Overview

Omnibot

Browser infrastructure for AI agents.

Omnibot lets LLM clients read, inspect, operate, debug, and verify the user’s real Chromium browser: real tabs, real login state, real extensions, and explicit tab targeting through a local daemon plus CLI.

This page is a product-level overview of the current Omnibot surface and the roadmap. For detailed implementation status, see Product Roadmap.

Quick Start

Install the CLI:

npm install -g @omniaibot/omnibot

Check the local runtime and extension connection:

omnibot doctor
omnibot status
omnibot tabs

Install agent skills so AI clients learn the right operation patterns:

omnibot skills install --agent codex
omnibot skills install --agent claude
omnibot skills install --agent opencode
omnibot skills install --agent hermes --profile nuwa

Observe a tab, act once, then verify:

OMNIBOT_SESSION_TOKEN=checkout omnibot snapshot -i --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot click --tab-id <TAB_ID> @e4
OMNIBOT_SESSION_TOKEN=checkout omnibot snapshot -i --tab-id <TAB_ID>

What It Is

Omnibot is a local bridge between agent runtimes and a browser the user is already using. It is designed for tasks where ordinary HTTP fetches or static scraping are not enough:

• Authenticated apps and dashboards.
• Client-rendered pages.
• Sites with rich controls, popups, comboboxes, modals, dialogs, and editors.
• Workflows that need screenshots, console logs, network evidence, or tab-aware state.
• Agent operations that should be explicit, inspectable, and recoverable.

The core product rule is simple:

Observe -> Act -> Verify

Agents should inspect the page, take one intentional action, and verify the expected state change before continuing.

Current Capabilities

Runtime And Distribution

Omnibot v2 includes a local CLI and daemon runtime. The CLI can auto-start or connect to the daemon, while the browser extension maintains the browser-side connection.

Current surface:

• omnibot doctor, status, start, stop, and run.
• Local daemon API on 127.0.0.1:18764.
• Browser extension WebSocket connection on 127.0.0.1:18765.
• HTTP long-poll fallback on 127.0.0.1:18766.
• npm distribution through a main CLI package and platform-specific binary packages.
• Packaged agent skills and omnibot skills install.
• License status, persisted activation state, usage count, and extension status refresh.

Reading And Page Understanding

Omnibot gives agents several ways to inspect browser state, depending on intent.

Use read when the goal is content understanding:

OMNIBOT_SESSION_TOKEN=research omnibot read --screens 5 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot read --screens 3 https://example.com/article

Use snapshot -i when the next step is browser operation:

OMNIBOT_SESSION_TOKEN=checkout omnibot snapshot -i --tab-id <TAB_ID>

Current reading features include:

• Clean rendered-page reading with screen budgets and JSON/text output.
• Accessibility-tree snapshots with actionable @eN refs.
• Interactive filtering, compact/depth/url options, and selector scoping.
• Popup/modal controls and combobox option probing.
• Rich text editor refs for contenteditable editors.
• Narrow state reads through get and is.
• Semantic find locators for role, text, label, placeholder, alt, title, test id, and nth matching.

User-Like Browser Operations

Omnibot can operate the page through refs, semantic locators, selectors, DOM fallback, and coordinate fallback.

Common commands:

OMNIBOT_SESSION_TOKEN=checkout omnibot click --tab-id <TAB_ID> @e4
OMNIBOT_SESSION_TOKEN=checkout omnibot fill --tab-id <TAB_ID> @e2 "a@b.com"
OMNIBOT_SESSION_TOKEN=search omnibot type --tab-id <TAB_ID> @e3 "omnibot"
OMNIBOT_SESSION_TOKEN=checkout omnibot wait --url "/dashboard" --tab-id <TAB_ID>

Implemented operation surface includes:

• click, dblclick, fill, type, press, hover, focus.
• select, check, uncheck, scroll, scrollintoview.
• Element-level drag.
• File upload through CDP, DOM node fallback, and JS DataTransfer fallback.
• Navigation, tab lifecycle, window creation, frame targeting, history, reload, and pushstate.
• Waits for selector, text, URL, load state, JavaScript function, and fixed time.
• JSON and file-driven batch commands.

Debugging And Evidence

Omnibot treats evidence as part of the browser workflow, not an afterthought.

Implemented evidence tools include:

• Screenshots: normal, full-page, annotated, PNG/JPEG, base64, and file output.
• Console capture: logs, error path, and clear.
• JavaScript dialogs: logs, clear, accept, dismiss, and prompt text handling.
• Network capture: start, stop, clear, logs, and summary.
• Raw CDP calls for last-resort inspection.

Example:

OMNIBOT_SESSION_TOKEN=debug omnibot screenshot --tab-id <TAB_ID> -o /tmp/page.png
OMNIBOT_SESSION_TOKEN=debug omnibot console logs --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug omnibot network summary --tab-id <TAB_ID>

Fallback Surfaces

The happy path is semantic locator or snapshot ref. When that fails, Omnibot exposes lower-level surfaces instead of forcing agents directly into raw JavaScript.

Current fallback tiers include:

• Selector operations.
• DomCUA visible-node operations.
• Coordinate mouse operations through CDP input events.
• Clipboard read/write through page context.
• Viewport get/set.
• Asset metadata list/export.
• Browser/session metadata.
• Site-skill exploration and generated operation memory.

Architecture

Omnibot has four main pieces:

1. The agent client calls omnibot commands.
2. The CLI connects to or starts the local daemon.
3. The daemon maintains browser sessions and dispatches actions.
4. The Chromium extension connects real browser tabs to the daemon.

Page-state commands require explicit tab targeting:

OMNIBOT_SESSION_TOKEN=<workflow> omnibot snapshot -i --tab-id <TAB_ID>

OMNIBOT_SESSION_TOKEN isolates workflow state such as refs, traces, recordings, aliases, and temporary resources. --tab-id chooses the actual browser tab. Refs such as @e4 are tab-scoped and should not be reused across tabs.

Typical Workflows

Inspect And Click

omnibot tabs
OMNIBOT_SESSION_TOKEN=checkout omnibot snapshot -i --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot click --tab-id <TAB_ID> @e4
OMNIBOT_SESSION_TOKEN=checkout omnibot snapshot -i --tab-id <TAB_ID>

Read An Authenticated Page

OMNIBOT_SESSION_TOKEN=research omnibot read --screens 5 --tab-id <TAB_ID>

Debug A Broken Flow

OMNIBOT_SESSION_TOKEN=debug omnibot console clear --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug omnibot network clear --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug omnibot network start --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug omnibot click --tab-id <TAB_ID> @e7
OMNIBOT_SESSION_TOKEN=debug omnibot network stop --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug omnibot console logs --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug omnibot network summary --tab-id <TAB_ID>

Operate Rich Controls

OMNIBOT_SESSION_TOKEN=editor omnibot snapshot -i --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=editor omnibot fill --tab-id <TAB_ID> @e12 "First paragraph.\n\nSecond paragraph."

snapshot -i can expose popup controls, custom combobox options, and rich text editors that are missing from the normal accessibility tree.

Command Examples

This section maps the product surface to concrete commands. Most page-state commands use both OMNIBOT_SESSION_TOKEN and --tab-id <TAB_ID> so agents keep workflow memory separate from the browser tab they are operating.

Runtime, Health, And Version

omnibot doctor
omnibot status
omnibot start
omnibot stop
omnibot run
omnibot version
omnibot license status

Tab Discovery And Targeting

omnibot tabs
OMNIBOT_SESSION_TOKEN=research omnibot tab list
OMNIBOT_SESSION_TOKEN=research omnibot tab new https://docs.example.com --label docs
OMNIBOT_SESSION_TOKEN=research omnibot tab close docs

Reading Page Content

OMNIBOT_SESSION_TOKEN=research omnibot read --screens 5 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot read --json --screens 5 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot read --screens 3 https://example.com/article

Snapshot Refs And UI Structure

OMNIBOT_SESSION_TOKEN=checkout omnibot snapshot -i --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot snapshot -i -c -d 4 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot snapshot -i -s "#main" -u --tab-id <TAB_ID>

Narrow State Reads

OMNIBOT_SESSION_TOKEN=research omnibot get title --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot get url --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot get text "#main" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot get html "#main" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot get value "input[name=email]" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot get attr "a.login" href --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot get count ".item" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot get box "#submit" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot get styles "#submit" --tab-id <TAB_ID>

Element State Checks

OMNIBOT_SESSION_TOKEN=checkout omnibot is visible "#submit" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot is enabled "#submit" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot is checked "#agree" --tab-id <TAB_ID>

Semantic Find

OMNIBOT_SESSION_TOKEN=checkout omnibot find role button --name "Submit" --action click --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot find text "Sign in" --action click --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot find label "Email" --action fill --action-value "a@b.com" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot find placeholder "Search" --action type --action-value "omnibot" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot find testid submit-button --action click --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot find nth ".card" 2 --action click --tab-id <TAB_ID>

Click, Fill, Type, And Keyboard

OMNIBOT_SESSION_TOKEN=checkout omnibot click --tab-id <TAB_ID> @e4
OMNIBOT_SESSION_TOKEN=checkout omnibot dblclick --tab-id <TAB_ID> @e7
OMNIBOT_SESSION_TOKEN=checkout omnibot fill --tab-id <TAB_ID> @e2 "a@b.com"
OMNIBOT_SESSION_TOKEN=search omnibot type --tab-id <TAB_ID> @e3 "omnibot"
OMNIBOT_SESSION_TOKEN=search omnibot press Enter --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=form omnibot keyboard type "hello" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=form omnibot keyboard inserttext "hello" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=form omnibot keydown Shift --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=form omnibot keyup Shift --tab-id <TAB_ID>

Hover, Focus, Select, And Checkbox State

OMNIBOT_SESSION_TOKEN=menu omnibot hover --tab-id <TAB_ID> @e5
OMNIBOT_SESSION_TOKEN=form omnibot focus "input[name=email]" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=form omnibot select @e5 "US" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=form omnibot check @e6 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=form omnibot uncheck @e6 --tab-id <TAB_ID>

Scroll And Drag

OMNIBOT_SESSION_TOKEN=read omnibot scroll down 500 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=read omnibot scroll down 500 --selector "#list" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=read omnibot scrollintoview @e7 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=board omnibot drag @e8 @e9 --tab-id <TAB_ID>

Upload Files

OMNIBOT_SESSION_TOKEN=form omnibot upload "input[type=file]" /path/to/file.png --tab-id <TAB_ID>

Navigation, History, Windows, And Frames

OMNIBOT_SESSION_TOKEN=research omnibot navigate https://example.com
OMNIBOT_SESSION_TOKEN=research omnibot navigate https://example.com --same-tab --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot open https://example.com
OMNIBOT_SESSION_TOKEN=research omnibot goto https://example.com --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot close <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot back --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot forward --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot reload --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot pushstate /dashboard --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot window new
OMNIBOT_SESSION_TOKEN=research omnibot frame main

Waits

OMNIBOT_SESSION_TOKEN=checkout omnibot wait "#ready" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot wait "#spinner" --state hidden --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot wait --text "Welcome" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot wait --url "/dashboard" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot wait --load domcontentloaded --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot wait --load networkidle --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot wait --fn "window.appReady === true" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=checkout omnibot wait 500 --tab-id <TAB_ID>

Batch Commands

OMNIBOT_SESSION_TOKEN=research omnibot batch '[{"cmd":"cdp","method":"Runtime.evaluate","params":{"expression":"document.title","returnByValue":true}}]' --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot batch --file /tmp/omnibot-batch.json --tab-id <TAB_ID>

Screenshots

OMNIBOT_SESSION_TOKEN=debug-checkout omnibot screenshot --tab-id <TAB_ID> -o /tmp/omni-shot.png
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot screenshot --annotate --tab-id <TAB_ID> -o /tmp/omni-annotated.png

Console Evidence

OMNIBOT_SESSION_TOKEN=debug-checkout omnibot console logs --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot console errors --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot console clear --tab-id <TAB_ID>

Network Evidence

OMNIBOT_SESSION_TOKEN=debug-checkout omnibot network clear --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot network start --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot network stop --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot network logs --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot network summary --tab-id <TAB_ID>

JavaScript Dialogs

OMNIBOT_SESSION_TOKEN=dialog omnibot dialog logs --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=dialog omnibot dialog clear --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=dialog omnibot dialog handle accept --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=dialog omnibot dialog handle dismiss --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=dialog omnibot dialog handle accept --text "prompt value" --tab-id <TAB_ID>

Clipboard

OMNIBOT_SESSION_TOKEN=research omnibot clipboard read --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=research omnibot clipboard write "text" --tab-id <TAB_ID>

Viewport

OMNIBOT_SESSION_TOKEN=visual omnibot viewport get --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=visual omnibot viewport set 1280 720 --tab-id <TAB_ID>

Assets

OMNIBOT_SESSION_TOKEN=extract omnibot assets list --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=extract omnibot assets export -o /tmp/assets.zip --tab-id <TAB_ID>

DomCUA Fallback

OMNIBOT_SESSION_TOKEN=repair omnibot dom visible --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=repair omnibot dom click n1 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=repair omnibot dom dblclick n1 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=repair omnibot dom scroll n1 --dy 800 --tab-id <TAB_ID>

Coordinate Mouse Fallback

OMNIBOT_SESSION_TOKEN=repair omnibot mouse click --x 100 --y 200 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=repair omnibot mouse click --x 100 --y 200 --click-count 2 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=repair omnibot mouse move --x 50 --y 60 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=repair omnibot mouse scroll --x 100 --y 200 --dy 500 --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=repair omnibot mouse drag --from-x 0 --from-y 0 --to-x 100 --to-y 0 --fast --tab-id <TAB_ID>

JavaScript And Raw CDP Fallback

OMNIBOT_SESSION_TOKEN=repair omnibot execute-js "return location.href" --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=repair omnibot execute-js --file /tmp/repair.js --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot cdp Runtime.evaluate '{"expression":"document.title"}' --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot cdp DOM.getDocument '{"depth":1}' --tab-id <TAB_ID>

Human Verification Inspection

OMNIBOT_SESSION_TOKEN=verify omnibot verify inspect --tab-id <TAB_ID>
OMNIBOT_SESSION_TOKEN=verify omnibot verify inspect --tab-id <TAB_ID> --no-image

Visibility Modes

omnibot visibility status
omnibot visibility set background
omnibot visibility set visible
omnibot visibility launch headless --user-data-dir /tmp/omnibot-headless

Browser And Session Metadata

OMNIBOT_SESSION_TOKEN=research omnibot browser list
OMNIBOT_SESSION_TOKEN=research omnibot browser current
OMNIBOT_SESSION_TOKEN=research omnibot browser claim 123
OMNIBOT_SESSION_TOKEN=research omnibot browser release 123
OMNIBOT_SESSION_TOKEN=checkout omnibot session name "checkout"
OMNIBOT_SESSION_TOKEN=checkout omnibot session list

Record, Replay, And Trace

OMNIBOT_SESSION_TOKEN=debug-checkout omnibot record start
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot record stop -o flow.json
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot replay flow.json
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot trace start
OMNIBOT_SESSION_TOKEN=debug-checkout omnibot trace stop -o trace.zip

Skills And Site Memory

omnibot skills path
omnibot skills install --agent codex
omnibot skills list-sites
omnibot skills install --agent opencode --site xhs
omnibot skills install --agent opencode --all-sites
omnibot explore run https://www.xiaohongshu.com --name xhs --goal "发布图文笔记"
omnibot explore list
omnibot explore show xhs
omnibot explore promote xhs --goal publish-note --version 2026-06-17T21-30-00Z

Partial Or Experimental Surfaces

These features exist, but should be treated with care:

• Human verification inspection can detect and extract metadata for NetEase Yidun captcha widgets, but does not solve captchas.
• Realistic mouse drag is experimental; the stable drag path is the fast linear mode.
• Console and network capture surfaces exist, but callers should verify filtering behavior and extension version compatibility.
• Record, replay, and trace commands have foundational payload/file shapes, but are not yet a complete durable macro workflow.
• Visibility modes model visible, background, dedicated-profile, and headless states, but dedicated/headless launch is not production-ready.
• Asset export currently focuses on metadata and zip output, not full offline mirroring or HAR-quality archives.

Roadmap

Omnibot’s next milestones expand from a stable local real-browser bridge into richer automation memory, observability, and platform depth.

M1: Stabilize The Local Browser Surface

Focus:

• Keep snapshot + click/fill/type/wait as the primary workflow.
• Expand end-to-end coverage for popup controls, rich text editors, upload, network/dialog capture, and explicit tab targeting.
• Keep skills, command docs, and release packages aligned.

M2: Finish Recording And Trace

Planned:

• Wire action dispatch into consistent record and trace events.
• Define a stable replay schema.
• Support more replayable browser actions.
• Add trace export validation and a readable trace report.

M3: Dedicated Profile And Visibility Launch

Planned:

• Implement actual dedicated-profile browser launch.
• Document login-state and extension-loading limits.
• Decide whether headless mode becomes supported or remains experimental.

M4: Observability And Archive Depth

Planned:

• Improve console and network filtering.
• Add richer asset export and HAR-style summaries.
• Evaluate whether trace viewing belongs in CLI form or a separate UI.

M5: Platform Expansion

Future product directions:

• Remote daemon as a supported mode.
• Hosted/team workflows and policy controls.
• Cloud browser providers and alternative browser backends.
• Mobile automation research.
• Dashboard, billing/payment integration, and account management.

Planned Capability Gaps

The following are intentionally not complete today:

• Full Playwright-compatible locator and action parity.
• Stable DomCUA node lifecycles across DOM mutations and frames.
• Full CUA policy layer and confirmation workflow.
• Robust cross-origin iframe introspection.
• Download management, PDF export, full page archive, and full HAR export.
• Persistent cookie/storage/state vault and auth profile management.
• Profiler, Web Vitals, React/Vue component introspection, and performance timelines.
• Streaming event feed for long-running operations.
• AI-assisted captcha solving.

When To Use Omnibot

Use Omnibot when browser runtime state matters:

• The user is already logged in.
• The page is rendered by client-side JavaScript.
• The task needs real extension behavior.
• The agent must click, fill, inspect, or verify UI state.
• Evidence such as screenshots, console logs, dialogs, or network activity matters.

Do not use Omnibot when a static file, an API call, or ordinary HTTP fetch is enough. Omnibot is for real browser state.

Design Principles

• Reliability over convenience.
• Explicit state over implicit state.
• Pattern over command memorization.
• Observe, act, then verify.
• Prefer semantic locators and refs before selectors.
• Fall back deliberately, with evidence.
• Keep structured results on stdout and diagnostics on stderr.