gosu-review
gosu-review
A plain review already finds the defects in what was built. gosu-review asks what that review does not: now that it exists, is it still worth it — to the people it was built for, to what the project is for, against the alternatives? Each seat is a real subagent with fresh context, looking from a vantage above the change itself. Leave line-level defects to a normal review; a seat that sees a critical one notes it in a line.
Target
Review /gosu-review <target> when given; otherwise the most recent artifact in the conversation — a feature, a change, a plan, a skill, a document. If several are plausible, ask one question, Do you want me to review X?, and start nothing else. Read "this" as the most recent artifact and "this repo" as the current repository state. On a broad target, continue but warn once that a narrower scope gives a sharper review.
Gather two things for every brief: the origin — the issue, plan, or request that asked for it and why — and the project's own statement of purpose, such as its README, charter, or spec. When either is missing, say so in the brief rather than inventing it.
Seats
Five seats always sit, one subagent each:
- User — meets it as the person it was built for. Would they notice it, reach for it, trust it? What do they do instead today?
- Premise — knowing what the build revealed, would we build this again, in this shape? Is it the right problem?
- Project thesis — does it strengthen what the project says it is for, or pull it sideways?
- Alternatives — what already does this: other tools, built-ins, the obvious manual way. Why would anyone pick this one?
- Subtract — what can go with nothing lost: an option, a mode, the whole thing.