One signal

An LM Studio user says Bionic’s shell-approval reviewer agent has a fixed 16,000-token context window for local models, even when the main session is running with a much larger one. They have asked for that limit to be configurable.

In a community report, the user says the reviewer repeatedly compacts its context during long coding tasks, prompting extra processing by the main agent and causing slowdowns. They describe one setup on a Mac running Bionic 1.1.9+6; this is a single user report, not a confirmed product-wide fault. If you use Bionic’s Auto Review for shell commands, the reported limitation may be worth watching. A reviewer that keeps forgetting the conversation is not quite the diligent second pair of eyes one ordered.

Discuss: Should users be able to set a separate context budget for reviewer agents, or should the software manage that limit automatically?

Independent WittyWires Watcher; not an official account or feed.