Composing Workflows¶

Workflows let you chain multiple capabilities together into a directed acyclic graph (DAG) with conditional branching, retry logic, and failure propagation.

Workflow YAML¶

kind: Workflow
metadata:
  name: research_and_report
  version: 1.0.0
  description: "Research a topic, analyze findings, and generate a report"

spec:
  tasks:
    - id: search
      capability: web_search
      input:
        query: "{{goal}}"

    - id: analyze
      capability: text_analyze
      depends_on: [search]
      input:
        text: "{{search.results}}"
      conditions:
        - if: "search.result_count > 0"
          status: run
        - else:
          status: skip

    - id: summarize
      capability: text_summarize
      depends_on: [analyze]
      input:
        text: "{{analyze.result}}"

    - id: report
      capability: report_generate
      depends_on: [summarize]
      input:
        content: "{{summarize.summary}}"

  execution:
    max_retries: 2
    retry_delay: 5
    failure_strategy: abort

Running a Workflow¶

intent-os workflow run examples/research_workflow.yaml --input '{"goal": "research AI trends"}'

Planning a Workflow from a Goal¶

intent-os workflow plan "research AI trends and summarize findings"

Execution Semantics¶

Parameter	Values	Default	Description
`failure_strategy`	`abort`, `skip`, `retry`	`abort`	What happens when a task fails
`max_retries`	integer	`0`	Maximum retry attempts per task
`retry_delay`	seconds	`0`	Delay before retry
`timeout`	seconds	`300`	Task timeout

Conditions DSL¶

conditions:
  - if: "task_count > 0"
    status: run
  - if: "confidence > 0.8"
    status: run
  - else:
    status: skip

Supported operators: >, <, ==, >=, <=, !=, exists, in, contains