release: v0.5.0

7 commits since v0.4.0 (released 22 days ago). Highlights:

HuggingFace surface (7 components, full coverage):
  - huggingface_pipeline (transformers.pipeline tasks — local + Inference API)
  - huggingface_chat_completion (HF router, OpenAI-compatible: Kimi-K2,
    Llama-3.3, Mistral, Qwen, DeepSeek, any chat model)
  - huggingface_text_to_image (FLUX / SDXL / etc; multi-provider routing
    via wavespeed / falai / replicate / together)
  - huggingface_dataset_asset (observable_source_asset: Hub dataset metadata)
  - huggingface_model_asset (observable_source_asset: Hub model metadata)
  - huggingface_inference_endpoint (paid dedicated endpoints)
  - huggingface_space_status_sensor (RunRequest + AssetMaterialization
    on Space stage transitions; pair with external_huggingface_space)

Precisely refactor (sensor-only, no submit pretense):
  - Dropped precisely_run_asset entirely (Precisely has no submit endpoint)
  - precisely_job_sensor tightened to verified Job Status endpoint only
  - Added external_precisely_job AssetSpec for catalog presence
  - Sensor upgraded with asset_key field → emits AssetMaterialization
    on terminal SUCCESS alongside RunRequest

External-asset pattern rollout:
  - external_precisely_job, external_huggingface_space,
    external_snowflake_openflow_flow — declare-only AssetSpec for
    things that run on external infra
  - Paired sensors (precisely_job_sensor, huggingface_space_status_sensor,
    snowflake_openflow_status_sensor) updated to materialize them

Snowflake single-entity components (counterparts to snowflake_workspace):
  - snowflake_task_execute_asset (EXECUTE TASK)
  - snowflake_stored_procedure_call_asset (CALL with auto-quoted args)
  - snowflake_dynamic_table_refresh_asset (ALTER DT REFRESH)
  - snowflake_task_completion_sensor (poll TASK_HISTORY; three modes:
    observe / trigger / both)
  - snowflake_openflow_status_sensor (poll TELEMETRY.EVENTS for OpenFlow)

Quality fixes:
  - HF demo setup script: install openai by default; remove placeholder
    endpoint asset and local-pipeline-with-torch from default scaffold
  - 3 demos relabeled honestly (content_moderation, local_nlp,
    nlp_utilities): OpenAI key now optional where possible; TOC restructured
    into "truly no auth" / "optional OpenAI" / "requires OpenAI key"
  - L1 validation harness shipped (tools/validate_manifest.py) — load +
    instantiate + build_defs per component; classifies fails vs skips
  - Manifest entry for precisely_job_sensor downgraded from level=live
    to level=code with explicit evidence about the validation gap

Total registry entries: 803 (up from 789 in v0.4.0).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: eric-thomas-dagster

dagster_community_components/__init__.py

@@ -21,7 +21,7 @@
 import importlib.util
 from pathlib import Path
 
-__version__ = "0.4.0"
+__version__ = "0.5.0"
 
 _PACKAGE_ROOT = Path(__file__).resolve().parent
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "dagster-community-components"
-version = "0.4.0"
+version = "0.5.0"
 description = "Community-maintained Dagster components — ingestion, transforms, IO managers, sensors, sinks, resources, and more."
 readme = "README.md"
 requires-python = ">=3.10"