What errors does your agent hit most, and for what use case?

[SebConejo]

We're working on automatic error recovery in the routing layer. Before building anything, we want to know what you actually deal with.

If you run an agent through Manifest, what errors do you see most? And what's your setup? (coding agent, support bot, data pipeline, something else)

From our logs (1M+ requests):

• 49% rate limits or insufficient balance
• 12.5% deprecated models
• 4.5% malformed parameters (max_tokens vs max_completion_tokens)
• 2.5% malformed message structures
• 2% malformed tool schemas
• 2% context length exceeded

Does this match your experience? What would actually save you time?