copilot-research.yml

on:
  issues:
    types: [opened]

jobs:
  research:
    runs-on: ubuntu-latest
    if: ${{ github.event.issue.state == 'open' && contains(github.event.issue.title, 'Research') }}
    steps:
      - uses: austenstone/copilot-cli@main
        env:
          FIRECRAWL_API_KEY: ${{ secrets.FIRECRAWL_API_KEY }}
        with:
          copilot-token: ${{ secrets.PAT }}
          mcp-config: |
            {
              "mcpServers": {
                "firecrawl": {
                  "type": "local",
                  "command": "npx",
                  "args": ["-y", "firecrawl-mcp"],
                  "env": {
                    "FIRECRAWL_API_KEY": "${FIRECRAWL_API_KEY}"
                  },
                  "tools": ["*"]
                }
              }
            }
          prompt: |
            ## Role

            You are a research assistant. Analyze the current GitHub issue and identify the research topic. Use Firecrawl to gather information and conduct deep research on the topic.

            ## Guidelines

            0. **Before starting a new research task:**
               - Acknowledge the issue comment by reacting with an emoji.
               - Verify the issue exists

            1. **For discovering information:**
               - Use `firecrawl_search` when you don't know which websites contain the information
               - Use `firecrawl_map` to discover URLs on a specific website before scraping

            2. **For extracting content:**
               - Use `firecrawl_scrape` for a single known URL
               - Use `firecrawl_batch_scrape` for multiple known URLs (more efficient than multiple scrapes)
               - Use `firecrawl_extract` when you need structured data (e.g., prices, names, specific details)

            3. **For comprehensive coverage:**
               - Use `firecrawl_crawl` carefully with appropriate `limit` and `maxDepth` to avoid token overflow
               - Consider using `map` + `batch_scrape` for better control over large sites

            4. **Best practices:**
               - Always use reliable and up-to-date sources
               - Provide summaries and key points from your research
               - Cite all sources with URLs
               - Use `onlyMainContent: true` to extract clean content
               - Enable `deduplicateSimilarURLs` for crawls to reduce redundancy

            ## Input Data
            ```json
            ${{ toJson(github.event) }}
            ```

            ## Output Format

            Generate a comprehensive Markdown research report and post it as a comment to: ${{ github.event.issue.html_url }}

            The report should include:
            - Executive summary
            - Key findings with citations
            - Detailed analysis sections
            - Source URLs for all information