Hi all, hoping someone can help or a Railway team member can take a look.

The issue: My nixpacks.toml at the repo root defines:

[phases.setup]

nixPkgs = ["nodejs_20"]

[phases.install]

cmds = ["npm install"]

The build log confirms the setup phase is correctly picked up (shows nodejs_20 instead of the old nodejs_18), but the install phase still runs npm ci instead of my configured npm install. I've confirmed:

The file is correctly pushed and live on the GitHub repo connected to this service (verified via direct SHA comparison between local and remote)

There is no .nixpacks/ directory committed in the repo

There is no Install Command override set in the service's Settings → Build tab (only Build Command and Start Command fields exist there, both empty/default)

Why this matters: npm ci is crashing partway through install with npm error Exit handler never called!, a known npm-internal bug, which then causes the build step to fail with tsx: not found since the install never completed. Switching to npm install was meant to work around this npm bug, but the override isn't being honored.

Additional clue: the build log shows inconsistent Docker stage numbering within the same single build, some lines show [stage-0 2/8] and others show [stage-0 3/15]. This made me suspect a stale cached build plan is being partially reused instead of fully regenerated from the current nixpacks.toml.

What I've tried: confirmed no repo-level conflicts, confirmed no dashboard override exists, triggered multiple fresh deploys, all show the same npm ci behavior despite the file change.

Is there a way to force-invalidate a cached Nixpacks build plan for a service?