cuggino

WARNING: This tool runs Claude in dangerous mode (--dangerously-skip-permissions). Only use it inside a sandbox or virtual machine. It grants the AI agent unrestricted access to execute commands, modify files, and interact with your system without confirmation prompts.

Why "cuggino"? The name comes from an Italian saying — when you have an app idea, there's always a cugino (cousin) who can build it for cheap.

An autonomous coding loop CLI that orchestrates AI agents to plan, implement, and review code changes against specs.

Specs-driven workflow: define your project specifications in a .specs/ folder, and the system autonomously plans, implements, and reviews changes to match them. Tasks are queued as markdown files in .cuggino/backlog/ and picked up in filename order.

Commands

cuggino run

Executes a single coding loop for a given focus.

Options: --focus, --specs, --max-iterations, --check-command, --commit

cuggino watch

Continuously processes backlog items, watches for new items, and handles spec issues.

Options: --specs, --max-iterations, --check-command, --commit, --audit

cuggino plan

Spawns an interactive Claude session for discussing and writing specifications.

Options: --specs

cuggino setup

Interactively configures the project (specs path, max iterations, check command, commit, audit). Writes configuration to .cuggino.json.

Specs-Driven Workflow

Specifications live in the .specs/ folder. The backlog system (.cuggino/backlog/) queues tasks as markdown files, processed in filename order.

The agent loop iterates through: Planning → Implementing → Reviewing — repeating until the implementation matches the specs or the iteration limit is reached.

For detailed specifications, see the .specs/ folder.