How do I install the Deep Research skill?

Run npx skillmds@latest add chuk-development/deep-research in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Deep Research safe to use?

This skill has not completed SkillMD's automated safety review yet. Independent scanners report: SkillSpector: PASS, Skill Scanner: PASS. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Deep Research?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Deep Research free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Deep Research?

chuk-development (@chuk-development) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Deep Research by chuk-development — Agent Skill

Name: Deep Research
Author: chuk-development

Deep Research

Multi-step research discipline — search, crawl primary sources, cross-check from a second angle, then answer. Use when a question needs current or contested facts, when the user claims something is new or just released, or when one search would only give a shallow answer.

File contents

Deep research

A single search plus a summary of its snippets is a shallow answer. It reads as confident and is wrong often enough to matter. Use this loop instead.

The loop

Search — web_search to find candidate sources. Read the extra_snippets bullets (on by default) before deciding you need more.
Crawl — web_crawl the 1-3 best results for full detail and context. Snippets are lossy; the article is the source.
Cross-check — if gaps or contradictions remain, search again from a different angle (different wording, different source type, the primary party's own page).
Compile — build the answer from crawled content, not from search snippets. Say what you could not confirm.

Search tuning

German queries or DE-specific facts (prices in EUR, DE law, local events): pass country: "DE" and search_lang: "de" so the SERP is localized.
freshness (pd/pw/pm/py) applies to web mode too, not just news — use it whenever recency matters ("latest release notes", "current version").

Fresh releases

Search engines take hours to days to index new content. When the user says something was JUST released, web_search alone is not evidence of absence. Crawl primary sources directly:

HuggingFace org pages (e.g. https://huggingface.co/MiniMaxAI, https://huggingface.co/Qwen)
GitHub org pages and release pages
Official blogs and announcement pages

If search results contradict the user's claim about a very recent release, crawl the source directly before concluding it does not exist.

Rules

Never fabricate facts, URLs, prices, or figures to fill a gap. An unanswered sub-question stated plainly beats a fluent guess.
Never stop with intention-only text ("I will now search"). Either emit the next tool call, or give the complete final answer.
Distinguish what a source states from what you inferred. When sources disagree, say so and name them rather than silently picking one.

---name: deep-researchdescription: Multi-step research discipline — search, crawl primary sources, cross-check from a second angle, then answer. Use when a question needs current or contested facts, when the user claims something is new or just released, or when one search would only give a shallow answer.---# Deep researchA single search plus a summary of its snippets is a shallow answer. It readsas confident and is wrong often enough to matter. Use this loop instead.## The loop1. **Search** — `web_search` to find candidate sources. Read the `extra_snippets` bullets (on by default) before deciding you need more.2. **Crawl** — `web_crawl` the 1-3 best results for full detail and context. Snippets are lossy; the article is the source.3. **Cross-check** — if gaps or contradictions remain, search again from a different angle (different wording, different source type, the primary party's own page).4. **Compile** — build the answer from crawled content, not from search snippets. Say what you could not confirm.## Search tuning- German queries or DE-specific facts (prices in EUR, DE law, local events): pass `country: "DE"` and `search_lang: "de"` so the SERP is localized.- `freshness` (`pd`/`pw`/`pm`/`py`) applies to web mode too, not just news — use it whenever recency matters ("latest release notes", "current version").## Fresh releasesSearch engines take hours to days to index new content. When the user sayssomething was JUST released, `web_search` alone is not evidence of absence.Crawl primary sources directly:- HuggingFace org pages (e.g. `https://huggingface.co/MiniMaxAI`, `https://huggingface.co/Qwen`)- GitHub org pages and release pages- Official blogs and announcement pagesIf search results contradict the user's claim about a very recent release,crawl the source directly before concluding it does not exist.## Rules- Never fabricate facts, URLs, prices, or figures to fill a gap. An unanswered sub-question stated plainly beats a fluent guess.- Never stop with intention-only text ("I will now search"). Either emit the next tool call, or give the complete final answer.- Distinguish what a source states from what you inferred. When sources disagree, say so and name them rather than silently picking one.

Frequently asked questions