Enforce Style Contracts with a Deterministic Linter
Go beyond passive analysis. Define explicit style targets for your content and audit for AI-style markers. Built for programmatic quality control in high-end editorial pipelines.
Deterministic style-contract engine. Detect 'assistant residue' and enforce structural variety in your automated writing workflows.
Overview
Go beyond passive analysis. Define explicit style targets for your content and audit for AI-style markers. Built for programmatic quality control in high-end editorial pipelines.
Define strict targets for sentence length, variety, and scannability. Get binary pass/fail results based on mathematical rules, not probabilistic guesses.
Identify common LLM patterns including stock phrases, assistant residue, and repetitive structural markers that flag content as 'AI-generated'.
Receive prioritized, ranked advice on how to resolve specific style violations. Turn raw metrics into specific editorial instructions for your team or agents.
Current workflow
Standard proofreading tools lack the structural rigor needed for large-scale content governance and AI-assisted workflows.
- Inconsistent style standards make it difficult to maintain a single 'voice' across multiple authors or AI models.
- Grammar checks miss specific structural issues like 'wall-of-text' risks and lack of sentence variety.
- Identifying 'AI-ish' content by feel is subjective and impossible to scale across thousands of articles.
- Probabilistic style checkers provide different results each time, making them unsuitable for CI/CD guardrails.
- Feedback loops for writing agents are often vague, leading to repetitive or uninspired content revisions.
Where it breaks
These gaps lead to content drift and higher manual editing costs for editorial teams.
- High manual overhead to ensure AI-generated drafts meet internal quality and style standards.
- Difficulty enforcing scannability rules across complex technical documentation or long-form guides.
- Risk of publishing content with visible 'assistant residue' that harms your brand's authority.
- Lack of numeric 'fit scores' makes it difficult to track content quality improvements over time.
The Style Governance Engine
@veldica/prose-linter provides a comprehensive framework for linting prose like code.
Define explicit targets (e.g., avg words per sentence <= 15) and enforce them programmatically.
Scan for lexical fingerprints and stock transitions common in major LLM providers.
Calculate a 0-100 compliance score for every document based on your custom style profile.
Automatically identify which structural changes will have the highest impact on your scannability and fit.
Verified request
# Install the package
npm install @veldica/prose-linter
# Usage in your project
import { lintText } from '@veldica/prose-linter';
const profile = {
targets: {
sentence_metrics: {
avg_words_per_sentence: { value: 15, operator: 'at_most' }
}
},
track_ai_patterns: true
};
const result = lintText("In this guide, we will explore...", profile);
console.log(result.summary.score); // 85
console.log(result.violations); // Specific failed checksVerified response
Get structured feedback on style violations and AI markers for direct integration.
{
"summary": { "score": 85, "overall_pass": false },
"violations": [
{
"metric": "avg_words_per_sentence",
"current_value": 18.5,
"target_value": 15,
"explanation": "Average sentence length exceeds target."
}
],
"ai_analysis": {
"style_band": "low",
"markers": [
{ "text": "in this guide, we will", "type": "stock_transition" }
]
}
}Output interpretation
The linter is designed for programmatic feedback, turning math into better writing.
- Binary Enforcement: Move beyond 'tips' to hard style contracts that must be met before publishing.
- Marker Detection: Find and remove phrases like 'delve into', 'tapestry', and other assistant-specific tell-tales.
- Structural Fit: Measure how well a draft matches your target 'shape' for specific audiences.
- CI/CD Integration: Fail builds when content quality drops below your mandatory style thresholds.
- Agent Feedback: Provide precise, numeric instructions to AI agents to guide more accurate revisions.
Practical Usage: Quality Guardrails
Integrate the linter into your editorial pipeline or build-system to automate quality control.
- Define your organization's 'Style Profile' as a shared JSON schema.
- Run the linter on every pull request containing new or updated documentation.
- Block merging if the 'Fit Score' is below your minimum quality bar.
- Automatically generate revision instructions for writers based on the failed checks.
- Audit legacy content to identify drafts with high AI-marker density for manual review.
import { inventoryMarkers } from '@veldica/prose-linter';
const audit = inventoryMarkers(text);
if (audit.style_band === 'high') {
console.log("Warning: High density of AI writing markers detected.");
}Choosing a Style Linter
Deterministic linting provides the stability needed for automated content systems.
Suggest corrections but don't enforce standards. Hard to integrate into automated pipelines.
Probabilistic AI detectors that change their mind often. High false-positive rates and opaque logic.
Rule-based and repeatable. Provides specific, explainable proof for every violation and marker.
Keep Exploring
Use the Workflow Library to browse more guides, comparisons, and integration examples to continue your evaluation.
See the solutions, comparisons, and integration guides collected in one place.
Review grounded audit, compare, fix-plan, and report excerpts before you wire the API into anything.
Jump from the workflow page into the quickstart, endpoint guides, and full OpenAPI reference.
Automate your editorial standards
Explore the package on GitHub or install via NPM. Lint prose with the same discipline you use for code.