Devloop

Go back one entry in the active pane's history (no-op if there's nothing to go back to). Pairs with browser_forward; use browser_navigate to go to a new URL.

Clear the active page's storage — cookies, localStorage, IndexedDB, cache, service workers — for the current origin. Set allOrigins to wipe the whole browser session. Use to log out or test a fresh-user flow; reload afterward.

Click the element matching a CSS selector (real mouse click). For keyboard keys use browser_press; to only hover use browser_hover.

Emulate a device/viewport on the active page. Pass a `device` preset, or custom `width`+`height`. `reset: true` restores a desktop viewport. For network conditions use browser_throttle instead.

Evaluate a JavaScript expression in the active page (via CDP) and return its value. Runs in the page's main world and can mutate page state. Prefer browser_snapshot for reading structure and browser_click/type for interactions.

Go forward one entry in the active pane's history (no-op if there's nothing to go forward to). Pairs with browser_back; use browser_navigate to go to a new URL.

Hover the pointer over an element to trigger hover menus/tooltips. Does not click — use browser_click for that.

Navigate the active pane to a URL (full page load). Returns the resolved URL and HTTP status. Use browser_back / browser_forward to move through history, browser_reload to refresh the same URL.

Press a single key (e.g. Enter, Escape, Tab, ArrowDown) on the page. For typing text use browser_type.

Reload the active pane's current page. Use browser_navigate to change URL instead.

Capture a PNG screenshot of the active page. To find/target elements, prefer browser_snapshot (returns selectors).

Scroll an element into view (pass selector), or scroll the window to coordinates (pass x and y). Provide a selector OR x+y, not both.

Set the value of a <select> (or input) and fire input/change events. For typing free text use browser_type.

Capture a structured snapshot of the active page: url, title, and the interactive + landmark elements (role, accessible name, value/state, heading level), each with a CSS selector `ref`. Pass a returned `ref` to browser_click / browser_type. Caps at 250 elements by default (raise `limit` for dense pages); `truncated:true` means the cap was hit. Prefer this over browser_screenshot to find and target elements reliably.

Throttle the active page's network conditions to a named profile. Use `none` to clear. For viewport/device emulation use browser_emulate.

Type text into the element matching a CSS selector (focuses, then types). For single keys use browser_press; for <select> use browser_select.

Wait until a CSS selector appears or text is present on the page (e.g. after a navigation or async render). Returns { ok, waitedMs } — ok=false on timeout. To wait for the network to go quiet instead, use browser_wait_for_idle.

Wait until network activity settles (no requests for idleMs). Returns { ok }. To wait for a specific element/text use browser_wait_for.

Clear the event buffer (irreversible). Call before reproducing an issue for a clean window. Note: repro clears for you unless clear=false.

Start a dev server and tee its logs into the buffer. Three ways to specify it: (1) `project` — a saved registry project (resolves cmd/cwd); (2) explicit `cmd`+`cwd`; (3) neither — `cwd` defaults to the server's directory and `cmd` is auto-detected from package.json. Explicit cmd/cwd override the project's. Stop it with dev_stop; check it with dev_status.

Report whether the dev server is running, plus its cmd/cwd/pid. Start/stop with dev_start/dev_stop.

Stop the running dev server (SIGTERM). Returns whether one was running. Start one with dev_start.

Triage what's broken right now: group repeated errors (browser console/page errors + server errors) with counts, list failed/4xx-5xx network requests, and return a one-line summary. Start here before digging through get_logs.

Export a shareable bug-report bundle as one JSON object: the diagnose summary, the timeline (logs), captured screenshots, a HAR of network, and repro steps if provided. For just the network log use export_har; for a quick triage use diagnose.

Export captured network requests as a HAR 1.2 document (importable into Chrome DevTools / Charles). Covers ALL requests (the full network ring, independent of DEVLOOP_NET_THRESHOLD — bodies kept for the curated subset: failures + status ≥ threshold). To browse requests in JSON use get_network.

Install a Chrome extension from a Web Store id or URL (downloads from the Web Store). Returns the updated list. (Cockpit only.) Remove with ext_remove.

List Chrome extensions (loaded + disabled): id, name, version, enabled. (Cockpit only.) Install with ext_install, toggle with ext_set_enabled.

Remove (uninstall/unload) a Chrome extension by id (irreversible for store extensions). (Cockpit only.) To keep it but turn it off, use ext_set_enabled.

Enable or disable a Chrome extension by id without uninstalling. Returns the updated list. (Cockpit only.) To uninstall entirely, use ext_remove.

Return recent events from the unified buffer (server stdout/stderr + browser console/network/pageerror), newest last. Filter by source, stream, grep, and tail incrementally with sinceSeq. Scope to one project's logs with `app`. For events around a specific moment use get_logs_around; to triage errors first use diagnose.

Return ALL events (server + browser) within +/- windowMs of a timestamp, time-ordered — the correlation tool. E.g. the browser console error and the backend stack trace from the same moment. Timestamps come from the `ts` field on any event.

List captured network requests (the full ring — every request, independent of DEVLOOP_NET_THRESHOLD). Each row has method/url/status/timing/headers (bodies for the curated subset). Unlike get_logs (curated timeline only); for an importable HAR file use export_har.

Build + launch the native dev build for the active pane (`expo run:ios` / `expo run:android`); output streams to the timeline. Cockpit-only, local build. Needs the native toolchain (Xcode for iOS; Android SDK + `adb` + a JDK + `$ANDROID_HOME` for Android). If the Android toolchain isn't set up it returns `started:false` with a checklist of exactly what to install, instead of failing cryptically. After it boots, use native_open to drive it.

Close the active native target; browser_* route back to the pane's web content. (Cockpit only.) Open one with native_open.

Report native readiness without building or opening anything: iOS simulator interactions, Android device interactions, and the Android local build toolchain — each a ✓/✗ checklist with the fix for anything missing. Cockpit-only, read-only. Re-probes live each call — run it after installing a missing tool to confirm it's resolved before native_build / native_open, or proactively before a long build. Returns { ios, androidInteractions, androidBuild }, each { ready, checks, summary }. native_build to build; native_open to drive a device.

Open a native target for the active pane (Expo/React Native): the iOS simulator or the Android device mirror. browser_* (snapshot/click/type/scroll/press/screenshot) then drive the native app via idb/adb, and JS + native logs stream to the timeline. Cockpit-only (needs the Electron app + a booted simulator/emulator). Returns ok:false with a reason if the device/tooling isn't ready. Build the app first with native_build; close with native_close.

Close a browser pane by id (irreversible). To detach a pane into its own window instead, use pane_pop. (Cockpit only.)

List browser panes (multi-target): each has id, url, active. The active pane is what browser_*/repro act on. Switch with pane_select. (Cockpit only.)

Open a new browser pane and make it active. Optionally navigate it to a URL and/or scope it to a project. Pass a saved `project` (resolves cwd/cmd/url/name), or explicit `cwd`/`cmd`/`label`, to scope the pane — cwd isolates its storage partition and names it. With none of these it's a blank, unscoped pane. To switch panes use pane_select; to close use pane_close. (Cockpit only.)

Detach a pane into its own standalone window (to view targets side-by-side). Closing that window re-docks the pane. pane_close removes it. (Cockpit only.)

Make a pane active so subsequent browser_*/repro calls target it. Find ids with pane_list. (Cockpit only.)

Set a pane's display label (the tab name). Also what `app` filters match in get_logs/diagnose. (Cockpit only.)

Save (or replace) a project in the registry so you can dev_start it by name. Replaces any existing project with the same name.

List saved projects (name, cwd, cmd, url) from the registry. Add with project_add, remove with project_remove.

Remove a saved project from the registry by name (irreversible; doesn't stop a running server). List names with project_list.

One-shot reproduce-and-correlate: clear the buffer (unless clear=false), perform one action OR a sequence in order, wait for async console/network/server events to land, then return EVERYTHING that happened on both sides plus per-step results and an errors summary. Use a sequence for flows like navigate → click → type → click submit.