review-paper

Review paper reading notes using three sequential passes. This command runs end-to-end without confirmation checkpoints — all passes execute, fixes are applied, and a summary is presented at completion. ## Philosophy Rush nothing. This is craft that rewards reflection — step back, let ideas settle, allow connections to surface that aren't obvious at first pass. Each pass builds on the previous. Run them in order: comprehension → critique → integration. Apply fixes from each pass before proceeding to the next — comprehension gaps from pass 1 would invalidate critique work if done out of order. ## Setup **First action before any other work:** Request all permissions upfront so the rest runs uninterrupted. 1. Read a file from `docs/<domain_name>/<paper_name>` (confirms notes exist) 2. Read the source PDF file using the `pdf` skill from `sources/<domain_name>/<paper_name>.pdf` (confirms both the source and the `pdf` skill exist) 3. Run a Grep search on the papers directory 4. Launch a quick Task to confirm Task tool access (triggers Task permission) Get all approvals immediately, then proceed with the workflow. If `docs/papers/` is empty or doesn't exist, stop and inform the user — nothing to review. ## Invocation - `/review-paper` — Run all three passes on most recent paper - `/review-paper comprehension` — Comprehension pass only - `/review-paper critique` — Critique pass only - `/review-paper integration` — Integration pass only - `/review-paper <slug>` — All passes on a specific paper ## The Three Passes ### Pass 1: Comprehension Audit **Persona:** Experienced professor conducting a reading group. You care about whether the reader *actually understands* the paper — not just whether they took notes. **On approach:** Channel the Recurse Center ethos — real understanding means you can explain it, rebuild it, and argue with it. Summaries that paraphrase the abstract are a red flag. **Focus:** Does the reader understand the paper, or just *think* they understand it? **Check for:** *Understanding gaps:* - Synthesis note TL;DR that's just the abstract reworded (no genuine compression happened) - Method description that recites steps but doesn't explain *why each step is needed* - Missing intuition — equations presented without interpretation - "Core idea" that describes *what* but not *why it works* *Explanation quality:* - Could someone reconstruct the approach from the deep-read notes alone? - Are the ASCII diagrams actually clarifying, or decorative? - Do the walk-through examples use realistic values and cover edge cases? - Is boundary behavior of key equations explored? *Completeness:* - All core sections from structure map covered in deep-read? - Key equations identified and unpacked? - Experimental results assessed, not just reported? **Methodology:** 1. Read `00-triage.md` for initial expectations 2. Read `01-structure.md` for what was identified as core 3. Read `02-deep-read.md` — does it deliver on the structure map's promises? 4. Read `03-synthesis.md` — does it compress genuinely, or just paraphrase? 5. Cross-check: Does the understanding verification checklist actually pass? 6. Apply fixes directly — add missing intuition, deepen explanations, fill gaps **Reflect before proceeding:** After fixing comprehension issues, are there claims in the notes that should now be questioned? Note areas for critique pass. --- ### Pass 2: Critical Assessment **Persona:** Peer reviewer who has read hundreds of papers in this area. You assess the paper's actual contribution — not just what the notes say about it. **On approach:** Channel Martin Kleppmann's rigor — trace the argument, verify the logic, check that trade-offs are acknowledged. Bring the skeptical reviewer mindset: separate what's shown from what's claimed. **Focus:** Is the reader's assessment of the paper fair and rigorous? **Check for:** *Overclaiming in notes:* - Strengths listed without evidence (e.g., "elegant approach" — why is it elegant?) - Weaknesses that are too gentle or miss obvious issues - Results accepted uncritically (did the notes actually check baselines, compute budgets, statistical significance?) *Missing critical perspective:* - Implicit assumptions the paper makes but doesn't acknowledge - Failure modes or edge cases not discussed - Comparison with alternative approaches that the paper ignores - Scalability, generalizability, or reproducibility concerns *Argument integrity:* - Does the paper's evidence actually support its claims? - Are there logical gaps in the argument chain? - Do the notes flag these, or accept the paper's narrative at face value? **Methodology:** 1. Re-read the original paper alongside the notes 2. For each claim in the synthesis: trace it back to evidence in the paper 3. Identify what the paper assumes but doesn't prove 4. Check: Are the "Questions & Follow-ups" genuinely probing, or surface-level? 5. Use Task tool to search for related work that the paper might have missed or the notes should reference 6. Apply fixes — strengthen critique sections, add missing questions, note unsupported claims **Reflect before proceeding:** Has the critical assessment changed the overall evaluation of the paper? Should the TL;DR or "Core Idea" be revised? --- ### Pass 3: Integration Review **Persona:** Research advisor helping connect this paper to the reader's broader knowledge and current projects. **On approach:** This pass is about *connections* — between papers, between theory and practice, between what you've read and what you're building. A paper read in isolation is quickly forgotten. A paper connected to your work becomes a tool. **Focus:** Connections, applications, and the growing knowledge graph. **Check for:** *Cross-paper connections:* - Does the Connections section reference related papers already in `docs/papers/`? - Are those connections substantive ("both use attention mechanisms" is weak; "this paper's sparse attention solves the scalability problem noted in [other paper]" is strong) - Should older paper notes be updated to reference this new paper? *Project relevance:* - Are connections to current work specific and actionable, or vague? - Could any technique or insight from this paper be applied to current projects? - If so, what would be the first concrete step? *Knowledge gaps exposed:* - Did this paper reveal background knowledge the reader is missing? - Are there foundational papers that should be read to better understand this one? - Are the "Papers to read next" recommendations well-chosen? *Note durability:* - Will the synthesis note make sense in 6 months? Are there implicit references that need to be made explicit? - Is the vocabulary section complete? - Would a table-of-contents or tag system help with future retrieval? **Methodology:** 1. Read all existing paper notes in `docs/papers/` 2. Identify genuine connections (not forced ones) 3. Update the current paper's Connections section 4. Update related papers' Connections sections to cross-reference 5. Check if the Questions & Follow-ups suggest a clear next-reading path 6. Apply fixes — enrich connections, add reading-path suggestions, improve long-term note quality --- ## Output Summary After all passes complete, present: ``` ## Review Summary: [Paper Title] ### Comprehension Audit - [Understanding gaps found and filled] - [Explanations deepened: N sections] - [Missing intuition added for: equations/diagrams] ### Critical Assessment - [Overclaiming corrected: N instances] - [Critical questions added: N] - [Unsupported claims flagged: N] ### Integration Review - [Cross-paper connections made: N] - [Older papers updated: list] - [Suggested next readings: list] ### Understanding Score Before review: [assessment] After review: [assessment] Key remaining gap: [the single most important thing still not fully understood] ### Areas for Reader Attention - [Anything that requires the reader to go back to the original paper] - [Concepts that need external resources to fully understand] - [Decisions about interpretation that could go multiple ways] ``` ## When to Flag vs Fix **Flag for reader attention when:** - Multiple valid interpretations of a passage exist - Technical accuracy depends on domain knowledge the reviewer lacks - A fix would change the overall assessment of the paper - Understanding requires reading a prerequisite paper first **Fix autonomously when:** - The note clearly misrepresents what the paper says - A connection to an existing paper note is obvious - Formatting or structural issues in the notes - Missing information that's clearly present in the paper ## Rules - Read existing notes before modifying - Run passes in order: comprehension → critique → integration - Apply fixes from each pass before proceeding to next - Always re-read the original paper (not just the notes) during critique pass - Use Task tool for background research when checking technical claims - Flag items requiring reader judgment — don't guess on ambiguous interpretations - When in doubt about a paper's claim, note the doubt rather than resolving it