[mikeland73]

Summary

With DEVBOX_COREPACK_ENABLED set, every devbox run / devbox shell can fail in the nodejs plugin's init hook with:

Internal Error: ENOENT: no such file or directory, lstat '<project>/.devbox/virtenv/nodejs/corepack-bin'
    at Object.realpathSync (node:fs:...)
    at EnableCommand.execute (.../corepack.cjs:...)

corepack-bin is an empty directory the plugin creates through create_files. Devbox only runs create_files during a full recompute. If the directory goes missing while .devbox/state.json is still up to date, devbox doesn't recreate it, and corepack enable --install-directory fails because the directory doesn't exist. One way this happens is a .devbox copied into a new git worktree: the copy keeps setup-corepack.mjs but drops empty directories.

Changes

• plugins/nodejs/setup-corepack.mjs: create the install directory (mkdirSync(..., { recursive: true })) before running corepack enable.
• plugins/nodejs.json: bump the plugin version to 0.0.6.
• testscripts/plugin/nodejs_corepack_autodetect.test.txt: new case that deletes corepack-bin while state is up to date, then checks that devbox run succeeds, prints no ENOENT, and installs the pnpm shim.

Testing

• go test ./testscripts -run 'TestScripts/nodejs_corepack_autodetect' passes. With the script change reverted, the new case fails.
• Tested by hand with a local build and nodejs@24: after deleting corepack-bin, devbox run reproduced the error before the fix. After the fix, it recreates the directory, and which pnpm resolves into corepack-bin.

🤖 Generated with Claude Code

[claude]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

[mikeland73]

@claude ptal

[claude]

Claude finished @mikeland73's task in 14s —— View job

---

Review of PR #3042

I read the diff and found nothing that should block merging.

• Fix (plugins/nodejs/setup-corepack.mjs): mkdirSync(corepackBinDir, { recursive: true }) runs right before corepack enable, which matches the root cause. It does nothing when the directory already exists. It also sits inside the existing try, so if mkdirSync throws (for example on permissions), the error is handled like any other corepack enable failure and doesn't block shell init.
• Version bump (plugins/nodejs.json): 0.0.5 to 0.0.6 is needed so existing projects pick up the changed script.
• Test (nodejs_corepack_autodetect.test.txt): Case 3b covers the regression. It deletes corepack-bin while state is current, then checks that devbox run succeeds, stderr has no ENOENT, and the pnpm shim exists. You report that it fails with the fix reverted, which is the check that matters.

Optional nits:

• The ! stderr 'ENOENT' check only works if the test doesn't redirect or swallow stderr elsewhere. That looks fine here.
• The code comment is a bit long. It's still useful context.

I did not run the tests myself. I relied on the results in the PR description.
· branch claude/corepack-enoent-error-34fa75