| name | pr-review |
| description | Gemini-native comprehensive PR review — strict priority-based organization and merge readiness assessment with 2M+ token grounding |
| capability | skill.pr_review |
| node_name | node_skill_pr_review_orchestrator |
| dot_color | 208 |
PR Review Skill (Gemini Edition)
Comprehensive PR review that leverages Gemini's massive context window to perform "whole-project" analysis, identifying architectural regressions and structural flaws that local diff-based reviews miss.
Workflow
- Seam Contract Check: Gemini extracts the ticket ID and verifies PR compliance against the corresponding
ModelTicketContract before starting the main review.
- Multi-Endpoint Data Fetch: Gemini pulls PR data from all 4 GitHub endpoints (comments, review comments, reviews, CI status) to ensure no feedback is missed.
- Claude Bot Detection: Specifically identifies and triages feedback from other automated agents using Gemini's superior pattern recognition.
- Priority Classification: Categorizes all findings into a strict priority system:
- CRITICAL: Security vulnerabilities, data loss, or system crashes (BLOCKING).
- MAJOR: Performance problems, incorrect behavior, or missing tests (BLOCKING).
- MINOR: Code quality improvements or missing documentation (BLOCKING).
- NIT: Stylistic preferences or optional refactoring (NON-BLOCKING).
- Merge Readiness: Gemini provides a definitive "Ready to Merge" verdict based on the resolution of all blocking issues.
Gemini Advantages
- Deep Architectural Grounding: Gemini sees the entire PR in the context of the entire codebase, identifying breaking changes across repository boundaries.
- Superior Intent Matching: More accurately determines the severity of a comment by understanding the technical intent of the reviewer and the impact on the system.
- Whole-Project Pattern Reuse: Automatically identifies and suggests internal APIs or models that would improve the PR's implementation.
Arguments
PR_NUMBER: GitHub PR number to review.
--strict: Fail if any Critical/Major/Minor issues are found.
--claude-only: Focus only on feedback from Claude-based bots.