deep-research

Conducts multi-phase web research and synthesizes verified findings into structured reports.

10|2|Updated May 6, 2026
One-click install
npx skills add https://github.com/jcrabapple/hermes-skills --skill deep-research-jcrabapple
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: deep-research
Source: https://github.com/jcrabapple/hermes-skills/tree/main/deep-research
Command: npx skills add https://github.com/jcrabapple/hermes-skills --skill deep-research-jcrabapple

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Answering complex questions from a single search or general knowledge produces shallow, unverified output. This Skill enforces a systematic multi-source research methodology with corroboration checks, source authority weighting, and citation verification before any report is delivered. ## Core Features & Use Cases - Five-Phase Research Methodology: Broad exploration, deep dive per dimension, diversity validation, synthesis check, and a deterministic URL verification pass. - Depth Modes and Category Templates: Surface, moderate, and exhaustive run budgets, plus output templates for product, comparison, how-to, fact-check, and general questions. - Delegation and Parallel Decomposition: Dispatch subagents via delegate_task or run Mixture-of-Agents parallel research, with documented fallback to inline execution when delegation fails. - Use Case: Ask for a deep dive on a technology market, and the Skill runs multi-angle searches, tallies independent source confirmations per claim, verifies every citation URL with curl, and delivers a structured markdown report with a sources section. ## Quick Start Use the deep-research skill to produce a comprehensive, well-sourced report on solid-state battery commercialization.

Frequently Asked Questions about deep-research

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a deep research report on a topic?▼

Provide the topic and optionally a depth hint like "quick research" or "deep dive". The Skill runs broad exploration, targeted deep dives per dimension, diversity validation, and synthesis, then verifies citation URLs before delivering a markdown report.

What is the difference between surface, moderate, and exhaustive research modes?▼

Surface runs one round with up to three subtopics in about two minutes. Moderate runs two rounds with URL verification in about ten minutes. Exhaustive runs four rounds with a critic filter and three verification retries in about thirty minutes.

When should I use deep-research instead of a quick answer engine?▼

Use deep-research for comprehensive reports, deep dives, and multi-faceted investigations needing 15+ sources. For quick factual answers, recent social sentiment, or proprietary academic datasets, route to answer-engine, last30days, or valyu instead.

How does the Skill prevent fabricated or low-quality citations?▼

Every citation URL is checked deterministically with curl for dead links, and each claim is tallied by independent confirming sources. Single-source claims are flagged explicitly, and citation laundering across articles quoting one study is traced back to the primary source.

What happens if subagent delegation fails during research?▼

The Skill falls back to inline execution using nanogpt_web_search and nanogpt_web_extract directly, or terminal-based Kagi CLI as a last resort. It never silently scrapes search engines with curl, and output quality is maintained in fallback mode.

Can research findings be saved to a personal wiki or blog?▼

Yes. Reference documents cover ingestion into a Karpathy-style LLM wiki at ~/wiki/ with schema, index, and log updates, plus an end-to-end pipeline from research to Obsidian notes, prose.sh blog publishing, and Mastodon posts.