Model naming issues

[buraktamturk]

I current have a custom provider is configured with name "somename" and this "somename" provider has models configured let's say "glm-5.1" or "kimi-2.7-code". This "somename" is openai compatible provider.

When I create a new session, the model name sent to the this "somename" is "somename:glm-5.1". On previous releases, I was able to switch model to "glm-5.1" instead of "somename:glm-5.1" using /model command. However, right now this doesn't work. I can't configure the jcode to use "glm-5.1" instead of "somename:glm-5.1".

Additionally, this issue happens when a subagent is spawned. The model calls subagent tool with { ..., "model": "kimi-2.7-code", ... } (this is visible in /observe pane), however, the subagent fails with "model not found" error because it sends "somename:glm-5.1" to the provider.

I had similar issues on previous releases but I was able to overcome them by using "/model glm-5.1", now even if I pick that one, the agent is using "somename:glm-5.1" as model.

I am looking for a solution to this naming issue.

[1jehuang]

Triage: confirmed bug, and a duplicate of the OpenRouterProvider::set_model prefix-handling gap already tracked in #363 / described in detail in #382. Verified in code: OpenRouterProvider::set_model (crates/jcode-base/src/provider/openrouter_provider_impl.rs:291) stores the model id verbatim for custom OpenAI-compatible profiles and never strips a non-builtin <profile>: prefix, so somename:glm-5.1 leaks to the wire. The subagent failure has the same root cause (the routed spec is restored/applied through the same path). Consolidating on #382 which has the full repro + fix plan.

[1jehuang]

Fixed in c0c0138 (master). OpenRouterProvider::set_model now strips a session-routing <profile>: prefix when the prefix is not a built-in routing prefix and matches either the provider's own profile id or a known built-in OpenAI-compatible profile. So a restored/subagent model like somename:glm-5.1 normalizes to glm-5.1 before hitting the wire, while built-in prefixes (claude:, openai-api:, ...) still round-trip. Will close on next release.