Scientific writing & document integrity guard for AI-assisted research - prompt-guided argument economy with deterministic scientific, delivery, journal, and Word integrity checks.
The plugin will be installed here. Keep web if you are unsure.
npx -y @deepseek-ai/dsh plugin --profile web add dsh-plugin-writing-guard@2.0.1
Compatibility and provenance
Plugin Writing Guard is published as dsh-plugin-writing-guard and currently resolves to version 2.0.1. The Hub verifies its manifest and preserves the exact installation source for reproducible installs.
Scientific Writing & Document Integrity Guard for AI-assisted research
Less explanation. More argument. Preserve the science.
Writing Guard combines prompt-guided scientific prose discipline with deterministic integrity checks for AI-assisted research.
It protects your manuscript across five layers:
Writing Guard does not write your paper for you.
It makes AI-assisted writing safer to ship.
New in 2.0 — Argument Economy & Control-Plane Separation
Writing Guard 2.0 moves semantic style decisions before generation and keeps deterministic code focused on integrity.
Critique is not content — reviewer/user/guard wording is control context, not manuscript evidence.
Prefer CUT over REWRITE — if a sentence only prevents criticism, reassures reviewers, advertises importance, or repeats an obvious implication, delete it.
Do not close every semantic loop — stop after evidence + necessary interpretation; keep non-obvious statistical interpretation and reproducibility detail.
Style-only revisions default to same length or shorter — no explanatory inflation without new supported content.
Auto-audit no longer reinjects snippets or rewrite suggestions — only structured rule/action metadata is sent back to the agent.
New English/Chinese cues for defensive-purpose framing, semantic-closure markers, and content-free evaluation.
Example:
Avoid: To prevent data leakage, normalization was carefully performed using only the training data.
Prefer: Normalization parameters were estimated from the training data.
The second sentence states the method fact without importing the defensive reason for mentioning it.
🛡️ Word Document Guard (introduced in 1.8)
Change what was requested. Verify what wasn't.
AI-assisted document editing introduces a subtle risk: the requested text may improve while unrelated parts of the manuscript silently change.
When you ask an AI agent to "rewrite paragraph 3 only," it may inadvertently modify:
equation structures
table borders
numbering
styles
section geometry
relationships
embedded media
OOXML package parts
Writing Guard 1.8 introduces a document-integrity layer for Word manuscripts:
Feature
What it does
Safe scoped editing
Only the specified range can change
Package validation
Verifies DOCX/OOXML package integrity after editing
