evidence-driven-writing

Guides evidence-grounded academic writing with citation traceability and draft-mode experiment gating.

2|Updated Aug 9, 2026
One-click install
npx skills add https://github.com/DeepJH/doubao-skill-and-info --skill evidence-driven-writing-deepjh
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: evidence-driven-writing
Source: https://github.com/DeepJH/doubao-skill-and-info/tree/main/skills/doubao-academic-polish/sub-skills/paper-write-zh/references/evidence-driven-writing
Command: npx skills add https://github.com/DeepJH/doubao-skill-and-info --skill evidence-driven-writing-deepjh

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Academic drafts often contain fabricated citations, unsupported strong claims, literature-list padding, and leftover AI or prompt artifacts. This reference enforces an evidence-driven workflow so every claim in introductions, related work, discussions, and conclusions traces back to verifiable sources or clearly marked planning data. ## Core Features & Use Cases - Evidence Map and Paragraph Blueprint: Build a source-to-claim mapping and per-paragraph argument roles before writing any literature-based section. - Experiment Gating (D0-D5): Stage-gate data, metrics, baselines, and results so Draft Mode uses clearly labeled PLANNING DATA while Final Mode requires real experimental evidence. - Anti-Pattern Firewalls: Block fabricated references, per-paper listing, prompt leakage, and unsupported claims like "significantly outperforms" without evidence. - Use Case: When writing a Chinese journal paper introduction on defect detection, first build an evidence map from verified literature, then draft a problem-pressure to contribution argument chain with GB/T 7714 in-text citations instead of a paper-by-paper summary. ## Quick Start Use the evidence-driven-writing reference to draft the introduction and related work of my paper, building an evidence map first and marking any mock experiment tables as PLANNING DATA.

Frequently Asked Questions about evidence-driven-writing

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I write a literature review without listing papers one by one?

Organize sources by method family or problem condition rather than chronology. Each paragraph should synthesize two to four sources, state their shared limitations, and bridge to the next approach, while keeping in-text citations like [1-3] attached to every literature-based claim.

How to avoid fabricated citations in AI-assisted paper writing?

Build an evidence map first that records each source's type, what it directly supports, and its risk level. Only use verified metadata, abstracts, DOI pages, or user-provided materials, and mark weak sources as metadata-only or needs-verification instead of writing full-text conclusions from them.

What is Draft Mode for experiment sections in paper writing?

Draft Mode lets you scaffold complete result sections using planning data and mock tables labeled PLANNING DATA - replace before submission. It forbids phrases like "experiments show" or "significantly outperforms" and requires boundary wording such as "based on planning data" until real results arrive.

Can metadata-only sources support claims in a research paper?

Metadata-level sources such as scholar search results can support citations and reference lists but not full-text claims about methods, results, or limitations. Those claims require abstracts, DOI or publisher pages, or user-verifiable materials as stronger evidence.

Why does AI-generated academic text feel like AI writing?

Common causes include per-paper listing, empty strong claims like "widely studied" or "great significance", process narration leaking into the body, and conclusions exceeding the evidence. The reference counters these with argument-chain blueprints, evidence-bound claim strength, and a contamination firewall checklist.