[codex exec] Add item.started and support it for command execution (#4250)
Adds a new `item.started` event to `codex exec` and implements it for
command_execution item type.
```jsonl
{"type":"session.created","session_id":"019982d1-75f0-7920-b051-e0d3731a5ed8"}
{"type":"item.completed","item":{"id":"item_0","item_type":"reasoning","text":"**Executing commands securely**\n\nI'm thinking about how the default harness typically uses \"bash -lc,\" while historically \"bash\" is what we've been using. The command should be executed as a string in our CLI, so using \"bash -lc 'echo hello'\" is optimal but calling \"echo hello\" directly feels safer. The sandbox makes sure environment variables like CODEX_SANDBOX_NETWORK_DISABLED=1 are set, so I won't ask for approval. I just need to run \"echo hello\" and correctly present the output."}}
{"type":"item.completed","item":{"id":"item_1","item_type":"reasoning","text":"**Preparing for tool calls**\n\nI realize that I need to include a preamble before making any tool calls. So, I'll first state the preamble in the commentary channel, then proceed with the tool call. After that, I need to present the final message along with the output. It's possible that the CLI will show the output inline, but I must ensure that I present the result clearly regardless. Let's move forward and get this organized!"}}
{"type":"item.completed","item":{"id":"item_2","item_type":"assistant_message","text":"Running `echo` to confirm shell access and print output."}}
{"type":"item.started","item":{"id":"item_3","item_type":"command_execution","command":"bash -lc echo hello","aggregated_output":"","exit_code":null,"status":"in_progress"}}
{"type":"item.completed","item":{"id":"item_3","item_type":"command_execution","command":"bash -lc echo hello","aggregated_output":"hello\n","exit_code":0,"status":"completed"}}
{"type":"item.completed","item":{"id":"item_4","item_type":"assistant_message","text":"hello"}}
```
This commit is contained in:
@@ -16,6 +16,7 @@ use codex_exec::exec_events::ConversationEvent;
|
||||
use codex_exec::exec_events::ConversationItem;
|
||||
use codex_exec::exec_events::ConversationItemDetails;
|
||||
use codex_exec::exec_events::ItemCompletedEvent;
|
||||
use codex_exec::exec_events::ItemStartedEvent;
|
||||
use codex_exec::exec_events::PatchApplyStatus;
|
||||
use codex_exec::exec_events::PatchChangeKind;
|
||||
use codex_exec::exec_events::ReasoningItem;
|
||||
@@ -156,7 +157,20 @@ fn exec_command_end_success_produces_completed_command_item() {
|
||||
}),
|
||||
);
|
||||
let out_begin = ep.collect_conversation_events(&begin);
|
||||
assert!(out_begin.is_empty());
|
||||
assert_eq!(
|
||||
out_begin,
|
||||
vec![ConversationEvent::ItemStarted(ItemStartedEvent {
|
||||
item: ConversationItem {
|
||||
id: "item_0".to_string(),
|
||||
details: ConversationItemDetails::CommandExecution(CommandExecutionItem {
|
||||
command: "bash -lc 'echo hi'".to_string(),
|
||||
aggregated_output: String::new(),
|
||||
exit_code: None,
|
||||
status: CommandExecutionStatus::InProgress,
|
||||
}),
|
||||
},
|
||||
})]
|
||||
);
|
||||
|
||||
// End (success) -> item.completed (item_0)
|
||||
let end_ok = event(
|
||||
@@ -178,9 +192,9 @@ fn exec_command_end_success_produces_completed_command_item() {
|
||||
item: ConversationItem {
|
||||
id: "item_0".to_string(),
|
||||
details: ConversationItemDetails::CommandExecution(CommandExecutionItem {
|
||||
command: "bash -lc echo hi".to_string(),
|
||||
command: "bash -lc 'echo hi'".to_string(),
|
||||
aggregated_output: "hi\n".to_string(),
|
||||
exit_code: 0,
|
||||
exit_code: Some(0),
|
||||
status: CommandExecutionStatus::Completed,
|
||||
}),
|
||||
},
|
||||
@@ -202,7 +216,20 @@ fn exec_command_end_failure_produces_failed_command_item() {
|
||||
parsed_cmd: Vec::new(),
|
||||
}),
|
||||
);
|
||||
assert!(ep.collect_conversation_events(&begin).is_empty());
|
||||
assert_eq!(
|
||||
ep.collect_conversation_events(&begin),
|
||||
vec![ConversationEvent::ItemStarted(ItemStartedEvent {
|
||||
item: ConversationItem {
|
||||
id: "item_0".to_string(),
|
||||
details: ConversationItemDetails::CommandExecution(CommandExecutionItem {
|
||||
command: "sh -c 'exit 1'".to_string(),
|
||||
aggregated_output: String::new(),
|
||||
exit_code: None,
|
||||
status: CommandExecutionStatus::InProgress,
|
||||
}),
|
||||
},
|
||||
})]
|
||||
);
|
||||
|
||||
// End (failure) -> item.completed (item_0)
|
||||
let end_fail = event(
|
||||
@@ -224,9 +251,9 @@ fn exec_command_end_failure_produces_failed_command_item() {
|
||||
item: ConversationItem {
|
||||
id: "item_0".to_string(),
|
||||
details: ConversationItemDetails::CommandExecution(CommandExecutionItem {
|
||||
command: "sh -c exit 1".to_string(),
|
||||
command: "sh -c 'exit 1'".to_string(),
|
||||
aggregated_output: String::new(),
|
||||
exit_code: 1,
|
||||
exit_code: Some(1),
|
||||
status: CommandExecutionStatus::Failed,
|
||||
}),
|
||||
},
|
||||
@@ -234,6 +261,27 @@ fn exec_command_end_failure_produces_failed_command_item() {
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn exec_command_end_without_begin_is_ignored() {
|
||||
let mut ep = ExperimentalEventProcessorWithJsonOutput::new(None);
|
||||
|
||||
// End event arrives without a prior Begin; should produce no conversation events.
|
||||
let end_only = event(
|
||||
"c1",
|
||||
EventMsg::ExecCommandEnd(ExecCommandEndEvent {
|
||||
call_id: "no-begin".to_string(),
|
||||
stdout: String::new(),
|
||||
stderr: String::new(),
|
||||
aggregated_output: String::new(),
|
||||
exit_code: 0,
|
||||
duration: Duration::from_millis(1),
|
||||
formatted_output: String::new(),
|
||||
}),
|
||||
);
|
||||
let out = ep.collect_conversation_events(&end_only);
|
||||
assert!(out.is_empty());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn patch_apply_success_produces_item_completed_patchapply() {
|
||||
let mut ep = ExperimentalEventProcessorWithJsonOutput::new(None);
|
||||
|
||||
Reference in New Issue
Block a user