How do I install the Browse Web skill?

Run npx skillmds@latest add hiai-gg/browse-web in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Browse Web safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Browse Web?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Browse Web free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Browse Web?

HiAi-gg (@hiai-gg) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Browse Web by hiai-gg — Agent Skill

Name: Browse Web
Author: HiAi-gg

Browse Web

Browse and interact with web pages using the agent-browser CLI: open pages, read accessible snapshots, click by ref, fill forms, navigate, and read state. Use when the agent needs to inspect or operate a real web page or web application.

HiAi-gg Updated Sep 17, 2026

File contents

Browse and Interact with Web Pages

Use this skill when an agent must open, read, or operate a real web page with the agent-browser CLI. This skill is deliberately thin: the CLI already provides an agent-native workflow (snapshots, refs, state), so this skill teaches when and how to use it safely — it does not re-document the CLI.

Workflow

1. Confirm the browser is installed

Run agent-browser doctor once before starting. If it reports a missing browser, run agent-browser install (the CLI's own supported install mechanism). Do not install system Chrome or use sudo.

2. Open the target page

agent-browser open "<url>"

The page loads in a persistent browser session. Confirm the title/page rendered before proceeding.

3. Read the page snapshot

agent-browser read

The output is an accessible text snapshot with refs (e.g. [12]). Treat refs as stable selectors for the current session. Read the snapshot before acting — never guess selectors from memory.

4. Interact by ref or selector

agent-browser click "<ref-or-selector>"
agent-browser fill "<ref-or-selector>" "<value>"
agent-browser type "<ref-or-selector>" "<text>"
agent-browser press "Enter"

Prefer the refs from the snapshot you just read. After each interaction, re-read the page (agent-browser read) to confirm the state changed as expected.

5. Navigate

agent-browser open "<next-url>"
# or click a link/button, then re-read

6. Read state and finish

Re-read the page to confirm the final state.
Close the session when done: agent-browser close.

Guardrails

Read the snapshot before acting; never guess selectors.
Default to reading (agent-browser read) over mutating.
Only fill/submit forms the user asked about; never submit destructive or payment forms without explicit intent.
Close the browser session when finished.
Do not print credentials or session cookies into responses.

---name: browse-webdescription: Browse and interact with web pages using the agent-browser CLI: open pages, read accessible snapshots, click by ref, fill forms, navigate, and read state. Use when the agent needs to inspect or operate a real web page or web application.---# Browse and Interact with Web PagesUse this skill when an agent must open, read, or operate a real web page withthe `agent-browser` CLI. This skill is deliberately thin: the CLI alreadyprovides an agent-native workflow (snapshots, refs, state), so this skillteaches *when and how* to use it safely — it does not re-document the CLI.## Workflow### 1. Confirm the browser is installedRun `agent-browser doctor` once before starting. If it reports a missingbrowser, run `agent-browser install` (the CLI's own supported installmechanism). Do not install system Chrome or use sudo.### 2. Open the target page```bashagent-browser open "<url>"```The page loads in a persistent browser session. Confirm the title/page renderedbefore proceeding.### 3. Read the page snapshot```bashagent-browser read```The output is an accessible text snapshot with refs (e.g. `[12]`). Treat refsas stable selectors for the current session. Read the snapshot before acting —never guess selectors from memory.### 4. Interact by ref or selector```bashagent-browser click "<ref-or-selector>"agent-browser fill "<ref-or-selector>" "<value>"agent-browser type "<ref-or-selector>" "<text>"agent-browser press "Enter"```Prefer the refs from the snapshot you just read. After each interaction, re-readthe page (`agent-browser read`) to confirm the state changed as expected.### 5. Navigate```bashagent-browser open "<next-url>"# or click a link/button, then re-read```### 6. Read state and finish- Re-read the page to confirm the final state.- Close the session when done: `agent-browser close`.## Guardrails- Read the snapshot before acting; never guess selectors.- Default to reading (`agent-browser read`) over mutating.- Only fill/submit forms the user asked about; never submit destructive or payment forms without explicit intent.- Close the browser session when finished.- Do not print credentials or session cookies into responses.

Frequently asked questions