Manul Browser for VS Code

Write browser automation in plain English and run it from the editor.

STEP 1: Sign in
    NAVIGATE to https://www.saucedemo.com/
    FILL 'Username' field with 'standard_user'
    FILL 'Password' field with 'secret_sauce'
    CLICK 'Login' button
    VERIFY that 'Products' is present

This is the editor side of manul-browser: one engine, written in Go, that resolves elements by the label a person sees rather than by selector. The extension finds that engine, runs .hunt files with it, and shows what happened on the lines it happened to.

Works with whichever way you installed the engine

There is one engine and three ways to get it. The extension looks for all of them, in this order, and uses the first that actually runs:

manulBrowser.enginePath overrides the search. Manul: Select Engine lists everything that was found and lets you pick; the status bar shows which one is in use.

Different engine versions are handled by asking, not by assuming: the extension reads the installed engine's --help and leaves out options that build does not have, and it merges whatever manul schema reports into the command list, so a newer engine's verbs show up in completion without an extension update.

Hooks are picked up for you

A hook script holds what a .hunt file cannot say: custom controls, CALL handlers, and suite hooks such as before_all.

# manul_hooks.py
import manul

@manul.before_all
def login(ctx):
    ctx.set("token", fetch_token())     # {token} in every hunt

manul.serve_hooks()

The engine only uses a hook script it is given with --hooks. Run a hunt without the flag and {token} stays {token}. The extension passes the flag on every run:

• Which script — the nearest manul_hooks.py, manul_hooks.mjs, manul_hooks.js or manul_hooks.cjs at or above the hunt file, no higher than the workspace folder. manulBrowser.hooks.path names one explicitly; manulBrowser.hooks.enabled turns pickup off.
• Which interpreter — a Python script runs under the environment the engine came from, or the first environment in the workspace that has the manul package, not under whatever python happens to be on PATH. That mismatch is the usual reason a hook script "does not work": it dies on import manul before it can say anything.
• What it registers — the Hooks view lists each script's suite hooks, CALL handlers and custom controls, and jumps to them. In a hunt, CALL HOST <name> completes from that list and Go to Definition opens the handler; { completes the variables the script publishes.
• What is wrong with it — a script that never calls serve_hooks() / serveHooks() is flagged before you run it, because the engine would start it, wait, and give up.

Manul: Create Hook Script writes a starting point in Python or JavaScript.

One thing to know: each hunt file is run as its own engine process, so before_all and after_all run once per file, not once for a whole Test Explorer run. To run a folder as a single suite, right-click it in the explorer and choose Manul: Run in Terminal.

Running

• Run / Debug in the editor title, the CodeLens on the first line, the explorer context menu, or Ctrl+Alt+R.
• The Test Explorer lists every .hunt file with its STEP blocks, runs them — several at once if manulBrowser.run.workers says so — and reports each block.
• After a run every executed line carries its outcome and duration. Hover a line for the detail: the error, the candidates the engine ranked for that step and why the winner won, and the screenshot if one was taken.
• The Run panel in the Manul Browser side bar shows the engine, the hook script and the last result, and holds the options: browser, visible or headless, screenshots, retries, HTML report.

Options left on From config are not passed at all, so manul.config.json decides. That file gets completion and validation.

Debugging

Set breakpoints in the gutter and choose Debug. With no breakpoints the run pauses before every step.

Debug runs always show the browser window.

Writing

• Syntax highlighting, outline and formatting for .hunt files.
• Completion for every command, header and block, with placeholders.
• The Step Palette view: the same list, searchable, one click from the cursor. Manul: Insert Step… is the keyboard version.
• New Hunt File, Record a Hunt from a URL and Scan a URL into a Draft Hunt create files in your tests folder.

When something does not work

Run Manul: Check Setup (Doctor). It checks, and says in plain words: which engines were found and which one runs, whether that build has the options the extension needs, which hook script a run would get, and whether the interpreter for it can actually import the binding.

Settings

Requirements

• VS Code 1.90 or later.
• A manul-browser engine, 0.1.2 or later for everything described here. Older builds run, with the options they lack left out.
• Chrome, Chromium, Edge or Firefox for the engine to drive.

Development

npm install
npm test            # unit tests, no browser needed
npm run package     # builds the .vsix

A push to main whose package.json version has no GitHub Release yet gets one, with the .vsix attached: bumping the version is what releases.

src/core has no dependency on VS Code and holds everything worth testing in isolation: finding the engine, finding and reading hook scripts, mapping the engine's results back to lines. src/test/integration runs inside a real VS Code against a real engine; its header says how to start it.

License

Apache 2.0 — Copyright 2026 Oleksii Poliakov. See LICENSE.