| name | referee-report |
| description | Produce a multi-perspective referee report on a working paper PDF. Spawns three reviewer subagents (empirical, theory, methods), then consolidates their memos into a single editor-facing report. Triggers on commands like /referee-report or when the user uploads a PDF and asks for a "referee" / "review". |
| args | {"paper":"path to a PDF","output":"optional path for the consolidated report (default review/<paper-stem>.md)"} |
Multi-perspective referee report
Phase 1 -- prepare
Read the PDF (use the Read tool with appropriate page ranges). Extract:
- Title, authors, abstract
- Section headings
- Key tables / figures (capture captions; flag the page numbers)
If the PDF is > 30 pages, surface the structure first and ask the user to confirm the focus before spawning subagents.
Phase 2 -- spawn reviewers in parallel
Issue THREE Task tool calls in the same turn so they run in parallel:
Task(subagent_type="empirical-econ-reviewer",
description="Empirical referee memo",
prompt="Read the paper at <paper>. Produce a referee memo
per your persona. Cite page numbers.")
Task(subagent_type="theory-reviewer",
description="Theory referee memo",
prompt=...)
Task(subagent_type="methods-critic",
description="Methods critic memo",
prompt=...)
Each subagent returns a single memo following its persona's template.
Phase 3 -- consolidate
Produce a single editor-facing report with these sections:
- Contribution claim (one sentence; agreement across all three?)
- Agreements -- bullet points all three reviewers raised
- Disagreements -- where reviewers diverge (the interesting part)
- Per-reviewer summaries -- 3-5 sentence summary of each memo
- Suggested decision -- revise / reject / accept, with the strongest argument for and against
- Citations -- page-level pointers from each reviewer
Phase 4 -- emit
Write to output path (default review/<paper-stem>.md).
Print the path. Do not commit unless asked.
