nsfc-qc_skill

Overview

This skill performs read-only quality control (QC) for NSFC grant proposals, producing a standardized QC report and structured artifacts while never modifying source files. It runs deterministic prechecks, parallel independent thread analyses, and a consolidated final report with prioritized findings and executable recommendations. All intermediate outputs are written into project_root/.nsfc-qc/ for traceability.

How this skill works

The skill validates inputs, creates a timestamped run directory, and runs a deterministic precheck that indexes citations, checks .bib completeness, measures section lengths, and detects basic formatting issues. It then runs multiple independent QC threads (via parallel-vibe or a single-thread fallback) where each thread inspects writing tone, citation authenticity risks with evidence packages, structure and logic, and other QC items. Finally it aggregates thread results into a single nsfc-qc_report.md plus nsfc-qc_metrics.json and nsfc-qc_findings.json, and optionally compiles a disposable copy to obtain page counts.

When to use it

• When you explicitly request “标书QC/质量控制/润色前质检/引用真伪核查/篇幅与结构检查”.
• Before submitting an NSFC proposal to catch citation issues and major logical gaps.
• When you need a reproducible, auditable QC trail written into project_root/.nsfc-qc/.
• If you want prioritized, evidence-backed fix items without automatic edits.
• When you need a structured JSON output for human review or further automation.

Best practices

• Always provide project_root and, if nonstandard, main_tex to avoid ambiguous detection.
• Prefer execution=serial unless you explicitly request parallel runs and understand resource implications.
• Review precheck artifacts (reference_evidence.jsonl, tex_lengths.csv) before acting on recommendations. These contain the evidence the QC uses.
• Treat P0 findings as mandatory fixes; P1 as primary revisions; P2 as optional polish.
• Use the isolation compile option at the end to get accurate page counts and include compile logs in the run artifacts.

Example use cases

• Pre-submission audit of an NSFC Young Investigator proposal to catch missing or dubious references and imbalanced sections.
• Team review where multiple independent QC threads surface differing perspectives; the skill consolidates and highlights conflicts.
• Automated gatekeeping: run the QC as part of a checklist to ensure no P0 issues remain before submission.
• Evidence-driven citation check: produce a citation evidence pack mapping bibkeys to retrieved metadata, links, and in-text contexts for manual verification.
• Generating a prioritized, file/section-targeted minimal edit roadmap for an editor or author to implement.