scholaraio: Skill for Claude Code

Scrub Metadata

Use this skill when the library contains already-ingested papers whose metadata is still clearly low quality after ingest or enrich, especially for non-standard documents that MinerU or fallback parsers converted successfully but described poorly.

scrub is a review-and-repair workflow, not a blind batch rewrite. It should reuse existing ScholarAIO repair and rename primitives, and it should treat .scrubbed as the durable marker for "reviewed and currently acceptable."

When To Use

Use this skill when the user wants to:

• clean bad metadata after enrich
• repair placeholder or garbled titles
• fix suspicious author names
• fill in missing years when the paper content supports it
• incrementally review a large library without reprocessing already-reviewed papers

Do not use this skill for:

• normal ingest
• DOI or citation-count refresh
• paper-content enrichment such as TOC/L3 extraction
• directory normalization when metadata is already trustworthy and rename alone is enough

Workflow

1. Find unreviewed candidates

Skip papers that already contain .scrubbed.

You can list suspicious, unreviewed papers with a Python helper that resolves papers_dir from the active ScholarAIO config:

python - <<'PY'
from scholaraio.services.audit import list_scrub_suspects
from scholaraio.core.config import load_config

cfg = load_config()

for issue in list_scrub_suspects(cfg.papers_dir):
    print(f"{issue.paper_id}\t{issue.rule}\t{issue.message}")
PY

If the user asked for a broad quality pass, it is also reasonable to start with:

scholaraio audit

Then narrow to papers that are both:

• not already .scrubbed
• obviously bad enough to justify manual review

2. Inspect one paper at a time

For candidates with readable metadata, inspect:

scholaraio show "<paper-id>" --layer 1

Before changing anything, record the stable paper UUID shown in the L1 header as stable_id. repair preserves this UUID even when the directory name changes.