SpringAI export

[corradodebari]

SpringAI Settings page, in the release/1.0-beta,
raises this error:

yaml.scanner.ScannerError: mapping values are not allowed here in "<unicode string>", line 45, column 42: ... Follow these strict rules: ^ Traceback: File "/Users/cdebari/Documents/GitHub/oaim-sandbox-langgraph/src/oaim_sandbox.py", line 140, in <module> main() File "/Users/cdebari/Documents/GitHub/oaim-sandbox-langgraph/src/oaim_sandbox.py", line 129, in main pg.run() File "/Users/cdebari/Documents/GitHub/oaim-sandbox-langgraph/.venv/lib/python3.11/site-packages/streamlit/navigation/page.py", line 306, in run exec(code, module.__dict__) File "/Users/cdebari/Documents/GitHub/oaim-sandbox-langgraph/src/sandbox/content/config/settings.py", line 336, in <module> main() File "/Users/cdebari/Documents/GitHub/oaim-sandbox-langgraph/src/sandbox/content/config/settings.py", line 328, in main data=spring_ai_zip(spring_ai_conf, ll_model), # Generate zip on the fly ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/Users/cdebari/Documents/GitHub/oaim-sandbox-langgraph/src/sandbox/content/config/settings.py", line 249, in spring_ai_zip yaml_content = spring_ai_obaas(src_dir, "obaas.yaml", provider, ll_model) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/Users/cdebari/Documents/GitHub/oaim-sandbox-langgraph/src/sandbox/content/config/settings.py", line 214, in spring_ai_obaas yaml_data = yaml.safe_load(formatted_content) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

[corradodebari]

The presence of ":" in the prompt description, i.e.:
Follow these strict rules:

breaks the yaml parsing:

raise ScannerError(None, None, yaml.scanner.ScannerError: mapping values are not allowed here in "<unicode string>", line 45, column 42: ... Follow these strict rules:

[corradodebari]

Work around: don't use ":" character in the prompt definition.
Searching a way to avoid the exception.

[gotsysdba]

There's got to be a way to deref before putting in config. Because we allow modification of the prompt by end-user, there's no real way to enforce the workaround. FWIW; probably should look to implement example prompts-per-API. Ollama does not behave well with prompts that work for OpenAI, etc. The current example prompts are a compromise and verbose to get both models to behave similarly (yet Ollama still "acts-up" and hence the "strict rules")