Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commitfc7700a

Browse files
authored
fix: improve reliability of app statuses (#18622)
We were discarding all "working" updates from the screen watcher becausewe cannot tell the difference between the agent or user changing thescreen, but it makes sense to accept it as the very first update,because the agent could be working but neglected to report that fact, soyou would never get an initial "working" update (it would justeventually go straight to "idle").Also messages can start at zero, so I made a fix for that as well,although the first message will be from the LLM and we ignorethose anyway, so this probably has no actual effect, but seems moretechnically correct.And it seems I forgot to actually update the last message ID, which also does not actually matter for user messages (since I think theSSE endpoint will not re-emit a user message it has already emitted),but seems more technically correct to check.Lastly, if we have the screen watcher, ignore the agent's self-reportedstate and always use "working" since it is unreliable. The idle state willeventually be caught by the watcher.
1 parentad67733 commitfc7700a

File tree

2 files changed

+390
-225
lines changed

2 files changed

+390
-225
lines changed

‎cli/exp_mcp.go‎

Lines changed: 60 additions & 26 deletions
Original file line numberDiff line numberDiff line change
@@ -362,11 +362,19 @@ func (*RootCmd) mcpConfigureCursor() *serpent.Command {
362362
}
363363

364364
typetaskReportstruct {
365-
linkstring
366-
messageIDint64
365+
// link is optional.
366+
linkstring
367+
// messageID must be set if this update is from a *user* message. A user
368+
// message only happens when interacting via the AI AgentAPI (as opposed to
369+
// interacting with the terminal directly).
370+
messageID*int64
371+
// selfReported must be set if the update is directly from the AI agent
372+
// (as opposed to the screen watcher).
367373
selfReportedbool
368-
state codersdk.WorkspaceAppStatusState
369-
summarystring
374+
// state must always be set.
375+
state codersdk.WorkspaceAppStatusState
376+
// summary is optional.
377+
summarystring
370378
}
371379

372380
typemcpServerstruct {
@@ -388,31 +396,48 @@ func (r *RootCmd) mcpServer() *serpent.Command {
388396
return&serpent.Command{
389397
Use:"server",
390398
Handler:func(inv*serpent.Invocation)error {
391-
// lastUserMessageID is the ID of the last *user* message that we saw. A
392-
// user message only happens when interacting via the AI AgentAPI (as
393-
// opposed to interacting with the terminal directly).
394-
varlastUserMessageIDint64
395399
varlastReporttaskReport
396400
// Create a queue that skips duplicates and preserves summaries.
397401
queue:= cliutil.NewQueue[taskReport](512).WithPredicate(func(reporttaskReport) (taskReport,bool) {
398-
// Use "working" status if this is a new user message. If this is not a
399-
// new user message, and the status is "working" and not self-reported
400-
// (meaning it came from the screen watcher), then it means one of two
401-
// things:
402-
// 1. The AI agent is still working, so there is nothing to update.
403-
// 2. The AI agent stopped working, then the user has interacted with
404-
// the terminal directly. For now, we are ignoring these updates.
405-
// This risks missing cases where the user manually submits a new
406-
// prompt and the AI agent becomes active and does not update itself,
407-
// but it avoids spamming useless status updates as the user is
408-
// typing, so the tradeoff is worth it. In the future, if we can
409-
// reliably distinguish between user and AI agent activity, we can
410-
// change this.
411-
ifreport.messageID>lastUserMessageID {
412-
report.state=codersdk.WorkspaceAppStatusStateWorking
413-
}elseifreport.state==codersdk.WorkspaceAppStatusStateWorking&&!report.selfReported {
402+
// Avoid queuing empty statuses (this would probably indicate a
403+
// developer error)
404+
ifreport.state=="" {
414405
returnreport,false
415406
}
407+
// If this is a user message, discard if it is not new.
408+
ifreport.messageID!=nil&&lastReport.messageID!=nil&&
409+
*lastReport.messageID>=*report.messageID {
410+
returnreport,false
411+
}
412+
// If this is not a user message, and the status is "working" and not
413+
// self-reported (meaning it came from the screen watcher), then it
414+
// means one of two things:
415+
//
416+
// 1. The AI agent is not working; the user is interacting with the
417+
// terminal directly.
418+
// 2. The AI agent is working.
419+
//
420+
// At the moment, we have no way to tell the difference between these
421+
// two states. In the future, if we can reliably distinguish between
422+
// user and AI agent activity, we can change this.
423+
//
424+
// If this is our first update, we assume it is the AI agent working and
425+
// accept the update.
426+
//
427+
// Otherwise we discard the update. This risks missing cases where the
428+
// user manually submits a new prompt and the AI agent becomes active
429+
// (and does not update itself), but it avoids spamming useless status
430+
// updates as the user is typing, so the tradeoff is worth it.
431+
ifreport.messageID==nil&&
432+
report.state==codersdk.WorkspaceAppStatusStateWorking&&
433+
!report.selfReported&&lastReport.state!="" {
434+
returnreport,false
435+
}
436+
// Keep track of the last message ID so we can tell when a message is
437+
// new or if it has been re-emitted.
438+
ifreport.messageID==nil {
439+
report.messageID=lastReport.messageID
440+
}
416441
// Preserve previous message and URI if there was no message.
417442
ifreport.summary=="" {
418443
report.summary=lastReport.summary
@@ -600,7 +625,8 @@ func (s *mcpServer) startWatcher(ctx context.Context, inv *serpent.Invocation) {
600625
case agentapi.EventMessageUpdate:
601626
ifev.Role==agentapi.RoleUser {
602627
err:=s.queue.Push(taskReport{
603-
messageID:ev.Id,
628+
messageID:&ev.Id,
629+
state:codersdk.WorkspaceAppStatusStateWorking,
604630
})
605631
iferr!=nil {
606632
cliui.Warnf(inv.Stderr,"Failed to queue update: %s",err)
@@ -650,10 +676,18 @@ func (s *mcpServer) startServer(ctx context.Context, inv *serpent.Invocation, in
650676
// Add tool dependencies.
651677
toolOpts:= []func(*toolsdk.Deps){
652678
toolsdk.WithTaskReporter(func(args toolsdk.ReportTaskArgs)error {
679+
// The agent does not reliably report its status correctly. If AgentAPI
680+
// is enabled, we will always set the status to "working" when we get an
681+
// MCP message, and rely on the screen watcher to eventually catch the
682+
// idle state.
683+
state:=codersdk.WorkspaceAppStatusStateWorking
684+
ifs.aiAgentAPIClient==nil {
685+
state=codersdk.WorkspaceAppStatusState(args.State)
686+
}
653687
returns.queue.Push(taskReport{
654688
link:args.Link,
655689
selfReported:true,
656-
state:codersdk.WorkspaceAppStatusState(args.State),
690+
state:state,
657691
summary:args.Summary,
658692
})
659693
}),

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp