| name | audit |
| description | Audit paper data quality in the knowledge base. Checks for missing fields, filename issues, DOI duplicates, title mismatches, and more. Supports LLM-based deep diagnosis for title mismatches and automated repair. Use when the user wants to check data quality, find problems, or fix metadata issues. |
Paper Audit
Check the data quality of papers already in the knowledge base. Runs in stages: rule-based checks (automatic) → LLM deep diagnosis (on suspicious items) → automated repair.
Stage 1: Rule-Based Checks
autor audit [--severity error|warning|info]
Issues are classified by severity:
- Error: missing title, missing MD file, JSON parse failure, duplicate DOI
- Warning: missing DOI / abstract / year / authors / journal, MD too short, title mismatch, filename year mismatch
- Info: filename does not follow the standard format
Stage 2: LLM Deep Diagnosis (title_mismatch items)
For each title_mismatch paper, use the Read tool to load meta.json and the first 80 lines of paper.md, then determine:
- Whether the actual subject/title in the MD body is consistent with the JSON metadata
- Whether it is harmless (a MinerU H1 recognition issue) vs. a genuine content mismatch
Stage 3: Repair
For confirmed mismatches, use the repair command:
autor repair "<paper-id>" --title "Correct Title" [--author "First Author"] [--year YYYY] [--doi "10.xxx/..."] --dry-run
autor repair "<paper-id>" --title "Correct Title" [--author "First Author"] [--year YYYY] [--doi "10.xxx/..."]
autor pipeline reindex
Check Rules
| Rule | Level | Description |
|---|
missing_title | error | Title is absent |
missing_md | error | No MD file corresponding to the JSON |
duplicate_doi | error | Duplicate DOI |
missing_doi | warning | DOI is absent |
missing_abstract | warning | Abstract is absent |
title_mismatch | warning | JSON title does not match MD H1 |
nonstandard_filename | info | Filename does not follow the standard format |
Examples
User says: "Check whether there are any problems in my paper library."
→ Run Stage 1 rule-based check
User says: "Run a deep check."
→ Run Stage 1 + Stage 2 (LLM diagnosis on each title_mismatch paper)
User says: "Fix the mismatched papers."
→ Run Stage 3
