defense-doc-classifier

Tier: Enterprise | Type: Tool | Category: Security | Version: 1.0.0

AI-powered document classification tool for French defense organizations operating under IGI 1300. Suggests security classification levels (NP/DR/CD/SD/TSD), detects misclassifications, and generates classification audit reports.

1. Overview

Plugin Name: defense-doc-classifier Slug: defense-doc-classifier Required Tier: Enterprise Plugin Type: Tool (REST API) Category: Security Author: Dryade License: DSUL

What It Does

Analyzes document content against IGI 1300 classification criteria to suggest appropriate security classification levels. Detects both over-classification (unnecessary access restriction) and under-classification (security risk). Supports batch audit across document repositories.

Key Capabilities

Content analysis against IGI 1300 classification criteria (NP, DR, CD, SD, TSD)
Classification suggestion with confidence score and detailed reasoning
Misclassification detection (over-classification and under-classification)
Batch classification audit across multiple documents
Classification criteria reference API

2. User Stories

Primary User Stories

US-1: New Document Classification

As a document manager, I want to get a classification suggestion for a new document so that I can apply the correct IGI 1300 marking.

Acceptance Criteria:

[x] Classification level suggested with confidence score
[x] Trigger keywords identified with reasoning
[x] Classification reasoning provided in French

US-2: Periodic Classification Review

As a security officer, I want to audit classifications across my document repository so that I can detect misclassifications before the ANSSI audit.

Acceptance Criteria:

[x] Batch audit across multiple documents
[x] Over-classifications and under-classifications identified
[x] Risk assessment for each misclassification

3. Architecture

Component Diagram

+------------------+     +------------------+     +------------------+
|   REST Client    | --> |  Doc Classifier  | --> |  Data Provider   |
|  /api/plugins/   |     |  plugin.py       |     |  (mock / real)   |
+------------------+     +------------------+     +------------------+
                                |
                          +-----v------+
                          |  Demo Data |
                          |  data/*.json|
                          +------------+

Dependencies

Internal: core.plugins.EnterprisePluginProtocol, core.plugin_config_store.PluginConfigStore, fastapi, pydantic
External: None
Plugin: None

4. API Spec

REST Endpoints

| Method | Path | Description | Auth | |--------|------|-------------|------| | POST | /classify | Classify a document per IGI 1300 | Yes | | POST | /audit | Audit classifications across documents | Yes | | GET | /status | Plugin health and status | No | | GET | /criteria | IGI 1300 classification criteria reference | Yes |

Request/Response Examples

POST /classify

// Request { "content": "Plan operationnel secret defense...", "current_classification": "NP", "document_type": "general" }

// Response { "suggested_classification": "SD", "confidence": 0.85, "triggers": [{"keyword": "secret defense", "criterion": "...", "category": "defense_nationale"}], "reasoning": "Classification suggeree: Secret Defense (SD)...", "misclassification": {"type": "sous-classification", "current": "NP", "suggested": "SD", "risk": "..."} }

5. Data Flow

Demo Data Description

classification_criteria.json: 13 classification criteria across NP/DR/CD/SD/TSD levels
sample_documents.json: 15 sample documents at all classification levels with expected classifications
igi1300_levels.json: Complete IGI 1300 level definitions with handling/storage/transmission rules
classification_review_template.json: Review template structure
misclassification_examples.json: 3 example misclassifications (2 over, 1 under)

Total: 5 demo files covering IGI 1300 classification framework.

6. Security Considerations

Data Handling

PII: No - all demo data uses fictional scenarios
Encryption: N/A - on-premise only
Data Retention: Stateless - no document content persisted

Isolation

On-premise deployment only
No external network calls
No document content stored or transmitted

Threat Model

Input injection: Keyword-based analysis resistant to prompt injection
Data leakage: Stateless design, no persistence of classified content

7. Test Plan

Test Classes

| Class | Tests | Coverage Target | |-------|-------|----------------| | TestPluginAttributes | 9 tests | Manifest consistency | | TestPluginConfig | 2 tests | Mock/real toggle | | TestDemoData | 6 tests | Data presence and structure | | TestDocClassifier | 13 tests | Classification, misdetection, audit |

Running Tests

cd dryade-plugins
python -m pytest enterprise/defense-doc-classifier/tests/ -x -v --tb=short

8. Deployment Notes

Requirements

No external packages beyond FastAPI and Pydantic (already in core).

Configuration

{
    "data_source": "mock"
}

Compatibility

Min Dryade Version: 1.0.0
Python: >=3.11
Notes: On-premise deployment only. Designed for classified environments.

9. User Guide

Getting Started

Ensure your Dryade instance has an Enterprise tier license
Install the plugin via the marketplace or dryade-pm push
Use the REST API or ask the AI: "Classify this document per IGI 1300"

Common Workflows

Workflow 1: Single Document Classification

POST document content to /classify
Review suggested classification, triggers, and reasoning
Security officer validates and applies classification

Workflow 2: Batch Classification Audit

POST list of documents to /audit
Review misclassifications (over and under)
Prioritize corrections (under-classifications first)

10. Screenshots

Screenshots will be added after UI integration.

11. Changelog

1.0.0 (2026-03-05)

Initial release
IGI 1300 classification engine (NP/DR/CD/SD/TSD)
Content analysis with trigger keyword detection
Misclassification detection (over/under)
Batch audit endpoint
Classification criteria reference API
15 sample documents for testing and demo

Future Roadmap

[ ] NLP-powered content analysis for nuanced classification
[ ] PDF document ingestion
[ ] Classification marking automation
[ ] Integration with document management systems

Defense-Doc-Classifier

Description

Screenshots

Details