Spawns one reviewer per configured model to adversarially review a change, then synthesizes a verdict. “Adversarial review”, “multi-model review”, “challenge this”, “stress test this code”, “find blind spots”, “tear this apart”.
Its defining design decision is stated as a negative: the adversarial signal comes from model diversity, not assigned personas. Every reviewer gets the same prompt and the same rubric; the four default reviewers differ only in which model runs them. That is a direct rejection of the approach gstack, bmad, and addy-agent-skills take, where a review panel is diversified by giving each seat a named role (pattern-parallel-persona-review). pstack argues the variance you want is between models, not between costumes, which makes this the wiki’s purest pattern-cross-model-review instance — gstack-codex gets a second opinion from one other vendor, interrogate makes the whole panel cross-vendor by construction.
The flow is short. Determine scope (pointed-at files, or git diff main...HEAD on a feature branch). State the intent explicitly in one paragraph derived from the user’s message, commit messages, the PR description, and the code itself, asking the user if unsure — reviewers judge against stated intent rather than guessing it. Spawn all reviewers in a single message, readonly: true, one per configured entry, each filling a shared reviewer-prompt template with the intent, the diff, and a shipped rubric. Then synthesize.
The deliverable is a synthesized verdict, and the skill explicitly does not auto-apply changes. It also ships a graceful-degradation rule that reveals how fast the model landscape moves under a framework like this: if a model slug is rejected as unresolvable, pick the closest equivalent from the error message (preferring the highest-reasoning tier of the same family), spawn with the valid slug, and open a separate PR to fix the configured value — do not block the review on the slug issue.
See Also
- pstack-blast-radius — the single-question review; interrogate is the broad one.
- pstack-comment-sicko — the framework’s single-axis persona reviewer, the one place it does use a costume.
- pstack-architect — runs interrogate on a synthesized sketch for pre-implementation pressure.
- gstack-codex · mp-code-review · addy-code-review · sp-requesting-code-review · ce-code-review — the review cluster.
- stage-review — the canonical stage this implements.