Skillsresearch
R

research

Investigate a question against high-trust primary sources and capture the findings as a Markdown file in the repo. Use when the user wants a topic researched, docs or API facts gathered, or reading legwork delegated to a background agent.

research - Conduct first-hand research and distill it into Markdown notes in the repository

Skill Overview

research is a deep-research skill: whenever you want to understand a question, verify the facts about an API, or simply delegate reading work, it launches a background agent to consult official documentation, source code, specifications, and first-party APIs, then writes the findings into a cited Markdown file and saves it in your repository.

Use Cases

  1. Gathering information before making technical choices: When preparing to introduce a library or service, you need to confirm exactly what it supports, what its limitations are, and how versions differ. research reads the original official documentation and specifications instead of relying on retellings in second-hand blog posts.
  2. Verifying facts before integrating a third-party API: A single mistake in parameter names, authentication methods, rate limits, or error codes can make an integration fail. Have the background agent read through the first-party API documentation and record the details before you start writing code.
  3. Delegating reading while continuing your own work: Research is a classic task that involves extensive reading and little writing. The skill runs as a background agent, so you do not need to watch it finish reading; you can continue with your current work and wait for it to return the notes.

Core Features

  1. Use only first-hand sources and trace citations back to the origin: Research is limited to primary sources such as official documentation, source code, technical specifications, and first-party APIs, rather than secondary summaries of them. Every conclusion must be traced through its citation chain to the actual material that establishes the fact.
  2. Produce a single Markdown research file with sources cited for each item: The research results will not be scattered throughout the conversation. They are consolidated into one Markdown file, with the source of each conclusion noted afterward, making it easy to verify later and for others to pick up the work.
  3. Determine the storage location according to existing repository conventions: The file will not be dropped arbitrarily in the root directory. The skill first checks where existing research notes are stored and how they are named, then follows those conventions. If the repository has no such notes yet, it chooses a reasonable location and clearly tells you where the file was saved.

Frequently Asked Questions

What is the difference between the research skill and simply asking an AI?

The difference lies in the source hierarchy and the form of the output. When you ask a direct question, the answer often comes from the model’s memory or second-hand articles found through search, making it difficult to determine which statements are supported by evidence. research requires tracing every conclusion back to the original material that establishes the fact and recording it in a cited Markdown file. Its value lies not in answering quickly, but in making the conclusions verifiable.

Where are the research results saved?

The skill first follows existing repository conventions—checking which directory existing research notes or technical documents are stored in and what naming scheme they use—and then saves the file accordingly. If the repository has no such notes, it chooses a reasonable location and explains where it saved the file. In either case, it will not place the file in the repository root.

Does every conclusion really have a source? What happens when no first-hand material can be found?

The skill requires sources to be cited item by item, so any conclusion that can be traced to a primary source will include a citation. However, be aware of its limitations: not every question has publicly available first-hand material. Private services, internal systems, and closed-source implementations without documentation may have no authoritative sources to consult. In such cases, it can only provide lower-level evidence and explain the limitations, rather than inventing a plausible-looking citation. For important decisions, treat its notes as a starting point and source of leads, not as the final authority.

Additional note: The research work is performed by a background agent, so the agent needs permission to read files and access the network. What it can actually find depends on the tools available to it.