feature/acp-session-list-cursor
claude/github-issue-40-0au5ua
claude/linked-issue-acp-tui-u4vvpa
claude/minor-release-cut-vkugow
claude/sigit-acp-local-chat-cx6380
claude/sigit-cloud-agent-expansion-reox0a
claude/tool-permission-system
claude/zen-feynman-0u78dk
development
feature/acp-cancel-request
feature/acp-session-list-cursor
feature/acp-v1-conformance
feature/agent-tools-multiedit-glob-todos-remember
feature/cargo-workspace
feature/hooks
feature/load-local-model-explicitly
feature/sigit-agent-profile
feature/sigit-code-cloud
feature/xcode-mcp-server
fix/acp-replay-loaded-session-history
fix/acp-replay-reasoning-and-attachments
fix/dsml-doubled-bar-markers
main
release/v1.5.10
release/v1.5.6
release/v1.6.1
| 1 | //! `run_command` runs in the ACP client's terminal when it offers one. |
| 2 | //! |
| 3 | //! An editor that advertises `terminal` can run a command itself and show it |
| 4 | //! live in the tool call. These tests spawn `sigit` in ACP mode against a |
| 5 | //! scripted endpoint and play that editor: it answers `terminal/*` with |
| 6 | //! output the command could not have printed locally, so each assertion can |
| 7 | //! tell where the command ran. |
| 8 | |
| 9 | use std::collections::VecDeque; |
| 10 | use std::io::{BufRead, BufReader, Read, Write}; |
| 11 | use std::net::TcpListener; |
| 12 | use std::path::{Path, PathBuf}; |
| 13 | use std::process::{Child, ChildStdin, Command, Stdio}; |
| 14 | use std::sync::mpsc::{Receiver, channel}; |
| 15 | use std::sync::{Arc, Mutex}; |
| 16 | use std::time::{Duration, Instant}; |
| 17 | |
| 18 | use serde_json::{Value, json}; |
| 19 | |
| 20 | const TIMEOUT: Duration = Duration::from_secs(60); |
| 21 | |
| 22 | // ── Scripted OpenAI-compatible endpoint ───────────────────────────────────── |
| 23 | |
| 24 | fn sse_body(events: &[Value]) -> String { |
| 25 | let mut body = String::new(); |
| 26 | for event in events { |
| 27 | body.push_str("data: "); |
| 28 | body.push_str(&event.to_string()); |
| 29 | body.push_str("\n\n"); |
| 30 | } |
| 31 | body.push_str("data: [DONE]\n\n"); |
| 32 | body |
| 33 | } |
| 34 | |
| 35 | fn sse_tool_call(id: &str, name: &str, arguments: Value) -> String { |
| 36 | sse_body(&[json!({ |
| 37 | "choices": [{"delta": {"tool_calls": [{ |
| 38 | "index": 0, |
| 39 | "id": id, |
| 40 | "function": {"name": name, "arguments": arguments.to_string()}, |
| 41 | }]}}] |
| 42 | })]) |
| 43 | } |
| 44 | |
| 45 | fn sse_text(text: &str) -> String { |
| 46 | sse_body(&[json!({"choices": [{"delta": {"content": text}}]})]) |
| 47 | } |
| 48 | |
| 49 | /// Serves one scripted SSE response per request and records each request body. |
| 50 | struct FakeEndpoint { |
| 51 | port: u16, |
| 52 | requests: Arc<Mutex<Vec<Value>>>, |
| 53 | } |
| 54 | |
| 55 | impl FakeEndpoint { |
| 56 | /// The result the model was given for the tool call `id`. |
| 57 | fn tool_result(&self, id: &str) -> String { |
| 58 | let requests = self.requests.lock().unwrap(); |
| 59 | requests |
| 60 | .iter() |
| 61 | .flat_map(|request| request["messages"].as_array().cloned().unwrap_or_default()) |
| 62 | .find(|message| message["role"] == "tool" && message["tool_call_id"] == id) |
| 63 | .and_then(|message| message["content"].as_str().map(str::to_string)) |
| 64 | .unwrap_or_else(|| panic!("no tool result for {id} reached the endpoint")) |
| 65 | } |
| 66 | } |
| 67 | |
| 68 | fn start_fake_endpoint(responses: Vec<String>) -> FakeEndpoint { |
| 69 | let listener = TcpListener::bind("127.0.0.1:0").expect("bind fake endpoint"); |
| 70 | let port = listener.local_addr().unwrap().port(); |
| 71 | let requests: Arc<Mutex<Vec<Value>>> = Arc::default(); |
| 72 | let recorded = Arc::clone(&requests); |
| 73 | let queue = Mutex::new(VecDeque::from(responses)); |
| 74 | |
| 75 | std::thread::spawn(move || { |
| 76 | for stream in listener.incoming() { |
| 77 | let Ok(mut stream) = stream else { continue }; |
| 78 | let mut reader = BufReader::new(match stream.try_clone() { |
| 79 | Ok(clone) => clone, |
| 80 | Err(_) => continue, |
| 81 | }); |
| 82 | let mut content_length = 0usize; |
| 83 | loop { |
| 84 | let mut line = String::new(); |
| 85 | if reader.read_line(&mut line).unwrap_or(0) == 0 { |
| 86 | break; |
| 87 | } |
| 88 | let line = line.trim(); |
| 89 | if line.is_empty() { |
| 90 | break; |
| 91 | } |
| 92 | if let Some(length) = line.to_ascii_lowercase().strip_prefix("content-length:") { |
| 93 | content_length = length.trim().parse().unwrap_or(0); |
| 94 | } |
| 95 | } |
| 96 | let mut body = vec![0u8; content_length]; |
| 97 | if reader.read_exact(&mut body).is_err() { |
| 98 | continue; |
| 99 | } |
| 100 | if let Ok(request) = serde_json::from_slice::<Value>(&body) { |
| 101 | recorded.lock().unwrap().push(request); |
| 102 | } |
| 103 | let payload = queue |
| 104 | .lock() |
| 105 | .unwrap() |
| 106 | .pop_front() |
| 107 | .unwrap_or_else(|| sse_text("out of scripted responses")); |
| 108 | let response = format!( |
| 109 | "HTTP/1.1 200 OK\r\ncontent-type: text/event-stream\r\n\ |
| 110 | content-length: {}\r\nconnection: close\r\n\r\n{}", |
| 111 | payload.len(), |
| 112 | payload |
| 113 | ); |
| 114 | let _ = stream.write_all(response.as_bytes()); |
| 115 | let _ = stream.flush(); |
| 116 | } |
| 117 | }); |
| 118 | |
| 119 | FakeEndpoint { port, requests } |
| 120 | } |
| 121 | |
| 122 | // ── ACP client over the binary's stdio ────────────────────────────────────── |
| 123 | |
| 124 | /// What the editor's terminal does, and what it was asked. |
| 125 | #[derive(Default)] |
| 126 | struct Editor { |
| 127 | /// Answer `terminal/create` with an error, the way a client with |
| 128 | /// terminals turned off would. |
| 129 | refuse_create: bool, |
| 130 | /// Every `terminal/*` request the agent sent, in order. |
| 131 | requests: Vec<Value>, |
| 132 | /// Every `session/update` the agent sent, in order. |
| 133 | updates: Vec<Value>, |
| 134 | } |
| 135 | |
| 136 | impl Editor { |
| 137 | fn methods(&self) -> Vec<&str> { |
| 138 | self.requests |
| 139 | .iter() |
| 140 | .filter_map(|request| request["method"].as_str()) |
| 141 | .collect() |
| 142 | } |
| 143 | |
| 144 | fn request(&self, method: &str) -> &Value { |
| 145 | self.requests |
| 146 | .iter() |
| 147 | .find(|request| request["method"] == method) |
| 148 | .unwrap_or_else(|| panic!("no {method} request in {:?}", self.requests)) |
| 149 | } |
| 150 | |
| 151 | /// The `tool_call_update`s for one tool call, in order. |
| 152 | fn call_updates(&self, tool_call_id: &str) -> Vec<&Value> { |
| 153 | self.updates |
| 154 | .iter() |
| 155 | .map(|update| &update["params"]["update"]) |
| 156 | .filter(|update| { |
| 157 | update["sessionUpdate"] == "tool_call_update" |
| 158 | && update["toolCallId"] == tool_call_id |
| 159 | }) |
| 160 | .collect() |
| 161 | } |
| 162 | } |
| 163 | |
| 164 | /// What the editor's terminal printed, which no local run could. |
| 165 | const EDITOR_OUTPUT: &str = "printed in the editor\n"; |
| 166 | |
| 167 | struct AgentUnderTest { |
| 168 | child: Child, |
| 169 | stdin: ChildStdin, |
| 170 | incoming: Receiver<Value>, |
| 171 | next_id: u64, |
| 172 | } |
| 173 | |
| 174 | fn spawn_agent(port: u16, config_dir: &Path, client_terminal: Option<&str>) -> AgentUnderTest { |
| 175 | let mut command = Command::new(env!("CARGO_BIN_EXE_sigit")); |
| 176 | command |
| 177 | .env("OPENAI_BASE_URL", format!("http://127.0.0.1:{port}")) |
| 178 | .env("OPENAI_API_KEY", "test-key") |
| 179 | .env("SIGIT_MODEL", "scripted-model") |
| 180 | .env("SIGIT_CONFIG_DIR", config_dir) |
| 181 | .env("SIGIT_MCP", "off") |
| 182 | .env("SIGIT_PERMISSIONS", "allow") |
| 183 | .env_remove("SIGIT_LOCAL_INFERENCE") |
| 184 | .stdin(Stdio::piped()) |
| 185 | .stdout(Stdio::piped()) |
| 186 | .stderr(Stdio::null()); |
| 187 | match client_terminal { |
| 188 | None => command.env_remove("SIGIT_CLIENT_TERMINAL"), |
| 189 | Some(value) => command.env("SIGIT_CLIENT_TERMINAL", value), |
| 190 | }; |
| 191 | let mut child = command.spawn().expect("spawn sigit in ACP mode"); |
| 192 | |
| 193 | let stdout = child.stdout.take().unwrap(); |
| 194 | let (message_tx, incoming) = channel(); |
| 195 | std::thread::spawn(move || { |
| 196 | for line in BufReader::new(stdout).lines() { |
| 197 | let Ok(line) = line else { break }; |
| 198 | if let Ok(message) = serde_json::from_str::<Value>(&line) |
| 199 | && message_tx.send(message).is_err() |
| 200 | { |
| 201 | break; |
| 202 | } |
| 203 | } |
| 204 | }); |
| 205 | |
| 206 | let stdin = child.stdin.take().unwrap(); |
| 207 | AgentUnderTest { |
| 208 | child, |
| 209 | stdin, |
| 210 | incoming, |
| 211 | next_id: 0, |
| 212 | } |
| 213 | } |
| 214 | |
| 215 | impl AgentUnderTest { |
| 216 | fn send(&mut self, message: Value) { |
| 217 | let mut line = message.to_string(); |
| 218 | line.push('\n'); |
| 219 | self.stdin.write_all(line.as_bytes()).expect("write stdin"); |
| 220 | self.stdin.flush().expect("flush stdin"); |
| 221 | } |
| 222 | |
| 223 | fn request(&mut self, method: &str, params: Value) -> u64 { |
| 224 | self.next_id += 1; |
| 225 | let id = self.next_id; |
| 226 | self.send(json!({"jsonrpc": "2.0", "id": id, "method": method, "params": params})); |
| 227 | id |
| 228 | } |
| 229 | |
| 230 | /// Wait for the response to request `id`, serving the agent's |
| 231 | /// `terminal/*` requests from `editor` on the way. |
| 232 | fn wait_for_response(&mut self, id: u64, editor: &mut Editor) -> Value { |
| 233 | let deadline = Instant::now() + TIMEOUT; |
| 234 | loop { |
| 235 | let remaining = deadline.saturating_duration_since(Instant::now()); |
| 236 | let Ok(message) = self.incoming.recv_timeout(remaining) else { |
| 237 | panic!("timed out waiting for the response to request {id}"); |
| 238 | }; |
| 239 | if message["id"] == id && message.get("method").is_none() { |
| 240 | assert!( |
| 241 | message.get("error").is_none(), |
| 242 | "request {id} failed: {message}" |
| 243 | ); |
| 244 | return message; |
| 245 | } |
| 246 | let Some(method) = message["method"].as_str() else { |
| 247 | continue; |
| 248 | }; |
| 249 | if method == "session/update" { |
| 250 | editor.updates.push(message.clone()); |
| 251 | continue; |
| 252 | } |
| 253 | if !method.starts_with("terminal/") { |
| 254 | continue; |
| 255 | } |
| 256 | editor.requests.push(message.clone()); |
| 257 | let request_id = message["id"].clone(); |
| 258 | let result = match method { |
| 259 | "terminal/create" if editor.refuse_create => { |
| 260 | self.send(json!({ |
| 261 | "jsonrpc": "2.0", |
| 262 | "id": request_id, |
| 263 | "error": {"code": -32603, "message": "terminals are off"}, |
| 264 | })); |
| 265 | continue; |
| 266 | } |
| 267 | "terminal/create" => json!({"terminalId": "term-1"}), |
| 268 | "terminal/wait_for_exit" => json!({"exitCode": 0}), |
| 269 | "terminal/output" => json!({ |
| 270 | "output": EDITOR_OUTPUT, |
| 271 | "truncated": false, |
| 272 | "exitStatus": {"exitCode": 0}, |
| 273 | }), |
| 274 | _ => json!({}), |
| 275 | }; |
| 276 | self.send(json!({"jsonrpc": "2.0", "id": request_id, "result": result})); |
| 277 | } |
| 278 | } |
| 279 | |
| 280 | /// `initialize` with or without the `terminal` capability, then a session |
| 281 | /// in `cwd`. |
| 282 | fn open_session(&mut self, terminal: bool, cwd: &Path) -> String { |
| 283 | let mut editor = Editor::default(); |
| 284 | let id = self.request( |
| 285 | "initialize", |
| 286 | json!({"protocolVersion": 1, "clientCapabilities": {"terminal": terminal}}), |
| 287 | ); |
| 288 | self.wait_for_response(id, &mut editor); |
| 289 | |
| 290 | let id = self.request("session/new", json!({"cwd": cwd, "mcpServers": []})); |
| 291 | self.wait_for_response(id, &mut editor)["result"]["sessionId"] |
| 292 | .as_str() |
| 293 | .expect("session id") |
| 294 | .to_string() |
| 295 | } |
| 296 | |
| 297 | /// Send a prompt, then `session/cancel` as soon as the agent starts |
| 298 | /// waiting on the editor's terminal. Returns the prompt's response. |
| 299 | fn prompt_cancelled_mid_command(&mut self, session_id: &str, editor: &mut Editor) -> Value { |
| 300 | self.prompt_stopped_mid_command(session_id, editor, |_prompt_id| { |
| 301 | json!({ |
| 302 | "jsonrpc": "2.0", |
| 303 | "method": "session/cancel", |
| 304 | "params": {"sessionId": session_id}, |
| 305 | }) |
| 306 | }) |
| 307 | } |
| 308 | |
| 309 | /// Send a prompt and, once the agent is waiting on the editor's terminal |
| 310 | /// (which never answers), send whatever `stop` builds from the prompt's |
| 311 | /// request id. Returns the prompt's response. |
| 312 | fn prompt_stopped_mid_command( |
| 313 | &mut self, |
| 314 | session_id: &str, |
| 315 | editor: &mut Editor, |
| 316 | stop: impl Fn(u64) -> Value, |
| 317 | ) -> Value { |
| 318 | let id = self.request( |
| 319 | "session/prompt", |
| 320 | json!({"sessionId": session_id, "prompt": [{"type": "text", "text": "go"}]}), |
| 321 | ); |
| 322 | let deadline = Instant::now() + TIMEOUT; |
| 323 | loop { |
| 324 | let remaining = deadline.saturating_duration_since(Instant::now()); |
| 325 | let Ok(message) = self.incoming.recv_timeout(remaining) else { |
| 326 | panic!("timed out: the cancel did not end the prompt"); |
| 327 | }; |
| 328 | if message["id"] == id && message.get("method").is_none() { |
| 329 | return message; |
| 330 | } |
| 331 | let Some(method) = message["method"].as_str() else { |
| 332 | continue; |
| 333 | }; |
| 334 | if method == "session/update" { |
| 335 | editor.updates.push(message.clone()); |
| 336 | continue; |
| 337 | } |
| 338 | if !method.starts_with("terminal/") { |
| 339 | continue; |
| 340 | } |
| 341 | editor.requests.push(message.clone()); |
| 342 | let request_id = message["id"].clone(); |
| 343 | let result = match method { |
| 344 | "terminal/create" => json!({"terminalId": "term-1"}), |
| 345 | "terminal/wait_for_exit" => { |
| 346 | // The command is still running; the user presses stop. |
| 347 | self.send(stop(id)); |
| 348 | continue; |
| 349 | } |
| 350 | "terminal/output" => json!({ |
| 351 | "output": "partial\n", |
| 352 | "truncated": false, |
| 353 | }), |
| 354 | _ => json!({}), |
| 355 | }; |
| 356 | self.send(json!({"jsonrpc": "2.0", "id": request_id, "result": result})); |
| 357 | } |
| 358 | } |
| 359 | |
| 360 | fn prompt(&mut self, session_id: &str, editor: &mut Editor) { |
| 361 | let id = self.request( |
| 362 | "session/prompt", |
| 363 | json!({"sessionId": session_id, "prompt": [{"type": "text", "text": "go"}]}), |
| 364 | ); |
| 365 | self.wait_for_response(id, editor); |
| 366 | } |
| 367 | } |
| 368 | |
| 369 | impl Drop for AgentUnderTest { |
| 370 | fn drop(&mut self) { |
| 371 | let _ = self.child.kill(); |
| 372 | let _ = self.child.wait(); |
| 373 | } |
| 374 | } |
| 375 | |
| 376 | /// A scratch project: `config/` for the agent and `work/` as the session root. |
| 377 | struct Scratch { |
| 378 | root: PathBuf, |
| 379 | config: PathBuf, |
| 380 | work: PathBuf, |
| 381 | } |
| 382 | |
| 383 | impl Scratch { |
| 384 | fn new(name: &str) -> Self { |
| 385 | let root = |
| 386 | std::env::temp_dir().join(format!("sigit_acp_terminal_{name}_{}", std::process::id())); |
| 387 | let _ = std::fs::remove_dir_all(&root); |
| 388 | let config = root.join("config"); |
| 389 | let work = root.join("work"); |
| 390 | std::fs::create_dir_all(&config).unwrap(); |
| 391 | std::fs::create_dir_all(&work).unwrap(); |
| 392 | Self { root, config, work } |
| 393 | } |
| 394 | } |
| 395 | |
| 396 | impl Drop for Scratch { |
| 397 | fn drop(&mut self) { |
| 398 | std::fs::remove_dir_all(&self.root).ok(); |
| 399 | } |
| 400 | } |
| 401 | |
| 402 | /// A `run_command` call printing a marker, in `cwd`. Quote-free so the same |
| 403 | /// command line works under `sh -c` and `cmd /C`. |
| 404 | fn echo_call(id: &str, cwd: &Path, background: bool) -> String { |
| 405 | sse_tool_call( |
| 406 | id, |
| 407 | "run_command", |
| 408 | json!({"command": "echo ran-locally", "cwd": cwd, "run_in_background": background}), |
| 409 | ) |
| 410 | } |
| 411 | |
| 412 | // ── The tests ─────────────────────────────────────────────────────────────── |
| 413 | |
| 414 | #[test] |
| 415 | fn run_command_runs_in_the_client_terminal_and_shows_it() { |
| 416 | let scratch = Scratch::new("terminal"); |
| 417 | let endpoint = start_fake_endpoint(vec![ |
| 418 | echo_call("call_1", &scratch.work, false), |
| 419 | sse_text("done"), |
| 420 | ]); |
| 421 | let mut agent = spawn_agent(endpoint.port, &scratch.config, None); |
| 422 | let session_id = agent.open_session(true, &scratch.work); |
| 423 | |
| 424 | let mut editor = Editor::default(); |
| 425 | agent.prompt(&session_id, &mut editor); |
| 426 | |
| 427 | assert_eq!( |
| 428 | endpoint.tool_result("call_1"), |
| 429 | format!("Exit code 0:\n{EDITOR_OUTPUT}") |
| 430 | ); |
| 431 | assert_eq!( |
| 432 | editor.methods(), |
| 433 | [ |
| 434 | "terminal/create", |
| 435 | "terminal/wait_for_exit", |
| 436 | "terminal/output", |
| 437 | "terminal/release", |
| 438 | ] |
| 439 | ); |
| 440 | |
| 441 | let create = &editor.request("terminal/create")["params"]; |
| 442 | assert_eq!(create["sessionId"], session_id.as_str()); |
| 443 | #[cfg(unix)] |
| 444 | assert_eq!(create["command"], "sh"); |
| 445 | #[cfg(windows)] |
| 446 | assert_eq!(create["command"], "cmd"); |
| 447 | assert_eq!(create["args"][1], "echo ran-locally"); |
| 448 | let cwd = PathBuf::from(create["cwd"].as_str().unwrap()); |
| 449 | assert_eq!( |
| 450 | cwd.canonicalize().unwrap(), |
| 451 | scratch.work.canonicalize().unwrap() |
| 452 | ); |
| 453 | for method in [ |
| 454 | "terminal/wait_for_exit", |
| 455 | "terminal/output", |
| 456 | "terminal/release", |
| 457 | ] { |
| 458 | assert_eq!(editor.request(method)["params"]["terminalId"], "term-1"); |
| 459 | } |
| 460 | |
| 461 | // The terminal is in the card while the command runs and after it ends. |
| 462 | let terminal = json!([{"type": "terminal", "terminalId": "term-1"}]); |
| 463 | let updates = editor.call_updates("call_1"); |
| 464 | let shown = updates |
| 465 | .iter() |
| 466 | .position(|update| update["content"] == terminal && update["status"].is_null()) |
| 467 | .unwrap_or_else(|| panic!("terminal never embedded: {updates:?}")); |
| 468 | let finished = updates |
| 469 | .iter() |
| 470 | .position(|update| update["status"] == "completed") |
| 471 | .expect("call_1 never completed"); |
| 472 | assert!(shown < finished, "embedded after completion: {updates:?}"); |
| 473 | // The terminal was released by then, so the closing update does not name |
| 474 | // it again; the card keeps the one already embedded. |
| 475 | assert!( |
| 476 | updates[finished]["content"].is_null(), |
| 477 | "the final update must not re-send the released terminal: {}", |
| 478 | updates[finished] |
| 479 | ); |
| 480 | assert_eq!( |
| 481 | updates[finished]["rawOutput"], |
| 482 | format!("Exit code 0:\n{EDITOR_OUTPUT}") |
| 483 | ); |
| 484 | } |
| 485 | |
| 486 | #[test] |
| 487 | fn a_client_without_a_terminal_gets_the_command_run_locally() { |
| 488 | let scratch = Scratch::new("noterminal"); |
| 489 | let endpoint = start_fake_endpoint(vec![ |
| 490 | echo_call("call_1", &scratch.work, false), |
| 491 | sse_text("done"), |
| 492 | ]); |
| 493 | let mut agent = spawn_agent(endpoint.port, &scratch.config, None); |
| 494 | let session_id = agent.open_session(false, &scratch.work); |
| 495 | |
| 496 | let mut editor = Editor::default(); |
| 497 | agent.prompt(&session_id, &mut editor); |
| 498 | |
| 499 | let result = endpoint.tool_result("call_1"); |
| 500 | assert!(result.contains("ran-locally"), "run_command said: {result}"); |
| 501 | assert!(editor.requests.is_empty(), "{:?}", editor.requests); |
| 502 | } |
| 503 | |
| 504 | #[test] |
| 505 | fn a_background_command_stays_local() { |
| 506 | let scratch = Scratch::new("background"); |
| 507 | let endpoint = start_fake_endpoint(vec![ |
| 508 | echo_call("call_1", &scratch.work, true), |
| 509 | sse_text("done"), |
| 510 | ]); |
| 511 | let mut agent = spawn_agent(endpoint.port, &scratch.config, None); |
| 512 | let session_id = agent.open_session(true, &scratch.work); |
| 513 | |
| 514 | let mut editor = Editor::default(); |
| 515 | agent.prompt(&session_id, &mut editor); |
| 516 | |
| 517 | let result = endpoint.tool_result("call_1"); |
| 518 | assert!(result.contains("task"), "run_command said: {result}"); |
| 519 | assert!(editor.requests.is_empty(), "{:?}", editor.requests); |
| 520 | } |
| 521 | |
| 522 | #[test] |
| 523 | fn a_directory_outside_the_session_roots_stays_local() { |
| 524 | let scratch = Scratch::new("outside"); |
| 525 | let outside = scratch.root.join("elsewhere"); |
| 526 | std::fs::create_dir_all(&outside).unwrap(); |
| 527 | let endpoint = |
| 528 | start_fake_endpoint(vec![echo_call("call_1", &outside, false), sse_text("done")]); |
| 529 | let mut agent = spawn_agent(endpoint.port, &scratch.config, None); |
| 530 | let session_id = agent.open_session(true, &scratch.work); |
| 531 | |
| 532 | let mut editor = Editor::default(); |
| 533 | agent.prompt(&session_id, &mut editor); |
| 534 | |
| 535 | let result = endpoint.tool_result("call_1"); |
| 536 | assert!(result.contains("ran-locally"), "run_command said: {result}"); |
| 537 | assert!(editor.requests.is_empty(), "{:?}", editor.requests); |
| 538 | } |
| 539 | |
| 540 | #[test] |
| 541 | fn a_terminal_the_client_refuses_falls_back_to_running_locally() { |
| 542 | let scratch = Scratch::new("refused"); |
| 543 | let endpoint = start_fake_endpoint(vec![ |
| 544 | echo_call("call_1", &scratch.work, false), |
| 545 | sse_text("done"), |
| 546 | ]); |
| 547 | let mut agent = spawn_agent(endpoint.port, &scratch.config, None); |
| 548 | let session_id = agent.open_session(true, &scratch.work); |
| 549 | |
| 550 | let mut editor = Editor { |
| 551 | refuse_create: true, |
| 552 | ..Editor::default() |
| 553 | }; |
| 554 | agent.prompt(&session_id, &mut editor); |
| 555 | |
| 556 | let result = endpoint.tool_result("call_1"); |
| 557 | assert!(result.contains("ran-locally"), "run_command said: {result}"); |
| 558 | assert_eq!(editor.methods(), ["terminal/create"]); |
| 559 | } |
| 560 | |
| 561 | #[test] |
| 562 | fn sigit_client_terminal_off_keeps_commands_local() { |
| 563 | let scratch = Scratch::new("off"); |
| 564 | let endpoint = start_fake_endpoint(vec![ |
| 565 | echo_call("call_1", &scratch.work, false), |
| 566 | sse_text("done"), |
| 567 | ]); |
| 568 | let mut agent = spawn_agent(endpoint.port, &scratch.config, Some("off")); |
| 569 | let session_id = agent.open_session(true, &scratch.work); |
| 570 | |
| 571 | let mut editor = Editor::default(); |
| 572 | agent.prompt(&session_id, &mut editor); |
| 573 | |
| 574 | let result = endpoint.tool_result("call_1"); |
| 575 | assert!(result.contains("ran-locally"), "run_command said: {result}"); |
| 576 | assert!(editor.requests.is_empty(), "{:?}", editor.requests); |
| 577 | } |
| 578 | |
| 579 | #[test] |
| 580 | fn a_cancel_mid_command_kills_the_client_terminal_and_ends_the_turn_cancelled() { |
| 581 | let scratch = Scratch::new("cancel"); |
| 582 | let endpoint = start_fake_endpoint(vec![ |
| 583 | echo_call("call_1", &scratch.work, false), |
| 584 | sse_text("never asked for"), |
| 585 | ]); |
| 586 | let mut agent = spawn_agent(endpoint.port, &scratch.config, None); |
| 587 | let session_id = agent.open_session(true, &scratch.work); |
| 588 | |
| 589 | let mut editor = Editor::default(); |
| 590 | let response = agent.prompt_cancelled_mid_command(&session_id, &mut editor); |
| 591 | |
| 592 | assert_eq!(response["result"]["stopReason"], "cancelled", "{response}"); |
| 593 | assert_eq!( |
| 594 | editor.methods(), |
| 595 | [ |
| 596 | "terminal/create", |
| 597 | "terminal/wait_for_exit", |
| 598 | "terminal/kill", |
| 599 | "terminal/output", |
| 600 | "terminal/release", |
| 601 | ] |
| 602 | ); |
| 603 | assert_eq!( |
| 604 | editor.request("terminal/kill")["params"]["terminalId"], |
| 605 | "term-1" |
| 606 | ); |
| 607 | assert_eq!( |
| 608 | endpoint.requests.lock().unwrap().len(), |
| 609 | 1, |
| 610 | "a cancelled turn must not ask the model for another round" |
| 611 | ); |
| 612 | } |
| 613 | |
| 614 | /// `$/cancel_request` names the prompt by its JSON-RPC id instead of the |
| 615 | /// session. It has to stop the turn the same way, and the prompt is still |
| 616 | /// answered with a result, so the client keeps what was streamed. |
| 617 | #[test] |
| 618 | fn a_cancel_request_for_the_prompt_ends_the_turn_cancelled() { |
| 619 | let scratch = Scratch::new("cancel_request"); |
| 620 | let endpoint = start_fake_endpoint(vec![ |
| 621 | echo_call("call_1", &scratch.work, false), |
| 622 | sse_text("never asked for"), |
| 623 | ]); |
| 624 | let mut agent = spawn_agent(endpoint.port, &scratch.config, None); |
| 625 | let session_id = agent.open_session(true, &scratch.work); |
| 626 | |
| 627 | let mut editor = Editor::default(); |
| 628 | let response = agent.prompt_stopped_mid_command(&session_id, &mut editor, |prompt_id| { |
| 629 | json!({ |
| 630 | "jsonrpc": "2.0", |
| 631 | "method": "$/cancel_request", |
| 632 | "params": {"requestId": prompt_id}, |
| 633 | }) |
| 634 | }); |
| 635 | |
| 636 | assert_eq!(response["result"]["stopReason"], "cancelled", "{response}"); |
| 637 | assert_eq!( |
| 638 | editor.request("terminal/kill")["params"]["terminalId"], |
| 639 | "term-1" |
| 640 | ); |
| 641 | assert_eq!( |
| 642 | endpoint.requests.lock().unwrap().len(), |
| 643 | 1, |
| 644 | "a cancelled turn must not ask the model for another round" |
| 645 | ); |
| 646 | } |