remove agent-callable restart tool (blast radius in multi-user model)
Nightly Build / build (push) Successful in 6m37s
Nightly Build / build (push) Successful in 6m37s
This commit is contained in:
@@ -43,7 +43,6 @@ pub mod list_secrets;
|
||||
pub mod notify;
|
||||
pub mod set_secret;
|
||||
pub mod read_notification;
|
||||
pub mod restart;
|
||||
pub mod show_file;
|
||||
pub mod toggle_item;
|
||||
|
||||
|
||||
@@ -1,69 +0,0 @@
|
||||
use std::sync::OnceLock;
|
||||
|
||||
use anyhow::Result;
|
||||
use serde_json::{Value, json};
|
||||
use tracing::{info, warn};
|
||||
|
||||
use crate::tools::{Tool, ToolDescriptionLength};
|
||||
|
||||
/// How to restart, when exiting for a supervisor is not the answer.
|
||||
///
|
||||
/// A shell with no supervisor watching its exit code would need to tear itself
|
||||
/// down and respawn on its own. That is knowledge about the process shell, which
|
||||
/// the core does not have — so such a shell installs it here. The default server
|
||||
/// shell has a supervisor (`run.sh`) and installs no handler, so `restart` falls
|
||||
/// back to the supervisor protocol below.
|
||||
///
|
||||
/// Returns only on failure; a successful handler never comes back.
|
||||
pub type RestartHandler = Box<dyn Fn() -> Result<()> + Send + Sync>;
|
||||
|
||||
static HANDLER: OnceLock<RestartHandler> = OnceLock::new();
|
||||
|
||||
/// Called once by the process shell during startup, before any tool can run.
|
||||
pub fn set_restart_handler(handler: RestartHandler) {
|
||||
if HANDLER.set(handler).is_err() {
|
||||
warn!("a restart handler is already installed — ignoring this one");
|
||||
}
|
||||
}
|
||||
|
||||
pub struct Restart;
|
||||
|
||||
impl Tool for Restart {
|
||||
fn name(&self) -> &str { crate::tools::tool_names::RESTART }
|
||||
fn category(&self) -> crate::tools::ToolCategory { crate::tools::ToolCategory::Shell }
|
||||
|
||||
fn description(&self) -> &str {
|
||||
"Restart the skald process. Nothing is recompiled: the same binary is re-executed, \
|
||||
so this applies config.yml and database changes, which are only read at startup. \
|
||||
To load new code, build first (./build.sh), then restart. \
|
||||
Requires user approval."
|
||||
}
|
||||
|
||||
fn parameters_schema(&self) -> Value {
|
||||
json!({
|
||||
"type": "object",
|
||||
"properties": {}
|
||||
})
|
||||
}
|
||||
|
||||
fn describe(&self, _args: &Value, _length: ToolDescriptionLength) -> String {
|
||||
"restart skald".to_string()
|
||||
}
|
||||
|
||||
fn execute(&self, _args: Value) -> Result<String> {
|
||||
// A shell that installed its own teardown-and-respawn handles it here.
|
||||
// Normally this never returns.
|
||||
if let Some(handler) = HANDLER.get() {
|
||||
info!("restart requested — delegating to the installed handler");
|
||||
handler()?;
|
||||
warn!("the restart handler returned without restarting — falling back");
|
||||
}
|
||||
|
||||
// Headless: exit with code -1 (= 255 on Unix), which `run.sh` reads as
|
||||
// "re-execute me". `_exit()` rather than `exit()` skips C atexit handlers
|
||||
// — whisper-rs's Metal GPU cleanup aborts with SIGABRT, turning the exit
|
||||
// code into 134 and stopping the supervisor instead of restarting it.
|
||||
info!("restart requested — exit(-1) → supervisor re-executes the binary");
|
||||
unsafe { libc::_exit(-1) }
|
||||
}
|
||||
}
|
||||
@@ -1,6 +1,5 @@
|
||||
pub const EXECUTE_TASK: &str = "execute_task";
|
||||
pub const EXECUTE_SUBTASK: &str = "execute_subtask";
|
||||
pub const RESTART: &str = "restart";
|
||||
pub const UPDATE_SCRATCHPAD: &str = "update_scratchpad";
|
||||
pub const WRITE_TODOS: &str = "write_todos";
|
||||
pub const ASK_USER_CLARIFICATION: &str = "ask_user_clarification";
|
||||
|
||||
Reference in New Issue
Block a user