ultra-research

Builds a multi-file knowledge base through recursive multi-agent web research until saturation.

4|Updated Apr 17, 2026
One-click install
npx skills add https://github.com/hint-shu/deep-research --skill ultra-research-hint-shu
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ultra-research
Source: https://github.com/hint-shu/deep-research/tree/main/skills/ultra-research
Command: npx skills add https://github.com/hint-shu/deep-research --skill ultra-research-hint-shu

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Ad-hoc web research produces shallow, unverifiable answers with no audit trail, and one-off reports cannot support long-term expertise on a strategic topic. ## Core Features & Use Cases - Recursive exploration until saturation: Runs up to 8 iterations of planner, three parallel researchers, fact-checker, and critic agents, stopping when no new verified facts appear. - Full knowledge base vault: Produces a 10,000+ word main report plus executive summary, glossary, timeline, key players, actionable playbooks, steel-manned counter-arguments, annotated bibliography of 150+ sources, and PDF/EPUB exports. - Cross-model verification: Uses Codex CLI, Perplexity, and Exa deep-researcher channels alongside Tavily/Firecrawl to fact-check claims, with three independent peer-review passes. - Use Case: Ask to become an expert on a technology domain; the skill runs the L1–L4 research ladder as a foundation, then iterates until saturation and syncs the resulting vault to long-term memory. ## Quick Start Run /ultra-research on the topic you want to master and confirm the budget prompt to generate a complete knowledge base.

Frequently Asked Questions about ultra-research

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

FAQPage Schema
How do I run a deep multi-source research investigation with Claude Code?▼

Invoke /ultra-research with your topic and confirm the budget prompt. It executes the L1–L4 research ladder first, then runs recursive exploration iterations until saturation, producing a full knowledge base vault in .firecrawl/research/<slug>/L5/.

What is the difference between deep-research, academic-research, and ultra-research?▼

deep-research (L2) takes ~12 minutes with 20–30 sources, academic-research (L4) takes ~40 minutes with 80–120 sources and a bibliography, and ultra-research (L5) runs 1+ hours with 150+ sources, recursive iterations, peer review, and a full knowledge vault.

How much does an ultra-research run cost in credits and tokens?▼

A single L5 run uses roughly 1000 Tavily credits and 500k–1M LLM tokens, estimated at $5–20 per session. The skill always shows a cost and time warning and requires user confirmation before starting.

Can ultra-research export results to Obsidian or PDF?▼

Yes. It generates knowledge-base.pdf and knowledge-base.epub via pandoc, and optionally exports the vault to an Obsidian vault with wiki-links when the OBSIDIAN_VAULT environment variable is set.

What happens if ultra-research crashes mid-run?▼

All state is checkpointed to disk after each iteration, including crew outputs, sources, and saturation-log.md. You can resume with ultra-research --resume <slug>, which continues from the last completed iteration without touching L1–L4 artifacts.

When should I not use ultra-research?▼

Avoid L5 for quick fact-checks or standard overviews; use L0 quick-research or L1 research instead. L5 is only justified for one-time deep investments in topics you will return to, given its hour-plus runtime and multi-dollar cost.