pipeline field renamed to stream in pipeline bindings — v4.96.0. Old pipeline field still accepted but deprecated with warning. Update config from { "binding": "MY_PIPELINE", "pipeline": "my-stream-name" } to { "binding": "MY_PIPELINE", "stream": "my-stream-name" }. source

• BREAKING: schedule property renamed to schedules on Workflow bindings — v4.95.0. Configuration-only rename. Scheduled triggering of Workflow instances not yet available. source

• NEW: migrations_pattern field in D1 database bindings — v4.98.0. Optional glob (relative to wrangler config) allowing nested migration file layouts (e.g. ORM-generated migrations/0000_init/migration.sql). Defaults to ${migrations_dir}/*.sql. wrangler d1 migrations create now errors if generated filename would not match pattern. When no migrations match but drizzle-style layout exists, wrangler logs hint. source

• NEW: --version-tag support on wrangler versions deploy — v4.99.0. Deploy version by tag (e.g. commit SHA) instead of Version ID: wrangler versions deploy --version-tag <sha>@100%. Tag resolved against worker's deployable versions. Supports traffic splitting across multiple --version-tag values. source

• NEW: --path positional argument (generic) for wrangler deploy and wrangler versions upload — v4.98.0. Replaces [script] positional. Accepts file (Worker entry-point) or directory (static assets). Type auto-detected. Example: wrangler deploy ./public for static site. --script named option now hidden/deprecated (still works for files only). source

• DEPRECATED: --script named option on wrangler deploy and wrangler versions upload — v4.98.0. Use positional --path instead. Continues working for backwards compatibility (file paths only). Passing directory to --script now errors with suggestion to use --path or --assets. source

• NEW: getPlatformProxy() passes through workflow bindings with script_name — v4.98.0. Workflows with script_name now work in split-process setups (SvelteKit/Remix). Miniflare reroutes engine's USER_WORKFLOW binding through dev-registry-proxy. Workflows without script_name still stripped (internal workflow engine cannot run in empty proxy worker). source

• BREAKING: experimental.testMode option removed from unstable_dev() — v4.96.0. Previously only affected default logLevel (was "warn" when testMode: true). Callers needing quiet logs should now set logLevel: "warn" directly. Default logLevel now matches wrangler dev ("log"). source

• NEW: wrangler artifacts commands — v4.92.0 adds CLI for Artifacts control-plane workflows (previously API-only). Includes list/inspect namespaces, create/list/inspect/delete repos, issue repo-scoped tokens for git auth. Human-readable and --json output. source

• NEW: --keep-vars flag on wrangler versions upload — v4.92.0. Matches existing wrangler deploy behavior. Preserves dashboard-configured environment variables instead of deleting them before upload. source

• NEW: --from-step-name, --from-step-count, --from-step-type options for wrangler workflows instances restart — v4.97.0. Restart Workflow instance from specific step. Works for remote instances and local wrangler dev --local sessions. --from-step-count and --from-step-type optional for disambiguation. source

• NEW: R2 public bucket simulation in wrangler dev — v4.100.0. Local R2 bucket objects now publicly accessible under /cdn-cgi/local/r2/public/<bucket-id>/<key> on dev server. <bucket-id> is bucket's bucket_name if set, otherwise binding. Bindings with remote: true excluded. source

• NEW: --x-new-config flag (experimental) — v4.100.0. Gates upcoming TypeScript config support. When enabled, loads Worker config from cloudflare.config.ts / wrangler.config.ts instead of wrangler.json/wrangler.jsonc/wrangler.toml. See defineWorker()/defineWranglerConfig() above. Format and exports may change. source

Also changed: cf-wrangler dev experimental delegate entrypoint (v4.100.0) · browser type generation now uses BrowserRun instead of generic Fetcher (v4.96.0) · wrangler types --check fixed for multi-worker setups with secondary -c configs (v4.99.0) · Workflow schedules property renamed from schedule (v4.95.0, config-only) · --skip-confirmation flag for wrangler containers images delete (v4.96.0) · D1 error messages improved for d1 execute, d1 export, d1 time-travel restore, d1 insights (v4.100.0) · Cloud authentication error messages now explain failure reason + list fixes (v4.100.0) · R2 Sippy error messages improved with specific --flag guidance (v4.100.0) · Pages CLI error messages improved with actionable guidance (v4.95.0) · OAuth login now shows actual error instead of hanging on provider rejection (v4.98.0)

Best Practices

• Use wrangler types after changing bindings — do not manually define your Env interface. Run this command whenever you add or rename a binding to generate type definitions that match your configuration exactly, catching mismatches at compile time instead of deploy time source

• Configure environments deliberately in wrangler.jsonc — each environment (production, staging, etc.) deploys as a separate Worker and does not inherit bindings from the root. Declare bindings per environment and always deploy with --env to avoid accidentally deploying to the root Worker source

• Store secrets with wrangler secret put, never in config or source — secrets must be managed through the CLI. Use .env files for local development (added to .gitignore), and use wrangler secret put API_KEY in production source

• Keep your compatibility_date current — set it to today's date on new projects and periodically update existing ones. This unlocks new APIs and bug fixes without changing your code source

• Enable nodejs_compat compatibility flag — this gives Workers access to Node.js built-in modules like node:crypto, node:buffer, and node:stream. Many libraries depend on these modules; enabling the flag avoids cryptic import errors source

• Stream request and response bodies instead of buffering — use response.body directly and TransformStream for piping, never await response.text() or await request.arrayBuffer(). Workers have a 128 MB memory limit; buffering large payloads causes crashes source

• Use ctx.waitUntil() for background work after the response — it lets you perform work after responding to the client, such as analytics or cache writes. Do not destructure ctx (it loses the this binding); call ctx.waitUntil() directly source

• Use bindings for Cloudflare services, not REST APIs — bindings like R2, KV, D1, Queues are direct in-process references with no network hop, no authentication, and no extra latency. Calling the REST API from within a Worker wastes time and adds complexity source

• Use service bindings for Worker-to-Worker communication instead of HTTP requests — service bindings are zero-cost, bypass the public internet, and support type-safe RPC. Access them via env.SERVICE_NAME from a configured binding source

• Use Hyperdrive when connecting to remote PostgreSQL or MySQL databases — it maintains a regional connection pool that eliminates the per-request cost of TCP handshake, TLS negotiation, and connection setup. Create a new Client on each request; Hyperdrive manages the underlying pool source

• Enable observability (logs and traces) in production — add observability.enabled: true and configure head_sampling_rate in wrangler.jsonc to start collecting structured logs and traces. Use console.log with JSON objects, console.error for errors, and console.warn for warnings source

• Never store request-scoped state in global scope — Workers reuse isolates across requests, so a variable set during one request is still present during the next, causing cross-request data leaks. Pass state through function arguments or store it on env bindings source

• Always await or ctx.waitUntil() Promises — floating promises cause silent bugs and dropped results. Enable the no-floating-promises lint rule (@typescript-eslint/no-floating-promises or oxlint typescript/no-floating-promises) to catch these at development time source

• Test with @cloudflare/vitest-pool-workers to access real bindings — this test environment runs tests inside the Workers runtime, giving access to KV, R2, D1, and Durable Objects. Note: the Vitest pool automatically injects nodejs_compat, so confirm your wrangler.jsonc includes the flag if your code depends on Node.js modules source