hunt-papers

name	hunt-papers
description	Searches arxiv for papers directly relevant to the techniques and gaps identified in the codebase scan
license	MIT
allowed-tools	arxiv-search
metadata	{"author":"aaryan-guglani","version":"2.0.0","category":"research","risk_tier":"low","knowledge_refs":"arxiv-categories"}

Hunt Papers

Use the arxiv-search tool to find exact matching papers, foundational works, or recent advancements directly related to the contents of TECHNIQUES.md.

Instructions

1. Read the TECHNIQUES.md file generated by the scan-codebase skill.
2. Consult knowledge/arxiv-categories.md for category codes to narrow queries.
3. For each major architectural element or training pattern, form an optimal, targeted search query and call the arxiv-search tool. Use both the canonical technique name (from the taxonomy) and common variants.
4. For each notable absence listed in TECHNIQUES.md, also search for the missing technique — this is often where the highest-value gaps are found.
5. Compile a deduplicated list of the most relevant papers found across all queries. Deduplicate by arxiv ID.
6. Assign a relevance tier: HIGH (directly applies to detected code), MEDIUM (applies to similar architecture), LOW (background reading only).
7. Output findings as PAPERS.md in the root of the project.

Output Format (PAPERS.md)

# Papers

## HIGH Relevance

- **[arxiv ID]**: [Paper Title] ([Year])
  - Related to technique: [Technique Name from TECHNIQUES.md]
  - Relevance reason: [Why this paper directly applies]

## MEDIUM Relevance

- **[arxiv ID]**: [Paper Title] ([Year])
  - Related to technique: [Technique Name]
  - Relevance reason: [Why this is related but not a direct match]