mesoscope@TSRI

LLM recipe generation

See SKILLS.md for the Mesoscope LLM recipe skill. It defines the prompt, CSV schema, serialized JSON schema, validation steps, downloadable artifacts, and legacy loading workflows.

Start the legacy server:

python3 localCGIServer.py

Use the Skills menu in Mesoscope to copy the recipe skill prompt into another LLM. The copied prompt asks that LLM to generate the recipe files and a direct recipe_json URL.

Load the generated bridge recipe:

http://localhost:8080/?recipe=data/codex_recipe_serialized.json&recipe_format=serialized

Load a hosted recipe file by URL:

http://localhost:8080/?recipe_url=https%3A%2F%2Fexample.org%2Frecipe_serialized.json&recipe_format=serialized

For file-based server loading, pass a URL only. Do not pass local filesystem paths or file:// URLs to the server proxy. Local disk files should be loaded with the browser file picker, or posted as raw JSON:

curl -X POST http://localhost:8080/recipe_json \
  -H 'Content-Type: application/json' \
  --data-binary @data/codex_recipe_serialized.json

For small recipes, an LLM can provide a one-click URL with the JSON encoded directly in the query string:

http://localhost:8080/?recipe_json=<encoded-json>&recipe_format=serialized