定制化招标信息采集

Bidding Hunter — 投标发掘引擎

A config-driven, reusable bid discovery engine that automates the tedious process of monitoring Chinese government procurement platforms.

Compatible with any AI assistant that can execute CLI tools: OpenClaw, Claude Code, Codex CLI, Cursor Agent, and similar platforms. The skill is platform-agnostic — all operations are exposed through the bidding-hunter CLI and standard shell commands.

🚨 MANDATORY TOOL DISCIPLINE — READ FIRST

ANY bidding-hunter task MUST use the existing tools below. You are FORBIDDEN from writing ad-hoc scripts, inline Node.js, curl, web_fetch, or any other approach.

Existing Tools (the ONLY allowed tools for bidding-hunter work)

Rules (ZERO EXCEPTIONS)

1. Use tools, don't replace them. A bidding-hunter task is a tool-execution task, NOT a coding task.
2. No ad-hoc scripts. Do NOT write scrape_*.js, fetch_*.js, test_*.js, inline node -e "...", curl, web_fetch, or any one-off scraping code.
3. No browser shortcuts. Do NOT use Playwright/Selenium directly unless you are inside a platform adapter's scan() method. All browser orchestration is handled by the scanner.
4. No manual dedup/matching. Do NOT copy-paste from the matcher or database modules. Use bidding-hunter scan → bidding-hunter report.

Escalation Path (when something breaks)

Existing tool fails
  → Read the tool's source code
  → Fix the bug in the tool itself (edit src/*.js, bin/*.js, scripts/*.js, templates/*.js)
  → Run tests: node --test tests/*.test.js
  → Re-run the tool
  → If the tool is fundamentally broken beyond repair: refactor the tool, record in memory/YYYY-MM-DD.md

Key principle: Fix the tool, don't work around it. Every workaround creates technical debt that compounds across sessions.

Memory requirement: After ANY code change to the bidding-hunter project (including bug fixes, adapter changes, config schema updates), write a note to memory/YYYY-MM-DD.md recording what was changed and why.

---

When to Use This Skill

• You need to discover relevant bidding opportunities across multiple Chinese procurement platforms
• You want automated keyword matching against daily procurement announcements
• You need deadline tracking and reminders for approaching bid deadlines
• You want to add a new procurement platform and let the agent figure out the scraping

Quick Start

# Create initial config
bidding-hunter init

# Edit config with YOUR keywords
# vim ~/.bidding-hunter/config.yaml

# Run a scan
bidding-hunter scan

# View results
bidding-hunter report

Core Workflow

1. Configure Your Keywords

Edit ~/.bidding-hunter/config.yaml — this is where YOU define what to search for:

matching:
  search_queries:        # What to type into platform search boxes
    - "建筑工程"
    - "信息化"
    
  tiers:                 # Multi-tier keyword matching
    high:
      label: "L1"
      keywords: ["建筑工程", "土建施工", "信息系统"]
    medium:
      label: "L2"  
      keywords: ["系统集成", "运维服务"]
    low:
      label: "L3"
      keywords: ["市政工程", "园林绿化"]
      
  blacklist:             # Filter these out
    - "中标"
    - "成交"
    - "废标"
    - "更正"

Important: The agent should NOT decide keywords — the user provides them via config.

2. Choose Your Platforms

platforms:
  enabled:
    - beijing    # 北京公共资源交易平台
    - hebei      # 河北省招标投标公共服务平台
    - liaoning   # 辽宁省公共资源交易平台
    - dalian     # 大连市公共资源交易平台
    - national   # 全国公共资源交易平台 (6 data sources)

3. Run the Scan

bidding-hunter scan

This runs a deterministic pipeline: scan → match → ingest → detail fetch → reminders → report.

4. Review & Manage Matches

# List tracked entries
bidding-hunter list --status tracked

# Mark as tracked
bidding-hunter status --id 5 --status tracked

# Update bid progress  
bidding-hunter status --id 5 --bid-status submitted

# Check reminders
bidding-hunter remind

Agent Usage Pattern

When a user asks you to run or manage bidding discovery:

1. First time setup: Run bidding-hunter init, then guide the user to configure keywords
2. Daily scan: Run bidding-hunter scan — this is deterministic, no LLM needed
3. Review results: Run bidding-hunter report and present findings to the user
4. Status updates: When user says "track #5" or "discard #3", use bidding-hunter status
5. Adding platforms: When user wants to monitor a new site:

   # Let the agent explore
   node scripts/explore-platform.js --url https://new-site.ggzy.gov.cn/ --recon
   # Then create adapter from findings
   bidding-hunter create-adapter --name new-site
   # Implement extraction logic based on exploration results
   # Test it
   bidding-hunter test-platform --name new-site
   

Key Design Principles (For Agent)

• Deterministic core: Scanning, matching, and ingestion do not depend on LLM availability
• Config over code: All keywords, sites, and schedules are in user-editable YAML
• One adapter per platform: Adding a new site = writing one ~50-line adapter file
• Idempotent: Safe to re-run; URL-based dedup and checkpoint system
• Graceful degradation: One failing site doesn't block others
• No privacy leakage: All data stays local; no telemetry or cloud uploads

Built-in Platform Adapters

Adding a New Platform

1. Explore the platform structure:

   node scripts/explore-platform.js --url https://new-site.gov.cn/ --recon
   

   This navigates to the site and captures structural hints.

2. Create adapter template:

   bidding-hunter create-adapter --name new-site
   

3. Implement the scan() method with:

   • Navigation logic (URL-based or search-based)
   • extractItems() function with correct CSS selectors
   • Pagination handling
   • Error handling for site-specific edge cases

4. Test:

   bidding-hunter test-platform --name new-site
   

5. Enable in config:

   platforms:
     enabled:
       - new-site
   

Operations

Checking Status

# View database stats
bidding-hunter stats

# Export for analysis
bidding-hunter export --format csv > bids.csv

Scheduling

Add to crontab:

0 16 * * * bidding-hunter scan --config ~/.bidding-hunter/config.yaml

Or use OpenClaw cron with the bidding-hunter scan command.

Notifications

Configure webhook channels in config:

notifications:
  channels:
    - type: feishu
      webhook: "${FEISHU_WEBHOOK_URL}"
    - type: slack
      webhook: "${SLACK_WEBHOOK_URL}"

Troubleshooting

Files

• Config: ~/.bidding-hunter/config.yaml
• Database: ~/.bidding-hunter/data.db (SQLite)
• Scan results: ~/.bidding-hunter/scan_results/YYYY-MM-DD/
• Custom adapters: ~/.bidding-hunter/platforms/*.js

See Also

• DESIGN.md — Full architecture and extension guide
• README.md — User-facing documentation
• CONTRIBUTING.md — How to contribute adapters