About

Imported from OpenWorkai/openwork-skillhub (_pkgs/agent-browser-core/SKILL.md). Install upstream with npx skills add OpenWorkai/openwork-skillhub --skill agent-browser-core. Copyright stays with the author.

Agent Browser Core

Name: agent-browser-core
Author: OpenWorkai

What this skill is for

Drive a real browser from the command line to automate web work — navigation, form filling, data extraction, screenshots — using structured commands instead of brittle selectors. The CLI produces DOM snapshots and stable element refs so the agent can target precisely and verify outcomes.

How to get oriented

references/agent-browser-overview.md — install, architecture, core concepts.
references/agent-browser-command-map.md — command categories and flags.
references/agent-browser-safety.md — high-risk controls and safe-mode rules.
references/agent-browser-workflows.md — recommended automation patterns.
references/agent-browser-troubleshooting.md — common failures and fixes.

Working model

Snapshot first. Capture page state before acting so you know what exists.
Act via refs. Target elements by their stable ref, not raw CSS/XPath.
Snapshot again. Re-read the DOM after any change to confirm it happened.
Use --json when scripting or piping output to other tools.
Wait deliberately. Check load state before interacting; never assume readiness.
Clean up. Close tabs/sessions when finished to free the browser.

Safe-mode guardrails (default)

Do not enable eval, --allow-file-access, a custom --executable-path, or arbitrary --args without explicit user approval. Avoid network route, set credentials, and cookie/storage writes unless the task truly needs them. Prefer an allowlist of target domains and block localhost / private-network destinations.

Sessions & authentication

By default the CLI launches a fresh, isolated browser instance — it does not inherit the user's existing logins.
To reuse a login across runs, persist state with --user-data-dir.
To attach to an already-running Chrome, use --cdp (handy for auth that's hard to automate).
If you hit OAuth or CAPTCHA walls that block automation, stop and hand off to the user rather than retrying forever.

Orchestrating complex tasks

State a clear plan, then execute step by step, reporting each step's outcome before continuing (batch downloads, multi-page scraping, long forms).
Pace actions and add natural delays to avoid anti-bot triggers.
Ask for explicit confirmation before sensitive acts: entering passwords, bulk downloads, deletions.

Verifying results

If extracted fields come back empty, 0.00, or otherwise abnormal, prompt the user to confirm the source or page state.
After any write (click/fill/submit), re-snapshot or get to prove the DOM actually changed.
Never claim "done" or "success" unless the underlying state substantively changed.

Use it

Copy one of these into your project. Installing also returns the manifest and these snippets.

yaml

targets:
  - https://api.opensmartroute.ai/api/v1/registry/openworkai-openwork-skillhub-agent-browser-core/manifest   # or paste the manifest below

Manifest

An Open Capability Manifest: the router reads it to know what this does, what it costs and when to pick it.

openworkai-openwork-skillhub-agent-browser-core.ocm.jsonjson

{
  "ocm": "1",
  "id": "openworkai-openwork-skillhub-agent-browser-core",
  "kind": "skill",
  "name": "agent-browser-core",
  "description": "AI-friendly web automation driven by the agent-browser CLI — DOM snapshots, stable element refs, and structured commands for reliable browser control.",
  "publisher": "OpenWorkai",
  "version": "1.0.0",
  "capabilities": {
    "domains": [
      "general"
    ],
    "tags": [
      "skill-md",
      "web-automation",
      "browser",
      "scraping",
      "ui-interaction",
      "github"
    ],
    "languages": [
      "en"
    ]
  },
  "quality_prior": 0.6,
  "examples": [
    "AI-friendly web automation driven by the agent-browser CLI — DOM snapshots, stable element refs, and structured commands for reliable browser control."
  ],
  "primary": false,
  "metadata": {
    "source": {
      "provider": "github",
      "repository": "https://github.com/OpenWorkai/openwork-skillhub",
      "path": "_pkgs/agent-browser-core/SKILL.md",
      "ref": "b23bbf0769378ab31220ea9d0680aac1946df721",
      "url": "https://github.com/OpenWorkai/openwork-skillhub/blob/b23bbf0769378ab31220ea9d0680aac1946df721/_pkgs/agent-browser-core/SKILL.md",
      "key": "OpenWorkai/openwork-skillhub/_pkgs/agent-browser-core/SKILL.md"
    }
  },
  "instructions": "# Agent Browser Core\n\n## What this skill is for\nDrive a real browser from the command line to automate web work — navigation, form filling, data extraction, screenshots — using structured commands instead of brittle selectors. The CLI produces DOM **snapshots** and stable element **refs** so the agent can target precisely and verify outcomes.\n\n## How to get oriented\n- `references/agent-browser-overview.md` — install, architecture, core concepts.\n- `references/agent-browser-command-map.md` — command categories and flags.\n- `references/agent-browser-safety.md` — high-risk controls and safe-mode ",
  "cost": {
    "context_tokens": 678
  }
}

Fetch it by URL: GET /api/v1/registry/openworkai-openwork-skillhub-agent-browser-core/manifest?version=1.0.0

Reviews

Star ratings from people who tried it. One review per account; edit yours any time.

No reviews yet. Install it, try it, and be the first to rate it.