webwright
Drive a local Playwright browser through a Roder-owned Webwright workspace with plan, final script, screenshots, logs, and verification evidence.
- License 许可
- MIT
- 所属仓库
- RoderAI/roder
- 最近更新
- 2026-11-13
- Stars
- ★ 9
- Forks
- 1
01 前言 Front Matter
- name
- webwright
- exposure
- direct_only
- metadata
- { "short-description": "Browser automation" }
- description
- Drive a local Playwright browser through a Roder-owned Webwright workspace with plan, final script, screenshots, logs, and verification evidence.
03 SKILL.md 正文
Use this skill for browser tasks that need durable evidence, repeatable Playwright scripts, screenshots, logs, or reusable CLI automation.
Modes
/webwright:run <task>: solve the literal task values supplied by the user./webwright:craft <task>: create a parameterized Python CLI where concrete task values become defaults and--helpdocuments how to reuse it.
Workspace Contract
Create or use one Webwright workspace under .roder/webwright/<task-id>/ unless the user gives a scoped output directory.
Required artifacts:
plan.mdwith# Critical Pointsand one independently verifiable checklist item per user constraint.final_script.pyat the workspace root.final_runs/run_<id>/final_script.pyfor each clean execution.final_runs/run_<id>/screenshots/final_execution_<step>_<label>.pngfor constraint-relevant evidence.final_runs/run_<id>/final_script_log.txtwith onestep <n> action:line per constraint-relevant action and the final datum at the end.
Use viewport={"width": 1280, "height": 1800} for Playwright screenshots. Do not use page.screenshot(full_page=True).
Roder Tools
Prefer Webwright helper tools for setup and checks:
webwright.prepare_workspacewebwright.allocate_runwebwright.lint_scriptwebwright.run_scriptwebwright.list_artifactswebwright.read_log_tailwebwright.verify_runwebwright.summarize_verification
Use normal Roder file, edit, shell, media, and artifact tools for the actual code-as-action loop. Keep generated files inside the Webwright workspace.
If local Playwright dependencies are missing, ask the user to run roder webwright setup --browser firefox or the selected browser (chromium or webkit). webwright.run_script uses the managed setup runtime automatically when no python override is supplied.
Verification
Only mark the task done when every critical point is checked and backed by a screenshot path or exact log line from the latest run. If evidence is ambiguous, fix the script and rerun in the next run_<id> directory.
webwright.run_script reports an outcome class (ok, nonzero_exit, signaled, timeout, launch_failed), the elapsed time, the first error line and a redacted stderr tail. Fix the script from that text before rerunning. A run executed through webwright.run_script that does not exit 0 fails webwright.verify_run even if its log has a final datum: line and its screenshots exist; a run executed outside that tool has no recorded exit status, so check its exit status yourself.
For /webwright:craft, also verify:
- The reusable function is import-safe and does not launch a browser at import time.
python final_runs/run_<id>/final_script.pysucceeds with no arguments.python final_runs/run_<id>/final_script.py --helpshows the user-facing argument contract.
04 同仓库更多 Skills
repo-tour
Use when a user asks for an orientation tour of an unfamiliar repository. Produces a structured tour of entry points, key modules, and how to run tests.
review
Review code changes.
vcs-snapshot
Create a scoped version-control snapshot.
vcs-snapshot
Create scoped VCS snapshots from the current workspace state.
roder-config
Configure and live-reload Roder from pasted config snippets, especially MCP server JSON, without leaking secrets or enabling executable integrations without approval.
chrome
Drive the user's logged-in Chrome through the Roder browser extension to inspect tabs, read DOM/console/network, and interact with pages.