What problem does it solve? Building Vision AI pipelines with DX-Stream involves many decisions—model selection, input sources, tracking, tiling, and output sinks—and jumping straight into code often produces broken or mismatched GStreamer pipelines. This Skill enforces a structured brainstorm-and-plan phase so every pipeline build starts from a validated, user-approved plan. ## Core Features & Use Cases - Pre-flight context checks: Silently queries model_list.json, lists existing pipelines, and verifies GStreamer plugin registration (gst-inspect-1.0 dxinfer) before asking the user anything. - Guided decision gathering: Asks targeted questions about pipeline category, model, input source, output sink, tracking, tiling, and multi-stream configuration. - Structured build plan with hard gate: Produces a file layout, pipeline pattern, and pre-flight checklist, then blocks all code generation until the user approves, routing afterward to the appropriate build skill. - Use Case: A developer wants an RTSP-based multi-stream object detection app with MQTT output. The Skill checks model availability, confirms decisions, drafts the plan, and hands off to the MQTT/Kafka build skill only after approval. ## Quick Start Ask the agent to plan a DX-Stream pipeline for your use case, for example: plan a single-model YOLOv8 pipeline that reads a video file and displays annotated output.
