dag2langgraph

Compiles a DAG JSON into LangGraph JSON. Includes validation for structure and cycles.

Install / Run (Poetry)

• Ensure Poetry is installed: https://python-poetry.org/docs/#installation
• Create an in-project virtualenv and install deps (creates .venv):
  • make setup
• Run tests:
  • make test
• Convert a DAG file via console script:
  • poetry run dag2langgraph examples/sample_dag.json -o /tmp/langgraph.json
• Or via stdin/stdout:
  • cat examples/sample_dag.json | poetry run dag2langgraph - > /tmp/langgraph.json

Makefile Targets

• setup: creates/uses .venv and installs deps via Poetry.
• test: poetry run pytest -q.
• run-sample: run CLI on examples/sample_dag.json.
• lock: (re)generate poetry.lock to match pyproject.toml.
• patch: bump patch version via Poetry, commit, and tag vX.Y.Z.
• publish: push branch and tag to origin (triggers GitHub Actions publish to PyPI).
• clean: remove .venv and caches.

Publishing

• GitHub Actions workflow Publish to PyPI builds and uploads the package when:

  • A GitHub Release is published, or
  • You push a tag like v0.1.0, or
  • You manually dispatch the workflow.

• Required secret in the repo settings:

  • PYPI_API_TOKEN: PyPI API token (user: __token__).

Releasing (Patch)

• Use the Release Patch Version workflow (workflow_dispatch) to:
  • Bump the patch version via Poetry
  • Commit the change and create tag vX.Y.Z
  • Create a GitHub Release
  • This will trigger the publish workflow to upload to PyPI

Input DAG schema (summary)

• nodes: array of objects with id (string), type ("function" | "tool"), name (string)
• edges: array of objects with source (string), target (string), optional condition (string)
• entry_point: string (must match a node id)

Output LangGraph JSON (summary)

• nodes: object keyed by node id, each with { type, name }
• edges: array of { source, target } plus condition when provided
• entry_point: copied from input

Errors

• Entry point not specified. — when entry_point is missing/empty
• Invalid DAG structure or cycles detected. — for malformed fields, missing refs, duplicate ids, or cycles