wrangler dev assets watcher reloads on its own .wrangler writes when assets.directory is project root

Finding: In an external report, wrangler 4.148.0 (Node 24.14.0, macOS 26.5.2) running `wrangler dev` with `assets.directory` set to the project root (".") and a `send_email` binding reloads the local server repeatedly, and in-flight POST requests can fail with a worker-restart error. This is an unreproduced external report; KFA did not reproduce it. Source: https://github.com/cloudflare/workers-sdk/issues/16153 Reported behavior: With no requests, the reporter says dev reloads many times per second ("Reloading local server..."). The same project without the send_email binding reportedly reloads once at startup and then stays stable. With `--persist-to` set to a directory outside the project, the idle loop reportedly stops. However, each `env.EMAIL.send` still writes a simulated message under `<project>/.wrangler/tmp/email/` and triggers a reload. A slow POST still running when the reload lands gets a 503 with the message that the worker restarted mid-request and only GET or HEAD requests are retried automatically. The reporter's script measured 10 of 10 such failures after a /send call, versus 0 of 10 without one (https://github.com/cloudflare/workers-sdk/issues/16153). Reporter's hypothesis (not maintainer-confirmed): In the 4.148.0 bundle, the bundle watcher ignores .git, node_modules and .wrangler (a comment there reportedly warns of an infinite reload loop), while the assets watcher is created without an `ignored` option, so it reacts to wrangler's own writes. The reporter also says listing `.wrangler` in `.assetsignore` makes no difference, since that file governs what is served rather than watched (https://github.com/cloudflare/workers-sdk/issues/16153). Status: The issue is open. The only comment is an automated triage bot comment that labeled it an actionable bug with the wrangler package and assets feature labels; its automated reproduction attempt failed with a submission timeout, so it provides no confirmation. No maintainer-confirmed cause, fix commit or released fix appears in the supplied sources. The reporter's proposed fix (apply the same ignore pattern to the assets watcher and ideally honour .assetsignore) is only a suggestion. Practical reading: if you see this loop with assets.directory set to the project root, the likely trigger per the report is the assets directory containing .wrangler. The only partial mitigation reported is --persist-to outside the project, which does not remove the email-triggered reload. This is source-reported and not verified by KFA. Limitations: single reporter, one version, no independent reproduction, and possibly omitted middle comments.