web-research（网络研究）

Web Research Skill

Combines WebSearch with automatic Qdrant storage to build a searchable knowledge base.

Workflow

1. Check Qdrant first    → qdrant-find for existing knowledge
2. Search if needed      → WebSearch for current information
3. Store valuable finds  → qdrant-store with rich metadata
4. Return synthesized    → Combine stored + new knowledge

Step 1: Check Existing Knowledge

Before searching the web, check if the answer already exists:

Tool: qdrant-find
Query: "<user's question or topic>"

If sufficient information exists with recent harvested_at, use it directly.

Step 2: Web Search

When stored knowledge is insufficient or stale:

Tool: WebSearch
Query: "<refined search query>"

Step 3: Store Results

After getting valuable results, store with rich metadata:

Tool: qdrant-store
Information: |
  # <Topic/Question>

  ## Key Findings
  - Finding 1
  - Finding 2

  ## Details
  <Synthesized information from search results>

  ## Sources
  - [Title](URL)

Metadata:
  # Required fields
  source: "web_search"
  content_type: "text"
  harvested_at: "2025-01-04T10:30:00Z"

  # Search context
  query: "<original search query>"
  urls: ["https://example.com/1", "https://example.com/2"]

  # Classification (for filtering)
  category: "technology"
  subcategory: "databases"
  type: "documentation"

  # Technical context (when applicable)
  language: "python"
  framework: "fastapi"
  version: "0.100+"

  # Quality signals
  confidence: "high"
  freshness: "current"

  # Relationships
  related_topics: ["vector-search", "embeddings", "rag"]
  project: "reflex"

Rich Metadata Schema

Required Fields

Search Context

Classification (Enables Filtering)

Technical Context

Quality Signals

Relationships

Image References (URL only, no download)

Filtering Examples

Find Python documentation:

qdrant-find with filter:
  category: "technology"
  language: "python"
  type: "documentation"

Find recent troubleshooting:

qdrant-find with filter:
  type: "troubleshooting"
  freshness: "current"

Find project-specific knowledge:

qdrant-find with filter:
  project: "reflex"

When to Store

Always store:

• Technical documentation findings
• API patterns and examples
• Error solutions and workarounds
• Best practices and recommendations
• Tool comparisons and evaluations

Skip storing:

• Simple factual lookups (dates, definitions)
• Ephemeral information (current weather, stock prices)
• Information already in Qdrant with same content

Example: Full Research Flow

User asks: "How do I set up GitHub Actions for Python testing?"

# Step 1: Check existing
qdrant-find: "GitHub Actions Python testing setup"
→ No relevant results

# Step 2: Search
WebSearch: "GitHub Actions Python pytest workflow 2025"
→ Returns results with workflow examples

# Step 3: Store
qdrant-store:
  Information: |
    # GitHub Actions Python Testing Setup

    ## Key Findings
    - Use `actions/setup-python@v5` for Python environment
    - Matrix testing across Python versions: 3.9, 3.10, 3.11, 3.12
    - pytest with coverage using `pytest-cov`

    ## Workflow Template

name: Python Tests on: [push, pull_request] jobs: test: runs-on: ubuntu-latest strategy: matrix: python-version: ["3.9", "3.10", "3.11", "3.12"] steps: - uses: actions/checkout@v4 - uses: actions/setup-python@v5 with: python-version: ${{ matrix.python-version }} - run: pip install -e .[test] - run: pytest --cov

    ## Sources
    - [GitHub Actions Python Guide](https://docs.github.com/en/actions/...)

  Metadata:
    source: "web_search"
    content_type: "code"
    harvested_at: "2025-01-04T10:30:00Z"
    query: "GitHub Actions Python pytest workflow 2025"
    urls: ["https://docs.github.com/en/actions/..."]
    domain: "github.com"
    category: "technology"
    subcategory: "ci-cd"
    type: "documentation"
    language: "python"
    framework: "pytest"
    platform: "github-actions"
    confidence: "high"
    freshness: "current"
    depth: "detailed"
    related_topics: ["testing", "ci-cd", "yaml", "github"]

Integration with Other Skills

• research-patterns: Use web-research for external searches
• qdrant-patterns: Follows same metadata conventions
• knowledge-ingestion-patterns: Compatible chunking approach
• github-harvester: Similar metadata schema for GitHub content