Step Dependencies

Explicit dependencies

Use depends_on to declare that a step must wait for another:

steps:
  - id: build
    run: "go build -o app ."
  - id: test
    run: "go test ./..."
  - id: deploy
    run: "scp app server:/opt/app"
    depends_on: "build"

deploy waits for build to complete, while test runs independently.

Multiple dependencies

Pass a list to wait on several steps:

- id: release
  run: "gh release create v1.0.0"
  depends_on:
    - "build"
    - "test"

release runs only after both build and test succeed.

Implicit variable edges

Pipe automatically detects when a step references another step’s output via $PIPE_* environment variables and creates a dependency edge:

steps:
  - id: get-version
    run: "git describe --tags --always"
  - id: build
    run: "docker build -t app:$PIPE_GET_VERSION ."

Even without depends_on, Pipe sees that build references $PIPE_GET_VERSION (produced by get-version) and ensures get-version runs first.

Both $PIPE_FOO and ${PIPE_FOO} forms are detected.

How the DAG is built

1. Each step registers the env vars it produces (PIPE_<STEP_ID>, plus PIPE_<STEP_ID>_<SUBRUN_ID> for sub-runs).
2. Explicit depends_on edges are added.
3. Implicit edges are added by scanning all run commands for $PIPE_* references and matching them to producing steps.
4. Duplicate edges are removed.
5. The graph is checked for cycles using Kahn’s algorithm.

Cascade failure

When a step fails, all steps that depend on it (directly or transitively) are skipped. Pipe reports which steps were skipped and why.

Cycle detection

Pipe rejects pipelines with circular dependencies at parse time:

Error: dependency cycle detected among steps: build, test, deploy

Self-dependencies are also caught:

Error: step "build": self-dependency

Viewing the graph

Use pipe inspect <name> to see each step’s dependencies:

pipe inspect deploy

The output shows each step with its depends_on relationships.