nexau-agent

The nexau-agent skill serves as the foundational resource for developers building intelligent agents within the NexAU framework. It provides the procedural knowledge and structural templates necessary to transition from initial concept to a functional, tool-capable agent. This skill encompasses the entire agent development lifecycle, covering YAML-based configuration, system prompt engineering, custom tool definition, and entry point scripting. It is designed for software engineers and AI developers who need to implement specialized agents that integrate with LLMs, sandbox environments, and existing tool ecosystems.

• Full implementation workflow: guidance on requirements analysis, directory structure setup, and agent validation protocols.

• Extensive documentation references: integrated pointers to NexAU core concepts, including agents, tools, LLMs, transports, sessions, and advanced hooks.

• Tool definition templates: standardized schema for both builtin and custom tools, ensuring compatibility with NexAU's architecture for file operations, web fetching, and shell execution.

• Configuration best practices: specific instructions for environment variable substitution, system prompt types (Jinja2/string), and tool call modes (structured/xml).

• Verification and testing: access to built-in validation scripts, Python syntax checking, and troubleshooting techniques for agent runtime issues.

• Integration capabilities: instructions on how to incorporate skills, tracers (e.g., Langfuse), and middleware for enhanced observability and performance.

• Use this skill when initiating a new standalone agent project or adding custom toolsets to an existing configuration.

• Follow the defined repository layout carefully: ensure agent_name.yaml, systemprompt.md, and start.py are structured to support the NexAU loader.

• When implementing tools, always verify if a builtin tool (e.g., read_file, google_web_search, or shell_tools) meets your needs before attempting to write custom Python bindings.

• Leverage the provided validation tools (validate_agent.py) early in the development cycle to catch configuration errors before deployment.

• Refer to the examples/ directory for reference implementations of complex agent architectures, including sub-agent delegation and multi-agent coordination teams.