AgentSkillsCN

just-in-time-research

当需要借助外部知识作出决策时加载此模块——例如技术评估、标准与合规性审查、市场数据，或那些未纳入您训练数据的最佳实践。

SKILL.md
--- frontmatter
name: just-in-time-research
description: Load when making decisions that require external knowledge - technology evaluation, standards/compliance checking, market data, or current best practices not in your training data.

Just-In-Time Research

When you encounter a decision requiring external knowledge (standards, best practices, market data, technical evaluation), conduct research immediately rather than guessing.

When This Rule Applies

Load this rule when:

  • •Evaluating technology choices or comparing options
  • •Checking standards/specifications (WCAG, RFC, ISO)
  • •Assessing compliance or regulatory requirements
  • •Needing current market data or statistics
  • •Looking up integration patterns for external services
  • •Any decision where your training data may be outdated

Do NOT research for:

  • •Well-established patterns you're confident about
  • •Decisions already documented in loaded artifacts
  • •Basic programming concepts

Research Priority Order

Use research tools in this order:

  1. •Check existing research first - Look for *-research-report-*.md files in the project/epic directories
  2. •Perplexity MCP (if configured) - Use perplexity_search for quick facts, perplexity_research for deep analysis
  3. •Built-in web_search - Fallback when Perplexity unavailable
  4. •Generate manual prompt - If automated tools insufficient, create a prompt for the user

Research Workflow

Step 1: Check Existing Research

Before conducting new research, always check:

code
specs/projects/[project-id]/*-research-report-*.md
specs/projects/[project-id]/epics/[epic-id]/*-research-report-*.md

Also check if upstream artifacts (architecture.md, context.md, ux-strategy.md) already contain relevant "Research Informing" sections.

If relevant research exists:

  1. •Load and extract relevant findings
  2. •Reference as "Reused Research" in your research trail
  3. •Only conduct new research if gaps remain

Step 2: Identify the Knowledge Gap

Ask yourself:

  • •What specific information is missing?
  • •Can this be decided from well-known best practices?
  • •Do I need current/external information to decide wisely?

Step 3: Execute Research

For quick facts (standards, specs, simple lookups):

  • •Use perplexity_search or web_search

For analysis (trade-offs, comparisons, deep dives):

  • •Use perplexity_research

For complex decisions (multi-factor analysis):

  • •Use perplexity_reason

Step 4: Synthesize and Document

Extract key findings with source URLs. Add to the artifact:

markdown
## Research Informing This Section

- **[Topic]**: [Finding] (Source: [URL], [Date])

When Automated Research Fails

If research tools don't provide sufficient information:

  1. •Create a research prompt file: [command]-research-prompt-[topic].md
  2. •Tell the user: "Deep research needed for [topic]. Please run this prompt in your preferred research tool."
  3. •Pause and wait for the user to provide findings
  4. •Continue after incorporating results

Quality Guidelines

  • •Prefer official documentation over blog posts
  • •Use recent sources (2 years for standards, 6 months for market data)
  • •Cross-reference multiple sources for critical decisions
  • •Note when information may be outdated
  • •Document what research influenced what decision

Commands That Use This Pattern

Project Level: /project-ux, /project-context, /project-constitution, /project-architecture, /project-design-system, /project-plan

Epic Level: /epic-architecture, /epic-plan

Story Level: /story-plan