BrowserTrace is a local replay debugger for Browser Use failures. The roadmap keeps Browser Use as the primary path while making Stagehand, Skyvern, Playwright + LLM, and computer-use runs easier to inspect, export, and discuss as secondary integrations.
v0.1.20 is the current launch release.
Shipped:
- Local SDK and CLI for recording browser-agent steps.
- Local web UI for inspecting runs.
- Screenshots, URLs, actions, model input/output, status, and errors.
browsertrace demofor a deterministic first run without API keys.browsertrace doctorfor first-run environment checks before any trace database exists.browsertrace exportfor standalone HTML traces.browsertrace export --publicfor public-safe sharing without prompt/model I/O, screenshots, or URLs.- Basic Browser Use, Stagehand, and Skyvern-shaped integration examples.
- A generic computer-use loop example for custom observe-decide-act browser agents.
- GitHub Pages live demo, integration guides, launch kit, and public-safe demo asset.
These are owner-only account actions. They are important for growth, but they cannot be completed by a contributor PR.
| Issue | Status | Why it matters |
|---|---|---|
| #5 PyPI publishing | Completed | Enables normal pip install "browsertrace[ui]" instead of GitHub install URLs. |
| #15 Social preview | Completed | Makes shared repo links show the BrowserTrace card. |
| #16 Search console submission | Owner action needed for search consoles; IndexNow submitted | Starts long-tail indexing for the GitHub Pages docs and demos. |
| #9 Day 1 warm launch posts | Owner action needed | Publishes from the owner's X, LinkedIn, WeChat, and Jike accounts. |
| #10 Directory and newsletter submissions | Owner action needed | Reaches curated developer-tool directories and newsletters without repeated submissions. |
| #18 Awesome list submissions | Monitoring open PRs | One focused PR is merged/listed (Jenqyang/Awesome-AI-Agents#220); remaining focused PRs are open; E2B CLA check has passed; wait for maintainer feedback before any more list submissions. |
Completed launch prep:
- #13 GitHub profile README
now points the
aaronlabprofile at BrowserTrace. - GitHub Release
v0.1.20includes the wheel, sdist, full demo export, public-safe demo export, demo video, poster, and GIF. - PyPI is live as
browsertrace==0.1.20, withpip install "browsertrace[ui]"as the canonical persistent install path. v0.1.20keeps thebrowsertrace doctoronboarding fix, packagedbrowsertrace demopath, explicitbrowsertrace compareCLI path, and/api/compare/<failed_run_id>/<success_run_id>JSON endpoint for failed-vs-successful Browser Use runs.- IndexNow submission is prepared and submitted for the main GitHub Pages launch URLs.
| Track | Current issue | Good contribution shape |
|---|---|---|
| Browser Use trace fields | #11 | Add fields that real Browser Use failures expose safely, with no required API keys in tests. |
| Stagehand adapter depth | #8 | Capture act, extract, observe, or agent calls with small examples and focused tests. |
| Skyvern workflow depth | #4 | Capture task/workflow status, selected elements, and failure state where the API exposes them. |
| Playwright + LLM trace shape | #12 | Document and test fields that help LLM-guided Playwright scripts explain failures. |
| Multi-run comparison | Open an issue with a real regression case | Compare two runs enough to answer whether a failure appeared after a code or prompt change. |
| Export ergonomics | Open a focused issue | Improve redaction, attachment flow, or public-safe trace sharing without hosted infrastructure. |
Good roadmap PRs are narrow and testable:
- Keep optional framework dependencies optional.
- Avoid network access, real API keys, and hosted browser accounts in tests.
- Use
BROWSERTRACE_HOMEfor isolated local data in examples and tests. - Use
browsertrace export <run_id> --public -o public.htmlbefore sharing a real trace publicly. - Include the exact command used to verify the change.
Start with CONTRIBUTING.md#first-pr-recipe; it keeps the first contribution small and reviewable.
Then open an integration request if the work depends on a specific
browser-agent framework.
Stack-specific guide context:
- Browser Use guide: https://aaronlab.github.io/browsertrace/browser-use-debugging.html
- Stagehand guide: https://aaronlab.github.io/browsertrace/stagehand-debugging.html
- Skyvern guide: https://aaronlab.github.io/browsertrace/skyvern-debugging.html
- Playwright + LLM guide: https://aaronlab.github.io/browsertrace/playwright-llm-debugging.html
- Computer-use guide: https://aaronlab.github.io/browsertrace/computer-use-agent-debugging.html
The project is moving in the right direction when:
- New users can install with
pip install "browsertrace[ui]". - Browser-agent builders report which trace fields they need.
- External issues and PRs include real failure shapes instead of generic feature requests.
- Public posts and directory submissions ask for workflow feedback, not stars or upvotes.
- The live GitHub audit eventually reports more than 1000 stars.