- Home
- Skills
- Huangwb8
- Chineseresearchlatex
- Nsfc Ref Alignment
nsfc-ref-alignment_skill
- Python
1.3k
GitHub Stars
4
Bundled Files
2 months ago
Catalog Refreshed
3 months ago
First Indexed
Readme & install
Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.
Installation
Preview and clipboard use veilstrat where the catalogue uses aiagentskills.
npx veilstrat add skill huangwb8/chineseresearchlatex --skill nsfc-ref-alignment- CHANGELOG.md1.1 KB
- config.yaml2.5 KB
- README.md5.5 KB
- SKILL.md6.4 KB
Overview
This skill inspects LaTeX NSFC proposal projects to detect reference inconsistencies and semantic misalignment risks between in-text citations and BibTeX entries. It produces a read-only, traceable audit bundle and a human-review report that highlights missing bibkeys, BibTeX field problems, DOI/format issues, and per-citation semantic risk flags. The tool never modifies .tex/.bib files by default and focuses on delivering actionable findings for manual remediation.
How this skill works
The script parses the main TeX entry and its \input/\include tree to collect all citation occurrences and referenced .bib files, then extracts source file/line and surrounding sentence context for each \cite. It parses all BibTeX entries, validates core fields (title/author/year/DOI/URL), applies deterministic checks (missing keys, duplicate entries, malformed DOIs), and writes structured artifacts for downstream semantic review by a host AI. Finally, it compiles a read-only markdown report summarizing P0/P1 risks and per-citation evidence.
When to use it
- Preparing or finalizing an NSFC proposal to ensure citations are present and not misleading.
- Before peer review or submission when you need an audit trail of reference issues.
- When you suspect incorrect or opportunistic citations in a multi-file LaTeX project.
- To gather structured inputs for an AI to evaluate whether sentences legitimately cite the referenced work.
- When you want a safe, read-only assessment without automatic edits to source files.
Best practices
- Run the skill early to catch missing bibkeys and malformed fields before last-minute edits.
- Enable online verification only for final drafts or critical entries to limit external queries.
- Review the ai_ref_alignment_input.json with the host AI and a domain expert for semantic judgments.
- Treat the markdown report as authoritative for remediation planning, then apply changes manually or request explicit automated fixes.
- Keep backups; the tool writes all intermediate artifacts under .nsfc-ref-alignment with timestamped runs.
Example use cases
- Find all \cite occurrences whose bibkeys are absent from project .bib files and locate them by file/line.
- Detect BibTeX entries with missing year or malformed DOI and enumerate them for curator action.
- Generate structured per-citation inputs so an LLM can decide if a sentence overclaims relative to the cited work.
- Produce an audit-ready report for co-authors and grant managers summarizing P0 (must-fix) and P1 (review) items.
- Create a reproducible run directory that stores JSON, CSV, and markdown artifacts for compliance records.
FAQ
No. By default it never modifies source files. It only writes intermediate artifacts under the project .nsfc-ref-alignment run directory and writes the final report to the specified report directory.
What does P0 vs P1 mean in the report?
P0 indicates high-priority issues that likely require correction (missing keys, obvious metadata contradictions, clear semantic mismatch). P1 covers suspicious or incomplete evidence that needs human verification (missing DOI/year, weak support claims).