Fix/prompt shield parser

[Joel (aspire488)]

Description

Tests and Documentation

Description

Fixes #2837.

PromptShieldTarget._input_parser currently drops text that appears after a closing </document> tag. This means text between documents and text after the final document can be silently omitted from the userPrompt sent to Azure Prompt Shield.

For example:

q1 <document> d1 </document> q2 <document> d2 </document> q3

was parsed without q2 and q3.

Changes

• Preserve text following each closing </document> tag.
• Keep document extraction behavior unchanged.
• Add regression coverage for multiple documents with text between and after them.

Testing

Added focused unit coverage for the reported reproduction.

No live Azure Prompt Shield or external service calls were performed.

Closes #2837.

[Roman Lutz (romanlutz)]

Thanks for the fix. I am closing this as a duplicate of #2838, which was opened first and addresses the same Prompt Shield parsing bug in #2837. Let us continue the fix and review there.

[Joel (aspire488)]

Thanks for the fix. I am closing this as a duplicate of #2838, which was opened first and addresses the same Prompt Shield parsing bug in #2837. Let us continue the fix and review there.

Thanks Roman, understood.

I’ll continue the fix and review on #2838 instead. Thanks for pointing me to the earlier PR.