Autonomous Coding Agent Template

Golden path template for building autonomous coding agents with Claude SDK.

What's Included

Category	Components
Core	Agent harness, session management, progress tracking
Security	Bash allowlist, command validation, sandbox support
Tools	File operations, bash commands, browser automation
Testing	Security hook tests, pytest configuration
Prompts	Initializer and coding agent templates

Quick Start

1. Go to Backstage Software Catalog
2. Select "Autonomous Coding Agent"
3. Configure your agent capabilities
4. Submit and start building

Template Options

Component Info

Parameter	Description
name	Unique agent name (lowercase, alphanumeric with hyphens)
owner	Owning team from Backstage catalog
description	What the agent does

Runtime

Parameter	Options	Default	Description
pythonVersion	3.13, 3.12, 3.11	3.12	Python version for the agent
packageManager	uv, poetry, pip	uv	Package manager for dependencies

Model & SDK

Parameter	Options	Default	Description
defaultModel	claude-sonnet-4-5, claude-opus-4-5, claude-haiku-3-5	claude-sonnet-4-5	Default Claude model for agent tasks
maxTurns	10-1000	100	Maximum conversation turns per session

Tools & MCP Servers

Parameter	Default	Description
enableFileTools	true	Read, write, edit, glob, grep tools
enableBashTools	true	Command execution with allowlist
enableBrowserTools	false	Puppeteer MCP for web interaction
enableGitHubTools	false	GitHub MCP for PR/issue management
enableDatabaseTools	false	PostgreSQL MCP for database queries

Agent Behavior

Parameter	Options	Default	Description
agentPattern	two-phase, single-session, multi-agent	two-phase	How the agent manages long-running tasks
progressTracking	feature-list, git-commits, none	feature-list	How to track task completion
maxIterations	1-200	50	Maximum agent iterations per session

Security

Parameter	Options	Default	Description
sandboxMode	container, none	container	OS-level sandbox for command execution
allowedCommands	array	ls, cat, grep, git, npm, node, python, pip	Commands the agent can execute

What Gets Created

{name}/
├── src/
│   ├── __init__.py
│   ├── main.py           # Entry point
│   ├── agent.py          # Session logic
│   ├── client.py         # Claude SDK setup
│   ├── security.py       # Command validation
│   ├── prompts.py        # Prompt loading
│   └── progress.py       # Progress tracking
├── prompts/
│   ├── app_spec.txt      # Application spec
│   ├── initializer_prompt.md
│   └── coding_prompt.md
├── tests/
│   ├── __init__.py
│   └── test_security.py
├── docs/
│   ├── GETTING_STARTED.md
│   ├── architecture.md
│   ├── SECURITY.md
│   ├── PATTERNS.md
│   ├── EXTENDING.md
│   └── TROUBLESHOOTING.md
├── .github/
│   ├── workflows/ci.yaml
│   └── dependabot.yml
├── catalog-info.yaml
├── pyproject.toml
├── Makefile
├── README.md
└── .gitignore

Documentation

Document	Description
Getting Started	First steps
Architecture	How it works
Security	Safety guardrails

Based On

This template is based on Anthropic's autonomous-coding quickstart.

Support

• Slack: #platform-help
• Office Hours: Thursdays 2-3pm