feat(test): PTY-based interactive CLI snapshot harness

[fengmk2]

Implements the RFC included in this PR as rfcs/interactive-snapshot-tests.md.

• crates/vite_cli_snapshots: PTY snapshot harness (libtest-mimic). Every step runs in a real pseudo-terminal with vt100 grid capture; interactive steps script keystrokes synchronized on OSC 8 milestones; snapshots are Markdown compared with real pass/fail semantics (UPDATE_SNAPSHOTS=1 to accept, .md.new plus unified diff on mismatch). Built on the pty_terminal/snapshot_test crates from vite-task at the already-pinned rev.
• One fixture tree replaces the local/global split: each case declares vp = "local" | "global" | ["local", "global"]. The parity matrix already caught real drift between the two help outputs. Per-case VP_HOME/HOME/npm-prefix isolation removes serial and the bootstrap byte-match requirement; seed-runtime symlinks a provisioned managed runtime so cases do not download Node per case.
• vpt test multitool: vtt-aligned subcommands plus json-edit, chmod, and probe (interactive payload proving milestone round-trips before product commands are instrumented).
• tool migrate-snap-tests <dir> --vp <flavor> [filter]: one-click migration of old steps.json cases with a report; validated by migrating the four check-pass* cases, which pass under the new harness with equivalent assertions.
• packages/prompts: milestone emission (select/confirm/text, gated on VP_EMIT_MILESTONES=1, byte-identical to the vite-task protocol, unit-tested; render output unchanged when disabled).
• Entry points: just snapshot-test [filter], pnpm snapshot-test. CI runs the suite on the Rust test job with VP_SNAP_SKIP_FLAVORS=local until the JS build joins during migration.

Validation: 8 snapshot trials pass in compare mode (including the interactive milestone case), prompts suite 10/10, clippy/rustfmt/vp check clean, just test excludes the harness crate.

Follow-ups per the RFC phasing: local-registry case support, remaining prompt components (multiselect, password, spinner), migration batches and old-harness removal.

• Windows: the suite runs in the new cli-snapshot-test-windows job via a cross-compiled nextest archive (no Rust toolchain on the runner); first landing on this PR, watch its leg.

[netlify]

✅ Deploy Preview for viteplus-preview canceled.

Name	Link
🔨 Latest commit	bbf3a94
🔍 Latest deploy log	https://app.netlify.com/projects/viteplus-preview/deploys/6a4a5a362be0500008ccb379

[fengmk2]

• feat(cli): global -C flag for working-directory switching #2031
• feat(test): PTY-based interactive CLI snapshot harness #2052 👈 (View in Graphite)
• main

---

How to use the Graphite Merge Queue

Add the label auto-merge to this PR to add it to the merge queue.

You must have a Graphite account in order to use the merge queue. Sign up using this link.

An organization admin has enabled the Graphite Merge Queue in this repository.

Please do not merge from GitHub as this will restart CI on PRs being processed by the merge queue.

This stack of pull requests is managed by Graphite. Learn more about stacking.

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	cargo/​nix@​0.28.0
	cargo/​winreg@​0.10.1
	cargo/​cp_r@​0.5.2
	cargo/​toml@​1.1.2%2Bspec-1.1.0
	cargo/​libtest-mimic@​0.8.2

View full report

[fengmk2]

@wan9chi Please help me see how to make vt reusable or inheritable/extendable by vpt, so that vpt only needs to add additional commands without duplicating the code implementation.

[fengmk2]

Feedback from #2031's Windows run (https://github.com/voidzero-dev/vite-plus/actions/runs/28742954669/job/85229263838): 9 of 31 trials fail, all snapshot mismatches from the new app-command fixtures, and both root causes are Windows gaps in the harness rather than product or fixture bugs.

1. redact.rs: the Windows backslash normalization is gated on an absolute-path redaction having matched the same screen.

In redact_string, the cow_replace("\\", "/") pass lives inside the if let Cow::Owned(..) branch, so it only runs when one of the path redaction pairs (temp dir, VP_HOME, ...) matched that screen. Screens that contain only relative native paths are never normalized. tsdown prints OS-native separators, so every fixture that snapshots pack output fails:

-ℹ entry: src/index.ts        (recorded on macOS)
+ℹ entry: src\index.ts        (Windows actual)
-ℹ dist/index.mjs  0.10 kB │ gzip: 0.11 kB
+ℹ dist\index.mjs  0.10 kB │ gzip: 0.11 kB

Affected trials: app_root_default_package::default_package (both flavors), pack_default_entry (both), single_package::pack_in_place (both), cwd_flag (both). The old harness's replaceUnstableOutput normalized backslashes unconditionally, which is why the old-format versions of these same tests passed Windows CI. Suggested fix: make the per-screen backslash-to-slash normalization unconditional on Windows (independent of whether a redaction pair matched).

2. Renderer: ConPTY repaints a row padded to full grid width with explicit spaces when a second console client attaches.

app_root_auto_select::auto_select::global and app_root_default_package::default_package::global capture one line padded with trailing spaces to the terminal width:

-Tip: run this directly with `vp -C apps/web build`
+Tip: run this directly with `vp -C apps/web build`                    ...(~400 spaces)

The signature is precise: it occurs exactly at the boundary where vp's own stdout lines are followed by the spawned tool's first writes, and only in the global flavor, where vp.exe spawns node as a second ConPTY client (the local flavor, with node as the direct PTY child, passes the same fixtures). Suggested fix: trim trailing whitespace per rendered row before snapshot comparison.

Rider on 1 (cosmetic): dist/index.mjs reports 0.11 kB on Windows vs 0.10 kB recorded in two trials (single_package::pack_in_place), so the emitted bundle is a few bytes larger on Windows (line endings or an embedded path, not yet pinned down). Only visible because sizes are not redacted; worth deciding whether sizes should be.

With fixes 1 and 2 in the harness, all nine trials should pass against the already-recorded snapshots; nothing in the #2031 fixtures needs to change. Happy to re-run the Windows job on #2031 once this lands to confirm.