Chrome Web Automation

Use when the user wants browser automation in an existing Chrome session: click, fill, inspect tabs, capture screenshots, or debug web flows.

Security Scan

The skill's purpose (drive an existing Chrome session) matches its instructions at a high level, but it omits how it will access the user's browser and provides no privacy/consent guardrails — making its real-world behaviour unclear and potentially risky.

Purpose & Capability

Name, description and SKILL.md consistently describe browser automation of an existing Chrome session. However, the skill declares no dependencies, credentials, install steps, or connector details for accessing a live Chrome session (no extension/native host/Chrome DevTools Protocol endpoint). This is plausible if the platform already supplies a Chrome automation tool, but the skill itself is ambiguous about how access is achieved.

Instruction Scope

Runtime instructions direct the agent to inspect tabs, click, type, capture screenshots, and produce transcripts — all actions that can capture sensitive data. The SKILL.md does not instruct to obtain explicit user consent, avoid capturing credentials or other secrets, or sanitize/redact sensitive artifacts before storing/transmitting. It also does not specify any external endpoints for data exfiltration, but the lack of guardrails leaves broad discretion to the agent.

Install Mechanism

Instruction-only skill with no install spec or code — lowest install risk. Nothing is written to disk by the skill as provided.

Credentials

No environment variables, credentials, or config paths are requested. This is proportionate to the stated purpose; however, the omission of connector details (extension, CDP URL, etc.) reduces transparency about required access.

Persistence & Privilege

Skill is not always-enabled and does not request elevated/persistent privileges. It does allow normal autonomous invocation (platform default), which increases its potential impact combined with the instruction-scope concerns above.

What to consider before installing

This skill claims to drive your live Chrome session but doesn't say how it will connect (extension, CDP endpoint, native host, etc.) and gives no privacy rules. Before installing or running it: (1) confirm your platform provides a trusted Chrome integration the skill expects; (2) require explicit user consent before inspecting tabs or taking screenshots; (3) avoid running this on pages containing passwords, payment data, or other sensitive info; (4) ask the publisher for the exact connector mechanism and any network endpoints the skill will use; and (5) prefer skills that declare dependencies and include redaction/privacy guardrails.

Use this skill before any browser interaction that depends on the user's live Chrome session.

Workflow

1. Start by inspecting current tabs and browser context.
2. Navigate or select the right tab.
3. Snapshot the page before interacting.
4. Click, type, submit, and screenshot using stable references.
5. Re-snapshot after navigation or large DOM updates.

Best Fits

• Web app debugging
• Form filling
• Screenshot capture
• Console or tab inspection
• Reproducing UI issues in a user browser session

Guardrails

• Re-snapshot when element references go stale.
• Do not assume current tab state without checking.
• Prefer explicit browser actions over ambiguous natural-language leaps.

Example Requests

• Open the current Chrome session and reproduce this checkout bug.
• Fill the form, capture screenshots, and show me where validation breaks.

Inputs

• Target site or current tab context
• Desired browser actions
• Artifacts to capture

Outputs

• Browser interaction transcript
• Screenshots or evidence
• Reproduction notes

Success Criteria

• The correct tab or page was inspected.
• Interactions were completed against fresh page state.
• Useful artifacts were captured when needed.

Non-Goals

• Synthetic browser testing unrelated to the user browser context

Source Provenance

Derived from src/skills/bundled/claudeInChrome.ts.