Letterly Automation

This skill provides a complete workflow for importing and processing your voice transcriptions from Letterly into your Obsidian vault.

Workflow Summary

This master skill orchestrates the following independent sub-skills:

1. letterly-export: Uses a Playwright browser agent to log in and download the latest CSV export from web.letterly.app.
2. letterly-process: Extracts only notes marked as "magic" in the CSV, creates markdown files with custom frontmatter, preserves the CSV tags column as letterly_tags, and places them in unprocessed/.
3. generate-metadata: Agent-driven step that reads each new transcription, uses letterly_tags as source context when present, generates structured top-level frontmatter metadata, and uses the Python validator/merger to enforce the schema.
4. obsidian-semantic-linker: Analyzes metadata-ready notes and links them to existing vault content using the local semantic embedding database (.smart-env/).
5. Deliver: Moves notes with valid generated metadata to My Outputs/Transcriptions/ and leaves notes with missing or invalid metadata in unprocessed/.

Metadata Gate

Delivery is based on valid generated metadata, not on the presence of ## Related Notes.

The automation script can validate metadata, but it cannot generate semantic metadata by itself because no AI API is called from Python. When running this as an agent workflow, run the generate-metadata skill after letterly-process creates markdown files and before semantic linking/delivery.