slbrow-skill

Browser automation via SLBrow MCP and HTTP API. Use when the user needs to navigate pages, analyze content, manage tabs, search history, extract text, or con...

Security Scan

The skill's code, instructions, and requirements are consistent with a local browser-automation tool that talks to a local SLBrow server and browser extension; it does not request unrelated credentials or external installs.

Purpose & Capability

Name/description (browser automation via SLBrow MCP/HTTP API) matches what the files implement: local HTTP endpoints on localhost:5556, tools for navigation, extraction, tabs, history, and VAI player controls. No unrelated env vars, binaries, or external services are requested.

Instruction Scope

SKILL.md instructs the agent to call a local SLBrow server and use a browser extension to access pages, selected text, and history — which is exactly the stated purpose. Be aware these tools permit reading page content, browser history, and selected text (sensitive user data); that sensitivity is expected but worth noting. The instructions do not direct reading arbitrary local files or sending data to remote hosts other than localhost.

Install Mechanism

No install spec is included (instruction-only plus shipped helper script). The included CLI script (scripts/slbrow.js) and tools.json are bundled with the skill; there are no external downloads, package installs, or extract-from-URL steps that would write or execute remote code.

Credentials

The skill requires no environment variables or credentials. It does, by design, interact with the user's browser via a local server/extension, which gives it access to browsing data; that access is proportionate to the claimed functionality but is a privacy-sensitive capability to grant.

Persistence & Privilege

always is false and no special persistence or system-wide config changes are requested. The skill can be invoked autonomously by the model (platform default) but does not claim elevated or permanent privileges over other skills or system settings.

Assessment

This skill implements local browser automation and is internally consistent with its description. Before installing: 1) Verify you trust the SLBrow server and the browser extension it requires — they can read page content, selected text, and browser history. 2) Ensure the SLBrow server is bound to localhost and not exposed to remote networks. 3) Review the included script (scripts/slbrow.js) and tools.json if you want to confirm no external endpoints are contacted. 4) Be cautious using the skill with sensitive accounts or pages (banking, SSO flows) because the tools can extract page content and history. 5) If you prefer the agent not to call this automatically, disable autonomous invocation at the platform level or only enable this skill when needed.

SLBrow Browser Automation (Skills for Browser)

Skills for browser-side AI assistant. The browser plugin (aiassist) connects to slbrow-mcp via MCP protocol; these skills guide when and how to use the tools. Can be used via MCP tools or direct HTTP API calls with curl.

Prerequisites

1. SLBrow server running: npx slbrow or npm start in ai/slbrowmcp/slbrow-mcp/
2. Browser extension installed and connected (Chrome/Firefox)
3. MCP plugin enabled in aiassist: http://127.0.0.1:5556/mcp (视联浏览器MCP), http://127.0.0.1:5556/vaimcp (视联视频AI MCP)
4. Default ports: HTTP 5556, WebSocket 5555

Usage Methods

Method 1: MCP Tools (Browser Plugin + MCP)

When the MCP plugin is enabled, the AI has access to browser tools. Use these skills to guide tool selection and workflow:

• Navigation: Use page_navigate to open URLs
• Analysis: Use page_analyze with intent_hint (article, form submit, post_create) before extraction
• Tabs: Use tab_list first, then pass tab_id to target specific tabs

Method 2: HTTP API with curl (Direct Control)

Always available - Use curl commands to control browser directly without MCP plugin dependency:

Basic format (PowerShell-friendly — no temp JSON file):

curl.exe -X POST http://localhost:5556/api/execute -H 'Content-Type: application/json' -d '{\"tool\":\"TOOL_NAME\",\"args\":{\"PARAM\":\"VALUE\"}}'

Important notes for Windows PowerShell:

• Use curl.exe instead of curl (PowerShell aliases curl to Invoke-WebRequest)
• Put the JSON body in -d '...' with single quotes around the whole argument; backslash-escape every double quote in the JSON (\"). That way the shell passes valid JSON to curl without needing --data-binary @file

Common curl examples:

# Create new tab and navigate
curl.exe -X POST http://localhost:5556/api/execute -H 'Content-Type: application/json' -d '{\"tool\":\"tab_create\",\"args\":{\"url\":\"https://www.163.com\"}}'

# Navigate current tab
curl.exe -X POST http://localhost:5556/api/execute -H 'Content-Type: application/json' -d '{\"tool\":\"page_navigate\",\"args\":{\"url\":\"https://www.example.com\"}}'

# List all tabs
curl.exe -X POST http://localhost:5556/api/execute -H 'Content-Type: application/json' -d '{\"tool\":\"tab_list\",\"args\":{}}'

# Analyze page content
curl.exe -X POST http://localhost:5556/api/execute -H 'Content-Type: application/json' -d '{\"tool\":\"page_analyze\",\"args\":{\"intent_hint\":\"article\"}}'

# Extract article content
curl.exe -X POST http://localhost:5556/api/execute -H 'Content-Type: application/json' -d '{\"tool\":\"page_extract_content\",\"args\":{\"content_type\":\"article\"}}'

# Close specific tab
curl.exe -X POST http://localhost:5556/api/execute -H 'Content-Type: application/json' -d '{\"tool\":\"tab_close\",\"args\":{\"tab_id\":1234567890}}'

# Search browser history
curl.exe -X POST http://localhost:5556/api/execute -H 'Content-Type: application/json' -d '{\"tool\":\"get_history\",\"args\":{\"keywords\":\"search term\",\"max_results\":10}}'

Response format: API returns JSON with success, tool, result, and formatted fields for easy parsing.

Available Tools

Tool	Description
page_navigate	Navigate current tab to URL
page_analyze	Analyze page for elements (intent_hint: article, form submit, post_create, etc.)
page_extract_content	Extract content (content_type: article, search_results, posts)
page_wait_for	Wait for element or text (condition_type, selector, text)
tab_create	Create tab(s) (url, urls[], or count)
tab_close	Close tab(s) (tab_id, tab_ids[])
tab_list	List open tabs with IDs
get_history	Search browser history (keywords, start_date, end_date, domains, max_results)
get_selected_text	Get selected text (optional tab_id)
get_page_links	Get page links (link_type: all/internal/external, domains, max_results)

VAI Tools (Seelink)

Tool	Description
get_page_seelink_player_list	Get all players on current page
use_seelink_players_ai	Apply AI function to players (ai_function_name: reduce_fog, face_mosaic, dark_reduce, human_outline, vechicle_outline, none)

Workflow Patterns

Multi-tab workflow:

1. tab_list to get tab IDs
2. Add tab_id to any tool to target a background tab

Content extraction:

1. page_navigate to target URL
2. page_analyze with intent_hint to find elements
3. page_extract_content for article/search_results/posts

Form filling:

1. page_analyze with intent_hint "form submit" to discover form elements
2. Note the element IDs returned — use them with subsequent page_analyze or page_extract_content to confirm state

VAI (Seelink player AI):

1. get_page_seelink_player_list to get player IDs and count
2. Choose targets by player_position_list (zero-based index) or player_id_list (not both)
3. use_seelink_players_ai with chosen targets and ai_function_name
4. To apply different functions to different players, call use_seelink_players_ai separately for each
5. To disable AI on a player, set ai_function_name to "none"

Fetching Skills in Browser

The browser plugin can fetch skills from the server to inject into AI context:

GET http://127.0.0.1:5556/api/skills

Returns the full SKILL.md content (text/markdown). Use when slbrow MCP plugins are enabled to guide tool usage.

API Reference

For full parameter schemas, see references/api-reference.md.

Additional resources

• One-page workflow checklist: SKILL_COMPACT.md

Error Handling

• EXTENSION_DISCONNECTED: Ensure extension is installed and server is running. Check http://localhost:5556/health first.
• Tool call timeout: Operation took >30s; retry with simpler args or break into smaller steps.
• page_analyze returns 0 elements: Page may still be loading — wait a moment and retry, or page_navigate again.
• tab_id invalid (tab was closed): Call tab_list to refresh available tab IDs before retrying.
• Server restarted (session lost): Client will receive 404; re-initialize MCP connection (refresh the page).

Quick Reference for curl Commands

Tab Management:

• Create tab: tab_create with url parameter
• List tabs: tab_list with empty args
• Close tab: tab_close with tab_id parameter
• Navigate: page_navigate with url parameter

Content Analysis:

• Analyze page: page_analyze with intent_hint (article, form_submit, post_create)
• Extract content: page_extract_content with content_type (article, search_results, posts)
• Get links: get_page_links with link_type (all, internal, external)

History & Text:

• Search history: get_history with keywords, max_results
• Get selected text: get_selected_text with optional tab_id

Seelink AI:

• Get players: get_page_seelink_player_list
• Apply AI: use_seelink_players_ai with player_position_list or player_id_list and ai_function_name

Troubleshooting

Server not responding:

• Check if SLBrow server is running: curl.exe http://localhost:5556/health
• Verify server is started with npx slbrow or npm start

Curl command fails:

• Ensure using curl.exe in PowerShell
• For -d, use single-quoted JSON with \" around each JSON string delimiter (see Method 2 examples above)
• Verify server is accessible on port 5556

Tab operations not working:

• Use tab_list to get current tab IDs first
• Verify tab_id is valid and tab still exists
• Check if browser extension is connected