Code Reviewer Agent

AI-powered code review agent that analyzes pull requests and tickets to deliver comprehensive, context-aware code reviews.

Quick Start

agnt5 create --template python/code_reviewer
export GITHUB_TOKEN=ghp_...
export JIRA_EMAIL=your@email.com JIRA_API_TOKEN=... JIRA_DOMAIN=https://yourcompany.atlassian.net
agnt5 dev up

What You Can Build

Automated PR Reviews: Analyze pull requests for security vulnerabilities, performance issues, and code quality
Ticket-Aligned Reviews: Verify code changes match Jira/Linear ticket requirements and acceptance criteria
Security & Performance Audits: Detect SQL injection, XSS, N+1 queries, and memory leaks automatically

Installation

Prerequisites

Python 3.12+
AGNT5 SDK
GitHub access token
Jira or Linear credentials (optional)

Setup

# Clone or create from template
agnt5 create --template python/code_reviewer
cd code_reviewer

# Install dependencies
uv sync

# Configure environment variables
export GITHUB_TOKEN=ghp_your_github_token

# For Jira integration
export JIRA_EMAIL=your@email.com
export JIRA_API_TOKEN=your_jira_token
export JIRA_DOMAIN=https://yourcompany.atlassian.net

# For Linear integration
export LINEAR_API_TOKEN=your_linear_token

# Start the worker
agnt5 dev up

Usage

Via Workflow Client

Call the workflow programmatically:

import asyncio
from code_reviewer.workflow import code_reviewer_workflow
from agnt5 import with_entity_context

@with_entity_context
async def main():
    result = await code_reviewer_workflow(
        pr_url="https://github.com/owner/repo/pull/123",
        ticket_url="https://yourcompany.atlassian.net/browse/PROJ-456"
    )

    print(f"Review Status: {result['status']}")
    print(f"Report saved to: {result['report_file']}")
    print(result['report'])

asyncio.run(main())

Example Output

The workflow generates a structured markdown report saved to /app/reports/:

/app/reports/repo_pr_123_review.md

Report includes:

Security vulnerability analysis
Performance issue detection
Code quality assessment
Ticket requirement alignment
Actionable feedback with file/line references

Configuration

Environment Variables

Variable	Description	Required
`GITHUB_TOKEN`	GitHub personal access token	Yes
`JIRA_EMAIL`	Jira account email	For Jira tickets
`JIRA_API_TOKEN`	Jira API token	For Jira tickets
`JIRA_DOMAIN`	Jira instance URL	For Jira tickets
`LINEAR_API_TOKEN`	Linear API token	For Linear tickets

Workflow Parameters

code_reviewer_workflow(
    pr_url: str,        # GitHub PR URL
    ticket_url: str     # Jira or Linear ticket URL
)

Architecture

Multi-Agent System

The workflow orchestrates three specialized agents:

1. Context Builder Agent

Tools: pr_fetcher, jira_ticket_fetcher, linear_ticket_fetcher, detect_ticket_source
Role: Gathers PR metadata, file diffs, and ticket details
Output: Enriched context combining code changes and requirements

2. Reviewer Agent

Tools: None (analyzes context from previous step)
Role: Performs deep code analysis for security, performance, quality, and standards
Output: Structured findings with file/line references

3. Synthesizer Function

Type: Stateless function
Role: Compiles review findings into a final markdown report
Output: Formatted review report

Workflow Steps

1. Context Building
   └─> Fetch PR details (title, author, changed files, diffs)
   └─> Detect ticket source (Jira/Linear)
   └─> Fetch ticket details (requirements, acceptance criteria)
   └─> Combine into enriched context

2. Code Review
   └─> Analyze security vulnerabilities
   └─> Check performance patterns
   └─> Evaluate code quality
   └─> Verify ticket alignment

3. Report Synthesis
   └─> Generate markdown report
   └─> Save to /app/reports/
   └─> Return structured result

Entity State Management

The CodeReviewSession entity tracks workflow state:

initial → context_building → code_review → synthesizing → completed
Stores context, review results, and generated report
Enables resumable workflows and state inspection

Troubleshooting

Missing credentials error

EnvironmentError: Missing GITHUB_TOKEN in environment variables

Solution: Export required environment variables before starting the worker.

Invalid PR/ticket URL

ValueError: Invalid PR URL: https://...

Solution: Ensure URLs match expected patterns:

GitHub: https://github.com/owner/repo/pull/123
Jira: https://company.atlassian.net/browse/PROJ-123
Linear: https://linear.app/team/issue/PROJ-123

Worker connection failed

🔗 Connecting to AGNT5 Coordinator... [FAILED]

Solution: Ensure coordinator is running and project ID in agnt5.yaml is correct.

Large PR truncation

Large file diffs are automatically truncated to 100 lines per file to prevent token overflow. Adjust max_patch_lines parameter in pr_fetcher tool if needed.

Customization

Extend Review Criteria

Modify the review prompt in src/code_reviewer/workflow.py:135-151 to add custom analysis:

review_user_prompt = f"""Using the gathered context below, perform a thorough code review:

Focus on:
- Security vulnerabilities (SQL injection, XSS, auth flaws, secrets)
- Performance issues (N+1 queries, inefficient algorithms, memory leaks)
- Code quality (smells, duplications, complexity, maintainability)
- Standards (naming, formatting, documentation)
- Alignment with ticket requirements
- [YOUR CUSTOM CRITERIA HERE]
"""

Add New Ticket Sources

Implement a new fetcher tool in src/code_reviewer/tools.py:

@tool(auto_schema=True)
async def custom_ticket_fetcher(ctx: Context, ticket_url: str) -> dict:
    # Your implementation
    pass

Update detect_ticket_source to recognize the new platform.

Related Templates

weather-agent: Example of tool-based agentic workflows
text-to-sql: Multi-step reasoning workflows with validation
pdf-questions: Document analysis with context extraction

License

MIT License - see LICENSE for details