Neura Market
recursive-research logo

recursive-research

Free

Recursive research up to PhD level across any domain (science, tech, business, arts, humanities) with source tiering, WDM + Munger inversion for autonomous decisions, and disk checkpointing to survive context compaction. *By [@Anjos2](https://github.com/Anjos2)*

FreeFree tier
Type
Open Source
Visit Website GitHub (40 stars)

About recursive-research

Recursive Research is an open-source Claude Code plugin skill that performs autonomous, recursive research up to PhD level across any domain—science, tech, business, arts, and humanities. It uses source tiering (Tier 1/2/3/Rejected), a Weighted Decision Matrix (WDM) with Munger inversion for self-critical decision-making, and per-cycle disk checkpointing to survive context compaction. The skill interrogates the user in Phase 0 to determine mode (web/local/mixed), priority/excluded sources, and a cycle cap. It identifies 3-5 seed threads, detects available MCPs (Firecrawl, Context7, WebFetch, WebSearch), iterates in auto-regulated cycles, and closes when a 5-criteria PhD fitness function is met or upon hitting the cap. Every autonomous decision is transparent and logged. Created by Joseph Huayhualla (@Anjos2), licensed under MIT.

Key Features

Recursive research up to PhD level across science, tech, business, arts, humanities
Source tiering (Tier 1/2/3/Rejected) with explicit rejection criteria (no author, data-less marketing, SEO spam, unsupervised AI content)
Weighted Decision Matrix (WDM) + Munger inversion for self-critical autonomous decisions
Per-cycle disk checkpointing and --resume mode to survive context compaction
Phase 0 interrogation of the user (mode, sources, cycle cap)
Auto-regulated cycles selecting least-covered thread each iteration
Detects and prioritizes available MCPs (Firecrawl, Context7, WebFetch, WebSearch) by speed and quality
5-criteria PhD fitness function to determine research completion
Transparent decision logging for every non-trivial autonomous choice
Supports infinite research continuation upon user request

Pros & Cons

Pros
  • Works across any domain with generic source tiering, not limited to code
  • Automatically rejects garbage sources using explicit criteria
  • Survives context limits via checkpointing and resume support
  • Self-critical with Munger inversion applied to consolidated knowledge
  • Transparent autonomous decisions with WDM + Munger reasoning shown
  • User controls via Phase 0 interrogation before research begins
  • Open source (MIT) and free to use with Claude Code
Cons
  • Requires Claude Code (Anthropic's CLI tool) as a dependency
  • Research quality depends on user-provided seed and available MCPs
  • May consume significant context/tokens for deep recursive cycles

Best For

Academic research across disciplines (papers, academic books, official archives, raw data)Business intelligence and competitive analysisTechnology landscape exploration and trend analysisArts and humanities researchLiterature reviews and knowledge synthesisInvestigative research requiring source quality evaluation

FAQ

How does the tool survive context limits?
It saves per-cycle disk checkpoints and supports a --resume mode, allowing research to continue across new sessions.
How does it reject garbage sources automatically?
It applies explicit criteria: sources without author, data-less marketing, SEO spam, and unsupervised AI content are rejected.
What is Munger inversion and why is it used?
Munger inversion is applied to the consolidated knowledge to force self-criticism: 'what do I not know?', 'what bias do my sources share?', 'what's missing?' — ensuring deeper analysis.
Can I use this skill with local files?
Yes, during Phase 0 you can choose between web, local, or mixed mode and specify local paths if applicable.
How do I install recursive-research?
The easiest way is through Claude Code's plugin manager: run '/plugin marketplace add Anjos2/recursive-research' then '/plugin install recursive-research'. Alternative methods include copying the skill file manually.