Agent Cross-Review

Agent Cross-Review

Protocol for structured collaboration between specialized agents.

When to Use

• One agent reviewing another's output
• Handoff between feature and expert agents
• Validating cross-cutting concerns
• Resolving conflicting recommendations

Core Principle

Review in domain, defer on scope.

Each agent excels in their specialty. Cross-review catches blind spots without overstepping boundaries.

---

Review Protocol

Step 1: Scope Identification

Before reviewing, identify:

## Cross-Review Context

| Item | Value |
|------|-------|
| Reviewer | {agent-name} |
| Author | {agent-name} |
| Artifact | {file or output} |
| Review Type | Technical / Structural / Integration |

Step 2: Domain Check

Step 3: Calibrated Feedback

Use priority tiers:

### Cross-Review: {artifact}

**CRITICAL** (blocks merge)
- [Issue affecting correctness or security]

**MAJOR** (should fix)
- [Issue affecting maintainability]

**MINOR** (nice to have)
- [Improvement suggestion]

**DEFER** (out of scope for this review)
- [Valid concern but not reviewer's domain]

---

Role-Specific Protocols

feature-interface-cli Reviewing expert-python

Focus Areas:

• Test file naming follows project conventions
• Coverage targets CLI-critical paths
• Integration with existing command structure

Defer To expert-python:

• Fixture design decisions
• pytest marker selection
• Test helper organization

Template:

## CLI Feature Review

### Structure
- [ ] Test files in correct location
- [ ] Naming follows test_{feature}_cmd.py
- [ ] Coverage priorities align with CLI usage

### Concerns for expert-python
- [List items needing pytest expertise]

expert-python Reviewing feature-interface-cli

Focus Areas:

• Type hints complete and correct
• pytest patterns followed
• DRY violations identified

Defer To feature-interface-cli:

• CLI-specific testing approaches
• Typer/​Rich patterns
• Project structure decisions

Template:

## Python Quality Review

### Code Quality
- [ ] Type hints present on public functions
- [ ] No mutable default arguments
- [ ] Error handling is specific

### Test Quality
- [ ] Fixtures use appropriate scope
- [ ] No duplicate fixture definitions
- [ ] AAA pattern followed

### Concerns for feature-interface-cli
- [List items needing project context]

---

Handoff Protocol

From Feature Agent to Expert Agent

## Handoff: {feature} implementation

### Completed
- [What's done]

### Needs Review
- [Specific areas needing expert input]

### Context
- [Domain-specific decisions made and why]

### Questions
1. [Specific question for expert]

From Expert Agent to Feature Agent

## Technical Recommendations: {area}

### Recommendations
1. [Recommendation with rationale]

### Priority Assessment
- **Now**: [Must address before merge]
- **Soon**: [Address in follow-up]
- **Later**: [Nice to have]

### Scope Consideration
- [Note if recommendation needs project context validation]

---

Conflict Resolution

When agents disagree:

Priority Framework

Resolution Steps

1. State the conflict clearly
2. Identify domain - whose expertise applies?
3. Check project context - what do existing patterns show?
4. Propose compromise - can both concerns be addressed?
5. Escalate if needed - ask user for decision

---

Common Pitfalls

---

Output Format

Quick Review

## Cross-Review Summary

**Artifact**: {file/​output}
**Verdict**: Approve / Needs Changes / Discuss

### In My Domain
- [Finding 1]
- [Finding 2]

### Deferred (not my domain)
- [Observation for other agent]

Detailed Review

## Cross-Review: {artifact}

### Context
- Author: {agent}
- Reviewer: {agent}
- Type: {technical/​structural}

### Findings

**CRITICAL**
- None / [issues]

**MAJOR**
- [issues]

**MINOR**
- [issues]

### Deferred Items
| Item | Recommended Reviewer | Reason |
|------|---------------------|--------|
| X | expert-python | pytest expertise |
| Y | feature-cli | project conventions |

### Verdict
[Approve / Needs Changes]

Quality Checklist

• Identified review scope before starting
• Focused on domain expertise
• Used priority tiers appropriately
• Deferred out-of-domain concerns
• Provided actionable feedback
• Considered project context