Greptile Summary

This PR adds a Moonshot adapter targeting the kimi-k2.5 model via Moonshot's OpenAI-compatible API, along with documentation and a test scaffold. The implementation is functionally sound and follows the existing adapter patterns, but has two notable maintenance concerns in Moonshot.php.

• Moonshot::send() is almost entirely a verbatim copy of OpenAI::send() (the three actual differences are: schema embedded in system prompt text rather than json_schema transport, response_format using json_object, and temperature omitted from the payload instead of being reset to 1.0). Because the class already inherits from OpenAI, future fixes or enhancements to the parent's send() will silently diverge from Moonshot.
• usesMaxCompletionTokens() is overridden to return true, but this method is only consulted by OpenAI::send(), which is never reached because Moonshot::send() replaces it entirely. The override is therefore dead code and may mislead future maintainers.
• The test scaffold correctly reads LLM_KEY_MOONSHOT from the environment and delegates all assertions to ConversationBase.
• The README additions are clear and accurate.

Confidence Score: 4/5

• Safe to merge; no runtime bugs found, but the adapter carries maintenance debt from send() duplication.
• The adapter is functionally correct and consistent with how other OpenAI-compatible adapters (e.g. Deepseek) are structured. The two flagged issues — send() duplication and a dead usesMaxCompletionTokens() override — are maintenance concerns rather than runtime errors, hence a score of 4 rather than 5.
• src/Agents/Adapters/Moonshot.php — review the send() duplication and the unreachable usesMaxCompletionTokens() override.

Important Files Changed

_{Reviews (1): Last reviewed commit: "Add Moonshot adapter for kimi-k2.5" | Re-trigger Greptile}