Literature Review Agent

This is the citation-hunting workhorse from the PaperOrchestra paper. It takes your outline, runs 10-15 search queries across intro directions and related work clusters, then drags every candidate through Semantic Scholar with a 70% fuzzy title match and a temporal cutoff tied to your submission deadline. The cache layer and pre-dedup step are mandatory because verification runs at 1 QPS and duplicate papers eat 30-40% of your quota otherwise. You get a verified pool in citation_pool.json, a clean refs.bib, and drafted intro plus related work sections that use 90% of what survived. Runs in parallel with the plotting agent. Expect 20-30 LLM calls and real wall time if you're verifying 100+ candidates sequentially.

GitHub stars: 572
last updated: Jun 2, 2026

View on GitHub

Install

$ npx -y skills add ar9av/paperorchestra --skill literature-review-agent --agent claude-code

Installs into .claude/skills of the current project.

Files

Literature Review Agent (Step 3)

Faithful implementation of the Hybrid Literature Agent from PaperOrchestra (Song et al., 2026, arXiv:2604.05018, §4 Step 3, App. D.3, App. F.1 p.46).

Cost: ~20–30 LLM calls. This is one of the two longest steps (the other is plotting). Wall-time floor is set by Semantic Scholar's 1 QPS verification limit.

Inputs

workspace/outline.json — specifically intro_related_work_plan with the Introduction search directions and the 2-4 Related Work methodology clusters
workspace/inputs/conference_guidelines.md — used to derive cutoff_date
workspace/inputs/idea.md, workspace/inputs/experimental_log.md — for framing the Intro and grounding the Related Work positioning

Outputs

workspace/citation_pool.json — verified Semantic Scholar metadata for every paper that survived verification
workspace/refs.bib — BibTeX file generated from the verified pool
workspace/drafts/intro_relwork.tex — drafted Introduction and Related Work sections, written into the template, with the rest of the template preserved verbatim