paper-deepstudy

Augment analysis/05-prior-work.md with a new prior work entry that the auto-run pipeline missed. Accepts BibTeX entry, arXiv URL, or free-text "author + year + one-line description". Updates timeline, comparison table, and lineage diagram.

Head-to-head comparison of two papers. Outputs compares/vs-<other-slug>.md with sections for problem, formalization, method, experiments, strengths/weaknesses, and when-to-use-which decision guidance.

Produce a focused deep dive on a single user-specified topic in the paper, going beyond what the auto-run analysis/ files cover. Outputs deep-dives/<topic-slug>.md.

Apply a user edit instruction to an existing rendered note (xhs or wechat). Reads the current rendering, dispatches the matching renderer with EDIT_INSTRUCTION, writes back with a backup of the prior version.

Audit the paper's reproducibility along 7 dimensions (data, code, hyperparameters, seeds, hardware, evaluation scripts, wet-lab protocol). Each dimension rated ✓ / ✗ / partial with cited evidence. Suggests review-round if serious issues are found.

Re-run a specific stage of paper-deepstudy on the most recently studied paper.

Re-pick which figures get embedded in xhs.md and wechat.md. Shows the user the figures from analysis/06-figures.md with their importance scores and captions; user multi-selects per platform; renderers are re-dispatched with the new selections.

Regenerate 5 title candidates for an existing xhs or wechat rendering, let the user pick one, swap it into the rendering, and archive the prior title. Optional --style filter to bias the candidates.

Run an adversarial review round on a paper already studied via /paper:study. The user raises objections; defense + judge sub-Agents argue them out; the user has final say; accepted weaknesses or questions are appended to review.md.

Deep-study a paper for ML / computational biology. Produces analysis files, a review draft, and Chinese xhs/wechat learning notes.

Use when the user wants a head-to-head comparison of two papers. Auto-studies the other paper if not already in ~/claude-papers/papers/, then dispatches compare-agent with both analysis directories.

Use when the user wants an in-depth treatment of a single topic in a paper that is already studied. Dispatches deep-dive-agent with topic + paper text + analysis files. Output lands at deep-dives/<topic-slug>.md.

Use when the user wants to refine an existing xhs.md or wechat.md rendering with a specific edit instruction. Dispatches the matching renderer with EDIT_INSTRUCTION, backs up the prior version.

Use when the user wants a structured reproducibility audit of a studied paper. Dispatches reproduce-checker on 7 dimensions (or 6 for ml-pure papers — wet-lab is N/A). Output goes to reproduce-check.md. Suggests /paper:review-round if serious issues are found.

Use when the user wants to change which figures are embedded in xhs.md and wechat.md. Shows the user the figures with importance scores from analysis/06-figures.md, lets them multi-select per platform, re-runs both renderers with new selections.

Use when the user wants to regenerate title candidates for an existing xhs.md or wechat.md rendering and pick a new one. Re-dispatches title-generator with optional --style filter, lets the user choose, swaps the title into the rendering, archives the prior title.

Use when the user wants to run an adversarial review round on a paper that has already been studied via /paper:study. Solicits objections, dispatches defense+judge sub-Agents, gets the user's final verdict, and updates review.md with accepted weaknesses or questions.

Use when the user wants to deep-study a paper (PDF or URL) for ML or computational biology. Produces analysis/, review.md, and Chinese xhs/wechat notes. Layers on top of claude-paper:study.

Use when the user wants to add a missing prior-work entry to analysis/05-prior-work.md. Re-dispatches prior-work-historian with the existing file + the new reference, asks it to update timeline + comparison table + lineage diagram.