AgentSkillsCN

perplexity-researcher

利用 Perplexity Sonar API 进行带引用的实时网络调研,支持快速查询、深度分析,以及异步研究报告的生成。

SKILL.md
--- frontmatter
name: perplexity-researcher
description: Conduct real-time web research with citations using Perplexity Sonar API. Supports quick lookups, deep analysis, and async research reports.

Perplexity Researcher

AI-powered research with real-time web search and automatic citations via Perplexity's Sonar API.

When to Use

  • Researching current events, news, or recent developments (last 24-48 hours)
  • Technical documentation, API references, best practices
  • Market research, competitor analysis
  • Fact-checking and verification with authoritative sources
  • Deep research reports requiring extensive sourcing

Prerequisites

  • Perplexity MCP server installed: claude mcp add perplexity --env PERPLEXITY_API_KEY="pplx-xxx" -- npx -yq @perplexity-ai/mcp-server
  • Valid API key from https://sonar.perplexity.ai/

Research Modes

1. Quick Research (sonar)

For simple factual queries and summaries. Fast and cost-effective.

code
Use Perplexity to quickly look up: [topic]

2. Standard Research (sonar-pro)

For complex queries requiring deeper analysis and follow-up capability.

code
Research [topic] using Perplexity:
- Current state and recent developments
- Key findings and insights
- Actionable recommendations
Include all sources with URLs.

3. Reasoning Research (sonar-reasoning-pro)

For technical decisions and trade-off analysis.

code
Analyze the trade-offs between [option A] vs [option B] for [use case].
Use Perplexity reasoning mode for in-depth technical analysis.

4. Deep Research (sonar-deep-research)

For comprehensive reports with extensive citations. Async mode, takes 2-5 minutes.

code
Conduct deep research on [topic]:
- Comprehensive multi-source analysis
- Multiple perspectives and viewpoints
- Extensive citations and sourcing
Use sonar-deep-research model (async).

Workflow

Quick Lookup

Use the MCP perplexity_search tool directly for fast answers with citations.

Structured Research

  1. Define scope: Clear research question with specific deliverables
  2. Select model: Match complexity to appropriate Sonar model (see references/models.md)
  3. Execute query: Use Perplexity MCP tool
  4. Format output:
    • Summary: 2-3 sentence overview
    • Key Findings: Bullet points with evidence
    • Sources: URLs with titles and access dates
    • Next Actions: Specific follow-up tasks

Save Research

Store findings in project documentation:

code
docs/research/YYYY-MM-DD-topic.md

Prompt Templates

Technology Evaluation

code
Evaluate [TECHNOLOGY] for [USE CASE]:
1. What problems does it solve?
2. Current alternatives and comparisons
3. Pros and cons with specific examples
4. Community and maintenance status
5. Integration complexity and learning curve
6. Clear recommendation with reasoning
Include sources from 2024-2025.

Best Practices Research

code
Research current best practices for [DOMAIN]:
- Industry standards (2024-2025)
- Common implementation patterns
- Pitfalls to avoid with examples
- Recommended tools and frameworks
Cite authoritative sources (official docs, reputable blogs).

API/Library Investigation

code
Investigate [API/LIBRARY]:
- Official documentation links
- Key features and capabilities
- Usage examples and code snippets
- Known limitations and gotchas
- Version compatibility matrix
Prioritize official docs and recent community discussions.

Competitive Analysis

code
Analyze competitors in [MARKET]:
- Top 5 players with market position
- Feature comparison matrix
- Pricing models and tiers
- Strengths and weaknesses
- Recent news and announcements
- Market trends and predictions

Current Events / News

code
What are the latest developments in [TOPIC]?
Focus on the last [timeframe: 24 hours / week / month].
Include sources with publication dates.

Model Selection Guide

Query TypeModelWhen to Use
Quick facts, definitionssonarSimple lookups, fast answers
Complex analysissonar-proMulti-step queries, comparisons
Technical decisionssonar-reasoningTrade-offs, architecture choices
Advanced reasoningsonar-reasoning-proDeep technical analysis
Comprehensive reportssonar-deep-researchDocumentation, whitepapers

See references/models.md for detailed specifications.

Output Format

Always structure research output as:

markdown
# Research: [Topic]

**Date**: YYYY-MM-DD
**Model**: sonar-pro
**Query**: [Original research question]

## Summary
[2-3 sentence executive summary]

## Key Findings
- Finding 1 with evidence
- Finding 2 with evidence
- Finding 3 with evidence

## Detailed Analysis
[Expanded analysis organized by theme]

## Sources
1. [Title](URL) - Brief description
2. [Title](URL) - Brief description

## Next Actions
- [ ] Action item 1
- [ ] Action item 2

Tips

  • Recency: Perplexity excels at recent information; specify timeframes when relevant
  • Citations: Always include sources for verification; note when information may be outdated
  • Iteration: Use follow-up queries to drill deeper into specific aspects
  • Cross-reference: For critical decisions, verify with multiple queries
  • Cost awareness: deep-research is expensive (~$0.50/report); use for substantial research only
  • Confidence: Note when Perplexity hedges or expresses uncertainty

Troubleshooting

MCP Connection Issues

bash
claude mcp list  # Verify server is running
claude mcp remove perplexity && claude mcp add perplexity ...  # Reinstall

Rate Limits

Implement delays between queries. Use lower search modes for bulk research.

Timeout Errors

Increase timeout: set PERPLEXITY_TIMEOUT_MS=600000 (10 minutes)