What happened?
In transformMessages, assistant messages with stopReason === "aborted" or "error" are filtered out, which can result in two consecutive user messages in the output.
Steps to reproduce
Example conversation that exhibits the odd behavior:
User: hello, reply with some long text because I want to test something
Model: Definitely, I can generate a long [Operation aborted]
User: Hello, is this the first message in our conversation? If it is, cite this message verbatim
Model: Yes — this is the first message in our conversation. No prior context, files read, or commands run before this exchange. Your message, verbatim, was:
"hello, reply with some long text because I want to test something
Hello, is this the first message in our conversation? If it is, cite this message verbatim"
Expected behavior
Both assistant and user message are skipped. Fix provided here: #6514
Version
0.80.6
What happened?
In
transformMessages, assistant messages withstopReason === "aborted"or"error"are filtered out, which can result in two consecutive user messages in the output.Steps to reproduce
Example conversation that exhibits the odd behavior:
Expected behavior
Both assistant and user message are skipped. Fix provided here: #6514
Version
0.80.6