nsfc-reviewers_skill

Overview

This skill simulates expert peer review for NSFC (National Natural Science Foundation of China) proposals when the user explicitly requests a review, expert-simulation, or manuscript audit. It provides multi-dimensional critique from domain experts, prioritizes issues by severity, and delivers actionable revision steps. Outputs are structured for traceability and practical editing.

How this skill works

The skill reads a user-provided proposal (single .tex, directory, or zip) and runs a deterministic workflow: file validation, structural indexing, multi-expert independent reviews (parallel or single-group fallback), and aggregation. Reviews are organized by severity (P0/P1/P2), include evidence anchors, and end with concrete, verifiable modification instructions and a minimal change sequence. Parallel review uses a parallel-vibe integration when available; otherwise it simulates five expert personas in a single panel.

When to use it

• When you explicitly ask to "review NSFC proposal", "simulate expert review", or "audit NSFC application".
• When you need a prioritized list of problems with evidence anchors and executable fixes.
• When you want independent expert viewpoints plus a synthesized set of edits to improve fundability.
• When you will provide the proposal path/file/zip and want the report written to a specified output.
• Not for drafting or editing a specific section — use dedicated writer skills for that. Not for asking general evaluation criteria without an explicit review request.

Best practices

• Provide the proposal root, main .tex, or a zip so the skill can assemble a reliable snapshot.
• Specify any focus (innovation, feasibility, team strength) and desired tone (strict/constructive) up front.
• Set panel_count only if you want parallel multi-group reviews; otherwise accept the default.
• Avoid sending personal or institutional info unnecessarily; the skill treats inputs as sensitive and minimizes quoting.
• Confirm output_path early and ensure the directory is writable to prevent fail-fast errors.

Example use cases

• Full-proposal audit before submission to catch high-severity flaws and evidence gaps.
• Pre-submission triage to obtain a minimal-sequence of edits for rapid revision.
• Independent multi-expert viewpoint generation when internal collaborators disagree on novelty or approach.
• Regression check after major rewrite to verify previous P0/P1 problems were resolved.