feat: v0.5.0–v0.7.0 — smart zoom, short metadata, frame descriptions

[JRemitz]

Summary

• v0.5.0 — Smart zoom (zoom.py): Per-frame OpenAI vision analysis for dynamic crop panning, with fallback to center on failure. Writes ZoomPath to context.shared["smart_zoom"].
• v0.6.0 — Short metadata + frame descriptions (short_metadata.py, frames.py): LLM-generated title/description for short uploads via POST_RENDER hook. Frame description sends all extracted frames in a single vision call, caches play-by-play summary on plugin instance, and feeds {{frame_summary}} into short metadata prompts.
• v0.7.0 — Player/event context: Player name, assists, event type, and team level from POST_RENDER hook data passed as {{player}}, {{assists}}, {{event}}, {{team_level}} template variables for richer AI-generated metadata.

New modules

• zoom.py — analyze_frame_for_zoom(), _encode_frame()
• short_metadata.py — generate_short_metadata(), ShortMetadata
• frames.py — describe_frames(), FrameDescriptions

New config fields

• smart_zoom_enabled, smart_zoom_model
• short_metadata_enabled
• frame_description_enabled, frame_description_model

New prompt templates

• smart_zoom_detect, short_title, short_description, frame_describe

Stats

• 22 files changed, +2365 lines
• 226 tests, 100% line + branch coverage

Test plan

• make check passes (lint, mypy strict, 100% coverage)
• Manual test: installed into reeln-cli venv, rendered shorts with frame_description_enabled: true
• Verify frame descriptions appear in shared context
• Verify short metadata prompts include {{frame_summary}}, {{player}}, {{event}}
• Verify PyPI release triggered by v0.7.0 tag

🤖 Generated with Claude Code

[JRemitz]

Superseded — splitting into per-version commits