open-review

Set Open-Review provider preferences. Pick which configured opencode providers Open-Review is allowed to dispatch to.

Dispatch a task to opencode as a background teammate. Always asks 3 structured questions first.

Print the final output of an Open-Review job (defaults to most recent)

Verify opencode is installed and ready for Open-Review dispatch

Show running and recent Open-Review jobs in this workspace

Delegates file-editing opencode CLI work — refactors, codemods, scaffolding, applying fixes, implementing features — to save Claude tokens. Use proactively when the user says "have opencode refactor", "use opencode to rewrite", "scaffold tests with opencode", "apply these fixes via opencode", or any opencode mention combined with an instruction that requires writing files. Do NOT use for read-only review/audit (route to open-review:plan) or for jobs likely to exceed 9 minutes (route the user to /open-review:dispatch instead). <example> Context: explicit opencode mention, edits requested. user: "Have opencode refactor the auth module to use async/await" assistant: "Routing to open-review:build." <commentary>Explicit "opencode" + refactor (writes files) = exact match.</commentary> </example> <example> Context: scaffold/codemod request. user: "Use opencode to scaffold Jest tests for the utils folder" assistant: "Scaffolding work — delegating to open-review:build." <commentary>Test scaffolding writes files; build agent.</commentary> </example> <example> Context: NEGATIVE — read-only review. user: "Have opencode review my auth changes" assistant: "Routing to open-review:plan (read-only)." <commentary>open-review:build edits files; review is read-only — wrong agent.</commentary> </example>

Thin forwarder subagent that delegates a single task to the locally installed opencode CLI via the Open-Review helper. Returns opencode's output verbatim. Use when you need context isolation while dispatching to opencode — i.e. when the raw output would otherwise pollute the main conversation, or for long jobs that should run in the background without the 9-minute synchronous guard that open-review:plan and open-review:build enforce.

Delegates read-only opencode CLI work — code review, repo audits, exploration of unfamiliar code, long-context analysis — to save Claude tokens. Use proactively when the user says "use opencode", "have opencode look at", "delegate to opencode", asks for a code review/audit/exploration, OR when the task requires reading >10 files but produces no edits. Do NOT use for file edits (route to open-review:build) or for jobs likely to exceed 9 minutes (route the user to /open-review:dispatch instead). <example> Context: explicit opencode mention, read-only. user: "Have opencode review the auth changes on this branch" assistant: "Routing to open-review:plan." <commentary>Explicit "opencode" + review = exact match.</commentary> </example> <example> Context: long-context exploration, no opencode mention. user: "Find every place we touch session tokens across the monorepo" assistant: "Repo-wide read — delegating to open-review:plan to save tokens." <commentary>>10 files, read-only → ideal opencode plan job.</commentary> </example> <example> Context: NEGATIVE — user wants edits. user: "Refactor this module to use async/await" assistant: "Routing to open-review:build (writes files)." <commentary>open-review:plan is read-only; this is a build job.</commentary> </example>

Dispatch general-purpose work to a locally installed opencode CLI as a teammate agent. Use when offloading code review, repo-wide analysis, planning passes, refactors, broad exploration, or any long-context task to save Claude tokens. Triggers when the user says "use opencode", "have opencode look at", "ask opencode", "delegate to opencode", "open-review", "dispatch to the build agent", "dispatch to the plan agent", or names the opencode CLI explicitly.