[mikeland73]

Summary

Implements item 5 of specs/perf-improvements.md.

direnv runs devbox generate direnv --print-envrc on every reload. It only prints a fixed .envrc template, but it still:

• inherited ensureNixInstalled from generate's PersistentPreRunE, which runs nix --version, and
• recorded a telemetry event, which also spawned the devbox upload-telemetry process.

Changes:

• generate's PersistentPreRunE skips the nix check when --print-envrc is set. The root uses cobra.EnableTraverseRunHooks, so overriding the hook on the direnv subcommand wouldn't stop the parent's hook from running. That's why the check goes in the parent.
• New midcobra.SkipTelemetry(cmd) lets a command opt out of being recorded for the current run. The telemetry post-run hook checks for it. The --print-envrc path calls it.

Testing

• New TestGenerateDirenvPrintEnvrcSkipsNixCheck runs the command through RootCmd() and checks the output and that nix.Ensured() stays false. Without the fix it fails.
• New TestSkipTelemetry.
• go test ./internal/boxcli/... passes. golangci-lint run ./internal/boxcli/... reports 0 issues.
• Output is byte-for-byte the same as on main.
• Timing for generate direnv --print-envrc on macOS, averaged over 20 runs. Binaries were built with fake telemetry keys and network was blocked by a proxy.

	main	this PR
telemetry on	131.0ms	33.7ms
DO_NOT_TRACK=1	92.7ms	10.6ms

With telemetry on, almost all of the remaining ~23ms is telemetry.Start() running ioreg for the machine ID. #3021 caches that.

🤖 Generated with Claude Code

[mikeland73]

@claude ptal

[claude]

PR #3022 was updated after the trigger (2026-10-25T21:46:37Z >= 2026-10-25T21:44:10Z). Mention Claude again to run on the latest version.

[mikeland73]

@claude ptal

[claude]

Claude finished @mikeland73's task in 17s —— View job

---

Review of PR #3022

Looks good to me. I found no blocking issues. I read the diff and the surrounding code but didn't run the tests or build the binary.

• Read the diff and surrounding code
• Check the telemetry skip path and the nix-check skip

What I checked

• Nix check: generate's PersistentPreRunE now returns early when --print-envrc is set. The reasoning for putting it in the parent holds, because EnableTraverseRunHooks means overriding the hook on direnv wouldn't stop the parent's hook. isPrintEnvrc returns false when the flag isn't defined (GetBool errors), so the other generate subcommands still run ensureNixInstalled.
• Telemetry: getSubcommand in postRun uses cmd.Find(args), which returns the same *cobra.Command that ran. The annotation set by SkipTelemetry(cmd) in runGenerateDirenvCmd is therefore visible to telemetrySkipped(subcmd). postRun still defers telemetry.Stop() before the early return, so cleanup is unchanged.
• Tests: TestGenerateDirenvPrintEnvrcSkipsNixCheck exercises the real RootCmd() path. TestSkipTelemetry covers the helper.

Minor, non-blocking notes

1. TestGenerateDirenvPrintEnvrcSkipsNixCheck skips if nix.Ensured() is already true from an earlier test. That keeps it from failing spuriously, but if some test starts calling ensureNixInstalled earlier in the package, this one will silently stop running. It's fine for now.
2. telemetrySkipped checks != "", so any non-empty annotation value counts as skipped. That's fine, though comparing against "true" would be slightly stricter.
3. preRun still calls telemetry.Start() for skipped commands, which is the ~23ms ioreg cost you mention. The PR description says perf(telemetry): reuse opened project, cache device ID, throttle uploads #3021 handles that, so this is expected.
4. --print-envrc combined with --envrc-dir errors out in runGenerateDirenvCmd, and that happens after the nix check is skipped. This only changes behavior for an invalid flag combination, and the error is the same as before.

Overall, this is ready to merge once CI is green.