Moonshot AI’s Kimi Code 2.7 canary adds finer control over how coding agents show their work and respond to developer input. The release also changes how agent timeouts and conversation history are handled, making this a substantial preview for people building with Kimi Code.
Moonshot AI/Kimi Watch analysis
What happened
The 13 November release notes cover changes across the terminal interface and agent core. Users can choose whether pressing Enter queues a follow-up or steers the agent, while transcript history is preserved across trimming, folding and replay. The notes also keep cron and MCP authorisation calls out of compact transcript groups.
Other changes include disabling the subagent deadline when timeoutMs is set to zero, preserving raw bytes in task-output tails, and anchoring token counting and compaction to usage. The release notes also describe an opt-in mutable replay mode and a dependency-injection-free package for the agent loop, language model and tools. Read the Kimi Code 2.7 canary notes.
Why it matters
Coding agents are easier to work with when people can see what they are doing, interrupt or redirect them, and trust that the conversation history has not quietly wandered off. These changes target those practical controls rather than promising a more capable model.
This is a canary build, so it is a preview for testers, not a reason to rebuild a dependable workflow around new behaviour today. Still, there is enough here to make the next round of testing more interesting than a changelog full of decorative commas.
Our read
The strongest changes are the explicit choice between queuing and steering, and better preservation of transcript history. They address the day-to-day business of supervising an agent. Developers already testing Kimi Code can try the canary; everyone else can wait to see which changes reach a stable release.
What to watch
- Whether the Enter behaviour and transcript changes reach a stable release.
- How the opt-in mutable replay behaves during real coding sessions.
- Whether timeout and usage-based accounting changes improve agent reliability in practice.
Discussion spark: When working with coding agents, which matters more to you: a clear record of what the agent has done, or the ability to redirect it instantly?
Sources and evidence
- moonshot-ai/[email protected] (13 November 2026, 06:14 UTC)
not affiliated with or endorsed by Moonshot AI and Kimi