manage-lightdash-agents

Manage Lightdash Agents

Skill for managing the lifecycle of AI agents within Lightdash projects.

Purpose

Provides tools for creating, reading, updating, and deleting Lightdash AI agents. This is the foundation of AgentOps, allowing you to define agent personas, system prompts, and operational parameters.

Tools

Wraps the following MCP tools from the lightdash-tools server:

• ldt__list_project_agents
• ldt__get_project_agent
• ldt__create_project_agent
• ldt__update_project_agent
• ldt__delete_project_agent

Safety Mode Compliance

• Read Tools: list_project_agents, get_project_agent.
• Write-Safe Tools: create_project_agent, update_project_agent.
• Write-Destructive Tools: delete_project_agent.

Behavior

1. Bootstrap Workflow:
   • When asked to create an agent, first use ldt__list_projects to identify the correct projectUuid.
   • Check if an agent with a similar name already exists using ldt__list_project_agents.
2. Configuration:
   • When creating or updating an agent, ensure the systemPrompt is clear and follows Lightdash best practices.
   • Reference the eval-driven-agentops.mdc rule to ensure an evaluation suite is in place before making significant prompt changes.
3. Safety Enforcement:
   • Deletion of an agent requires LIGHTDASH_TOOLS_SAFETY_MODE=write-destructive.

Rules

• NEVER delete an agent without explicit confirmation.
• ALWAYS verify the projectUuid before performing any operation.
• Use ldt__get_project_agent to inspect the full configuration of an agent before updating it.