Skip to content
Back to skills

Browser

ASecurity

Fast, reliable browser automation via CDP relay. Use for real page interaction and extraction with minimal calls.

  • 866 stars
  • 0 votes
  • 0 copies
  • 8 views
  • Added February 8, 2026
devopsbashexpress

Works with

  • cli

Security analysis

A100/100

Scanned February 12, 2026

npx -y skills add wecode-ai/Wegent --skill browser --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Browser?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Browser
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/wecode-ai-browser/badge)](https://www.skillsdirectory.com/skills/wecode-ai-browser)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: browser
description: Fast, reliable browser automation via CDP relay. Use for real page interaction and extraction with minimal calls.
---

# Browser Control Skill

## Hard Rules

1. Start with the intended action directly (`navigate`/`open`/`act`/`evaluate`). Do not run `status` as a pre-check.
2. Keep calls minimal: target `<= 8` per task (hard cap `12`, unless user explicitly asks for deep exploration).
3. Prefer `evaluate` for extraction. Combine data into one comprehensive extraction when possible (avoid repeated partial extraction).
4. `snapshot` is for refs/structure only. Budget: at most `1 initial + 1 final` per page.
5. If `Ref not found`: do not retry stale ref. Take one fresh `snapshot`, retry once, then stop if still failing.
6. `act.wait` is strict: do not use `timeMs` by default. Use condition waits (`text`/`selector`/state). Use fixed-time waits only when user explicitly asks or no condition is possible.
7. `screenshot` is strict: default `0`. Use only if user explicitly requests it or visual evidence is strictly necessary after text/DOM extraction. Max `1` screenshot per task unless user asks for more.
8. Keep tab context stable: once `targetId` is known, pass it in subsequent actions when supported.

## Execution Order

1. Direct action first (`navigate`/`open` or immediate `act`/`evaluate`).
2. Use `snapshot` only if refs are needed for interaction.
3. Execute interactions with minimal `act` calls (merge related input when possible).
4. Use one final verification step only if needed (`evaluate` preferred; `snapshot` optional).

## Connection Handling

Connection recovery is built into the tool. On connection failure, let the tool auto-attach/launch/retry once. If still disconnected, stop and instruct the user to install/connect the extension.

## Minimal CLI Usage

```bash
~/.wegent-executor/bin/browser-tool '<json>'
```

## Quick Examples

```bash
# Navigate directly
~/.wegent-executor/bin/browser-tool '{"action":"navigate","url":"https://example.com"}'

# Snapshot only when refs are needed
~/.wegent-executor/bin/browser-tool '{"action":"snapshot","interactive":true}'

# Act on ref
~/.wegent-executor/bin/browser-tool '{"action":"act","request":{"kind":"click","ref":"e1"}}'

# Comprehensive extraction in one evaluate
~/.wegent-executor/bin/browser-tool '{"action":"evaluate","expression":"(() => ({title:document.title,url:location.href}))()"}'
```

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.