Adversarial-ревью спецификации задачи

★ 6.9 · code-review-automation

Оспаривание технических решений (Challenge) is a Claude Code skill that adversarially reviews a grooming spec before implementation begins and returns a verdict — APPROVED, NEEDS_REVISION, or BLOCKED — with findings classified by MoSCoW priority. Reach for it when a task looks straightforward but you want a second opinion before the team writes a single line; when you suspect unvalidated assumptions are buried in the spec; when missing dependencies could blow up the estimate mid-sprint. The skill targets the wp-media/wp-rocket repository: given an issue number N it locates `<N>.md` and `<N>-spec.md` under `.TemporaryItems/Issues/wp-rocket/issues/`, invokes a challenger sub-agent that optionally reads Section 13 of `AGENTS.md` for session learnings, and outputs a `## Challenge Report`. If the spec is missing, it instructs you to run `/groom <N>` first. After the report you can post it as a deduplicated GitHub issue comment tagged `<!-- ai-pipeline:challenge -->` via GitHub CLI. The direction is pre-implementation spec analysis, not review of existing code — other cards in code-review-automation cover that. Works for teams using Claude Code with GitHub CLI on wp-rocket; not applicable to other repositories or when no spec file has been generated yet.