Organize, rate, filter, and synthesize paper notes into a curated literature review. Run AFTER split-pdf or cheap-scan1 reading is complete. Single-pass pipeline replaces the old lit-filter + lit-synthesizer two-step.
Single-pass pipeline — collects all paper notes, scores relevance, identifies redundancy, and produces a synthesized narrative. Replaces the old /lit-filter + /lit-synthesizer two-step workflow.
Run this skill AFTER all (or most) papers have been read via split-pdf or cheap-scan1 and have notes.md files in their output directories.
$ARGUMENTS — one of:
notes.md files (from split-pdf or cheap-scan1)paper_skims.md or similar collection fileall — process everything in docs/ and articles/ directoriesIf no argument is given, default to scanning all available sources.
Read all available paper notes from:
articles/split_*/notes.md — deep-read notes from split-pdfarticles/scanned_*/notes.md — deep-read notes from cheap-scan1$RB/docs/paper_skims.md — the master skim file.md files in docs/ containing paper summariesIf a paper has both a skim AND deep-read notes, use the deeper notes. If both split-pdf and cheap-scan1 notes exist, use whichever has more complete section coverage.
Build a quick-reference table:
| # | Paper | Triage Score | Subtopic | Pages | Connection Summary |
Spawn the literature-organizer agent with:
study-parameters.mdarticles/literature_organized.mdAgent prompt template:
You are the literature-organizer agent. Read your specification at:
.claude/agents/literature-organizer.md
Study context: [Load from study-parameters.md — paper title, method, key findings, treatment variables]
[PAPER NOTES/SUMMARIES HERE]
Write the output to: $RB/articles/literature_organized.md
The agent will, in a single pass:
After the agent returns:
articles/literature_organized.mdThe organizer MUST compress these patterns:
| Pattern | Action |
|---|---|
| Multiple papers making the same empirical point | Merge into one sentence citing all |
| Repeated "Connection to Our Study" sections | State relevance ONCE per subtopic |
| Abstract-length summaries of similar papers | Condense to 1-2 sentences within narrative |
| Redundant institutional background | One paragraph max |
| Repeated theoretical mechanism explanations | One theoretical paragraph, cite all |
| Papers with similar methods and findings | Lead with highest-impact, parenthetically cite rest |
| Input | Target Output |
|---|---|
| Many individual paper summaries (~300 lines each) | ~150-200 lines of synthesized narrative |
| Repeated per-paper "Connection" sections | ~20 lines of grouped connection text |
| Multi-dimension extraction per paper | Selective dimensions only |
Before finalizing, verify:
authorYYYYdescriptor format| Task | Model | Why |
|---|---|---|
| Globbing/collecting notes | haiku | Pure file lookup |
| Scoring, filtering, synthesizing | sonnet | Requires domain judgment |
| Applying to QMD narrative | opus | Restructuring into argumentative prose |
articles/literature_organized.md — the single artifact this skill produces.