Create Course Study Podcast

Turn course material into a story the learner can follow on a walk or during steady exercise. Audio is for preview, explanation, and spoken recall — not a substitute for diagrams, math, or writing code. Treat driving and attention-demanding chores as preview-only listening.

Keep momentum

Use available context and safe, reversible defaults before asking for input. Bundle non-blocking questions and continue useful work. Interrupt only when a missing choice would change the result, privacy, cost, authority, safety, or something hard to undo. Never repeat an answered question.

Sound like a story, not a lecture

• Use the strengths of narrative journalism — curiosity, scenes, gradual reveals, a clear ending — without imitating any named show, host, or catchphrase.
• Open inside a concrete situation. Bring in the formal term only after the idea is already clear.
• Introduce one unfamiliar term at a time. Translate it immediately into ordinary words and reuse that same translation later. Two easily confused terms may follow each other as an explicit contrast.
• Use short spoken sentences with one idea each. Never stack acronyms, definitions, or lists in a single breath.
• Keep one recurring person, goal, or problem as the episode's spine.
• Simplify the wording, never the underlying fact.

Read narrative-recall-style.md before writing the script, prompt, or producer brief.

Make the episode

1. Agree on the course, topic, listener, length, and allowed sources.
2. Gather the sources and separate course facts from corrections, analogies, and teaching choices. Reuse the course's shared source folder (approved sources, key terms and pronunciations, and what is graded), or create one for this project.
3. Write a one-paragraph story spine, plus a short list of required terms with the plain-English translation and one concrete example for each. Drop terms the episode does not need.
4. Write the script or producer brief using producer-brief-template.md. Build each teaching beat as: situation, the listener's guess, the formal term, one exact example, the common mistake, why it matters.
5. Add a few recall pauses: ask one short question, leave a real 3–5 second pause, say once that pausing the player for longer counts, accept "I do not know yet," then give the plain answer and the reason. Answer every question the episode raises. From the second episode on, open with one question about the previous episode.
6. Add pronunciation notes for names, acronyms, and code. Speak code by meaning; do not read underscores or punctuation aloud.
7. Run scripts/audit_spoken_density.py <script-or-brief> --strict and rewrite long sentences and acronym stacks until it passes.
8. Generate the audio with the user's approved tool and browser, one job at a time, and let it finish. A spinner is not completion.
9. Check the finished episode, not just the plan: download it, confirm it plays all the way through, and review the transcript or full audio for the planned pauses, answered questions, approved terms, source-accurate claims, and zero leaked quiz or assignment answers. Use local speech-to-text only with the user's permission.
10. Keep the audio together with its sources, brief, and audit notes. If the user asked for a copy on another device, deliver it and confirm the copy matches.

Reject or regenerate when

• a section opens with a definition instead of a situation;
• a term or acronym appears before it is explained;
• a recall question is asked before its answer was made understandable;
• the episode contradicts the approved sources or leaks graded answers;
• a planned pause or invited question is missing from the finished audio.

A complete download proves the file is intact, not that the teaching is right — check both. See established-example.md for the generic end-to-end flow and project layout.