test(evals): exercise the workspace V4 CLI

[shrey150]

Summary

• make evals resolve and execute the workspace browse package so the V4 stack tests the CLI source under review
• port the deterministic adapter to the current command tree, flags, JSON output, and stable tab identifiers
• support --verbose consistently in parsing, configuration precedence, help, and one-shot execution
• share a socket-safe session-name contract across deterministic and prompted adapters
• restrict the prompted wrapper to browser-driver commands and harden temporary-directory, timeout, signal, daemon, and Browserbase cleanup
• adapt the harness to the current v4-spike agent-tool runtime and observation contracts

Refreshed on 2026-08-10 against the current parent and v4-spike commit d7556e9a.

Contract

Surface	Result
Public CLI commands and result JSON	Unchanged
Eval dependency	Uses browse: workspace:* instead of a published CLI build
Deterministic startup	Uses open about:blank --local or open about:blank --remote
Prompted wrapper	Pins mode/session, rejects non-driver commands, and cleans up on success, failure, or process signal
Session names	Stay within the Unix socket byte limit; covered against the actual socket path builder

E2E Test Matrix

Command / flow	Observed output	Confidence / sufficiency
2026-08-10 cumulative stack refresh on Chrome Stable 151	pnpm build passed 6/6 tasks; pnpm check passed 11/11 tasks with 0 errors; the full pnpm test graph exited 0, including 17/17 protocol browser tests and 15/15 local SDK browser tests	Proves the updated six-PR stack is compatible with v4-spike d7556e9a across build, static checks, unit tests, and real-browser suites.
Clean coordinated-release tarball consumer at the cumulative tip	Packed workspace Stagehand and Browse, installed both into an empty npm project, opened example.com, produced a snapshot, clicked through to IANA, filled an injected input, and read back published-v4-ok	Proves the intended co-published artifacts work outside the monorepo with a real Chrome session; it does not substitute for a credentialed Browserbase smoke.
Frozen install with Node 24.18.0 + pnpm 11.10.0	Supply-chain policy passed for the refreshed 1,655-entry top lockfile	Proves reproducible installation of the current stack.
corepack pnpm --filter @browserbasehq/stagehand-evals test:unit	55 files passed; 428/428 tests passed	Covers deterministic and prompted adapters, wrapper restrictions, verbose parsing, socket length, cleanup, reporting, and current harness contracts.
node packages/evals/dist/cli/cli.js run core --tool browse_cli --env local --trials 1 --concurrency 1 --verbose	18/18 tasks passed, 100%	Exercises every deterministic core category through the exact workspace CLI build and real Chrome Stable 151.
Real local CLI workflow outside the eval harness	Snapshot, fill, select, upload, selector/coordinate click, screenshot, public navigation, tab lifecycle, and idempotent stop all passed	Guards against the eval harness masking CLI or daemon behavior.
Full JavaScript workspace unit graph	13/13 tasks; 1,788 passed and 10 todo	Covers CLI, evals, SDK, extension, protocol, docs, integrations, and repository rules/scripts.
Python CI-equivalent suite	455 passed, 1 skipped; generation, lock, Ruff, and ty checks passed	Confirms cross-language V4 protocol compatibility.
Go 1.26.5 CI-equivalent suite with Chrome Stable 151	Generation/extension drift, gofmt, vet, build, generator tests, and live Chrome tests passed	Confirms the rebuilt V4 extension and Go client remain compatible.
npm browse@0.9.6 parity and release audit	Exact 70-command manifest, identical 142-file list, bundled skill, help text, and version. A clean registry-backed install resolved Stagehand 4.0.0 and passed open/snapshot/status; a clean coordinated source-tarball install also passed click/fill	Proves published surface parity and the intended co-release path; PR #2526's Stagehand patch must publish with/before Browse.
Remote Browserbase deterministic/prompted rerun	Not run: this environment has no Browserbase or model/verifier credentials	Remaining live verification. GitHub CI is the next credentialed gate; prior pre-refresh remote results are intentionally not counted for the new head.
Final corepack pnpm build and corepack pnpm check	6/6 build tasks; 11/11 checks; 400 warnings and 0 errors	Proves the exact refreshed top stack passes repository gates.

No changeset is included because this pull request changes the private eval harness rather than a published package.

Final release validation (2026-08-10)

At the cumulative stack tip:

• release validation now recognizes public browse changesets
• a Changesets version dry run resolves browse 0.9.7, Stagehand 4.0.1, extension 1.0.1, and Go 4.0.1
• the packed Browse 0.9.7 manifest rewrites workspace:* to exact Stagehand 4.0.1
• a clean npm consumer installed those versioned tarballs and completed real-Chrome open, snapshot, model-free click, URL read, and stop with zero audit vulnerabilities

[changeset-bot]

⚠️ No Changeset found

Latest commit: 858b52e

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

This PR includes no changesets

When changesets are added to this PR, you'll see the packages that this PR includes changesets for and the associated semver types

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[shrey150]

Superseded by #2838 at the top of the clean stack. The replacement points evals at the workspace Browse package, updates command mappings and lifecycle cleanup, adds focused adapter tests, and carries the Browse release changeset. The unrelated eval CLI --verbose UX work from this draft was intentionally not carried into the migration stack. Closing this draft.