dsh-codex-review

Configuration

Two writers touch this row. In the Web UI, open Plugins → dsh-codex-review: the card picks the route and the reviewer model (its list is the running composition's own model catalog, so subscription models a route plugin registered appear there too), the memory mode, the reasoning effort, and the command name. The same fields belong in your profile's cordis.patch.yml (see cordis.patch.yml for the shipped defaults and the commented rationale):

field	default	meaning
backend	spawn	subagent backend: spawn (fresh in-process child), fork (inherits this conversation), or a plugin-provided one such as codex
provider	openai-codex	LLM route the reviewer is pinned to
model	gpt-5.6-sol	model id on that route
conversation	shared	shared keeps one reviewer session per chat that remembers earlier reviews; fresh starts a one-shot reviewer per call
reasoningEffort	null	one of off…max, or null to let the route decide
childMaxDepth	null	0 forbids the reviewer from delegating further; null sends nothing (safe for backends without the depthLimit capability)
persona	null	per-child persona text, if the backend supports it
command	review	the slash command name
instruction, diffCommand, statusCommand	see patch	what the reviewer is told and what it runs first
maxOutputChars	40000	cap on the review text handed back

A patch's config replaces the row's config wholesale — it is not deep merged. When you override one field, restate the ones you want to keep.