Sanity check execution result content

[rayruizhiliao]

These are the 3 last messages:

• system tells agent to validate the routine
• agent uses the execute routine tool
• the tools results come back

But i dont see the part where the agent decides if this routine execution was successful or not?

Looks like we check the output itself for exec_result.ok and failed_placeholders. but it is possible that the routine itself structurally correct, but returns nothing useful in the body. We can have the agent looks at a preview of the output and make sure all is well before truly exiting.

This might be an overkill/too much for this PR tho, lmk what you think!

Originally posted by @dimavrem22 in #121 (comment)