# Using Bashly with AI

There are three practical ways to use AI while building Bashly projects. Choose the one that matches your workflow.

# Quick chooser

Route	Best for	Setup
Chat with any AI	You already know Bashly and want fast iteration	None
Bashly Chat	You want Bashly-focused guidance with minimal prompt tuning	Open link
Bashly Skill	You use coding agents and want repeatable project workflows	Install skill

# Route 1: Chat with your favorite AI

This is the simplest route if you are already familiar with Bashly.

Give the model:

Your CLI goal.
Your current bashly.yml (or desired command tree).
The behavior you want for each command.

Example starter prompt:

I am building a Bashly CLI.
Please design or update my bashly.yml and command partials under src/.
Keep names and help text concise.
After editing, list the exact bashly generate command and 3 test commands.

# Route 2: Use Bashly Chat

Open Bashly Chat

Use this when you want a Bashly-oriented assistant without setting up a local agent skill.

Tip

The bashly AI chat is available at bashly.dev/chat.

It is a good fit for:

Learning Bashly conventions.
Drafting command trees quickly.
Reviewing or improving existing bashly.yml.

# Route 3: Use the Bashly Skill

Bashly AI Kit

The Bashly Skill is intended for agent-based workflows (for example Codex or Claude Code), where the agent edits project files directly and follows a defined Bashly workflow.

Install in Codex using the installer prompt shown in the repository README:

install the skill from https://github.com/bashly-framework/bashly-ai-kit/tree/main/skills/bashly
(master branch)

# Prompting tips for all routes

To get better results, include these in your request:

The exact command UX you want (tool action ARG --flag).
Required vs optional arguments and flags.
Validation rules and error behavior.
Expected output examples.
Your current file layout (src/, settings, overrides).

When you have edits, run:

bashly generate

Then test at least one --help path, one success path, and one error path.