Inputs:

• EPISODE_NUMBER: Episode folder number (directory: episodes/<EPISODE_NUMBER>)

Output files:

• episodes/<N>/cover_art.png
• episodes/<N>/cover_art_prompt.txt
• episodes/<N>/cover_art_debug/<RUN_ID>/... (request/response/debug artifacts)

1. Ensure EPISODE_NUMBER is provided.
2. Ensure episodes/${EPISODE_NUMBER}/metadata.json exists.
3. Ensure ElevenLabs API key is present in env:
   • ELEVEN_LABS_API_KEY, or
   • ELEVENLABS_API_KEY
4. Run: ./.agents/skills/episode-generate-cover-art/scripts/generate_cover_art_elevenlabs.sh --episode-dir "episodes/${EPISODE_NUMBER}"

Behavior guarantees:

• Uses ElevenLabs API endpoints only.
• Prompt format is exact:
  • Create cover art for the following podcast episode based on this metadata:
  • followed by raw metadata.json content.
• Input source is episodes/<N>/metadata.json.
• Discovers/ranks nano-banana model candidates (from /v1/models when available), then falls back through nano-banana aliases.
• Tries multiple endpoint and payload-schema variants for API compatibility.
• Parses image response from:
  • binary image body,
  • JSON URL payload,
  • JSON/base64 payload.
• Rejects invalid outputs and retries fallbacks:
  • non-image content,
  • tiny files,
  • too-small dimensions,
  • low-variance placeholder-like outputs.

Debug artifacts:

• Every request/response attempt is written into episodes/<N>/cover_art_debug/<RUN_ID>/.
• Final selected endpoint/model/output metadata is written to:
  • episodes/<N>/cover_art_debug/<RUN_ID>/final_selection.txt

Validation:

• cover_art.png must be non-empty and image-readable.
• Final output dimensions and file size are printed by the script.
• For batch verification across episodes:

bash
1for n in 3 4 5; do
2  ./.agents/skills/episode-generate-cover-art/scripts/generate_cover_art_elevenlabs.sh --episode-dir "episodes/$n"
3done

• Fail the run if any episode fails to produce valid cover art.