miyabi-agent-workflow

GitHub workflow automation agents: PR creation, Issue analysis, and Deployment for the Miyabi framework.

📋 Overview

miyabi-agent-workflow provides three specialized agents for automating GitHub-based development workflows:

• PRAgent (まとめるん): Automatic Pull Request creation with Conventional Commits
• IssueAgent (みつけるん): Issue analysis, classification, and label inference
• DeploymentAgent (はこぶん): CI/CD automation with build, test, deploy, and rollback

Key Capabilities:

• 📝 Conventional Commits: Auto-generates PR titles following feat(scope): description format
• 🏷️ Smart Labeling: Infers appropriate labels from Miyabi's 57-label system
• 🤖 Agent Assignment: Automatically assigns tasks to the right agent (Coordinator, CodeGen, etc.)
• 🚀 CI/CD Pipeline: End-to-end deployment from build to health check
• 🔄 Auto-Rollback: Automatic rollback on deployment failure
• 📊 Impact Assessment: Evaluates issue severity and business impact

🚀 Features

PRAgent (まとめるん)

• Conventional Commits Compliance: Generates feat, fix, refactor, docs, test, ci prefixes
• Scope Detection: Extracts scope from title (e.g., "Auth: login bug" → fix(auth): login bug)
• PR Body Generation: Includes overview, changes, test results, checklist, and related issues
• Label Assignment: Automatically applies labels based on task type and priority
• Reviewer Assignment: Assigns appropriate reviewers based on code owners

IssueAgent (みつけるん)

• Type Inference: Detects Feature, Bug, Refactor, Docs, Test, or Deployment
• Severity Assessment: Evaluates Critical, High, Medium, or Low severity
• Impact Evaluation: Determines System, User, or Cosmetic impact level
• Agent Routing: Assigns to Coordinator, CodeGen, Review, PR, or Deployment agents
• Duration Estimation: Predicts task completion time (e.g., Feature: 60 min, Bug: 30 min)
• Dependency Extraction: Parses dependencies from issue body (e.g., Depends on #123)
• Label Generation: Selects from 57-label system across 11 categories

DeploymentAgent (はこぶん)

• Build Execution: Runs cargo build --release with error handling
• Test Execution: Runs cargo test --all with pass/fail tracking
• Environment Support: Staging (auto-deploy) and Production (approval-required)
• Health Checks: Validates deployment success with HTTP health checks
• Auto-Rollback: Reverts to previous version on failure
• Deployment Status: Tracks Pending → Building → Testing → Deploying → HealthChecking → Success

📦 Installation

Add to your Cargo.toml:

[dependencies]
miyabi-agent-workflow = "0.1.0"

Or install the CLI:

cargo install miyabi-cli

🔧 Usage

PRAgent Example

use miyabi_agent_workflow::PRAgent;
use miyabi_agent_core::BaseAgent;
use miyabi_types::{AgentConfig, Task, TaskType};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let config = AgentConfig {
        github_token: std::env::var("GITHUB_TOKEN")?,
        repo_owner: Some("your-org".to_string()),
        repo_name: Some("your-repo".to_string()),
        ..Default::default()
    };

    let pr_agent = PRAgent::new(config);

    let task = Task {
        id: "task-001".to_string(),
        title: "Auth: Add Google OAuth support".to_string(),
        description: "Implement Google OAuth 2.0 authentication".to_string(),
        task_type: TaskType::Feature,
        ..Default::default()
    };

    // Generates PR:
    // Title: feat(auth): Add Google OAuth support
    // Body: Overview, changes, checklist
    // Labels: ✨ type:feature, 🏗️ state:implementing
    let result = pr_agent.execute(&task).await?;

    println!("PR created: {:?}", result.data);
    Ok(())
}

IssueAgent Example

use miyabi_agent_workflow::IssueAgent;
use miyabi_types::{AgentConfig, Issue};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let config = AgentConfig::default();
    let issue_agent = IssueAgent::new(config);

    let issue = Issue {
        number: 42,
        title: "Critical: Database connection timeout".to_string(),
        body: "Users cannot login. Database times out after 5s.".to_string(),
        labels: vec!["bug".to_string()],
        ..Default::default()
    };

    let analysis = issue_agent.analyze_issue(&issue)?;

    // Output:
    // Type: Bug
    // Severity: Critical
    // Impact: System
    // Agent: CodeGenAgent
    // Duration: 30 minutes
    // Labels: ["🐛 type:bug", "🔥 priority:P0-Critical", "🤖 agent:codegen"]

    println!("Analysis: {:?}", analysis);
    Ok(())
}

DeploymentAgent Example

use miyabi_agent_workflow::{DeploymentAgent, Environment};
use miyabi_agent_core::BaseAgent;
use miyabi_types::{AgentConfig, Task, TaskType};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let config = AgentConfig {
        firebase_staging_project: Some("miyabi-staging".to_string()),
        staging_url: Some("https://staging.miyabi.dev".to_string()),
        ..Default::default()
    };

    let deploy_agent = DeploymentAgent::new(config);

    let task = Task {
        id: "deploy-001".to_string(),
        title: "Deploy to staging".to_string(),
        task_type: TaskType::Deployment,
        ..Default::default()
    };

    // Executes:
    // 1. cargo build --release
    // 2. cargo test --all
    // 3. Deploy to staging
    // 4. Health check (https://staging.miyabi.dev/health)
    // 5. Rollback if health check fails

    let result = deploy_agent.execute(&task).await?;

    println!("Deployment: {:?}", result.status);
    Ok(())
}

📊 Label System Integration

IssueAgent integrates with Miyabi's 57-label system across 11 categories:

See LABEL_SYSTEM_GUIDE.md for full documentation.

🔄 Deployment Workflow

┌─────────────────────┐
│  cargo build        │ → BuildResult
└─────────────────────┘
          ↓
┌─────────────────────┐
│  cargo test         │ → TestResult
└─────────────────────┘
          ↓
┌─────────────────────┐
│  Deploy (Staging)   │ → Automatic
└─────────────────────┘
          ↓
┌─────────────────────┐
│  Health Check       │ → GET /health
└─────────────────────┘
          ↓
┌─────────────────────┐
│  Success / Rollback │
└─────────────────────┘

🏗️ Architecture

PRAgent

PRAgent::execute(task)
  ├── generate_pr_title()         → feat(scope): description
  ├── generate_pr_body()          → Markdown body
  ├── infer_labels()              → [type:feature, state:implementing]
  └── create_pull_request()       → GitHub PR

IssueAgent

IssueAgent::analyze_issue(issue)
  ├── infer_issue_type()          → Feature/Bug/Refactor...
  ├── assess_severity()           → Critical/High/Medium/Low
  ├── evaluate_impact()           → System/User/Cosmetic
  ├── determine_agent()           → CodeGenAgent/ReviewAgent...
  ├── estimate_duration()         → 30-120 minutes
  ├── extract_dependencies()      → [#123, #456]
  └── generate_labels()           → [57-label system]

DeploymentAgent

DeploymentAgent::deploy(task)
  ├── execute_build()             → cargo build --release
  ├── execute_tests()             → cargo test --all
  ├── deploy_to_environment()     → Staging/Production
  ├── run_health_check()          → GET /health (3 retries)
  └── rollback_if_failed()        → git revert + redeploy

🔗 Dependencies

• Core: miyabi-agent-core, miyabi-types, miyabi-core, miyabi-github
• Runtime: tokio, async-trait
• GitHub: octocrab, reqwest
• Serialization: serde, serde_json
• Utilities: chrono, regex, thiserror, tracing

🧪 Testing

# Run all tests
cargo test --package miyabi-agent-workflow

# Test specific agent
cargo test --package miyabi-agent-workflow pr_agent
cargo test --package miyabi-agent-workflow issue_agent
cargo test --package miyabi-agent-workflow deployment_agent

# Integration tests (requires GitHub token)
GITHUB_TOKEN=ghp_xxx cargo test --package miyabi-agent-workflow --test integration

📚 Related Crates

• miyabi-agent-coordinator - Task orchestration and DAG planning
• miyabi-agent-codegen - AI-powered code generation
• miyabi-agent-review - Code quality review and scoring
• miyabi-github - GitHub API client wrapper
• miyabi-types - Shared type definitions

🤝 Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

📄 License

Licensed under the MIT License. See LICENSE for details.

🔖 Version History

• v0.1.0 (2025-10-25): Initial release
  • PRAgent with Conventional Commits support
  • IssueAgent with 57-label system integration
  • DeploymentAgent with build/test/deploy/rollback workflow
  • Automatic agent routing and label inference

Part of the Miyabi Framework - Autonomous AI Development Platform