Literature Review

Run a structured literature review with primary-source synthesis, consensus mapping, and gap analysis.

The literature review workflow produces a structured survey of the academic landscape on a given topic. It focuses on mapping the state of the field — what researchers agree on, where they disagree, and what remains unexplored — and delivers a cited, verified output to outputs/.

Invocation

CLI

feynman lit "<topic>"

REPL

/lit <topic>

Examples

feynman lit "scaling laws for language model performance"
feynman lit "diffusion models for protein structure prediction"

/lit mechanistic interpretability survey
/lit "retrieval-augmented generation benchmarks"

Workflow stages

1. Plan

   The lead agent outlines the review scope: key questions, source types to search (papers, web, repositories), time period, expected sections, and a small task ledger with a verification log.

   The plan is written to outputs/.plans/<slug>.md. The lead agent presents the plan and waits for your confirmation before proceeding.

2. Gather

   For wide sweeps, the researcher subagent is used for delegated paper triage before synthesis. For narrow topics, the lead agent searches directly.

   Researcher outputs are saved to <slug>-research-*.md files. Assigned tasks are never silently skipped — each is marked done, blocked, or superseded.

3. Synthesize

   Findings are organized into three categories: consensus, disagreements, and open questions. Where useful, concrete next experiments or follow-up reading suggestions are proposed.

   Quantitative comparisons across papers are rendered as charts using pi-charts; taxonomies and method pipelines use Mermaid diagrams. Before finishing the draft, every strong claim is swept against the verification log and downgraded if it is inferred or single-source on a critical point.

4. Cite

   The verifier subagent adds inline citations to the draft and verifies every source URL.

5. Verify

   The reviewer subagent checks the cited draft for unsupported claims, logical gaps, zombie sections (text that references material no longer in scope), and single-source critical findings.

   • FATAL issues are fixed before delivery. If FATAL issues are found, one additional verification pass is run after fixes.
   • MAJOR issues are noted in Open Questions.
   • MINOR issues are accepted.
6. Deliver

   The final literature review is saved to outputs/<slug>.md. A provenance record is written alongside it at outputs/<slug>.provenance.md.

Outputs

Artifact	Path
Research plan	outputs/.plans/<slug>.md
Intermediate research files	<slug>-research-*.md
Final literature review	outputs/<slug>.md
Provenance record	outputs/<slug>.provenance.md

Provenance record

The <slug>.provenance.md sidecar records:

• Date of the run
• Sources consulted vs. accepted vs. rejected
• Verification status
• Intermediate research files used

Every /lit output requires a .provenance.md sidecar. This is enforced by the workspace contract in AGENTS.md.

Subagents used

Subagent	Role
researcher	Gathers and triages source material when the sweep is wide enough to benefit from delegation
verifier	Adds inline citations and verifies source URLs
reviewer	Checks the cited draft for logical gaps and unsupported claims

When to use /lit

Use /lit when you need a map of the research landscape rather than a deep dive into one specific question. It is particularly useful when:

• Starting a new research project and needing to understand what has already been done
• Preparing a related work section for a paper
• Evaluating whether a research direction is sufficiently novel
• Identifying open problems in a field

For a deeper investigation of a single focused question, use Deep Research instead.

Related

• Deep Research — thorough multi-agent investigation on a specific question
• Source Comparison — structured agreement/disagreement matrix across sources
• Paper Draft — turn review findings into a paper-style draft