domains / software-engineering / multi-agent-code-review / performance-testing-review-multi-agent-review

performance-testing-review-multi-agent-review

Skill card

combined

The two copies describe the same task (coordinating a multi-perspective code review) but disagree on the core procedure. The first copy is a high-level orchestration framework that mandates dynamic agent routing, parallel/sequential execution, weighted conflict resolution, and quality-score validation. The second copy is a bounded, safety-focused procedure that explicitly forbids spawning agents without authorization, rejects weighted voting and confidence scores, and requires central reproduction of findings. Because the first copy's 'intelligent synthesis' and 'weighted scoring' directly conflict with the second copy's 'resolve disagreement through code or tests' and 'do not invent confidence scores,' crowning one would hide a real, safer alternative. The combined instruction must reconcile these by making delegation optional and prioritizing evidence-based resolution over algorithmic scoring.

Repositories

Other copies

Combined skill

# Multi-Agent Code Review Orchestration

## When to Use
Use this skill when a codebase or diff needs review across multiple disciplines such as correctness, security, architecture, and performance.

## Inputs and Scope
1. **Define Scope**: Pin the repository, base/head commits, changed paths, and intended behavior.
2. **Authorization Check**: Determine if the user has authorized delegation to independent agents and if the host environment supports it. If not, conduct the review sequentially.

## Review Process
1. **Context Analysis**: Read the complete diff and directly affected call paths. Identify relevant perspectives (e.g., security, performance) based on the change.
2. **Agent Selection (Optional)**: If delegation is authorized, select specialized reviewers (Code Quality, Security, Architecture, Performance). Assign each a bounded question, owned paths, and a time/effort limit.
3. **Execution**: 
   - **Parallel**: Run independent reviews concurrently if authorized.
   - **Sequential**: Run reviews one by one if delegation is not authorized or for dependent insights.
4. **Evidence Collection**: Record findings with exact locations, triggers, consequences, and reproduction steps. Keep hypotheses separate from demonstrated failures.

## Synthesis and Resolution
1. **Deduplication**: Merge insights by root cause, not wording.
2. **Conflict Resolution**: Resolve disagreements through code inspection or test execution. Do not use weighted voting or confidence scores as evidence of correctness.
3. **Validation**: Reproduce important findings centrally. Preserve original failing results; a rerun does not erase them.

## Output
Return a consolidated report prioritizing:
1. Blockers (critical issues with demonstrated impact).
2. Material improvements.
3. Untested areas or hypotheses.

## Limitations
- This procedure does not prove whole-repository safety from a diff.
- Parallel review adds cost and may duplicate blind spots.
- No compliance validation or automated quality scoring is performed.