Research Podcast

Create a traceable chain from research evidence to spoken dialogue.

Paths

• SKILL_DIR: directory containing this file.
• DATA_DIR: ~/.research-podcast.

Read QUALITY_RULES.md before drafting and PROMPT.md for dialogue style.

Workflow

1. Read all supplied research material.
2. Build DATA_DIR/runs/<run-id>/source_map.json with stable source IDs, source labels, locations when known, factual summaries, and supporting evidence.
3. Use the requested mode: deep-dive, exam-review, defense, or briefing; default to deep-dive.
4. Follow PROMPT.md and save DATA_DIR/runs/<run-id>/script.json. Facts and interpretations require supporting source_ids.
5. Validate before audio:

python SKILL_DIR/scripts/validate_script.py --script DATA_DIR/runs/<run-id>/script.json --sources DATA_DIR/runs/<run-id>/source_map.json

6. For audio, create DATA_DIR/venv, install scripts/requirements.txt, copy config/config.example.yaml to DATA_DIR/config.yaml, set both voice IDs, and ensure ffmpeg is installed.
7. Generate:

DATA_DIR/venv/bin/python SKILL_DIR/scripts/speak.py --script DATA_DIR/runs/<run-id>/script.json

The audio tool requests the provider credential interactively and writes an MP3 under DATA_DIR/episodes/.

Completion checks

Confirm the source map exists, the script validates, factual claims are traceable, and requested audio is non-empty.