NoveltyAudit v0.3.1 — Alpha

NoveltyAudit v0.3.1 — Alpha

No single paper kills your novelty? Maybe three do.

NoveltyAudit is an evidence-first Agent Skill that finds the smallest set of prior papers that can collectively challenge a scholarly novelty claim—and checks whether those papers were meaningfully connected before the historical cutoff.

Highlights

• Minimal Prior Sets of one to three papers
• Strict earliest-public-date cutoff handling
• OpenAlex, Semantic Scholar, arXiv, and Crossref integration
• Multi-provider backward and forward citation expansion
• Bridge-aware multi-paper analysis
• Full-text evidence bound to claim facets
• Deterministic report validation and runtime provenance
• Explicit INCONCLUSIVE state when obligations are incomplete
• No fabricated novelty percentage

Install

Download both release assets:

• scholarly-novelty-audit-v0.3.1.zip
• scholarly-novelty-audit-v0.3.1.zip.sha256

Verify the archive before extracting it:

sha256sum -c scholarly-novelty-audit-v0.3.1.zip.sha256       # Linux
shasum -a 256 -c scholarly-novelty-audit-v0.3.1.zip.sha256  # macOS

Then copy the extracted scholarly-novelty-audit folder into the skills directory discovered by your Agent Skills-compatible host and install its requirements.txt.

Alpha status

The deterministic pipeline and validation contracts are implemented and tested. End-to-end reviewer-grounded Recall@5, MRR, and reviewer-prediction performance have not yet been established, and schemas or CLI interfaces may evolve before 1.0.

See the README for the Quick Start, reproducible report example, methodology, and scope limits.

Full Changelog: https://github.com/niansia/NoveltyAudit/commits/v0.3.1