prompting within a charm: a lot of low hanging fruit to improve

[seefeldb]

e.g.

• it's confused about user and assistant role, e.g. putting instructions in user role but then the data in assistant role
• or it asks llm to download something from a url even though it already has the contents

[linear]

from jordan:

Issue has grown stale, and may no longer be relevant. Please reopen if this is still applicable