Description
Full auto: nothing is asked, nothing is sandboxed, and every capability this build names is switched on. ui.permission_mode is always-approve, ui.yolo is set -- the product's own separate switch, carried in the binary as yoloMode, yolo_mode, a yolo_toggle and a --yolo flag alongside the approval modes rather than as an alias of one -- and sandbox.profile is off with sandbox.auto_allow_bash set. Beyond that it sets the six [features] toggles, plus subagents.enabled, memory.enabled and managed_mcps.enabled. What is claimed about defaults, and what is not. subagents.enabled restates the documented default -- the vendor's own subagents page says *"Enabled by default when the setting is unset"* -- and is kept so the posture states what it wants rather than depending on that staying true, the same reason cursor keeps sandbox.mode and pi keeps enableSkillCommands. For the six features and for memory and managed_mcps, this repository can source that the keys exist and not what they default to: they were read out of the pinned artifact rather than off a page, with an invented control key that appears zero times in the same bytes, and the vendor's settings reference does not list them at all. An earlier version of this description said four of them were off by default; nothing here could source that, so it is gone rather than softened. Telemetry and feedback are toggles too and are deliberately left alone: they send data outward rather than granting you a tool. This is a setup posture -- keys in this product's own configuration file. It is not an execution profile and it grants no environment: what it changes is what the product asks you and which of its own tools exist, not what anyone is permitted to run. The key measurement behind this posture is recorded in references/grok-baseline.json, against the artifact that baseline pins -- and the version lives there rather than here, because this sentence named 1.0.5 while the baseline pinned 1.0.13.