Academic Paper Writing Assistant

The Problem It Solves

Transitioning from a structured outline to a finished academic paper involves multiple error-prone and labor-intensive steps. Manual workflows—such as literature retrieval, data verification, statistical analysis, writing, and formatting—are not only time-consuming but also risky; unverified data can lead to flawed conclusions, and AI-generated content may trigger high similarity scores. This skill automates and standardizes these processes, ensuring papers are grounded in verified data and citable sources while preempting common pitfalls.

Core Capabilities and Key Steps

The skill orchestrates the entire workflow through these key stages:
- Outline Parsing and Data Validation: Reads user-provided outline files to extract variables, analysis methods, and data types. It mandatorily checks for unverified data markers (e.g., (估), ~) and pauses to initiate a real data search strategy—prioritizing official sources like national statistics websites—thus preventing foundational errors.
- Intelligent Literature Retrieval and Parallel Research: Employs a domain-aware smart router to select optimal retrieval channels based on topic (e.g., cnki-unified via browser automation for Chinese social sciences, arxiv API for international technical literature). It supports background fan-out acceleration for parallel multi-source literature surveys.
- Statistical Pre-check and Reasoning Anchors: Before content generation, it runs a mandatory pre-check script (e.g., _tmp/precheck.py) to compute key statistics (e.g., r, R², p). If results deviate significantly from outline expectations (e.g., p-value threshold crossing, large R² discrepancy), it generates a reasoning anchor file (_tmp/reasoning_anchor.md), requiring user input on next steps to avoid silent masking of issues.
- Conditional Writing and AIGC Prevention: Writing logic dynamically branches based on statistical outcomes—for instance, significance statements use if...else structures to avoid hardcoded assertions. Meanwhile, the HUMANIZE_SNIPPETS dictionary embedded in build.py automatically replaces AI high-frequency terms (e.g., ‘与’ to ‘和’) during paragraph generation, reducing AIGC detection risks at the source.
- Chart Generation and Word Formatting: Uses Matplotlib to create charts, addressing Chinese font compatibility issues (e.g., avoiding SimHei for Unicode superscripts). Generates properly formatted Word documents via python-docx, adhering to standards for margins, font styles, equation numbering, and table layouts.
- AIGC Diagnosis and Project Review: After paper generation, it automatically runs dual-tool AIGC diagnostics (e.g., PaperPass and CNKI) and executes lightweight rewriting. Post-project, it triggers an automated review to archive lessons learned in LESSONS.md, enabling future projects to load and avoid recurring mistakes.

Applicability and Considerations

This skill is designed for scenarios requiring rapid conversion of structured outlines into academic papers, but with clear boundaries:
- Requires High-Quality Input: Outlines must specify variables, analysis methods, and data sources; if data is placeholder, explicit user authorization is needed for simulation.
- Environment and Skill Dependencies: It operates within the Hermes environment and requires loaded skills (e.g., cnki-unified, academic-pre-review-committee). Some functions (e.g., literature retrieval) may degrade to manual guides due to network or CAPTCHA issues.
- Statistics-Driven Output: Paper narratives are entirely based on pre-check statistical results; deviations from expectations necessitate user decisions (e.g.,如实报告, adjust data, or reconstruct), with no implicit statistical corrections.
- Not a Panacea for Rewriting: AIGC rewriting strategies target common AI terms but may not bypass all detection tools; manual review is recommended.