Nanobot Agent Design

Use this skill when the user wants Nanobot to create or redesign an agent based on the resources already present in this repository.

Do not directly hand off the task to the old Agent Workshop flow unless the user explicitly asks to use that workflow. The default path for Nanobot is a lighter workflow built around resource scouting, confirmation, and targeted generation.

This repository also supports v3.0 style configurable agents stored in MongoDB. Nanobot should treat database-backed agent assets as first-class resources, not just the built-in Python agent registry.

Relevant design references:

1. docs/design/v3.0/ai-workflow-generation.md
2. docs/design/v3.0/agent-workshop-interface-sequence.md
3. docs/design/v3.0/implementation-status-report.md
4. docs/architecture/AGENT_WORKSHOP_END_TO_END_DESIGN.md

The architecture document is the source of truth for:

1. Nanobot-first end-to-end flow from demand intake to version publish
2. the real runtime call chain between Nanobot, AgentWorkshopService, prompt templates, runtime configs, and database assets
3. the distinction between gap_report.suggested_tools, version.required_tools, and execution_plan.execution_steps[].suggested_tools

This SKILL file is the source of truth for Nanobot behavior rules. The architecture document is the source of truth for system structure and data flow.

Core Principle

Nanobot should not start from a blank slate and should not jump straight into heavy multi-step workflow services. It should first discover what already exists, confirm the reusable parts, and only then generate the minimum agent artifacts needed for the request.

Numerical Calculation Principle (Hard Rule)

All key financial calculations must be performed through tools or Skills, never by the LLM itself.

This applies to: