literature-reviewer

name	literature-reviewer
description	Use when you want Cursor Agent to review candidate related-work papers and write decisions to assets/notes/agent_artifacts/literature/reviewer_output.json so you can update references.bib and prose safely.

Literature reviewer

This skill is for screening candidate papers, not for writing prose.

Use it when:

• assets/notes/agent_artifacts/literature/provider_output.json exists (from related-work-scout), or you have merged candidates into one {"papers":[...]} file at that path
• the goal is to decide which papers you may cite in Related Work
• the model must behave like a reviewer, not an author

Do not use it to:

• rewrite main.tex without your explicit follow-up
• write any explanation outside the reviewer artifact

Output contract

Write exactly one JSON artifact to:

assets/notes/agent_artifacts/literature/reviewer_output.json

Read the schema in references/reviewer-output-schema.json before producing output.

Allowed top-level shapes:

{"reviews": []}

or a bare JSON array.

Each review item must contain:

• paper_id
• decision
• risk_level
• usable_in_draft
• rationale
• issues

Allowed decisions:

• auto_approve
• auto_reject
• needs_review

Workflow

1. Read assets/notes/project.md, the candidate papers JSON at assets/notes/agent_artifacts/literature/provider_output.json, and any assets/refs/*.json for extra context.
2. Read references/reviewer-output-schema.json.
3. Review every candidate paper.
4. Write decisions to assets/notes/agent_artifacts/literature/reviewer_output.json.
5. Stop.

Constraints

• Review every paper in the candidates file; do not skip ids.
• Use auto_approve only when the paper is safe to cite in Related Work.
• Use auto_reject for clear mismatches, excluded items, or unusable metadata.
• Use needs_review for ambiguous relation, shaky metadata, or uncertain setting overlap.
• Keep issues short and concrete.
• Do not output Markdown, prose explanation, or comments in the artifact.

Prompts

For Codex, reuse references/codex-prompt.md.

For Claude, reuse references/claude-prompt.md.