lit-organizer

name: lit-organizer description: 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. allowed-tools: Read, Write, Edit, Glob, Grep, Task argument-hint: [path-to-paper-notes-directory or "all"]

Literature Organizer: Rate, Filter, and Synthesize

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.

When to Use

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.

Input

$ARGUMENTS — one of:

• Path to a directory containing notes.md files (from split-pdf or cheap-scan1)
• Path to paper_skims.md or similar collection file
• all — process everything in docs/ and articles/ directories

If no argument is given, default to scanning all available sources.

Step 1: Collect All Paper Notes (haiku — pure lookup)

Read all available paper notes from:

1. articles/split_*/notes.md — deep-read notes from split-pdf
2. articles/scanned_*/notes.md — deep-read notes from cheap-scan1
3. $RB/docs/paper_skims.md — the master skim file
4. Any other .md files in docs/ containing paper summaries

If 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 |

Step 2: Launch Literature Organizer Agent (sonnet)

Spawn the literature-organizer agent with:

• The collected table from Step 1
• The study parameters (key findings, treatment tiers) from study-parameters.md
• Instructions to produce articles/literature_organized.md

Agent 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:

1. Score each paper 1-5 by relevance
2. Assign inclusion tier (A-D) and subtopic codes
3. Identify redundancy clusters
4. Write shortened summaries and synthesized narrative by subtopic
5. Flag coverage gaps

Step 3: Apply to QMD (main conversation)

After the agent returns:

1. Read articles/literature_organized.md
2. Use tier assignments to structure the literature sources QMD chapter
3. Use synthesized subtopic narratives as prose basis
4. Use gap analysis to target further literature searches

Bloat Detection Rules

The organizer MUST compress these patterns:

Target Compression

Output Quality Check

Before finalizing, verify:

• Every Tier A paper appears prominently in the synthesis
• No subtopic section exceeds ~40 lines of prose
• No two consecutive paragraphs make the same point
• All BibTeX keys use authorYYYYdescriptor format
• Effect sizes and p-values are preserved exactly

Model Routing

Output

articles/literature_organized.md — the single artifact this skill produces.