AGENTS

Code Review Agent

Multi-source + code-graph driven code review, outputting structured defect reports and evidence chains.

Core Pain Points

1. Low manual review efficiency: Code review requires simultaneously reading commit logs, requirement documents, code diffs, and changed source files — reviewing file by file is time-consuming.
2. Incomplete context: Looking only at local diffs tends to miss call relationships, data flows, and requirement constraints, leading to incomplete issue identification.
3. Inconsistent, hard-to-trace results: Different reviewers apply different standards for issue classification, severity levels, and improvement suggestions; defects lack code evidence, making re-verification and remediation tracking costly.

Solution & Technical Implementation

1. Multi-source input parsing: Consumes commit logs, requirement documents, code diffs, and changed source files in a unified pipeline, extracting both requirements and code structure.
2. Code graph + diff correlation: Builds a code graph that associates changed lines with functions, modules, and call context — filling the engineering context gap that a raw diff cannot provide.
3. Template-guided review: Uses predefined review templates to produce issue classification and severity, defect location, code evidence chain, and improvement suggestions as a structured report.

Highlights

• Commit diff + code graph: Balances local changes with global dependency relationships.
• Requirement-constrained parsing: Uses the requirements document as a review baseline, reducing subjective judgments that are detached from requirements.
• Evidence-chain output: Each issue is pinpointed to specific code and change content for easy re-verification.

Business Value

• Automates document consolidation, context retrieval, and initial issue classification, letting engineers focus on decision-making and fixes.
• Template-driven checks and evidence-based conclusions reduce omissions and improve consistency of review results.
• Review templates, code graphs, and report structures can be reused across teams and projects, making them suitable for continuous integration or quality-gate pipelines.

---