fix(evals): use workspace Browse to unblock release builds

[shrey150]

Superseded by #2963, which now targets main and includes every change from this PR together with the adapter fixes. No separate merge of this PR is needed.

Release versioning rewrites evals' pinned Browse dependency without updating the lockfile, causing the next frozen install to fail (failed release). Use browse: "workspace:*" so evals consume the current checkout and their dependency remains stable across release bumps.

This build-fix PR changes only the eval dependency, its lockfile importer, and workspace path/version assertions (four files). Native adapter runtime fixes are in the stacked child #2963 and are not required to merge this release fix. This implements the workspace approach as an alternative to #2961's Changesets flag.

Stack

1. This PR → main: workspace dependency and build fix.
2. fix(evals): use workspace Browse and repair CLI adapter #2963 → this branch: repair the pre-existing native Browse adapter and add runtime regression coverage.

Merge this PR first. The child preserves the already-reviewed adapter fixes separately.

E2E Test Matrix

Verified on a7711cac6 with pnpm 11.10.0. The parent was checked out and tested independently of the adapter fixes. Current head 468ae0a70 merges main's README cleanup; all files except README are identical to the verified head, and README exactly matches main and passes formatting.

Command / flow	Observed output	Confidence / sufficiency
GitHub checks on 468ae0a70	46 passed, 9 skipped; none failed or pending	Updated build-only parent is conflict-free and passes independently; no outstanding Cubic findings
CI=true pnpm install --frozen-lockfile	Passed; evals link to workspace Browse	Confirms the dependency and lockfile agree
CI=true pnpm exec changeset version, CI=true pnpm exec tsx scripts/release/consolidate-changelogs.ts, then CI=true pnpm install --frozen-lockfile	Passed; Browse became 0.9.7; eval manifest and lockfile were unchanged	Reproduces the release sequence that previously failed; generated release files were restored afterward

| pnpm exec turbo run build --filter=@browserbasehq/stagehand-evals | 14/14 tasks passed | Builds the workspace CLI and eval bundles |
| pnpm exec vitest run --config packages/evals/vitest.config.ts --maxWorkers=4 | 90 files / 761 tests passed | Full eval unit suite, with the inherited devbox OpenAI key unset |
| Actual generated Claude/Codex Browse wrapper against local Chromium | Opened a synthetic form, filled/clicked it, and read the expected title; daemon files cleaned up | Proves the existing external wrapper works with the workspace dependency without child adapter fixes |
| Evals typecheck and repository formatting check | Passed | No type or formatting errors |

The existing native adapter's obsolete-command failure also occurs with published Browse 0.9.5. It is intentionally repaired in #2963. This PR does not claim native runtime compatibility or perform a live package publish.

[changeset-bot]

⚠️ No Changeset found

Latest commit: 468ae0a

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

[cubic-dev-ai]

No issues found across 5 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

Architecture diagram

sequenceDiagram
    participant EvalPkg as @browserbasehq/stagehand-evals
    participant Pnpm as pnpm (workspace resolver)
    participant BrowseCli as Browse CLI package (workspace)
    participant Lockfile as pnpm-lock.yaml
    participant Release as Release tooling (changeset version)
    participant Test as Eval test suite

    Note over EvalPkg,BrowseCli: Workspace Dependency Resolution Flow

    EvalPkg->>EvalPkg: package.json declares browse: "workspace:*"
    Pnpm->>Pnpm: Resolve workspace dependency
    Pnpm->>BrowseCli: Link local package (packages/cli)
    BrowseCli-->>Pnpm: Confirmed available
    Pnpm->>Lockfile: Record workspace importer
    Lockfile-->>Pnpm: Lockfile entry synced

    Note over Release,Lockfile: Release Versioning Sequence
    
    Release->>Pnpm: Run changeset version
    Pnpm->>BrowseCli: Bump version (e.g., 0.9.7)
    Pnpm->>Lockfile: Update CLI package version
    alt PR pre-fix (published version pinned)
        Pnpm->>Pnpm: Rewrite eval dependency to new version
        Pnpm->>Lockfile: Mismatch - no lockfile update
        Note over Pnpm,Lockfile: Frozen install fails on next run
    else CHANGED: workspace protocol
        Pnpm->>EvalPkg: Keep workspace:* reference
        Pnpm->>Lockfile: Lockfile remains consistent
    end
    Pnpm-->>Release: Versioning succeeds

    Note over Test,BrowseCli: Runtime Eval Flow (unchanged external wrapper)
    
    Test->>Test: Import workspace Browse CLI
    Test->>BrowseCli: Resolve tool metadata
    BrowseCli-->>Test: Version + entrypoint (packages/cli/bin/run.js)
    Test->>BrowseCli: Execute Browse wrapper (external)
    BrowseCli->>Browser: Drive local Chromium
    Browser-->>BrowseCli: Page interactions complete
    BrowseCli-->>Test: Test results pass

    alt Frozen install verification
        Pnpm->>Lockfile: pnpm install --frozen-lockfile
        alt Lockfile consistent
            Lockfile-->>Pnpm: Success
            Pnpm-->>EvalPkg: Linked workspace dependency
        else Lockfile mismatch
            Lockfile-->>Pnpm: Failure (blocked release)
        end
    end

Loading

_{Re-trigger cubic}