Plugin System

AI Shell Suite supports tool plugins that provide specialized interactive assistants.

Built-in Plugins

ID	Name	Requires
metasploit	Metasploit Assistant	msfconsole
wapiti	Wapiti Assistant	wapiti
nmap	Nmap Assistant	nmap
docker	Docker Assistant	docker
podman	Podman Assistant	podman
kubectl	Kubectl Assistant	kubectl
helm	Helm Assistant	helm
git	Git Assistant	git
ansible	Ansible Assistant	ansible / ansible-playbook
terraform	Terraform Assistant	terraform or tofu
aws	AWS Assistant	aws
trivy	Trivy Assistant	trivy
systemd	Systemd Assistant	systemctl
network	Network Assistant	ip / ss / ifconfig

Using a Plugin

ai-shell --mode kubectl
ai-shell -m git -p local
ais -m terraform
ais -m ansible --dry-run

Or run ai-shell and pick from the interactive menu (plugins appear after the core modes). Unavailable tools are marked “(not installed)”.

Creating a New Plugin

1. Create a file under ai_shell/plugins/, e.g. mytool.py:

from .base import ToolPlugin, register_plugin_class

MYTOOL_SYSTEM_PROMPT = (
    "You are an expert assistant for mytool. "
    "When you provide a command, enclose it in a ```bash ... ``` block."
)

@register_plugin_class
class MyToolPlugin(ToolPlugin):
    id = "mytool"
    name = "MyTool Assistant"
    description = "Short description shown in the menu"
    system_prompt = MYTOOL_SYSTEM_PROMPT
    start_command = ["bash"]          # or ["mytool"] for a direct shell
    requires_pty = True
    color_key = "info"

    def check_available(self) -> bool:
        import shutil
        return shutil.which("mytool") is not None

1. Restart ai-shell. The plugin is auto-discovered and appears in the menu.

API Overview

Symbol	Purpose
ToolPlugin	Abstract base class
@register_plugin_class	Decorator – registers on import
list_plugins()	All registered plugin instances
list_plugin_info()	Lightweight metadata for menus
get_plugin(id)	Lookup by id
discover_plugins()	Scans the package (called automatically)

Plugins may override:

• check_available() – detect whether the binary is installed
• on_start() / on_stop() – lifecycle hooks
• preprocess_command(cmd) – transform LLM suggestions before execution