AgentSkillsCN

research

对技术主题、库、API 或概念进行深度研究。当您被要求研究、调查、深入探索,或全面收集某一主题的相关信息时,可选用此方案。将学习成果保存至 .agents/research/。

SKILL.md
--- frontmatter
name: research
description: Deep research on technical topics, libraries, APIs, or concepts. Use when asked to research, investigate, explore deeply, or gather comprehensive information on a topic. Saves learnings to .agents/research/.

Research Skill

Conducts deep research on technical topics, saving findings for future reference.

Capabilities

  • Web search for documentation, articles, and examples
  • Deep analysis using Oracle for complex reasoning
  • Cross-reference multiple sources
  • Save structured learnings to .agents/research/

Workflow

1. Clarify Scope

Before starting, confirm:

  • Topic: What exactly needs researching?
  • Depth: Quick overview or comprehensive deep-dive?
  • Focus: Implementation details, best practices, comparisons, or concepts?

2. Research Phase

Use these tools in combination:

ToolUse For
web_searchFind documentation, articles, tutorials
read_web_pageExtract detailed content from URLs
oracleDeep analysis, synthesizing findings, reasoning about trade-offs
librarianExplore open-source implementations
gh (via Bash)Fetch repo content, issues, PRs, releases when you have a direct GitHub URL

Note: librarian accesses GitHub repositories (public repos and private repos you've authorized). For local codebase searches, use finder or Grep instead. Use gh CLI for targeted queries like gh repo view, gh issue list, or gh release list.

Research strategy:

  1. Start with web_search to find authoritative sources
  2. Use read_web_page to extract key information
  3. Consult oracle for analysis and synthesis
  4. Use librarian to find real-world implementations

3. Document Findings

Save research to .agents/research/[topic-slug].md:

markdown
# Research: [Topic Name]

**Date:** YYYY-MM-DD
**Status:** Draft | Complete
**Tags:** [relevant, tags]

## Summary

[2-3 sentence overview of key findings]

## Key Learnings

- Learning 1
- Learning 2
- Learning 3

## Details

### [Subtopic 1]

[Detailed findings]

### [Subtopic 2]

[Detailed findings]

## Sources

- [Source Title](url) - Brief description of what was learned
- [Source Title](url) - Brief description

## Open Questions

- [ ] Question that needs further research
- [ ] Another question

## Related Research

- [[other-topic.md]] - How it relates

4. Synthesize & Report

Provide the user with:

  1. Executive summary (3-5 bullet points)
  2. Recommendations based on findings
  3. Link to saved research file

File Naming Convention

Use kebab-case slugs: .agents/research/[topic-slug].md

Examples:

  • react-server-components.md
  • firebase-auth-patterns.md
  • zod-vs-yup-comparison.md

Deep Research Mode

For comprehensive research, use Oracle with focused prompts:

text
task: "Analyze [topic] considering:
- Current best practices
- Common pitfalls
- Performance implications
- Security considerations"

Quick Reference

Start research:

text
Research [topic] - [specific focus]

Check existing research:

bash
ls -la .agents/research/

Update existing research: Add new sections or update status from Draft to Complete.