About novelty-matrix
Novelty Matrix creates a claim-by-claim positioning table for paper reviews, matching manuscript claims with related work and recording overlap, differences, and supporting evidence. It uses available references rather than retrieving a new literature corpus.
Key features
- 01Builds a claim-to-work comparison table
- 02Records overlap and contribution differences
- 03Includes evidence notes for each row
- 04Exports Markdown and TSV
Use cases
- 01Assessing novelty during paper review
- 02Comparing claims with supplied related work
Install
Copy the command and run it in a terminal.
$
npx skills add https:/ / github.com/ willoscar/ research-units-pipeline-skills --skill novelty-matrixAfter installing, type/novelty-matrixin Claude Code to confirm it loaded.
Bundled filesscripts/run.py
Details
From the project READMENovelty Matrix
Triggers & routing
- Trigger: novelty matrix, prior-work matrix, overlap/delta, 相关工作对比, 新颖性矩阵.
- Use when:
paper-review中评估 novelty/positioning,需要把贡献与相关工作逐项对齐并写出差异点证据。
Transforms a claim ledger plus related-work surface into a novelty positioning table for paper-review.
Input
Required:
output/CLAIMS.md
Optional:
- manuscript reference list
- user-provided related work list
Output
output/NOVELTY_MATRIX.mdoutput/NOVELTY_MATRIX.tsv(review-novelty-row.v1)
Contract
Each matrix row must expose:
- claim
- closest related work
- overlap
- delta
- evidence note
Script boundary
scripts/run.py should:
- extract candidate related works from available sources
- generate stable overlap/delta rows
- write a table-shaped artifact
Keep matching heuristics and markdown rendering in shared tooling.
Acceptance
- output exists
- includes at least one row per claim
- includes at least three unique related works across the matrix
- if fewer than three related works are available, preserve that limitation in the artifact and block Completion so the user can provide a reference surface
- TSV rows preserve
claim_id, overlap, delta, and evidence as separate fields
Non-goals
- final novelty judgment prose
- retrieval of a new literature corpus
Comments
Markdown supported