Struct CommonAgentState 

pub struct CommonAgentState {
    pub system_prompt: String,
    pub model: String,
    pub root: Option<String>,
    pub skip_permissions: bool,
    pub output_format: Option<String>,
    pub add_dirs: Vec<String>,
    pub capture_output: bool,
    pub sandbox: Option<SandboxConfig>,
    pub max_turns: Option<u32>,
    pub env_vars: Vec<(String, String)>,
    pub on_spawn_hook: Option<OnSpawnHook>,
}

Shared configuration state for CLI-based agent providers.

Embed this struct in each provider to avoid duplicating field declarations and trivial setter implementations.

Fields

system_prompt: String

model: String

root: Option<String>

skip_permissions: bool

output_format: Option<String>

add_dirs: Vec<String>

capture_output: bool

sandbox: Option<SandboxConfig>

max_turns: Option<u32>

env_vars: Vec<(String, String)>

on_spawn_hook: Option<OnSpawnHook>

Optional callback invoked with the OS pid of the spawned agent subprocess. Threaded through from AgentBuilder::on_spawn / Agent::set_on_spawn_hook. Providers call CommonAgentState::notify_spawn right after Command::spawn so callers can capture the child pid before the terminal wait.

Implementations

impl CommonAgentState

pub fn new(default_model: &str) -> Self

pub fn notify_spawn(&self, child: &Child)

Invoke the registered on_spawn hook with the pid of a freshly spawned child, if any. Safe to call even when no hook is set.

pub fn get_base_path(&self) -> &Path

Get the effective base path (root directory or “.”).

pub fn make_command( &self, binary_name: &str, agent_args: Vec<String>, ) -> Command

Create a Command either directly or wrapped in sandbox.

Standard pattern used by Claude, Copilot, and Gemini. Sets current_dir, args, and env vars. Providers with custom sandbox behavior (Codex, Ollama) keep their own make_command().

pub async fn run_interactive_command( cmd: &mut Command, agent_display_name: &str, ) -> Result<()>

Execute a command interactively (inheriting stdin/stdout/stderr).

Returns ProcessError on non-zero exit.

pub async fn run_interactive_command_with_hook( cmd: &mut Command, agent_display_name: &str, on_spawn: Option<&OnSpawnHook>, ) -> Result<()>

Same as [run_interactive_command], but invokes on_spawn once with the child’s OS pid right after spawn and before awaiting the child’s exit — that window is what lets callers register the child pid with an external process store (e.g. so zag ps kill self can SIGTERM the agent child rather than the parent zag process).

pub async fn run_non_interactive_simple( &self, cmd: &mut Command, agent_display_name: &str, ) -> Result<Option<AgentOutput>>

Execute a non-interactive command with simple capture-or-passthrough.

If capture_output is set, captures stdout and returns Some(AgentOutput). Otherwise streams stdout to the terminal and returns None.

Used by Copilot, Gemini, Ollama. Providers with custom output parsing (Claude, Codex) keep their own non-interactive logic.