AI Agent Workflows

An autonomous AI agent system that executes multi-step workflows with tool calling, reasoning traces, and real-time progress visualization. Demonstrates the agentic AI pattern: plan, act, observe, reason, repeat.

Why This Exists

AI agents are the next wave beyond simple chatbot wrappers. This project demonstrates the core patterns: multi-step reasoning, tool calling with structured inputs/outputs, workflow orchestration, and transparent decision traces. The UI shows every step of the agent's reasoning so you can see exactly how it plans and acts.

Workflows

1. Triage GitHub Issues

Agent fetches open issues, analyzes severity, classifies by type and priority, assigns to team members, and generates a summary report.

2. Review Pull Request

Agent examines code diffs, runs static analysis, identifies bugs and code quality issues, and submits a structured review with inline comments.

3. Incident Response

Agent queries error logs, checks infrastructure health, identifies root cause, executes remediation runbook, and generates an incident report with timeline.

4. Dependency Audit

Agent scans for security vulnerabilities, checks for outdated packages, analyzes license compatibility, and produces a prioritized action plan.

Architecture

User selects workflow
        |
        v
+-------------------+
|   Agent Runner     |
|                    |
|  1. Think          | <- Plan next action based on goal + observations
|  2. Tool Call      | <- Execute tool with structured input
|  3. Observe        | <- Process tool result
|  4. Repeat         | <- Loop until goal achieved
|  5. Respond        | <- Final structured output
|                    |
+-------------------+
        |
        v
  Real-time step UI
  (thinking, tools, results)

Agent Step Types

Type	Icon	Description
Thinking	Brain	Agent's internal reasoning and planning
Tool Call	Wrench	Executing a tool with structured input
Tool Result	Clipboard	Response from a tool execution
Response	Check	Final structured output to the user

Tools Available

Tool	Category	Description
github.list_issues	GitHub	Fetch open issues from a repository
github.get_issue	GitHub	Get detailed issue information
github.add_labels	GitHub	Apply labels to an issue
github.assign_issue	GitHub	Assign an issue to a team member
github.get_pull_request	GitHub	Fetch PR details and metadata
github.get_pr_diff	GitHub	Get the code diff for a PR
github.create_review	GitHub	Submit a code review
code.analyze	Analysis	Run static analysis on code files
logs.query	Observability	Search and filter application logs
infra.check_service	Infrastructure	Check service health status
infra.check_disk	Infrastructure	Inspect disk usage breakdown
runbook.execute	Operations	Execute a predefined remediation runbook
npm.audit	Security	Scan for known vulnerabilities
npm.outdated	Packages	Check for outdated dependencies
license.check	Compliance	Scan dependency licenses

Tech Stack

Component	Technology	Why
Frontend	Next.js 15 + Tailwind CSS	Modern React with dark theme
Agent Engine	TypeScript async generators	Clean step-by-step execution
Rendering	Dynamic import (no SSR)	Client-side only for interactivity

Quick Start

git clone https://github.com/ctonneslan/ai-agent-workflows.git
cd ai-agent-workflows
npm install
npm run dev

Open http://localhost:3000 and select a workflow.

Project Structure

src/
├── app/
│   ├── globals.css         # Dark theme styles
│   ├── layout.tsx          # Root layout
│   └── page.tsx            # Main page
├── components/
│   ├── WorkflowRunner.tsx  # Workflow selector + step renderer
│   └── AgentStep.tsx       # Individual step display
└── lib/
    └── agent.ts            # Agent engine, workflows, tool definitions

License

MIT