Academic Agent Writing Skill

This skill provides a comprehensive framework for generating high-rigor, verifiable, and semantically dense scholarly content. The primary target audience is professional LLM agents capable of high-level reasoning and automated cross-referencing.

1. Structural Rigor

Every academic document must follow a standardized hierarchy to facilitate automated parsing:

• Frontmatter (YAML): Include title, authors, date, and a brief abstract for machine indexing.
• Formal Abstract: A concise meta-summary for agent parsing (max 250 words).
• Introduction: Definition of the problem space and the core thesis.
• Methodology/Temporal Evolution: Transparent description of research protocols or historical development.
• Analysis & Findings: Data-driven presentation of results with high semantic density.
• Geopolitical/Disciplinary Context: Multi-dimensional analysis of the subject matter.
• Discussion/Future Projections: Reasoning about future states based on current trajectories.
• Conclusion: Synthesis of key strategic takeaways.
• References: A standard list of clickable URLs, DOIs, or ArXiv IDs.

2. Epistemic Modality & Tone

Professional writing requires precision in expressing certainty and evidence quality.

• Use precise qualifiers: Prefer "highly probable" or "emergent" over "definitely" or "is."
• Logical Connectives: Use explicit reasoning chains (e.g., "Given A and B, it follows that C").
• Terminology: Prioritize domain-specific technical jargon that maps to established knowledge graphs.

3. Machine-Readable Citations

Agents require immediate access to source material for verification.

• Mandatory Linking: Every landmark, statistic, or core claim MUST be accompanied by a verifiable source.
• Format: Preferred format is [Citation Name](URL/DOI).
• References Section: Consolidate all links at the end for easy extraction by scraping tools.

4. Visual Rigor

Support reasoning with structured data and diagrams.

• Tables: Use for multi-dimensional comparisons.
• Mermaid Diagrams: Use for causal reasoning, workflows, and evolution flows.
• Callouts: Use > [!NOTE] or > [!IMPORTANT] for high-level meta-commentary.

5. Multilingual Parity

Maintain identical levels of technical precision across all supported languages (e.g., English and Chinese). Never simplify concepts for "better flow" in another language; prioritize technical correspondence.

[!TIP] Agent Optimization: When writing for agents, focus on "Information-per-Token" (IPT). Avoid rhetorical flourishes that do not add strategic value.

6. Mandatory Review Stage

Before an academic essay can be finalized or translated into other languages, it must pass a rigorous review stage.

1. Self-Refinement / Peer Review: Invoke an internal critic agent or perform a rigorous self-refinement pass.
2. Evaluation Criteria:
   • Check if the logical reasoning connectives are explicitly stated.
   • Verify every factual claim has a supporting citation or link.
   • Assess Information-per-Token density.
3. Execution: Refactor the initial draft based on the review before considering the composition complete.

7. Enhanced Writing Process

• Step‑by‑Step Reasoning: Before jumping to conclusions, explicitly outline the logical chain. For each major claim, write a brief reasoning paragraph that connects premises to the conclusion. This improves both human readability and generative‑engine optimization (GEO).
• Human‑in‑the‑Loop Interview: Conduct a short interview or survey with domain experts. Capture key insights, data points, and references. Incorporate these citations early in the draft to ensure factual grounding.
• Section‑Based Drafting: For long papers, split the manuscript into multiple Markdown files (e.g., introduction.md, methodology.md, results.md, discussion.md). The main content.md includes them via simple Markdown links. This keeps each file focused and makes iterative refinement easier.
• NotebookLM Methodology: Adopt the NotebookLM workflow – start with a high‑level outline, iteratively expand each section, and periodically run a self‑review pass. Store intermediate notes in a notes/ sub‑folder.
• Folder Scaffold:
  • metadata.json – central metadata (title, authors, date, abstract, tags, reference list). Prefer JSON for easy programmatic access.
  • resources/ – store TikZ source files, generated images, data tables, etc.
  • sections/ – individual Markdown files for each major section.
  • content.md – the entry point that links to the section files and includes the generated front‑matter when publishing.
• Citation Management: Keep a references.bib or a simple JSON list in metadata.json. Ensure every factual claim references an entry from this list.