Agent-Skills.md

Agent Skills: Technical Research Skill

Automatically determines the optimal research tool based on query type and available information. Use when answering technical questions requiring external verification, documentation lookup, or library/repo research. Do not use for statistical calculations, hypothesis testing, or data analysis → see analyzing-data skill.

UncategorizedID: Git-Fg/thecattoolkit/conducting-research

Repository

Git-Fg/thecattoolkit
Git-FgLicense: MIT
1

Install this agent skill to your local

pnpm dlx add-skill https://github.com/Git-Fg/thecattoolkit/tree/HEAD/plugins/sys-research/skills/conducting-research

Skill Files

Browse the full folder contents for conducting-research.

Download Skill

Loading file tree…

plugins/sys-research/skills/conducting-research/SKILL.md

Skill Metadata

Name
conducting-research
Description
"Automatically determines the optimal research tool based on query type and available information. Use when answering technical questions requiring external verification, documentation lookup, or library/repo research. Do not use for statistical calculations, hypothesis testing, or data analysis → see analyzing-data skill."

Technical Research Skill

Use this skill when the user asks technical questions requiring external verification or research. This skill automatically selects the optimal research tool based on your query.

Trigger Conditions

Automatically invoke this skill when you encounter:

  • "How do I..." questions about libraries/frameworks
  • Error messages requiring external documentation
  • "What is..." queries about specific technologies
  • Library comparisons or feature lookups
  • API documentation requests
  • Implementation examples for specific tools

Tool Selection Logic

Step 1: Unknown Terms? Use DuckDuckGo (Scout)

If you're unfamiliar with a term, library name, or need to find the correct repository:

Tool: mcp__plugin_sys-research_duckduckgo__search

Goal: Find:

  • Exact official library name
  • Primary GitHub repository owner/repo
  • Official documentation URL

Example: User asks about "that new Python async library everyone is using" → Search to identify the specific library name and repository

Step 2: Standard Libraries? Use Context7 (Librarian)

For established frameworks, SDKs, and well-documented libraries:

Tool Sequence:

  1. mcp__plugin_sys-research_context7__resolve-library-id with the library name
  2. CRITICAL: Use the exact libraryId from step 1 in step 2
  3. mcp__plugin_sys-research_context7__query-docs with the resolved ID

Best For:

  • React, Next.js, Vue.js, Angular
  • Python: pandas, numpy, requests, flask, django
  • AWS SDK, Azure SDK, GCP SDK
  • Node.js: express, mongoose, graphql
  • Java: spring, hibernate, junit
  • Go: gin, gorm, cobra
  • Rust: serde, tokio, actix

Example: User asks "How do I implement authentication in React?" → Resolve library ID for "react" then query for authentication patterns

Step 3: Specific Repos? Use DeepWiki (Code Expert)

For GitHub repositories, specific branches, or niche tools:

Tool: mcp__plugin_sys-research_deepwiki__ask_question or mcp__plugin_sys-research_deepwiki__read_wiki_structure

Best For:

  • GitHub repositories (e.g., "owner/repo-name")
  • Specific branches or commits
  • Open source project internals
  • Niche or experimental libraries
  • Understanding specific file structures

Example: User asks "How does the dev branch of vercel/next.js implement edge runtime?" → Use DeepWiki with repoName "vercel/next.js" and branch "dev"

Decision Tree

Query Type
├── Contains library name I recognize?
│   ├── YES → Is it a standard framework/library?
│   │   ├── YES → Context7 (resolve → query)
│   │   └── NO → DuckDuckGo (identify) → DeepWiki
│   └── NO → DuckDuckGo (identify)
├── Mentions specific GitHub repo?
│   └── YES → DeepWiki (ask_question)
└── Asking for general understanding?
    └── YES → DeepWiki (read_wiki_structure)

Failure Handling

Context7 Resolution Fails

Fallback: Use mcp__plugin_sys-research_duckduckgo__search to find:

  1. Correct library naming convention
  2. Official documentation URL
  3. Alternative names or aliases

Then retry Context7 with the correct name.

Insufficient Documentation Data

Fallback: Use mcp__plugin_sys-research_duckduckgo__fetch_content on:

  1. Official docs URL found via search
  2. GitHub README or wiki pages
  3. Community tutorials or guides

DeepWiki Returns Limited Info

Fallback: Use mcp__plugin_sys-research_duckduckgo__search to find:

  1. Additional resources about the repository
  2. Community discussions or issues
  3. Related projects or alternatives

Response Format

Structure your answer as:

## Research Summary

**Query:** [Restate the user's question]

**Tools Used:** [List the MCP tools invoked]

**Key Findings:**
- [Bullet point 1 with citation]
- [Bullet point 2 with citation]

**Solution/Code Example:**
```[language]
[Complete, working code example]

Explanation: [Why this solution works based on the retrieved documentation/repo structure]


## Examples

### Example 1: Standard Library Query
**User:** "How do I read a CSV file in Python?"

**Action:**
1. Recognize "pandas" as standard library
2. Call `resolve-library-id` for "pandas"
3. Call `query-docs` with resolved ID for CSV reading
4. Provide code example and explanation

### Example 2: Unknown Library
**User:** "What's the best way to do state management in Svelte?"

**Action:**
1. Not immediately familiar with Svelte state libraries
2. Call `duckduckgo__search` for "Svelte state management libraries"
3. Identify "svelte/store" as the official solution
4. If needed, use Context7 or DeepWiki for deeper dive
5. Provide comprehensive answer

### Example 3: GitHub Repository
**User:** "How does authentication work in the Auth.js NextAuth repo?"

**Action:**
1. Recognize this as GitHub repo-specific question
2. Call `deepwiki__ask_question` with repoName "next-auth/next-auth.js"
3. Ask specific question about authentication implementation
4. Provide detailed answer with file references

## Best Practices

1. **Always verify** the library name before using Context7
2. **Use the exact libraryId** returned from resolve-library-id
3. **Cite sources** in your response (documentation paths, file locations)
4. **Provide complete, working code examples**
5. **Explain the "why"** not just the "how"
6. **Chain tools intelligently** when one tool's output informs the next
7. **Document your research process** so the user understands the methodology