Humboldt

Inspired by Alexander von Humboldt — a true polymath who spent a lifetime crossing continents, measuring everything, and synthesizing the natural world into a unified picture of knowledge. This plugin's purpose is to do as much in one turn as he did in a lifetime.

Deep Cycle Research — an autonomous Claude Code plugin that drills any research question across multiple cycles, builds a linked wiki, and evaluates coverage gaps until the topic is exhausted.

How it works

Each research cycle has two phases:

1. Drill — 5–8 parallel agents each run one search query end-to-end: search → select best URLs → fetch full pages → extract knowledge → write/merge topic files in backlinks/
2. Synthesize — one pass reads all backlink files, updates wiki.md, evaluates coverage across 9 knowledge dimensions, and plans the next cycle's queries

The loop continues until coverage exceeds 90%, sources are depleted, or 30 cycles complete.

Output

File	Description
wiki.md	Main synthesis document — every concept links to a topic page
backlinks/*.md	Topic pages — one per concept, merged across cycles
research/log.jsonl	Full audit trail of every query and phase
research/directions.md	Gap analysis and queries planned for next cycle
archive/	Timestamped snapshots of completed research runs

---

Installation

Option A — Load for one session (no install required)

claude --plugin-dir /path/to/Humboldt

Use this to try the plugin or run it from within a specific research directory.

Option B — Install permanently via /plugin

1. Clone or download this repository:

   git clone https://github.com/fjfok/Humboldt.git

2. Open Claude Code and run:

   /plugin install /path/to/Humboldt
   

3. Verify the plugin loaded:

   /help
   

   You should see the humboldt skills listed.

Option C — Add to a team marketplace

Add a pointer to this repo in your .claude/marketplace.json:

{
  "plugins": [
    { "path": "/absolute/path/to/Humboldt" }
  ]
}

---

Usage

All skills are namespaced under /humboldt:.

/humboldt:autodeepsearch {question}

Run a full autonomous research cycle on any question. Runs until completion with no further input needed.

/humboldt:autodeepsearch What are the economic effects of universal basic income?

/humboldt:drill {query} | {research question}

Run a single targeted drill for one query. Useful for filling a specific gap after a full run.

/humboldt:drill UBI pilot programs Finland Norway | economic effects of universal basic income

If you omit the | separator, the query serves as both the search term and the research context.

/humboldt:synthesize {question}

Re-synthesize wiki.md from existing backlink files. Run this after one or more manual drills to incorporate new findings.

/humboldt:synthesize What are the economic effects of universal basic income?

/humboldt:archive {label}

Move current research output into archive/{label}--{date}/ so you can start a new run.

/humboldt:archive ubi-economic-effects

If no label is given, one is derived from the wiki.md title.

/humboldt:clean

Delete all current research output (wiki, backlinks, research state). Always asks for confirmation first. Use /humboldt:archive instead if you want to keep the results.

---

Tips

• Run /humboldt:autodeepsearch from the directory where you want output written — wiki.md and backlinks/ are created relative to your working directory.
• Use /humboldt:drill + /humboldt:synthesize together to manually guide research when you know exactly what gap to fill.
• After archiving, the directory is clean and ready for a completely new topic.