Research is the part of knowledge work that eats the most time and returns the least predictable results. You open twenty tabs, skim a dozen documents, take scattered notes, then try to assemble something coherent from the mess. Agent skills turn that process into a pipeline: one skill searches, another summarizes, a third structures the output into something you can actually hand to someone.

These skills cover the full arc from sourcing through synthesis, built for agents like Claude Code, Codex, and Cursor.

What separates a good research skill from a bad one

Source breadth. A research skill that only pulls from one search engine or one document format will leave gaps. The best ones search across multiple sources, handle PDFs and web pages equally, and let you scope the search to specific domains or time ranges.

Citation discipline. Research without sources is just opinion. Look for skills that attach URLs or document references to every claim. If you can’t trace a finding back to where it came from, the skill isn’t doing research. It’s generating plausible text.

Output structure matters too. A wall of text is not a summary. Good summarization skills produce structured output: executive summary, findings, open questions, action items. They adapt format to the document type, because a legal contract needs different treatment than a product spec.

And finally, composability. Research rarely ends at one step. A skill that produces output another skill can consume (structured Markdown, JSON, tagged sections) fits into a pipeline. Skills that produce opaque prose force you to bridge each step by hand.

Top agent skills for research and summarization

Research Assistant

The foundational research skill. Research Assistant takes a topic or question, searches the web across multiple sources, and returns a structured report with citations. It’s built for thoroughness over speed, so expect it to check 10+ sources before synthesizing.

What makes it useful is the synthesis step. It doesn’t just return a list of links. It reads each source, extracts relevant findings, cross-references them, and produces a report that highlights agreements, contradictions, and gaps. Every claim gets an inline citation so you can verify anything that looks off.

Compatible with: Claude Code, Universal | Category: Content gh skill install alirezarezvani/claude-skills/research-assistant

Document Summarizer

Different job than web research. Document Summarizer works on documents you already have. Feed it a 50-page legal agreement or a technical report and it returns a structured brief: executive summary, decisions, open questions, action items.

It adapts format based on what you give it. A legal document gets clause-level extraction with risk flags. A technical report gets architecture decisions and dependency callouts. A financial document gets metric highlights and variance explanations. That format awareness is what separates it from just asking an LLM to “summarize this.”

Compatible with: Claude Code, Cursor, Universal | Category: Content gh skill install alirezarezvani/claude-skills/document-summarizer

Meeting Notes Summarizer

Meetings generate transcripts. Transcripts sit unread. Meeting Notes Summarizer fixes that by turning raw transcripts or rough bullet notes into structured meeting notes with decisions, action items (with owners and due dates), and open questions.

The part that actually saves time is commitment extraction. It identifies who said what, pulls out every promise and deadline, and flags items discussed but never resolved. Run it after every call and your follow-up list writes itself.

Compatible with: Claude Code, Cursor, Universal | Category: Content gh skill install anthropics/skills/meeting-notes

Long-Form Writer

When research needs to become a deliverable. Long-Form Writer produces structured long-form content (reports, white papers, articles) with consistent voice, clear argument flow, and proper sourcing. It pairs well with Research Assistant: run the research first, then feed its output here to get a polished report with methodology, findings, and recommendations.

This is the output stage. It handles the structural work that turns raw findings into something you can send to stakeholders or publish.

Compatible with: Claude Code, Codex, Cursor, Universal | Category: Content gh skill install sickn33/antigravity-awesome-skills/long-form-writer

Knowledge Graph Builder

Research produces findings. Without structure, those findings scatter across files and sessions. Knowledge Graph Builder extracts entities, relationships, and facts from text and organizes them into a graph your agent can query later.

This is the skill for ongoing research projects: competitive analysis, market mapping, technical landscape reviews. Anything where findings accumulate over weeks. Instead of re-reading old notes, your agent queries the graph for relevant relationships. It supports entity extraction, relationship mapping, and fact versioning so the graph stays current as new information arrives.

Compatible with: Claude Code, Codex, Universal | Category: Memory gh skill install VoltAgent/awesome-agent-skills/knowledge-graph-builder

Feedback Synthesizer

A specialized skill for qualitative data. Feedback Synthesizer takes user interviews, survey responses, support tickets, or any batch of unstructured feedback and produces themed analysis: frequency counts, sentiment patterns, prioritized actions, and signal-versus-noise separation.

If your research involves understanding what people think (customer discovery, user testing, market validation), this replaces manual affinity mapping. Feed it 50 interview transcripts and get back the top themes, how often each appeared, representative quotes, and what to do about each one.

Compatible with: Claude Code, Cursor, Universal | Category: Product gh skill install alirezarezvani/claude-skills/feedback-synthesizer

How to choose

Start with the shape of your input. Researching a topic from scratch? Research Assistant is the entry point. Already have documents? Document Summarizer or Meeting Notes Summarizer handles the digestion. Need to turn findings into a deliverable? Long-Form Writer produces the output.

For ongoing research that spans multiple sessions, add Knowledge Graph Builder to retain structure across context windows. For qualitative data from people, Feedback Synthesizer is purpose-built.

Most research workflows chain two or three of these together. A common pattern: Research Assistant gathers sources, Document Summarizer digests each one, Long-Form Writer assembles the final report. The structured Markdown output of each skill makes those handoffs clean. And yes, you can chain them in a single agent session. Most coding agents support sequential skill invocation out of the box.

One thing worth noting: Document Summarizer handles PDFs, Word documents, and plain text, but for specialized formats like spreadsheets or slide decks, you may need a format-specific skill (like the DOCX Skill) as a pre-processing step before the summarizer can work with it.