Claude Flow AI Defence

Claude Flow AI Defence

AI Manipulation Defense System (AIMDS) providing prompt injection detection, jailbreak prevention, PII detection, and vector search-integrated threat analysis with self-learning capabilities.

Quick Command Reference

The AI Defence module is primarily accessed via the security defend CLI subcommand and MCP tools.

Core Commands

security defend

AI manipulation defense - detect prompt injection, jailbreaks, PII.

npx @claude-flow/cli@latest security defend

MCP Tool Usage

aidefence_analyze

Analyze content for manipulation attempts.

Tool: aidefence_analyze
Args: { "content": "text to analyze" }

aidefence_has_pii

Check if content contains personally identifiable information.

Tool: aidefence_has_pii
Args: { "content": "text to check" }

aidefence_is_safe

Check if content is safe from manipulation.

Tool: aidefence_is_safe
Args: { "content": "text to verify" }

aidefence_learn

Learn from new threat patterns to improve detection.

Tool: aidefence_learn
Args: { "pattern": "threat pattern", "category": "injection" }

aidefence_scan

Scan content for known threats.

Tool: aidefence_scan
Args: { "content": "text to scan" }

aidefence_stats

View defense statistics and detection rates.

Tool: aidefence_stats
Args: {}

Common Patterns

Content Safety Check

1. aidefence_is_safe -> { "content": "user input" }
2. If unsafe: aidefence_analyze -> { "content": "user input" }
3. aidefence_learn -> record the pattern

PII Detection Workflow

1. aidefence_has_pii -> { "content": "document text" }
2. If PII found: sanitize content before processing

Threat Monitoring

1. aidefence_stats -> {} (check current detection rates)
2. aidefence_scan -> { "content": "batch of inputs" }

Key Options

• --verbose: Enable verbose output
• --format: Output format (text, json, table)

Programmatic API

import { AIDefence, ThreatAnalyzer, PIIDetector } from '@claude-flow/aidefence';

// Initialize defense system
const defence = new AIDefence();

// Check safety
const isSafe = await defence.isSafe(content);

// Detect PII
const hasPII = await defence.hasPII(content);

// Analyze threats
const analysis = await defence.analyze(content);

// Learn from new patterns
await defence.learn({ pattern: 'new threat', category: 'injection' });

RAN DDD Context

Bounded Context: Security Related Skills: claude-flow-security, claude-flow

References

• Complete command reference: See references/commands.md
• Full README
• npm