Hubcap

Navigate & manage tabs: goto, back, forward, reload, new, close, tabs, version Read page info: title, url, info, source, meta, links, images, scripts, tables, forms, frames Query DOM: query, html, text, attr, value, count, visible, exists, bounds, styles, computed, layout, shadow, find, selection, caret Click & interact: click, dblclick, rightclick, tripleclick, clickat, hover, mouse, drag, fill, clear, type, press, focus, select, check, uncheck, setvalue, upload, dispatch, tap, swipe, pinch Scroll: scroll, scrollto, scrolltop, scrollbottom Wait: wait, waittext, waitgone, waitfn, waitload, waitnav, waiturl, waitidle, waitrequest, waitresponse Capture: screenshot, pdf Network: network, har, responsebody, intercept, block, throttle, console, errors Emulate: emulate, useragent, geolocation, offline, media, viewport, permission Storage: cookies, storage, session, clipboard Analyze: a11y, coverage, csscoverage, domsnapshot, listeners, metrics, stylesheets Profile: heapsnapshot, trace JavaScript: eval, evalframe, run, bridge Assert: assert (subcommands: text, title, url, exists, visible, count) Utility: retry, pipe, shell, record, setup, help Advanced: raw, dialog, highlight

Tab management

By default, all commands target the first tab. Use --target <id> to target a specific tab.

• hubcap new --wait <url> creates a new tab, navigates, and waits for load. Returns {"targetId": "...", "loaded": true}.
• Use --target <id> on subsequent commands to operate on that tab.
• When the user says "open a new tab and go to X", use hubcap new --wait <url> — do NOT use separate new + goto (goto would hit the first tab, not the new one).

# Open a new tab, wait for load, then work in it
TARGET=$(hubcap new --wait https://example.com | jq -r .targetId)
hubcap --target "$TARGET" screenshot --output page.png

Screenshots

• hubcap screenshot --output file.png captures the visible viewport
• hubcap screenshot --output file.png --full captures the full scrollable page
• hubcap screenshot --output file.png --selector '.element' captures a specific element

Use --full when the user wants to capture the entire page including content below the fold.

JavaScript evaluation

• hubcap eval '<expression>' supports top-level await (e.g. hubcap eval 'await fetch("/api/data")')
• hubcap run <file.js> also supports top-level await

Bridge (persistent two-way channel)

hubcap bridge keeps a bidirectional message channel open between stdin/stdout and JavaScript running in a page. The script gets send() and messages (async iterator) in scope. Messages flow as LDJSON.

# Echo messages back with transformation
echo '{"data":{"n":7}}' | hubcap bridge --target "$TARGET" '
  for await (const msg of messages) {
    send({doubled: msg.n * 2});
    break;
  }
'

Use --file to load the script from disk. Heartbeats ensure the JS exits if hubcap dies. Multiple bridges can coexist in the same tab.

Key patterns

# Chain commands
hubcap goto --wait https://example.com && hubcap screenshot --output page.png

# Parse JSON output
hubcap tabs | jq '.[].url'

# Wait then act
hubcap wait '.results' && hubcap text '.results'

# Mobile testing
hubcap emulate iPhone-12 && hubcap goto --wait https://example.com

# Retry flaky operations
hubcap retry click '#dynamic-button' --attempts 3 --interval 1s

Exit codes

Global flags

All commands accept: -port <n> (default 9222), -host <s> (default localhost), -timeout <d> (default 10s), -output <fmt> (json/ndjson/text), -quiet, -target <id>.

Docs

Full documentation: https://hubcap.tomyandell.dev/commands/