Menu
Create GitHub pull requests for browseruse-bench with clear conventional-commit style titles and a review-ready body.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Create and integrate new modular agents for browseruse-bench (BaseAgent in browseruse_bench/agents/, registry import in __init__.py, entry in configs/agent_registry.yaml, runtime config in root config.yaml agents section). Use when a user asks to add/register a new agent, scaffold a custom agent, or implement a new agent based on external docs.
Generate and maintain PR review checklists from GitHub review threads for the current branch PR or a specified PR. Use when you need reviewer comments with exact code locations, nearby code snippets, and raw thread chat logs, and when you want continuously updated .pr/<pr_number>/review_checklist.md and .pr/<pr_number>/unresolved_checklist.md.
Use this skill whenever the user describes a benchmark experiment using any combination of these parameters — agent (e.g. browser-use, skyvern), model (e.g. deepseek, minimax, claude, gemini, gpt), data/benchmark (e.g. LexBench-Browser), browser (e.g. Chrome-Local, lexmount), and tasks (specific IDs or "all"). Trigger phrases include "跑实验 agent=xxx model=xxx", "跑实验 agent xxx model xxx", "run [model] on [tasks]", "eval [model] results", "re-run [model]", or any request that specifies an agent + model + task set to execute or evaluate. Also triggers on requests to chain run→eval pipelines, check experiment progress, or read final results.
Update Mintlify documentation in docs/ for browseruse-bench, including adding new pages, enforcing .mdx files, and keeping English/Chinese navigation parity. Use when adding or editing docs pages, fixing missing/incorrect language variants, or updating docs/docs.json navigation.
Draft and track fixes for unresolved PR review threads. Use when you need to generate a solution draft markdown for the latest/current PR unresolved threads and keep `.pr/<pr_number>/review_checklist.md` plus `.pr/<pr_number>/unresolved_checklist.md` checkbox states synchronized as issues are fixed.
| name | create-pr |
| description | Create GitHub pull requests for browseruse-bench with clear conventional-commit style titles and a review-ready body. |
| allowed-tools | Bash(git:*), Bash(gh:*), Read, Grep, Glob |
Create a draft GitHub pull request for this repository using gh.
gh is installed and authenticated.Use conventional commit style:
<type>(<scope>): <summary>
feat: New featurefix: Bug fixperf: Performance improvementrefactor: Refactor without behavior changedocs: Documentation onlytest: Tests onlybuild: Build/dependency changesci: CI workflow/config changeschore: Maintenance taskscorecliagentsbenchmarksdocsskillsAdd, Fix, Refactor)git status
git diff --stat
git log origin/main..HEAD --oneline
git push -u origin HEAD
gh pr create \
--draft \
--base main \
--title "<type>(<scope>): <summary>" \
--body "$(cat <<'PRBODY'
## Summary
- What changed
- Why it changed
## Testing
- [ ] `uv run ruff check .`
- [ ] `uv run pytest`
## Related
- Optional: `closes #<issue-number>`
PRBODY
)"
Summary: explain behavior change and key files touched.Testing: list commands actually run.Related: reference GitHub issues when available.fix(cli): Propagate skills command exit code
docs(benchmarks): Correct LexBench split filenames
refactor(agents): Guard duplicate agent registration names
<type>(<scope>): <summary>.