Discussion: Roadmap and Ideas 💡

[cosinusalpha]

I am planning the next phase of development, and I have two specific features I really want to build. But beyond that, I want to know what you need.

My Current Focus

• "Session Recorder" (Human to Bash): I want to build a mode where you browse manually (handling logins, navigation, etc.), and webctl records your actions.
  • The Goal: You hit "Stop," and it outputs a ready-to-use Bash script containing the webctl commands to replay that exact session.
  • This would make creating reproducible agent workflows instant.
• Improved Output/Markdown: I'm not 100% happy with the current DOM-to-Markdown export.

What do YOU want?

I built this for my specific use cases, but I'm curious about yours.

• Are you using this for testing, scraping, or agentic workflows?
• Do you need support for other browsers (Firefox/WebKit), or is Chromium enough?
• Are there specific "Action" commands missing (e.g., drag-and-drop, file upload)?

Let me know in the comments below!

[jure-losthex]

Use case: driving a heavy internal SPA (property-valuation app) for both agentic workflows and E2E testing — Claude Code sessions navigate, fill forms, screenshot, extract. What makes webctl click for us is the shared, stateful session: a human can drive part of the flow (log in, clear a CAPTCHA, set up tricky state) and hand the same live session straight to the agent to finish — and back again. That human-in-the-loop handoff is the whole reason this beats raw Playwright MCP for our work. Thanks for building it.

A few things I hit and patched locally against 0.4.2 — happy to send either as a PR if you're interested:

1. Network-traffic (HAR) capture. For debugging flaky flows and verifying what the app actually sends, I wanted a full HTTP dump of a session. Playwright supports this natively at context creation (record_har_path / record_har_content), so it drops in cleanly:

• config keys har_enabled (bool) + har_content (omit/embed/attach), gated in create_session, **har_kwargs on the new_context(...) calls.
• HAR is per-session (<profile_dir>/traffic.har), flushed on context.close() — i.e. on webctl stop. Verified end-to-end: valid HAR 1.2 with embedded bodies.
• ~30 LOC across config.py / session_manager.py / cli/app.py. One limitation worth noting: Playwright decides HAR at context-creation, so it's enable-before-start, not a mid-session toggle. Live/on-demand capture would need CDP Network-domain buffering — bigger lift.

2. Attended-mode viewport. In attended mode the context is pinned to viewport=1280×720, so resizing the visible Chromium window grows the chrome around frozen content, and wide-layout screenshots come out narrow. Switching the attended branch to no_viewport=True lets content track the real window; plus an optional attended_window_size config key ("WxH") that appends --window-size. Default unset = upstream behavior, so it's non-breaking. (Open items before a clean PR: views/a11y.py reads page.viewport_size, which is nullable under no_viewport — needs a fallback — plus tests.)

Your questions:

• Browsers: Chromium is enough for us.
• Missing actions: current set covers our flows (upload included). No drag-and-drop need yet.
• Session Recorder: big +1 — human-to-Bash replay would be great for seeding reproducible agent scripts.
• Markdown output: also a pain point here; the DOM-to-Markdown is noisy on complex SPA layouts. Would gladly test iterations.

Let me know if you'd like PRs for either patch (HAR is the cleaner/smaller of the two).