Alert task author loudly for common failure heuristics: e.g. [object Object]

[pringshia]

How might we proactively alert task authors by detecting common failure modes?

This would provide ways to cut down on frustration and wasted spend.

For example:

• if the payload contains "[object Object]" this is usually due to a js serialization issue somewhere in the stack causing data loss.
• if payload outputs are consistently null this could also be an issue

---

The goal is to detect these failures early so that an intervention may occur.