code-review

name	code-review
description	Performs production-ready code reviews on git changes. Supports commit/range/file-scoped analysis, impact assessment, breaking-change detection, confidence-aware finding classification, and risk-weighted verdict generation.

Code Review Capabilities

Structured capability set for reviewing git changes with evidence-backed findings and risk-aware verdicts.

Tools

• Git CLI: Retrieves commit metadata, diffs, range context, and history graph information
• Search tools (rg, fd): Traces references, consumer impact, and related tests across repositories

Domains

• Change analysis: Diff interpretation, file classification, and change-scope characterization
• Impact assessment: Consumer tracing, publicly exposed contract impact, behavioral regression risk
• Risk evaluation: Severity classification, confidence modeling, and verdict generation

Input Contract

Review targets support three input shapes:

Target Type	Required Fields	Description
Single Commit	commit_hash	Reviews one commit and its patch-level changes
Commit Range	start_hash, end_hash	Reviews a contiguous range where start_hash is earliest and end_hash is latest
File-Scoped	target (commit_hash or range) + file_path[]	Limits analysis to selected files within the target

Preconditions

• Repository context is available and readable by git
• Target commit or range can be resolved from repository history
• Diff and metadata can be retrieved for the target
• If analysis preconditions fail (unknown hash, missing parent, ambiguous range), limitations are recorded in the review output

Core Capabilities

• Commit and Range Analysis: Collects metadata, patch details, and structural change information
• Change Classification: Categorizes files by role and change type for review prioritization
• Impact Detection: Traces direct and indirect consumers to estimate runtime and contract impact
• Breaking Change Detection: Identifies incompatible changes on externally consumable contracts
• Confidence-Aware Findings: Attaches evidence confidence and verification status to each finding
• Risk-Weighted Verdicts: Combines severity, confidence, and critical-domain context to determine outcome

Output Contract

Review output consists of:

• Findings: Each finding includes severity, evidence, impact, confidence, verification status, and remediation suggestion
• Verdict: APPROVE, COMMENT, or REQUEST_CHANGES with explicit decision rationale
• Analysis Limitations: Unverifiable areas and constraints discovered during review
• Risk Context: Critical-domain exposure and weighted risk interpretation

Supported Analysis Types

• Commit Range: Review changes across a start~end commit range (start = earliest, end = latest)
• Single Commit: Review changes in a specific commit hash
• File-Scoped: Review specific files within a commit or range

Validation Scenarios

• Root Commit Review: Handles commits without parents via root-safe diff retrieval
• Merge Commit Review: Preserves first-parent and combined-merge perspectives for consistent interpretation
• Breaking Change + Consumer Counting: Uses normalized consumer counting to reduce false positives
• Dynamic/Reflective Usage: Reports unverifiable linkage with reduced confidence
• Behavior Coverage Risk: Escalates risk when changed behavior lacks relevant test coverage
• Verdict Reproducibility: Produces stable verdicts from deterministic weighted risk rules

Severity Levels

Level	Scope
Critical	Security exposure, data integrity risk, service availability risk, externally breaking contract changes
Major	Behavioral defects, reliability degradation, significant performance or observability regressions
Minor	Maintainability, readability, and medium-term quality concerns
Nit	Low-impact consistency and polish suggestions

Technical References

• git_operations.md: Git retrieval patterns, range calculation, and fallback handling
• change_analysis.md: File classification, noise filtering, and large-change strategy
• impact_detection.md: Exposure analysis, consumer counting, and confidence modeling
• severity_criteria.md: Severity boundaries, confidence axis, and escalation conditions
• output_format.md: Finding schema and risk-weighted verdict specification