Search Skill

Execute search missions and compile structured results from various sources.

Purpose

Provide a unified search interface that can:

1. Search local files, codebases, and directories
2. Search web content (via WebSearch/WebFetch or future provider integrations)
3. Compile results into structured JSON output

This skill operates without user-provided input files - it receives a search mission and executes it autonomously.

Search Modes

Local Mode

Search the local environment:

Tool	Use Case
Glob	Find files by pattern (e.g., **/*.md, src/**/*.ts)
Grep	Search file contents for patterns
Read	Read matched files for content extraction
Bash	Execute find, ls, or other search commands

Web Mode

Search web content:

Tool	Use Case
WebSearch	Search the web for topics, news, documentation
WebFetch	Fetch and extract content from URLs

Future Providers (planned integration):

• jina.ai - AI-powered web search and extraction
• firecrawl - Web crawling and scraping
• exa.ai - Neural search engine

Process

Step 1: Parse Mission

Extract from the mission description:

• Search target: What to search for (keywords, patterns, topics)
• Search mode: local, web, or auto-detect
• Scope: Directories, file types, domains, date range
• Compile format: How to structure results

Step 2: Execute Search

Based on mode:

Local:

1. Use Glob to find matching files
2. Use Grep to search content patterns
3. Use Read to extract relevant sections
4. Compile matched results

Web:

1. Use WebSearch to find relevant pages
2. Use WebFetch to extract content
3. Parse and filter results
4. Compile into structured format

Step 3: Compile Results

Structure findings into the output schema with:

• Source attribution (file path or URL)
• Relevance scoring
• Extracted content snippets
• Metadata (date, author if available)

Input

The skill receives a mission description specifying:

• What to search for
• Where to search (mode/scope)
• What to extract and compile

Example missions:

Search for all TODO comments in the src/ directory and compile a list with file locations.

Search Hacker News for the top 5 trending AI stories today and summarize each.

Find all TypeScript files that import from '@looplia-core/core' and list their exports.

Output Schema

{
  "query": "original search query/mission",
  "mode": "local | web",
  "results": [
    {
      "source": "path/to/file.ts | https://example.com/page",
      "type": "file | url",
      "title": "Optional title or filename",
      "content": "Extracted content or snippet",
      "relevance": 0.95,
      "metadata": {
        "lineNumber": 42,
        "matchContext": "surrounding context",
        "date": "2025-12-24",
        "author": "optional"
      }
    }
  ],
  "summary": "Brief summary of findings",
  "totalMatches": 15,
  "compiledAt": "2025-12-24T10:30:00Z"
}

Usage in Workflows

This skill enables input-less workflow steps:

steps:
  - id: find-news
    skill: search
    mission: |
      Search Hacker News for today's top 3 AI stories.
      Extract title, URL, points, and brief summary for each.
    output: outputs/news.json

No input: field required - the skill operates autonomously.

Important Rules

1. Always attribute sources - Include file paths or URLs for all results
2. Respect scope - Only search within specified boundaries
3. Compile, don't dump - Structure results, don't return raw search output
4. Handle empty results - Return valid JSON even when no matches found
5. Rate limit web searches - Be mindful of API limits for web providers