AgentSkillsCN

research-trigger

研究命令的自然语言封装——当用户提出需要深度研究、查找最佳实践或收集多源信息的问题时，自动触发 /research

SKILL.md
--- frontmatter
name: research-trigger
description: Natural language wrapper for research command - automatically triggers /research when users ask questions requiring deep research, best practices lookup, or multi-source information gathering
schema_version: 1.0

research-trigger

Type: ANALYSIS-ONLY DAIC Modes: DISCUSS, ALIGN, IMPLEMENT, CHECK (all modes) Priority: High

Trigger Reference

This skill activates on:

  • •Keywords: "research", "find information", "look up", "investigate", "best practices", "how do I", "what are the options", "search for", "find examples", "discover", "explore"
  • •Intent Patterns: (research|investigate|find|lookup).*?, how.*(do I|implement|use|approach), what.*(options|alternatives|ways to), best practices.*?, (search|find|discover).*?(information|examples|docs)

From: skill-rules.json - research-trigger configuration

Purpose

Automatically trigger the /research slash command when users ask questions requiring deep research, best practices lookup, or multi-source information gathering using natural language. Eliminates the need to remember the /research syntax.

Core Behavior

In any DAIC mode:

  1. •

    Question Detection

    • •Detect when user asks research-oriented questions
    • •Examples: "How do I implement JWT auth?", "What are best practices for caching?"
    • •Automatically invoke /research with the question
  2. •

    Intent Classification

    • •Identify research intent patterns:
      • •How-to questions ("How do I...")
      • •Comparison questions ("What are the options for...")
      • •Best practices queries ("Best practices for...")
      • •Investigation requests ("Research X", "Investigate Y")
  3. •

    Research Invocation

    • •Call /research command with extracted question
    • •Pass through to parallel research agents
    • •Return comprehensive research findings with citations

Natural Language Examples

Triggers this skill:

  • •✓ "Research best practices for error handling"
  • •✓ "How do I implement OAuth2?"
  • •✓ "What are the options for state management?"
  • •✓ "Find information about GraphQL subscriptions"
  • •✓ "Investigate performance optimization techniques"
  • •✓ "Look up JWT token refresh patterns"
  • •✓ "Search for examples of Sentry integration"
  • •✓ "Discover alternatives to Express middleware"
  • •✓ "What are best practices for API versioning?"

Does NOT trigger:

  • •✗ "What is the current mode?" (DAIC mode question, not research)
  • •✗ "What's in progress?" (project status, not research)
  • •✗ "What files changed?" (git status, not research)

Safety Guardrails

ANALYSIS-ONLY RULES:

  • •✓ NEVER call write tools (Edit, Write, MultiEdit)
  • •✓ Only invokes /research command which is read-only
  • •✓ Safe to run in any DAIC mode
  • •✓ Provides research findings and recommendations only

Invocation Rules:

  • •Only invoke /research when clear research intent detected
  • •Extract the full question from user's message
  • •Handle "how", "what", "why" questions appropriately
  • •Distinguish research questions from status/info queries

Implementation Notes

This skill acts as a natural language wrapper around /research. The actual research logic lives in .claude/commands/research.md, which:

  • •Classifies query type (breadth-first vs depth-first)
  • •Launches parallel research subagents
  • •Synthesizes findings with automatic citations
  • •Returns comprehensive research report

This skill simply detects the intent and routes to that command.