What problem does it solve? AI-assisted coding sessions produce changes that are easy to close without a rigorous risk check, letting correctness bugs, unsafe assumptions, and hidden regressions slip into PRs. This Skill performs a skeptical senior-engineer review of the current task's changes and recommends the smallest safe next step. ## Core Features & Use Cases - Risk-ranked findings: Classifies issues as must-fix, should-fix, or optional, each tagged with impact and effort bands and ordered by value. - No-op discipline: Returns an explicit no-op when the diff and validation evidence are unchanged since the last meaningful review, avoiding review churn. - Optional consensus mode: Supports --consistency N to run N independent review passes merged by consensus, plus a refuter stage that attempts to disprove surviving findings. - Use Case: After completing an implementation slice in a Fhorja task, run the review before slice closure or PR packaging to surface contract mismatches, weak tests, and over-engineering, then route the findings to the right follow-up command. ## Quick Start Ask the assistant to run review-hard on the active task to check the current changes for real engineering risks before closing the slice.
