What problem does it solve? Human code review of an agent-written branch is slow and error-prone: the reviewer must reconstruct what was changed, what was already reviewed, what the diagrams and REST contract did, and which tests pin the behavior. This Skill assembles all of that into one self-contained, tabbed review.html page whose snippets are cut from the working tree at build time, so the page cannot drift from the code it describes. ## Core Features & Use Cases - Review record rendering: Reads the review-points.md file the coding agent committed with its fixes (what it fixed, declined, and assumed) and renders the three piles verbatim, never inventing findings of its own. - Deterministic evidence producers: One command runs diagram deltas, sequence diagrams, a C4 container view, a Code City snapshot, a narrated feature video, endpoint-complexity deltas, REST contract diffs, logging and design-system audits, code owners, and Playwright traces, each gated on its own prerequisites. - Deep-linked, incremental page: Every code reference clicks into the editor, snippets are extracted at build time, and refreshes re-run only the producers whose inputs moved. - Use Case: After running /implement-ticket on a branch, invoke /human-review origin/main to get a single page a human reviewer can open in the browser to approve a pull request in minutes. ## Quick Start Run /human-review origin/main in the repository to build and serve the tabbed review page for this branch against its base.
