lamm-mit/skills/idea-generation

Idea Generation Skill Summary

This workflow generates 5 research ideas grounded in literature, following this process:

Key Steps

1. Check Workspace – Audit existing papers in $W/papers/
2. Plan Strategy – Decide whether to use current collection or search for more papers
3. Acquire Resources – Either delegate to /research-collect (100+ papers) or run quick searches
4. Analyze Literature – Extract contributions, methods, limitations, and gaps from papers
5. Generate Ideas – Create 5 distinct concepts using different strategies (combination, simplification, generalization, constraint relaxation, architecture innovation)
6. Score & Enhance – Rate ideas by novelty/feasibility/impact; detail the top candidate with math and implementation roadmap
7. Code Mapping – Link concepts to reference implementations
8. Summarize – Document all findings and next steps

Core Principle

"Ideas MUST be grounded in actual papers, not generated from model knowledge." Each idea requires citations (≥2 papers by arXiv ID).

Output Location

Results save to $W/ideas/ as individual markdown files, plus consolidated summary.

Integration

Works with /research-collect (upstream) and /research-pipeline (downstream for implementation).