# Research

> Research a question from high-trust primary sources and capture the findings as Markdown in the repo. Use when the user says "do some research", "research about", "look into", or asks for options, comparisons, current capabilities, or evidence about a tool, workflow, or practice. Version-sensitive package, SDK, or API docs go to read-the-damn-docs instead.

- Skill: `caneff/research` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add caneff/research`
- Raw SKILL.md: https://api.skillmd.com/api/skills/caneff/research/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Docs & Writing
- Author: caneff (https://skillmd.com/u/caneff)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/caneff/research

---


Spin up a **background agent** to do the research, so you keep working while it reads.

Its job:

1. Investigate the question against **primary sources** (official docs, source code, specs, first-party APIs), not retail or aggregator pages (Amazon, Google Books, blog summaries) summarizing them. Follow every claim back to the source that owns it — one citation is enough even when several sources repeat a primary `.gov`/`.mil` text verbatim.
2. Read each source once per run; reuse it for every claim it supports.
3. Do the reading yourself — a sub-agent's completion reports to the top-level session, not to you, so nested delegation loses the work. If the question is too large for one agent, report that and stop.
4. Write the findings to a single Markdown file, citing each claim's source.
5. Save it where the repo already keeps such notes; match the existing convention, and if there is none, put it somewhere sensible and say where.