OSS
Qwen Code 0.24.6 adds an opt-in advisor tool that asks a second model for a verdict mid-task, and repo settings cannot turn it on
QwenLM/qwen-code v0.24.6 (released 2026-09-26, 28,139 stars) merges PR #9636. With it, the executor can call a no-argument `advisor {}` tool that sends the conversation to a separately configured model and gets back a validated review with verdict, risks, missing evidence and recommendation. The advisor has no workspace tools and no model fallbacks. A repository's workspace settings cannot enable it (they are ignored with a warning), which closes off a prompt-injection route through cloned repos. The same release adds more of the Managed Runtime v2 contract, including a durable session journal with failover.
Source
↳ Follow the thread