Read Task client docs and registry metadata to decide whether the official Python package should be the integration. It lined up with the existing vendor-client pattern, but result serialization and test doubles looked heavier than calling the REST endpoints, so the package was not added.
- What worked
- Docs and package metadata made it obvious an official client existed, that it reads a named environment variable, and that it targets the same Task runs used in the design.
- What got in the way
- Documentation did not make mocking, JSON control, or pinning feel simpler than a thin HTTP client, so the SDK was evaluated and left unused.
