What is the best agent skill stack for research?
Reviewed . Scores and counts come from the directory snapshot of .
Short answer: for a company, company-research from browserbase/skills (84/100) researches the product and ICP, discovers target companies through the Browserbase Search API and scores fit. For a market, manus from sanjay3290/ai-skills (87/100) delegates deep web research, market analysis and product comparisons to the Manus agent. For a paper, deep-reader from tronghieu/agent-skills (90/100) reads it without loading the whole text into one context window, citation-management validates the references, and chat-history, the top score in the set, recovers research you have already done.
The stack
Seven skills, in score order. Scores come from the directory snapshot of October 22, 2026, and install counts are the figures skills.sh published on November 2, 2026.
| Skill | Repository | Score | What it does |
|---|---|---|---|
| chat-history | mblode/agent-skills | 91/100 | Recovers decisions, previous fixes and research from past AI conversations with source evidence, reading local Claude Code, Codex, Grok and Cursor history. |
| youtube-full | ZeroPointRepo/youtube-skills | 90/100 | Provides YouTube access for transcripts, video and channel search, channel browsing, playlists and within-channel search, for summaries, quotes, translations and topic research. |
| deep-reader | tronghieu/agent-skills | 90/100 | Reads and answers questions about long books and papers without loading the whole text into one context window, keeping page-anchored notes. |
| citation-management | K-Dense-AI/claude-scientific-skills | 89/100 | Manages academic citations by searching OpenAlex, PubMed and Google Scholar, extracting metadata, validating references and generating formatted BibTeX entries. |
| seo | mblode/agent-skills | 87/100 | Audits and fixes technical SEO, researches search demand, creates content briefs, and measures SEO, AEO and GEO performance. |
| manus | sanjay3290/ai-skills | 87/100 | Delegates complex, long-running tasks to the Manus AI agent for autonomous execution, such as deep web research, market analysis, product comparisons and stock analysis. |
| company-research | browserbase/skills | 84/100 | Researches a company's product and ICP, discovers target companies with the Browserbase Search API, researches each deeply and scores ICP fit. |
How to use it
- Check what you already know with chat-history. It recovers decisions, previous fixes and research from past AI conversations, with source evidence, by reading local Claude Code, Codex, Grok and Cursor history. A good share of the research I have paid for twice was sitting in an old session.
- For a company, run company-research. It researches the product and ICP, discovers target companies with the Browserbase Search API, researches each one deeply and scores ICP fit. Scoring fit is the useful part: a list of two hundred companies is noise, a list of twenty with reasons is a plan.
- For a market, hand the long job to manus, which delegates deep web research, market analysis, product comparisons and stock analysis to the Manus agent for autonomous execution. Add seo for the demand side: search demand research, content briefs, and SEO, AEO and GEO measurement.
- For a paper, read it with deep-reader, which keeps page-anchored notes, then check the references with citation-management, which searches OpenAlex, PubMed and Google Scholar, validates references and produces formatted BibTeX.
- When the best source is a talk or an interview, youtube-full pulls transcripts, searches videos and channels, and supports quotes and topic research. Transcripts are searchable; a two-hour interview is not, until it is text.
- Write the brief with every claim pointing at a source from steps 2 to 5. A research note without sources is an opinion in a nicer font.
Install the stack
npx skills add https://github.com/mblode/agent-skills/tree/main/skills/chat-history
npx skills add https://github.com/ZeroPointRepo/youtube-skills/tree/main/clawhub/youtube-full
npx skills add https://github.com/tronghieu/agent-skills/tree/main/skills/deep-reader
npx skills add https://github.com/K-Dense-AI/claude-scientific-skills/tree/main/skills/citation-management
npx skills add https://github.com/mblode/agent-skills/tree/main/skills/seo
npx skills add https://github.com/sanjay3290/ai-skills/tree/main/skills/manus
npx skills add https://github.com/browserbase/skills/tree/main/skills/company-researchThe skills CLI works for Claude Code, Codex, Cursor and other agents, and each skill's page in the directory lists the copy-by-hand routes if you would rather not run npx.
What to check before you install
- Licenses: six of the seven are MIT; manus is Apache-2.0. Confirm the license file in each repository anyway.
- Services: company-research uses the Browserbase Search API, manus needs access to the Manus agent, and youtube-full needs YouTube access, so read each SKILL.md for what it expects. chat-history reads your local AI conversation history, which is as sensitive as anything on your machine.
- Read each SKILL.md before you enable it. The directory has not security scanned any skill.
FAQ
Where should I start if I only install one?
chat-history, at 91/100 the top score here, because it costs nothing in outside services and it tells you what you already found. Start there even if the answer is nothing; a clean slate you can prove is worth the five minutes. The second install depends on the job: company-research for a company, manus for a market, deep-reader for a paper.
Do these skills work in Codex and Cursor?
Yes. All seven use the Agent Skills format, so the same folder works in Claude Code, Codex, Cursor and any other agent that reads a SKILL.md. The skills CLI installs for each of them, and every skill page in the directory shows the commands.
Can the agent do the whole research job on its own?
manus is built for exactly that: its summary describes delegating complex, long-running tasks for autonomous execution. My rule is that an agent researches and a person decides. Read the sources it cites before you act on the conclusion, and keep citation-management in the loop for anything academic.