timeln-podcastlisted

# Timeln Podcast Turn **Timeln saves** into **one educational deep-dive MP3**. Saves are source material — like papers fed to NotebookLM. The episode **teaches the topics** so the listener learns something they didn't know. It does **not** narrate what was saved or why. **Outcome test:** After listening, could the user explain the mechanism behind a topic from their saves to someone else? If the script only says "you saved X about Y," it failed. **Deliverable:** `{slug}.mp3` in the shell **working directory** when `render.sh` runs (default slug: `timeln-podcast-{YYYY-MM-DD}`). Report the absolute path in chat. **Requires shell** for `engine/setup.sh` and `render.sh` (not listed in `allowed-tools`). --- ## Setup (once) ```bash cd .agents/skills/timeln-podcast/engine && ./setup.sh ``` Needs: Python 3.11, espeak, ffmpeg. macOS: `brew install espeak`. --- ## Workflow (4 steps) ### 1 — Pull ``` whoami get_recent_docs(window="weekly") ``` - Default **lookup window:** 7 days (`weekly`). User may override (`monthly`, or stated range). - Filter noise, duplicates, corrupted ingests. Count saves → **N**. - Cluster saves by **topic** (not by date saved). - For each candidate topic cluster, `get_document` on the richest saves — read full content, not titles. You need mechanisms, stats, and examples to teach from. Optional: `query_knowledge` / `get_topic_entities` for depth on concepts. **Topic selection:** Pick **1–3 topics** with enough substance for a 4–6 minute deep di