AgentSkillsCN

tech-research

综合运用多种情报来源开展全面的技术调研——包括 Grok(通过 Playwright 获取 X/Twitter 开发者讨论)、DeepWiki(基于 AI 的 GitHub 仓库分析),以及 WebSearch。为每种来源分配并行子代理,并将各路成果整合为一份统一的报告。当需要评估技术、比较各类库/框架、研究 GitHub 仓库、洞察开发者情绪,或探究技术架构决策时,应使用此技能。触发短语包括“技术调研”、“研究这项技术”、“技术调研”、“调研一下”、“比较库”、“评估框架”、“探究仓库”。

SKILL.md
--- frontmatter
name: tech-research
description: Comprehensive technical research by combining multiple intelligence sources — Grok (X/Twitter developer discussions via Playwright), DeepWiki (AI-powered GitHub repository analysis), and WebSearch. Dispatches parallel subagents for each source and synthesizes findings into a unified report. This skill should be used when evaluating technologies, comparing libraries/frameworks, researching GitHub repos, gauging developer sentiment, or investigating technical architecture decisions. Trigger phrases include "tech research", "research this technology", "技术调研", "调研一下", "compare libraries", "evaluate framework", "investigate repo".

Tech Research

Orchestrate multi-source technical research by dispatching parallel subagents to gather intelligence from X/Twitter (via Grok), GitHub repositories (via DeepWiki), and the web (via WebSearch). Synthesize all findings into a single actionable report.

Architecture: The main agent orchestrates research by dispatching up to 3 parallel subagents. Each subagent handles one data source to keep the main context clean.

When to Use

  • Evaluating a technology, library, or framework for adoption
  • Comparing alternatives (e.g., "Zustand vs Jotai vs Redux")
  • Investigating a GitHub repo's architecture and community reception
  • Gauging developer sentiment on a new API, tool, or announcement
  • Cross-language research (Chinese/Japanese developer communities)

Research Sources

SourceWhat It ProvidesBest For
Grok (X/Twitter)Real developer opinions, @handles, post URLsSentiment, expert discovery, niche recommendations
DeepWiki (GitHub)AI-powered repo analysis, architecture, API docsUnderstanding codebases, comparing repo internals. Only use ask_question — never read_wiki_structure or read_wiki_contents (they return massive dumps that easily exceed context limits)
WebSearchOfficial docs, blog posts, benchmarks, tutorialsFacts, performance data, official announcements

Workflow

1. Analyze the Research Question

Break the user's question into sub-queries for each source:

  • Grok query: Developer opinions, community sentiment, expert recommendations
  • DeepWiki query: Repository architecture, API design, code quality (requires owner/repo)
  • WebSearch query: Official docs, benchmarks, comparisons, recent announcements

Not every research task needs all 3 sources. Select sources based on the question:

Research TypeGrokDeepWikiWebSearch
"Should we use library X?"YesYes (if OSS)Yes
"What are devs saying about X?"YesNoMaybe
"How does repo X work internally?"NoYesMaybe
"Compare X vs Y performance"MaybeYes (both repos)Yes
"What's new in framework X?"YesNoYes

2. Dispatch Parallel Subagents

Launch subagents concurrently using Task. See references/subagent_templates.md for complete prompt templates.

Grok subagent:

code
Task(subagent_type: "general-purpose", description: "Ask Grok about [topic]", prompt: <grok_template>)

DeepWiki subagent:

code
Task(subagent_type: "general-purpose", description: "DeepWiki research [repo]", prompt: <deepwiki_template>)

WebSearch subagent:

code
Task(subagent_type: "general-purpose", description: "Web research [topic]", prompt: <websearch_template>)

3. Synthesize and Report

After all subagents return, merge findings into a unified report:

markdown
## Tech Research: [Topic]

### TL;DR
[2-3 sentence executive summary with clear recommendation]

### Community Sentiment (from X/Twitter)
- [Key opinions with @username attribution]
- [Verified post URLs]

### Repository Analysis (from DeepWiki)
- Architecture overview
- Code quality observations
- API design patterns
- Activity and maintenance status

### Web Intelligence
- Official documentation highlights
- Benchmark data
- Blog post insights
- Recent announcements

### Comparison Matrix (if comparing alternatives)
| Criteria | Option A | Option B |
|----------|----------|----------|
| [criterion] | [finding] | [finding] |

### Recommendation
[Clear, actionable recommendation based on all sources]

### Limitations
[What couldn't be verified or found]

Grok Query Strategies

Choose the right strategy based on the research goal. Strategy matters more than the topic itself. Every query MUST include X/Twitter-scoping keywords to avoid falling back to broad web search.

StrategyBest ForTemplate
Expert Discovery (highest success)Finding people to follow, active practitioners"Who are the most active [topic] developers on X worth following? Give @handles and link to recent posts"
Community Sentiment (high success)New API launches, framework releases"Search X posts about [topic]. What are developers saying about [aspect 1], [aspect 2]? Include @username and post URLs"
Technology Selection (medium)Choosing between libraries"What [type of tool] are developers currently recommending on X for [use case]? Include @username and post URLs"
CJK Research (medium)Chinese/Japanese community"搜索 X 上中文开发者关于 [topic] 的讨论。请给出推文链接和 @用户名"
Workflow DiscoveryHow devs use tools"How are developers on X using [tool] for [domain]? Include skills, workflows, @username and post URLs"

Avoid: Too-specific technical queries (X is not Stack Overflow) and searching for niche library names (search for the need instead).

Grok Pre-flight

Before dispatching a Grok subagent, run the pre-flight check:

bash
bash ${CLAUDE_PLUGIN_ROOT}/skills/tech-research/scripts/grok_preflight.sh
Exit CodeAction
0 READYDispatch Grok subagent
1 NEEDS_LOGINDispatch anyway; subagent will verify
2 NOT_CONFIGUREDSkip Grok source, note in report

If Grok is unavailable, proceed with DeepWiki + WebSearch only.

Tips

  • For CJK communities, query Grok in the target language directly
  • DeepWiki accepts up to 10 repos in a single query for comparisons
  • WebSearch is best for recent information (include current year in queries)
  • Always verify Grok post URLs before citing — accuracy is ~80%
  • Run subagents in parallel to minimize total research time