Skip to content

feat: add deep-research skill — autonomous multi-source research agent#13412

Open
vominh1919 wants to merge 1 commit into
NousResearch:mainfrom
vominh1919:feat/deep-research-skill
Open

feat: add deep-research skill — autonomous multi-source research agent#13412
vominh1919 wants to merge 1 commit into
NousResearch:mainfrom
vominh1919:feat/deep-research-skill

Conversation

@vominh1919

Copy link
Copy Markdown
Contributor

What

An autonomous deep research skill that takes any question and produces a comprehensive, multi-source research report with citations, confidence ratings, and actionable conclusions.

Why

Users frequently need to research complex topics but:

  • Web search alone gives shallow results
  • Manual research across 10+ sources takes hours
  • Synthesizing conflicting information is hard
  • No structured output with citations

This skill automates the entire research workflow.

How It Works

  1. Decompose: Breaks the question into 5-8 focused sub-questions
  2. Research: Searches web + arXiv for each sub-question
  3. Validate: Cross-validates findings across sources
  4. Synthesize: Identifies patterns, contradictions, gaps
  5. Report: Generates structured report with citations

Features

  • Multi-source research (web search + arXiv papers)
  • Cross-validation across sources
  • Confidence ratings (High/Medium/Low) per finding
  • Source quality assessment (A/B/C/D rating)
  • Structured report with executive summary
  • Knowledge gaps identification
  • Actionable recommendations
  • Methodology guide for consistent research quality
  • Report templates

Use Cases

Use Case Example
Technology evaluation "Compare Rust vs Go for microservices"
Market analysis "AI agent framework market landscape"
Literature review "Recent advances in chain-of-thought reasoning"
Competitive analysis "Cursor vs Copilot for enterprise"
Due diligence "Evaluate Nous Research ecosystem"

Usage

/deep-research What are the best practices for building AI agent frameworks in 2026?

Files

  • skills/research/deep-research/SKILL.md — Main skill documentation
  • skills/research/deep-research/references/methodology.md — Research methodology guide
  • skills/research/deep-research/templates/report.md — Report template

@vominh1919 vominh1919 force-pushed the feat/deep-research-skill branch from 5f4561b to 00fe4f8 Compare April 21, 2026 08:23
A comprehensive research skill that takes any question and produces
a structured report with citations, confidence ratings, and recommendations.

Features:
- Auto-decomposes questions into sub-questions
- Multi-source research (web + arXiv)
- Cross-validation across sources
- Confidence ratings per finding
- Structured report with citations
- Methodology guide with source quality ratings
- Report templates

Use cases: market analysis, tech evaluation, literature review,
competitive intelligence, due diligence, policy research.
@vominh1919 vominh1919 force-pushed the feat/deep-research-skill branch from 00fe4f8 to 672a0a1 Compare April 21, 2026 08:50
@alt-glitch alt-glitch added type/feature New feature or request P3 Low — cosmetic, nice to have comp/tools Tool registry, model_tools, toolsets tool/skills Skills system (list, view, manage) labels Apr 22, 2026
@Abd0r

Abd0r commented May 3, 2026

Copy link
Copy Markdown
Contributor

Hi @vominh1919 — opened #19341 today not knowing this PR existed (thanks @alt-glitch for the pointer). Going to defer to your prior work on the methodology side.

A few additive pieces in #19341 that are independent of this PR — I'm offering to coordinate so they don't collide:

  • tools/local_web_tools.py (free-tier local_web_search / local_web_extract, same JSON contract as web_tools.py but using SearXNG / ddgr / ddgs / Brave free / Tavily free instead of paid backends) — purely a tool, no overlap with the skill.
  • scripts/start-llama-server.sh — turnkey llama.cpp launcher, auto-detects Qwen3.5/3.6 from filename.
  • Qwen3.5 / Qwen3.6 ops notes: these models default to thinking mode and do not honor /think /no_think the way Qwen3 did — only chat_template_kwargs.enable_thinking=false works (per the official model card). If useful, I can send a follow-up PR adding a one-paragraph note to your SKILL.md so users on Qwen3.5/3.6 don't hit the silent-thinking-mode failure mode.

Whatever the maintainers decide on the methodology split, your PR has my support — happy to coordinate or just stay out of the way.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

comp/tools Tool registry, model_tools, toolsets P3 Low — cosmetic, nice to have tool/skills Skills system (list, view, manage) type/feature New feature or request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants