- Add --from-text/-t option to scan and process commands for
pre-formatted tracklists (e.g. from Perplexity)
- Refactor llm_parser to use Chat API instead of Generate API
- Reuse _extract_json() from vision_llm for robust JSON extraction
- Improve SYSTEM_PROMPT with strict rules (Various Artists, no
invented years, no composer info in titles, /no_think)
- Remove format:"json" constraint that caused empty responses
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Tesseract OCR fails on rotated/low-contrast CD back covers.
New vision_llm module sends images directly to qwen3-vl via
Ollama chat API, bypassing OCR entirely. Robust JSON extraction
handles thinking tags, markdown blocks, and empty responses.
CLI scan/process commands gain --vision flag.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Modular Python package with Typer CLI (scan/apply/process commands),
Pydantic data models, OCR via Tesseract, LLM-based tracklist parsing,
mutagen audio tagging, M3U playlist generation, and cover processing.
Includes 8 passing tests and ruff lint config.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
