Hypertopos Skills: An AI Agent Investigation Workflow Framework for Geometric Data Spaces

Hypertopos Skills is a structured behavior framework designed for AI Agents to perform exploration, investigation, detection, and monitoring tasks in geometric data spaces (GDS). Unlike a code library, it provides decision frameworks and workflows rather than fixed functions. It includes 8 core skill modules covering the full data investigation lifecycle, serving as a new paradigm for AI Agent-assisted data investigation.

Geometric Data Spaces (GDS) abstract data points as high-dimensional geometric entities, using properties like distance, density, and shape to understand data structure. However, effective exploration, investigation, and anomaly detection in GDS require systematic methodologies. Hypertopos Skills addresses this gap by offering a complete framework to guide AI Agents in GDS tasks.

Hypertopos Skills is a decision framework, not a code library. Its design is 'reactive' instead of 'instructional': it provides decision frameworks (judgment, scope, workflows) rather than rigid checklists. This approach leverages MCP for tools/data access while empowering Agents to apply frameworks flexibly. Advantages include flexibility (adaptable to different scenarios), extensibility (easy to add new skills), and maintainability (framework updates benefit all skills).

The framework includes 8 interrelated modules:

1. gds-analyst: Core coordinator (drives investigation, plans methods, delegates tasks).
2. gds-explorer: Space overview (builds basic understanding of data distribution/density/boundaries).
3. gds-investigator: Root cause analysis (deep dives, hypothesis validation).
4. gds-detective: Standard detection methods (event rate analysis, drift detection).
5. gds-scanner: Advanced detection (cross-mode, neighbor/trajectory analysis).
6. gds-monitor: Continuous monitoring (drift, state changes, health checks).
7. gds-fraud-investigator: AML/fraud detection (25 typical fraud patterns).
8. gds-sphere-designer: Data space design (end-to-end building/tuning from raw data).

The modules collaborate in a hierarchy:

• Design phase: gds-sphere-designer (builds GDS from raw data).
• Investigation phase: gds-analyst (core coordinator) delegates to:
  • gds-explorer (space orientation), gds-investigator (root cause), gds-detective (standard detection), gds-scanner (advanced detection), gds-monitor (continuous tracking), gds-fraud-investigator (AML/fraud).

gds-analyst acts as the 'investigation commander' to select tools based on tasks.

Installation methods:

1. Recommended: Use Skills CLI: npx skills add hypertopos/hypertopos-skills (compatible with Claude Code, Cursor, Copilot, etc.).
2. Manual: Copy gds-*/ folders to Agent's skill directory (e.g., .claude/skills/).

Dependencies:

• hypertopos-mcp (MCP server for tools/data access).
• hypertopos Python library.
• Pre-built spheres (for real-time data operations).

Hypertopos Skills applies to various data-intensive scenarios:

• Data exploration: gds-explorer + gds-analyst for quick GDS understanding.
• Anomaly detection: gds-detective + gds-scanner + gds-investigator for systematic discovery and root cause analysis.
• Continuous monitoring: gds-monitor for tracking data drift and state changes.
• Financial security: gds-fraud-investigator for AML and fraud detection.
• Data space building: gds-sphere-designer for end-to-end GDS construction.

Current status: Research phase (API may change, new skills added, existing ones optimized). Community: Apache 2.0 license, encouraging contributions. Conclusion: Hypertopos Skills fills the gap between 'what AI Agents can do' and 'how they should do it', serving as infrastructure for AI-assisted data investigation. It will become increasingly important as AI Agent capabilities advance.