quick

First Read ${CLAUDE_PLUGIN_ROOT}/skills/model-routing/SKILL.md and apply its routing, foreground agent, validation, persistence, quoting, and background-job rules. The rules below are step-specific.

Run both Stereo phases for one small task. Preserve the canonical /stereo:plan and /stereo:implement semantics with fixed quick safeguards and no approval gate between an approved plan and implementation.

Raw slash-command arguments: $ARGUMENTS

Arguments and role defaults

After reading the routing skill, parse all arguments before repository work:

• --planner <model> resolves as explicit flag > workspace planner default > claude:fable; the scope gate still runs inline in this session before any routed draft.
• --planner-effort <none|minimal|low|medium|high|xhigh|max> overrides effort for a Codex-routed planner.
• --plan-reviewer <model> resolves as explicit flag > workspace planReviewer default > codex:sol.
• --plan-reviewer-effort <none|minimal|low|medium|high|xhigh|max> overrides effort for a Codex-routed plan reviewer.
• --implementer <model> resolves as explicit flag > workspace implementer default > claude:opus. The latest Codex plan-review payload's model and effort never resolve the implementer; per the implementation routing below, a Codex-routed selection resumes planReviewThreadId only when it is the plan reviewer's resolved model.
• --implementer-effort <none|minimal|low|medium|high|xhigh|max> overrides effort for a Codex-routed implementer.
• --implementation-reviewer <model> resolves as explicit flag > workspace implementationReviewer default > codex:sol; the cross-ecosystem reviewer is independent of this orchestrating session and of the Claude-routed default implementer.
• --implementation-reviewer-effort <none|minimal|low|medium|high|xhigh|max> overrides effort for a Codex-routed implementation reviewer.
• --effort <none|minimal|low|medium|high|xhigh|max> is the command-wide default for Codex-routed roles that have no role effort flag. When no active role is Codex-routed, a command-wide --effort is inert: accept it, report it as inert, and never translate it into a Claude-side control.
• --slot <name> selects the durable plan slot this run stores into and defaults to default. Slot names are trimmed, lowercased, may contain only letters, digits, hyphens, and underscores, and must start with a letter or digit. Relay the CLI's validation error verbatim.
• --max-plan-rounds <n> defaults to 2 and must be an integer from 1 to 6.
• --max-fix-rounds <n> defaults to 2 and must be a positive integer. Direct gate-fix turns and review-driven fix turns both count toward it; the implementer's own inner-loop iterations never do.
• --isolated runs implementation, implementation review, and fixes in a throwaway detached git worktree outside the repository and hands the delta back as a user-confirmed patch. It never commits and never writes in the main working tree. The plan draft and every plan-review round always run against the main tree.
• Remaining text is the task. Ask for it if empty.

Reject missing values, duplicate role or role-effort flags, invalid effort or round values, unknown flags, unknown claude:* values, and claude:session as implementer. Accept claude:inherit alongside claude:session and the four explicit Claude aliases. Accept a Codex selection with or without the codex: prefix and reject codex:claude:*. Reject a role effort flag when its selected role is Claude-routed. Resolve every Codex role through role effort > command-wide effort > workspace role effort > the routing skill's pair default. Plan-review payload effort belongs to the payload model and is never borrowed by the resolved implementer. A delta is never gated by the model that produced it: when the resolved implementer and implementation reviewer are the same model and the reviewer came from the built-in default rather than a flag or workspace default, substitute the other ecosystem's review default (codex:sol for a Claude-routed implementer, claude:fable for a Codex-routed one) and report the substitution; a same-model reviewer selected by flag or workspace default is honored but called out as self-review. The removed --model flag is unknown; report the role-named alternatives. The renamed --impl-reviewer and --impl-reviewer-effort flags are unknown; report --implementation-reviewer and --implementation-reviewer-effort as their replacements. Reject --max-plan-rounds above 6 and point at /stereo:plan for a longer plan-review loop; Quick's absolute safeguard at 6 is fixed.