AgentSkillsCN

sparkgen-workflow

查看、验证并编辑ai_workflow.yaml配置文件。

SKILL.md
--- frontmatter
name: sparkgen-workflow
description: View, validate, and edit the ai_workflow.yaml configuration
user_invokable: true
auto_invokable: true
auto_invoke_hint: Invoke when the user discusses workflow, orchestration, handoffs, or ai_workflow.yaml
arguments: "<show|validate|add-handoff|set-pattern|env-override|edit>"

SparkGen Workflow

Manage the central workflow configuration in config/ai_workflow.yaml.

Dynamic Context

Before any action:

  1. •Read config/ai_workflow.yaml — the single source of truth
  2. •If server is running, fetch live config: curl -sf http://localhost:8000/v1/workflow -H "X-API-Key: ${API_KEY:-dev-local-key}"

Actions

Show Workflow (/sparkgen-workflow show)

Parse and display config/ai_workflow.yaml in a readable format:

  • •LLM: provider, model, region
  • •Embedding: provider, model
  • •Agents: name, role, tools (table format)
  • •Orchestration: pattern, entry_agent, max_iterations
  • •Handoffs: from → to (condition) for each rule
  • •Guardrails: enabled, active sets
  • •RAG: enabled, default mode, knowledge bases
  • •Environment overrides: list which environments have overrides

Validate Workflow (/sparkgen-workflow validate)

bash
python -c "from app.config.workflow_loader import load_workflow; w = load_workflow(); print(f'Workflow loaded: {w[\"name\"]}, {len(w.get(\"agents\", []))} agents')"

Check for:

  • •YAML syntax errors
  • •Missing required fields
  • •Agent references in handoffs match defined agents
  • •Tool references in agents match defined tools
  • •Prompt files referenced actually exist

Add Handoff (/sparkgen-workflow add-handoff <from> <to> <condition>)

Add a new routing rule to the handoffs: section:

yaml
- from: <from_agent>
  to: <to_agent>
  condition: "<condition_expression>"

Then validate the workflow.

Set Pattern (/sparkgen-workflow set-pattern <single_agent|router_manager>)

Update orchestration.pattern in the workflow YAML:

  • •single_agent: Only one agent, no routing
  • •router_manager: Main agent routes to specialists based on intent Validate that the pattern matches the agent count (single_agent requires 1 agent).

Environment Override (/sparkgen-workflow env-override <env_name> <section> <key> <value>)

Add or update an environment-specific override in the environments: section:

yaml
environments:
  <env_name>:
    <section>:
      <key>: <value>

Edit (/sparkgen-workflow edit)

Open config/ai_workflow.yaml for direct editing. After changes, always run validate.

Workflow Structure Reference

The YAML has 13 sections: version, name, description, llm, embedding, knowledge_bases, rag, tools, memory, guardrails, prompts, agents, orchestration, handoffs, observability, evaluation, environments.