Workflows and router

Multi-agent collaboration flow

Static workflows

Default path: agentic-orchestration-tool/config/workflows/workflow.yaml.

CLI:

python main.py --config config/workflows/workflow.yaml

Top-level structure (high level):

  • workflow.name, workflow.process (sequential hierarchical)
  • workflow.topic — substituted into task templates ({topic})
  • workflow.agent_providers — inline provider defs (legacy key workflow.providers still accepted)
  • workflow.tasks — tasks referencing agent_provider_id
  • workflow.task_sequence — ordered task ids

See agentic-orchestration-tool/README.md for schema details and Ollama provider options.

Routable workflows (meta)

Router mode scans config-dir/workflows/*.yaml (default config-dir = config). A file is eligible if it has both:

  • Top-level meta — routing metadata
  • Top-level workflow — the workflow body

meta fields

Field Purpose
id Unique workflow id returned by the router model
summary Short line for the routing prompt
description Longer text (optional; may default to summary)
good_for Non-empty list of strings — capabilities
router_include Optional; false excludes file from router while keeping it runnable via --config

Router run

Requires Ollama reachable at OLLAMA_HOST and a model at ROUTER_OLLAMA_MODEL (CLI --router-model).

python main.py "Your natural language task"

The task becomes the workflow topic.

Flags: --config-dir, --router-model, --router-host.

Example workflow files (repo)

File Notes
workflow.yaml Research-style brief (with meta if routable)
workflow_brainstorm.yaml Brainstorming-oriented meta
workflow_web_dev.yaml Web-dev oriented; Ollama-centric analyst → architect → implementer pattern
workflow_agent_skills_smoke.yaml Agent skills catalog smoke — task/backstory injection without LLM-heavy steps

Skills attach via workflow.skills / task.skills; see Agent skills.