Recursive Research

Features

• Recursive research up to PhD level
• Multi-domain applicability (science, tech, business, arts, humanities)
• Source tiering and automatic rejection of unreliable sources
• Autonomous decision-making via WDM and Munger inversion
• Disk checkpointing to survive context compaction and enable resuming research

Use Cases

• Deeply understanding a new field for informed decision-making
• Preparing for technical documentation, academic papers, or proposals
• Identifying the state-of-the-art and knowledge gaps in a domain
• Conducting comprehensive literature reviews

Non-Goals

• Performing arbitrary web scraping without a research focus
• Replacing structured academic databases for highly specialized queries
• Generating novel scientific hypotheses without user seed input

Workflow

1. User provides research seed, mode, local paths, and preferences.
2. Skill identifies initial research threads using WDM and Munger inversion.
3. Skill detects and prioritizes available MCPs and native tools.
4. Skill iterates through cycles, selecting threads, investigating sources, and consolidating findings.
5. Sources are tiered, and disk checkpoints are saved each cycle.
6. Research closes when PhD criteria are met, cycle cap is reached, or user pauses.
7. Final synthesis, action items, and knowledge gaps are generated.

Practices

• Source tiering
• Autonomous decision-making
• Knowledge consolidation
• Error handling

Prerequisites

• Claude Code environment
• Local project directory for output files