browser-network-inspector（浏览器网络检查器）

Browser Network Inspector

A single-skill browser debugging capture tool for inspecting page requests and turning them into readable reports.

Use this when you want to:

• see which APIs a page or button triggers
• inspect login, registration, or form-submit flows
• capture page-level fetch / XMLHttpRequest / basic WebSocket activity
• export a clean JSON + Markdown report after browser actions

This skill is packaged as one workflow. Internally it uses a local browser runtime, but the user-facing experience should be treated as one skill from start to finish.

Use this skill for

• Debugging requests from a web page you are authorized to inspect
• Understanding login / registration / submit flows at the page level
• Seeing which APIs a button click or form submit triggers
• Capturing and summarizing fetch / XMLHttpRequest activity from the current page
• Producing a redacted request summary for later analysis

Do not use this skill for

• Extracting or replaying third-party login sessions
• Capturing full system traffic
• Pulling access tokens, cookies, session secrets, or passwords for reuse
• Bypassing captchas, auth, or platform protections

About the capture model

This skill provides browser-side request capture, not raw packet sniffing.

It is designed to help with:

• page request inspection
• form submission debugging
• login / registration flow analysis
• browser API tracing around user actions

It can observe:

• page-level fetch
• page-level XMLHttpRequest
• basic WebSocket lifecycle + message events

It does not try to replace system/network tools such as:

• Wireshark
• mitmproxy
• Fiddler

It also does not inspect arbitrary native processes or full machine traffic.

In short: this skill is best described as a browser debugging capture and reporting tool.

Runtime expectation

• This skill expects a local agent-browser CLI runtime to be installed on the machine.
• The helper script scripts/capture-session.js auto-detects the local binary and drives the browser runtime for you.
• If agent-browser is missing, install it first before using this skill.

Workflow

1. Open the target page with the browser runtime managed by this skill.
2. Inject the network collector into the page:

$collector = Get-Content "$env:USERPROFILE\.openclaw\workspace\skills\browser-network-inspector\scripts\collect-network.js" -Raw
agent-browser eval $collector

1. Optional: configure include/exclude host filters before the flow:

agent-browser eval "window.__bniSetConfig({ includeHosts: ['example.com'], excludeHosts: ['ads.example.com'], captureWebSocket: true })"

1. Perform the desired browser actions (open, click, fill, submit, wait).
2. Read back the captured records:

agent-browser eval "JSON.stringify(window.__bniExport ? window.__bniExport() : [])"

1. Save the JSON output to a local file.
2. Run the summarizer:

node "$env:USERPROFILE\.openclaw\workspace\skills\browser-network-inspector\scripts\summarize-network.js" <input.json> [output.md]

1. Or use the bundled one-shot helper after you finish the browser actions:

node "$env:USERPROFILE/.openclaw/workspace/skills/browser-network-inspector/scripts/capture-session.js" --json-out ".capture/session.json" --md-out ".capture/session.md" --include-hosts "example.com,api.example.com"

v2 additions

• Host include/exclude filtering
• Basic WebSocket event logging
• One-shot export helper: scripts/capture-session.js
• Improved summary report with source breakdown

Polished workflow helpers

Additional JS-only helpers included:

• scripts/clear-session.js — clear the in-page capture buffer
• scripts/capture-and-report.js — create a timestamped report directory and save both JSON + Markdown
• Export helpers sanitize non-JSON wrapper output before writing capture files

Example:

node "$env:USERPROFILE/.openclaw/workspace/skills/browser-network-inspector/scripts/clear-session.js"
node "$env:USERPROFILE/.openclaw/workspace/skills/browser-network-inspector/scripts/capture-and-report.js" --include-hosts "example.com,api.example.com" --label login-flow --open-report

Notes

• This captures page-level JS network activity. It is not a full packet sniffer.
• fetch and XMLHttpRequest are supported in v1.
• v2 adds basic WebSocket event logging and a Node.js export helper.
• Request and response bodies are truncated and redacted before export.
• If a site uses service workers, native browser internals, or non-page network paths, results may be incomplete.
• Keep helper scripts in JavaScript or Python only for this environment.
• The bundled export helper is Node.js-based, auto-detects the local browser runtime executable, and injects the collector in chunks to avoid Windows command-length limits.

Redaction defaults

The collector and summarizer should redact or suppress obvious sensitive values such as:

• authorization
• cookie
• set-cookie
• password
• token
• accessToken
• refreshToken
• session
• csrf

Output expectation

The summary should tell you:

• how many requests were captured
• which endpoints were hit
• which requests failed
• rough timing and status distribution
• likely key requests in the user flow

If the user wants raw logs, save them locally first and avoid echoing full sensitive payloads into chat.