Migrate all Stagehand templates to V4 - #101
Conversation
|
Review the following changes in direct dependencies. Learn more about Socket for GitHub. |
|
Warning Review the following alerts detected in dependencies. According to your organization's Security Policy, it is recommended to resolve "Warn" alerts. Learn more about Socket for GitHub.
|
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
shrey150
changed the title
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit 7dad3ec. Configure here.
| resumeUploaded: review.resumeUploaded, | ||
| unfilledPresentFields, | ||
| })}`, | ||
| ); |
There was a problem hiding this comment.
Resume upload required when absent
Medium Severity
Success now requires resumeUploaded even when providedFieldStatus.resume.present is false. The prompt only asks to upload a resume when a file input exists, and the review schema already records that absence, so applications without a resume field fail after an otherwise verified open-and-fill.
Reviewed by Cursor Bugbot for commit 7dad3ec. Configure here.
| resumeUploaded: review.resumeUploaded, | ||
| unfilledPresentFields, | ||
| })}`, | ||
| ); |
There was a problem hiding this comment.
Empty fields fail human-review path
Medium Severity
Any provided field that exists on the form but is not filled now fails the run. visaStatus is an empty string, and the agent is told not to invent facts and to leave missing or ambiguous answers for review, so a present visa/work-authorization field cannot be filled and will always trip unfilledPresentFields.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit 7dad3ec. Configure here.
2 participants
Summary
Stagehand is the SDK for browser agents.
@browserbasehq/stagehand@4.0.0andstagehand==4.0.0for the core SDKsbrowserbase.launch()/localBrowser.launch(),Stagehand.create({ browser }),browser.context, V4{ data, metadata }envelopes,locator(), and explicit Stagehand-then-browser cleanupCompanion PR: browserbase/create-browser-app#40
Agent migration: 14 templates
Nine TypeScript templates use Vercel AI SDK
ToolLoopAgentwith Stagehand code mode's statefulcode_executetool:typescript/agent-with-human-in-looptypescript/browser-agent-demotypescript/business-lookuptypescript/company-address-findertypescript/dynamic-form-fillingtypescript/exa-browserbasetypescript/gemini-3-flashtypescript/gemini-cuatypescript/microsoft-cuaFive Python templates use Deep Agents with the Stagehand integration's stateful browser tools:
python/business-lookuppython/cerebras-docs-checkerpython/company-address-finderpython/exa-browserbasepython/gemini-cuaThe Go template's removed autonomous
Sessions.Executestep is now an explicit, verifiable V4Observe→Act→ typedExtractworkflow.Package state
@browserbasehq/stagehand@4.0.0stagehand==4.0.054302fc5f13be5ad8e717d8e1388502de22be2eduntil@browserbasehq/stagehand-codemodeis publishedstagehand/mainsource (packages/integrations/deepagents), with its server dependency on publishedstagehand==4.0.0v0.0.0-20260810174315-a46676658b4funtil the module is taggedThe code-mode Git dependency needs two temporary measures until publication: the pnpm subdirectory selector must be
path:packages/integrations(no leading slash), and pnpm overrides must resolve the upstream monorepo'sworkspace:/catalog:dependencies. These are now present in all nine manifests.Fresh E2E re-verification — 2026-08-18
This section replaces all historical E2E claims. Every executable entrypoint was freshly installed and invoked against the current PR head. No earlier run was counted as evidence.
There are 77 first-level language directories but 79 executable entrypoints: 68 Stagehand flows and 11 non-Stagehand flows. The previous description's “10 executable non-Stagehand flows” missed that both TypeScript and Python Playwright directories each contain two entrypoints.
Strict current result
7dad3ec; this is not equivalent to every documented quickstart being independently runnable because some TypeScript directories are manifestlessThe V3 side is a reconstructed behavioral baseline using the final V3 releases (
@browserbasehq/stagehand@3.7.1,stagehand==3.22.0). A literal clean install of several base-branch manifests is no longer reproducible because they usedlatestor open-ended constraints that now resolve V4.Legend: FIXED means the fresh pre-fix V4 run failed and the exact post-fix rerun passed. BLOCKED is not counted as a pass.
TypeScript Stagehand — 35 entrypoints
agent-with-human-in-loopamazon-global-price-comparisonamazon-product-scrapingbasic-cachingbasic-recaptchabrowser-agent-demobrowserbase-reductobusiness-lookupcompany-address-findercompany-value-prop-generatorcontextSF_REC_PARK_EMAIL/SF_REC_PARK_PASSWORD.council-eventspackage.json, so its documented install/start path is incomplete. With a minimal temporary ESM harness, the restoredgoto→act→act→extractsequence reaches the site: calendar-opening succeeds, but year selection returnsNo action found.download-financial-statementsdynamic-form-fillingexa-browserbaseextend-browserbaseform-fillinggemini-3-flashgemini-cuagift-findergoogle-trendsimage-url-downloadjob-applicationlicense-verification02237476.manual-mfa-with-contextsmfa-handlingmicrosoft-cuanurse-verification346, Active.pickleballpolymarket-researchproxiesproxies-weathersec-filing-researchsmart-fetch-scraperwebsite-link-tester/templates/amazon-price-comparison). Expected nonzero exit proves the checker caught the site issue.Python Stagehand — 32 entrypoints
amazon-global-price-comparisonamazon-product-scrapingbasic-cachingbasic-recaptchabrowserbase-reductobusiness-lookupcartesia-form-fillingcerebras-docs-checkercompany-address-findercompany-value-prop-generatorcontextcouncil-eventsdownload-financial-statementsexa-browserbaseextend-browserbaseform-fillinggemini-cuagift-findergoogle-trendsimage-url-downloadjob-applicationlicense-verificationmanual-mfa-with-contextsmfa-handlingnurse-verificationactcalls. Because the result table is client-rendered after submit, boundedextractretries returned Ronald Agee license346, Active, in two independent clean runs.pickleballpolymarket-researchproxiesproxies-weathersec-filing-researchsmart-fetch-scraperwebsite-link-testerGo Stagehand — 1 entrypoint
hackernewsNon-Stagehand — 9 directories / 11 entrypoints
getting-started-with-browserbaseplaywright/basic-recaptchaplaywright/quickstart-playwrightplaywright-mfa-handlingpuppeteer/quickstart-puppeteerselenium/quickstart-seleniumgetting-started-with-browserbaseplaywright/basic-recaptchaplaywright/quickstart-playwrightplaywright-mfa-handlingselenium/quickstart-seleniumFixes added from the fresh audit
Commits
f5a9cbd,22d2341,1672f82, and7dad3ec:act-based flow and uses boundedextractretries for its asynchronously rendered result tableRepository proof
git diff --checkpass.envfiles are includedRemaining work before claiming 100%
council-eventswithout replacing its Stagehand primitivescouncil-eventsso its documented quickstart is independently runnableexa-browserbaseat the default five-company workloadgemini-cuaCDP/TaskGroup failuregemini-3-flashand the Python link tester under repeated/stress runs@browserbasehq/stagehand-codemode@4.0.0, then remove the temporary Git pin/overridesNote
Cursor Bugbot will update its generated summary for the latest commit.