Skip to content

Agent Contracts

Agents in LazyAI are specialized actors. Each agent must have a clearly defined contract that dictates its scope, inputs, outputs, and handoff behavior.

Semantic Validation

When lazyai-cli validate --all runs, agents are audited for both structural validity and semantic completeness.

Structural Requirements (Errors)

  • Must be a readable file.
  • Must contain valid YAML frontmatter.
  • Must declare a name in frontmatter.
  • The body must not be empty.

Contract Requirements (Warnings)

An agent's markdown body should define its role and workflow completely: - Role/Purpose: What the agent does. - Trigger/Misuse: When to use and When not to use. - Workflow: The steps the agent takes. - Evidence Requirements: What must be proven before yielding. - Human Gates: When the agent must stop and ask the user (e.g., ⛔ Human gate). - Handoff: What to do when finished (e.g., pass output to another agent or user). - Output: The format of the final deliverable.

Agents that lack these explicit contract boundaries are likely to drift, hallucinate scope, or fail to interact correctly with the rest of the multi-agent system.