InfoSeek - Deep Web Search & Archival

Overview

InfoSeek performs comprehensive web research on any subject (person, organization, product) across multiple search engines, deduplicates results, extracts clean content, and archives everything with full metadata in organized folders.

Prerequisites

Before executing a search task, verify these skills are installed:

import os
from pathlib import Path

workspace = os.environ.get('OPENCLAW_WORKSPACE')
skills_dir = Path(workspace) / 'skills'

required = ['baidu-search', 'tavily', 'Multi-Search-Engine', 'agent-browser-clawdbot-0.1.0']
missing = [s for s in required if not (skills_dir / s).exists()]

If any are missing, instruct the user to install them:

openclaw skills install baidu-search
openclaw skills install tavily-search
openclaw skills install multi-search-engine

Workflow

Phase 0: Task Setup

1. Confirm the search subject — name, organization, or product
2. Collect optional context — background info, time range, output format (default: .md), special requirements
3. Check dependencies — run the prerequisite check above
4. Create archive folder — run:

   python scripts/infoseek_helper.py create-folder "\x3Csubject_name>"
   

Phase 1: Multi-Engine Deep Search

Execute searches across all available engines. Each engine runs independently.

1.1 Baidu Search (100+ pages)

Use the baidu-search skill:

• Query: "\x3Csubject> \x3Cbackground_context>"
• Depth: 100+ pages
• Record: URL, title, website name, publish date for each result

1.2 Tavily Search

Use tavily_search tool:

query: "\x3Csubject> \x3Cbackground_context>"
search_depth: advanced
max_results: 50

1.3 Multi-Search-Engine

Use the multi-search-engine skill across multiple engines simultaneously.

1.4 Browser Deep-Crawl

For discovered URLs, use the browser tool to:

1. Open each page
2. Extract body content (filter ads, sidebars, comments)
3. Extract metadata: title, author, editor, date, website name

Phase 2: Deduplication

Run URL deduplication on all collected results:

python scripts/infoseek_helper.py deduplicate "\x3Ctemp_results_file>"

The script normalizes URLs (remove www, tracking params, unify http/https, remove trailing slashes) and checks against the SQLite database to skip duplicates.

Phase 3: Content Extraction & Storage

For each unique URL:

1. Extract content using the browser tool — get title, body, metadata
2. Filter content — remove ads, sidebars, navigation, comments, related articles, footers
3. Generate filename:

   python scripts/infoseek_helper.py generate-filename \
     --date "\x3CYYYYMMDD>" --title "\x3Ctitle>" --website "\x3Csite>" --format "\x3Cext>"
   

   Format: YYYYMMDD-title-website.ext
4. Save the file:

   python scripts/infoseek_helper.py save-content \
     --folder "\x3Carchive_path>" --filename "\x3Cname>" --url "\x3Curl>" \
     --website "\x3Csite>" --source "\x3Csource>" --date "\x3Cdate>" \
     --title "\x3Ctitle>" --author "\x3Cauthor>" --editor "\x3Ceditor>" \
     --content "\x3Cbody>" --task "\x3Csubject>"
   

5. Record in database:

   python scripts/infoseek_helper.py add-url \
     --url "\x3Cnormalized_url>" --task "\x3Csubject>" --filename "\x3Cname>"
   

Phase 4: Task Report

Output a summary when complete:

InfoSeek Task Report
====================
Subject: {query}
Engines used: {engines}
Total found: {total} | Duplicates skipped: {dupes} | New archived: {new}
Files saved: {count}
Location: {path}
Database records: {db_total}

File Naming

Format: YYYYMMDD-title-website.ext

• Date: 8 digits (YYYYMMDD) from page metadata
• Title: page title (strip special chars \x3C>:"/\|?*)
• Website: domain or media name
• Extension: md (default), json, txt, csv, xlsx, html, docx

If filename exists, append 8-char hash to prevent overwrites.

Output Formats

All formats include full metadata (URL, website, source, date, title, author, editor) plus body content.

• .md — Markdown with metadata table
• .json — Structured JSON with metadata object and content field
• .txt — Plain text with header metadata
• .csv — One row per article, all metadata as columns
• .xlsx — Excel spreadsheet with metadata columns
• .html — Styled HTML page with metadata table
• .docx — Word document with metadata paragraph

Storage Structure

{workspace}/
├── infoseek-archives/
│   ├── \x3Csubject_1>/
│   │   ├── 20260404-title-website.md
│   │   └── ...
│   └── \x3Csubject_2>/
└── infoseek/
    ├── infoseek.db          # SQLite dedup database
    ├── infoseek.log         # Operation log
    └── backups/

Deletion Policy

Strict data retention — no permanent deletes without confirmation.

Process:

1. List files to delete (name, URL, date)
2. Ask user: "Confirm deletion? Files go to recycle bin and can be recovered."
3. On confirmation, move to recycle bin (Windows: PowerShell, Mac/Linux: system trash)
4. Update database, log the deletion, confirm to user

Never:

• Delete without user consent
• Permanently delete (bypass recycle bin)
• Delete without logging
• Delete without updating database

Configuration

Override defaults in task instructions:

• Search depth: default 100 pages, specify e.g. "150 pages"
• Time range: default unlimited, specify e.g. "2020-01-01 to 2026-04-07"
• Output format: default md, specify e.g. "xlsx"
• Storage path: default {workspace}/infoseek-archives/, specify custom path

Troubleshooting

Security & Privacy

• All searches use public channels only
• No personal data stored — only search results
• SQLite database is local, never uploaded
• Deletions use system recycle bin (recoverable)
• All operations logged and auditable
• No telemetry, no external data transmission

Version History