Research Paper Review

Quick start

1. Decide whether the user needs:
   • extraction only
   • a local viz server for existing result JSON
   • the default multi-agent / Claude-style paper-review pass
   • contextualized review that also assesses novelty, impact, and related-work coverage
2. Inside ChatGPT/Codex, default LLM review to the active chat model. Do not require external LLM API keys for normal review.
3. Default to Claude-style / multi-agent orchestration for paper review. Build the full-paper summary first, then spawn one worker per planned review pass when the runtime supports subagents.
4. Use serial single-agent execution only as a fallback when subagents are unavailable, disabled, or clearly impractical for the paper size or runtime.
5. Bootstrap the upstream engine with python3 scripts/install_engine.py only when deterministic helpers are needed for extraction, workspace preparation, viz output, or explicit upstream provider-backed review.
6. Use the /openaireview workspace convention as the canonical stage root: ./review_results/<slug>_review/. Record any non-canonical output paths in artifact-index.md.
7. Use references/chatgpt-native-review.md, references/rating-rubric.md, and the 1-5 issue/output scales.
8. When review quality depends on novelty, impact, or literature context, route through research-systematic-literature-review and research-novelty-review instead of limiting the critique to internal paper consistency.
9. When reusing an existing review workspace, infer its provenance from metadata.json, artifact-index.md, final_issues.json, review_summary.json, and any round-N/ folders before moving files or declaring artifacts missing.
10. Treat research-systematic-literature-review as an independent literature-review skill that can either run on its own or attach to this review workspace through <review_dir>/context/.
11. If the user asks for project-level sequencing, current-state inspection, or coordination across multiple research stages, invoke research-pipeline-planner first instead of treating paper review as the whole task.