Anthropic compatible api support

[CaiDingxian]

Currently, many LLM providers have implemented compatibility with Anthropic's protocol, similar to OpenAI-compatible APIs. However, in KiloCode's Anthropic adapter, the model names are hardcoded and do not allow custom model identifiers such as MinMax-M2, claude-haiku-4-5, or claude-sonnet-4-5-20250929.

In practice, I've found that the OpenAI-compatible API cannot apply token caching for certain models using the Anthropic protocol.

By manually editing packages\types\src\providers\anthropic.ts and src\api\providers\anthropic.ts to hardcode new models, I was able to get them working effectively.
However, I would prefer official support for custom models, which would enable better compatibility with the latest official models as well as third-party models.

[ramhaidar]

+1 This ☝️