simpleagentsbuilder

name	simpleagentsbuilder
description	This skill should be used when the user asks to create, improve, or validate YAML agent workflows, especially requests like "build an agent YAML", "design workflow YAML", "add routing in YAML", or "make interview/email workflow nodes and edges".

SimpleAgentsBuilder

Author YAML workflows that are deterministic, testable, and compatible with SimpleAgents graph execution.

Use This Skill For

• Create new graph YAML workflows (entry_node, nodes, edges)
• Refactor existing YAML agent flows
• Add routing, guardrails, and one-question-at-a-time behavior
• Add or tighten output_schema for llm_call nodes

Core Rules

1. Model the workflow as a graph, not a linear prompt script.
2. Define config.output_schema for every llm_call node.
3. Keep switch routing deterministic and explicit.
4. Keep chat systems one-question-at-a-time unless user asks otherwise.
5. Keep business policy in prompts/routing, not hidden in bindings.

Required Structure

Use this skeleton:

id: workflow-id
version: 1.0.0
entry_node: start_node

nodes:
  - id: start_node
    node_type:
      llm_call:
        model: gemini-3-flash
        messages_path: input.messages
    config:
      output_schema:
        type: object
        properties:
          state:
            type: string
        required: [state]
        additionalProperties: false
      user_input_prompt: |
        Return JSON only.

edges:
  - from: start_node
    to: next_node

Node Design Pattern

• detect_* node: classify intent/state with strict enum schema.
• route_* node (switch): deterministic branching by JSON-path conditions.
• worker/action nodes: llm_call for generation, custom_worker only when handler is intentional.
• terminal behavior: explicit node with final message/question.

Routing Pattern

For state-based routing:

node_type:
  switch:
    branches:
      - condition: '$.nodes.detect.output.state == "ready"'
        target: generate
    default: clarify

Keep conditions simple (==, !=) and tied to a stable node output path.

Prompting Pattern

• Instruct: Return JSON only.
• Put user instructions in config.user_input_prompt; use config.node_system_prompt for optional system-level guardrails.
• Give exact response shape.
• Encode policy as explicit bullet rules.
• Keep each node prompt single-responsibility.

Validation Checklist

Before finalizing YAML:

• IDs are unique
• entry_node exists
• Every switch target exists
• Every llm_call has output_schema
• Required fields align with routing conditions
• edges cover intended flow transitions
• No ambiguous multi-question prompts in interview/chat flows

For examples and reusable templates, read:

• references/patterns.md
• references/checklist.md

Working examples:

• examples/minimal-chat.yaml
• examples/email-classification.yaml