Run a research query, produce a structured markdown report, and file it so knowledge accumulates instead of dying in conversations. Use when the user asks to research a topic, investigate a question, compare options, analyze competitors, or any exploratory query where the output should be preserved. Trigger on: "research", "investigate", "deep dive", "compare", "analyze", "look into", "what do we know about", or any question that deserves a filed answer rather than a chat response.
Run a research query and file the output as a permanent knowledge base article. Research
outputs live in a research/ directory and can be indexed for future reference.
Every conversation with an LLM produces insights that evaporate when the conversation ends. If you research pricing models today and someone asks the same question next week, you start from zero. This skill makes research cumulative — each query adds to the knowledge base, and future queries can build on past research.
Before touching a single file or running a single search, articulate the specific research question. Research without a thesis is data hoarding.
The user provides a research question, either as:
If the input is vague (e.g., "research competitors", "look into our architecture"):
Before starting, check if this topic has already been researched:
ls research/ 2>/dev/null
Search for existing articles on the topic using Grep. If existing research covers it:
Use all available tools to investigate:
Cast a wide net first, then focus on what's most relevant. Take notes as you go — don't rely on remembering everything at the end.
Write a structured markdown report. The format depends on the query type:
---