polaris_shell

Shell command execution for Polaris agents.

Overview

Provides shell command execution with a safety-first permission model, directory sandboxing, timeout enforcement, and output truncation.

• ShellExecutor - Execution engine with permission checking (global resource)
• ShellTools - LLM-facing tool wrapper for the tool registry
• ShellPlugin - Plugin that registers both of the above
• ShellConfig - Configuration for allowed/denied commands and directories

Permission Model

Every command is matched against glob patterns, evaluated in precedence order:

Compound commands (&&, ||, ;, |) are split and each subcommand is evaluated separately. The most restrictive result applies.

Example

use polaris_shell::{ShellConfig, ShellPlugin};

let plugin = ShellPlugin::new(
    ShellConfig::new()
        .with_allowed_commands(vec!["cargo *".into(), "git *".into()])
        .with_denied_commands(vec!["rm -rf *".into(), "sudo *".into()])
        .with_allowed_dirs(vec!["/home/user/project".into()])
);

License

Apache-2.0