Aptoflow

Production-grade agentic workflow development system. Build, test, and deploy AI-powered workflows to Modal using OpenRouter for LLM access.

Getting Started

# 1. Bootstrap the environment
python bootstrap.py

# 2. Configure environment variables
cp .env.example .env
# Edit .env with your API keys

# 3. Create your first workflow
# In Claude Code, run: /new-workflow

Claude Code Skills

Skill	Description
/new-workflow	Scaffold a new workflow — gathers name, description, deploy target, recommends model
/build	Implement a workflow from its .agent/plans/ spec
/deploy	Deploy a workflow to Modal with pre/post validation
/test	Run tests for a specific workflow
/list-workflows	Show all workflows with status, model, and run commands
/onboard	Get oriented — scans project structure and workflow status

lib/ — Shared Modules

Module	Purpose
client.py	OpenRouter wrapper via OpenAI SDK (get_client(), chat())
agent.py	Agentic tool-calling loop with safety guardrails
logger.py	Structured JSON logging
models.py	Pydantic base models (WorkflowInput/Output, WebhookRequest/Response)
cost.py	LLM cost tracking and aggregation
auth.py	Bearer token verification and rate limiting for webhooks

Safety Features

• Max iterations — Agent loops capped at configurable limit (default 10)
• Timeouts — Agent loops have wall-clock timeouts (default 300s)
• Cost tracking — Every LLM call tracked and summarized
• Rate limiting — In-memory per-IP rate limiter on webhook endpoints
• Auth — Bearer token verification on all webhook endpoints
• Validation — Pydantic models for all inputs and outputs

Project Structure

Aptoflow/
├── lib/              # Shared modules
├── workflows/        # Individual workflows
├── tests/            # Centralized test suite
├── .agent/plans/     # Workflow implementation plans
├── .claude/skills/   # Claude Code skill definitions
├── CATALOG.md        # Workflow registry
├── MODELS.md         # OpenRouter model guide
└── bootstrap.py      # Environment setup