#Hermes agent runner broken (API issue)
elfin shore
urban pumice
sorry the other logs didnt make sense they were for yesterday
elfin shore
This is a Hermes-side message-format regression after the first tool call, not a Fireworks/API-key problem.
The key error is:
Extra inputs are not permitted, field: 'messages[3].tool_name', value: 'skill_view'
That means the first model request succeeds, Hermes runs a tool, then the next request sends the tool result back with an internal tool_name field still attached. Strict OpenAI-compatible providers reject that because tool_name is not part of the Chat Completions message schema. That also explains why it only shows up once a tool is called.
The kanban DB errors in the log are separate noise. The API failure here is the messages[*].tool_name leak.
There is already a matching fix open here:
https://github.com/NousResearch/hermes-agent/pull/28958
cc @slow geode or @ancient iron (review?)
hazy laurel
Not sure if related, but I'm being rate limited with just a single message after the latest update, using OpenCode Go subscription
slow geode
This is a Hermes-side message-format regression after the first tool call, not a...
this was merged @urban pumice