Pipeline Outputs

Pipeline outputs define named aliases for pipeline results, making pipelines composable. Parent pipelines can reference child pipeline outputs without knowing internal step or artifact details.

Defining Outputs

Add a pipeline_outputs block at the top level of a pipeline:

kind: WavePipeline
metadata:
  name: impl-feature

pipeline_outputs:
  pr_url:
    step: create-pr
    artifact: result
    field: ".url"
  branch_name:
    step: create-pr
    artifact: result
    field: ".branch"
  summary:
    step: analyze
    artifact: report

steps:
  - id: analyze
    persona: navigator
    exec:
      type: prompt
      source: "Analyze the feature request"
    output_artifacts:
      - name: report
        path: .wave/output/report.md
        type: markdown

  - id: implement
    persona: craftsman
    dependencies: [analyze]
    exec:
      type: prompt
      source: "Implement the feature"

  - id: create-pr
    persona: craftsman
    dependencies: [implement]
    exec:
      type: prompt
      source: "Create a pull request"
    output_artifacts:
      - name: result
        path: .wave/output/pr-result.json
        type: json

Output Fields

Field	Required	Description
step	yes	Source step ID that produces the artifact
artifact	yes	Artifact name from the source step's output_artifacts
field	no	JSON dot-notation path for field extraction (e.g., .url, .data.name)

When field is set, Wave extracts a specific value from a JSON artifact. When omitted, the entire artifact is exposed.

Consuming Outputs in Parent Pipelines

Parent pipelines access child outputs through artifact injection:

steps:
  - id: run-impl
    pipeline: impl-feature
    input: "Build the login page"

  - id: notify
    persona: navigator
    dependencies: [run-impl]
    memory:
      inject_artifacts:
        - pipeline: impl-feature
          artifact: pr_url
          as: pr_link
    exec:
      type: prompt
      source: "Notify the team about the new PR"

The pipeline field in inject_artifacts references the child pipeline by name, and artifact references the named output (not the raw step artifact).

When to Use Pipeline Outputs

• Composability: When a pipeline will be called as a sub-pipeline by others
• Abstraction: To hide internal step structure from consumers
• Field extraction: To expose a specific JSON field rather than the full artifact

See Also

• Pipeline Schema: Pipeline Outputs - Field reference
• Composition - Sub-pipelines, iterate, and other composition primitives
• Outcomes - Extracting deliverables for display (distinct from pipeline outputs)