Detect AI manipulation threats using temporal analysis, vector search, and formal verification.
npx skills add https://github.com/ruvnet/midstream --skill aimds
Please help me install this Agent Skill. Skill: AIMDS Source: https://github.com/ruvnet/midstream/tree/main/.claude/skills/AIMDS Command: npx skills add https://github.com/ruvnet/midstream --skill aimds
This Skill helps you build production-grade AI manipulation defense systems by integrating Midstream's temporal analysis, AgentDB's vector search, and lean-agentic's formal verification to detect, analyze, and prove safety of AI-driven workflows, reducing risk, manual review time, and overall complexity.
Create a new AIMDS project structure and install dependencies as described in the Quick Start section of the SKILL.md. Start the Midstream, AgentDB, and lean-agentic services, then feed a sample input to evaluate safety; review the result and the proof trace to confirm actions.
High-intent search queries and answers about installing and using this skill.
Real-time AI manipulation detection uses temporal analysis to identify prompt injections, jailbreaks, and policy violations as they occur. AIMDS combines Midstream's temporal tools, AgentDB's vector search, and lean-agentic's formal verification to analyze behavioral patterns, score risk, and block unsafe outputs with audit proofs.
Temporal analysis detects manipulation attempts by tracking behavioral sequences and timing anomalies with precision. Vector pattern matching rapidly scores risk by comparing input patterns against known threats in AgentDB. Combined, they enable both sequential detection and similarity-based threat recognition in production AI systems.
Yes. Formal verification with theorem proving ensures policy compliance and safety before actions execute. AIMDS integrates symbolic reasoning to generate proof traces that document why outputs are safe, providing auditable evidence for compliance reviews and security teams.
Yes. AIMDS requires Midstream for temporal analysis, AgentDB for vector search, and lean-agentic for formal verification as separate service dependencies. Start these services before feeding inputs to evaluate safety and review proof traces.
Vector search excels at pattern matching against known threats but may miss novel attack vectors without prior examples in memory. Temporal analysis complements this by detecting behavioral anomalies independent of historical patterns, reducing but not eliminating false negatives on unseen threats.
AIMDS automates threat detection across prompts, jailbreaks, and policy violations, then generates formal proofs justifying safe outputs. Security teams review only flagged cases and proof traces, eliminating manual analysis of routine safe outputs and accelerating audit cycles.
FilipcsupkaCoordinate distributed AgentDB deployments with QUIC synchronization for cross-node operations.
Nairon-AIDetect and document security threats in code repositories using STRIDE methodology.
HeshamFSGenerate DOE samples and rank parameter influence for simulation calibration.
ConsiliencyDetect project technology stacks by analyzing package manifests and configuration files.
