AgentSkillsCN

fact-grounding

当您需要从网络获取最新、准确的信息——验证各类说法、深入研究陌生主题、查询错误代码,或在大模型训练数据可能已过时、信息不足的情况下,可灵活调用相应 Perplexity 工具,按所需深度精准筛选。

SKILL.md
--- frontmatter
name: fact-grounding
description: "Use when you need current, accurate information from the web - verifying claims, researching unfamiliar topics, looking up error codes, or when LLM training data may be outdated or insufficient. Selects appropriate Perplexity tool based on depth needed."

Fact-Grounding with External Sources

Overview

Ground responses in verified, current information rather than relying solely on LLM knowledge. Use Perplexity MCP tools to search the web for facts.

Core principle: When in doubt, verify. When outdated, search. When complex, research.

When to Use

Error Resolution

  • Error codes/messages - Look up specific error codes to find working solutions
  • Stack traces - Search for others who encountered the same issue
  • Deprecation warnings - Find migration guides and updated approaches

Exploration & Discovery

  • New territory - Exploring unfamiliar frameworks, APIs, or domains
  • What's possible - Finding what others have tried, patterns that work
  • Best practices - Current recommendations (not outdated training data)

Verification

  • API signatures - Verify current method signatures, parameters
  • Version compatibility - Check which versions work together
  • Pricing/limits - Current pricing, rate limits, quotas
  • Release status - Is feature X stable? When was Y released?

Currency

  • Recent releases - Features in versions released after training cutoff
  • Current events - News, announcements, breaking changes
  • Documentation changes - APIs evolve, docs update

Tool Selection (by depth)

NeedToolUse Case
Quick answerperplexity_askSingle factual question, error code lookup, conversational follow-up
Find sourcesperplexity_searchMultiple sources needed, comparing solutions, finding alternatives
Deep diveperplexity_researchComplex topic, synthesized analysis with citations, exploring new ground

The Process

  1. Recognize the trigger - Is this something I should verify externally?
  2. Select depth - Quick answer, source search, or deep research?
  3. Formulate query - Clear, specific, searchable
  4. Execute search - Use appropriate Perplexity tool
  5. Synthesize results - Present findings with source attribution
  6. Acknowledge limitations - Note if information is incomplete or conflicting

Query Formulation Tips

For Error Lookups

  • Include the exact error message or code
  • Add technology context: framework, version, platform
  • Example: "ECONNREFUSED 127.0.0.1:5432" PostgreSQL Docker Windows

For Exploration

  • Be specific about what you're trying to achieve
  • Include constraints: language, framework, requirements
  • Example: "real-time collaboration React alternatives to Yjs 2024"

For Verification

  • Include version numbers when relevant
  • Ask for current or latest explicitly
  • Example: "Next.js 14 app router middleware current syntax"

Poor Queries (avoid)

  • Vague: "best framework"
  • Opinion-seeking: "should I use React"
  • Too broad: "how does authentication work"

Examples

Error Code Lookup

code
Query: "CS8602 possible null reference" C# nullable reference types fix
Tool: perplexity_ask

Exploring Possibilities

code
Query: "PDF generation libraries .NET 8 comparison 2024" features pricing
Tool: perplexity_search

Deep Research

code
Query: "implementing event sourcing with Azure Cosmos DB" patterns pitfalls production
Tool: perplexity_research

Red Flags

  • Presenting outdated info as current - Always verify versions, dates, pricing
  • Skipping verification for technical claims - API signatures, library features change
  • Not citing sources - When using researched info, attribute it
  • Over-researching simple facts - Use perplexity_ask for quick answers
  • Under-researching complex topics - Use perplexity_research for deep dives

Integration

Related workflows (shortcuts for specific depths):

  • /ask - Quick questions (uses perplexity_ask)
  • /search - Find sources (uses perplexity_search)
  • /research - Deep investigation (uses perplexity_research)

Invoke with @fact-grounding when you need guidance on which approach to use.