@evident-ai/cli 3.1.1-dev.6a74c3d → 3.1.1-dev.6a8f72d
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +45 -0
- package/dist/index.js +1749 -406
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
package/dist/index.js
CHANGED
|
@@ -38,11 +38,6 @@ function getApiUrl() {
|
|
|
38
38
|
function getTunnelUrl() {
|
|
39
39
|
return tunnelOverride ?? process.env.EVIDENT_TUNNEL_URL ?? defaults.tunnelUrl;
|
|
40
40
|
}
|
|
41
|
-
var config = new Conf({
|
|
42
|
-
projectName: "evident",
|
|
43
|
-
projectSuffix: "",
|
|
44
|
-
defaults
|
|
45
|
-
});
|
|
46
41
|
var credentials = new Conf({
|
|
47
42
|
projectName: "evident",
|
|
48
43
|
projectSuffix: "",
|
|
@@ -267,16 +262,28 @@ async function getToken() {
|
|
|
267
262
|
}
|
|
268
263
|
return null;
|
|
269
264
|
}
|
|
265
|
+
function toError(err) {
|
|
266
|
+
return err instanceof Error ? err : new Error(String(err));
|
|
267
|
+
}
|
|
270
268
|
async function deleteToken(options = {}) {
|
|
271
269
|
const keytar = await getKeytar();
|
|
270
|
+
const failures = [];
|
|
272
271
|
if (keytar) {
|
|
273
272
|
if (options.all) {
|
|
274
|
-
|
|
273
|
+
let accounts = [];
|
|
274
|
+
try {
|
|
275
|
+
accounts = await keytar.findCredentials(SERVICE_NAME);
|
|
276
|
+
} catch (err) {
|
|
277
|
+
failures.push({ type: "enumerate", error: toError(err) });
|
|
278
|
+
}
|
|
275
279
|
await Promise.all(
|
|
276
|
-
|
|
277
|
-
|
|
278
|
-
|
|
279
|
-
|
|
280
|
+
accounts.map(async (entry) => {
|
|
281
|
+
try {
|
|
282
|
+
await keytar.deletePassword(SERVICE_NAME, entry.account);
|
|
283
|
+
} catch (err) {
|
|
284
|
+
failures.push({ type: "delete", account: entry.account, error: toError(err) });
|
|
285
|
+
}
|
|
286
|
+
})
|
|
280
287
|
);
|
|
281
288
|
} else {
|
|
282
289
|
await keytar.deletePassword(SERVICE_NAME, keychainAccount());
|
|
@@ -287,6 +294,7 @@ async function deleteToken(options = {}) {
|
|
|
287
294
|
} else {
|
|
288
295
|
clearCredentials();
|
|
289
296
|
}
|
|
297
|
+
return { failures };
|
|
290
298
|
}
|
|
291
299
|
|
|
292
300
|
// src/utils/ui.ts
|
|
@@ -404,8 +412,10 @@ async function deviceFlowLogin(options) {
|
|
|
404
412
|
}
|
|
405
413
|
async function tokenLogin() {
|
|
406
414
|
console.log("Token login mode.");
|
|
407
|
-
console.log("
|
|
408
|
-
console.log(
|
|
415
|
+
console.log("Create a token under Settings \u2192 CLI tokens in the dashboard, then paste it below.");
|
|
416
|
+
console.log(
|
|
417
|
+
"(Alternatively, run `evident login` on a machine with a browser, or set EVIDENT_TOKEN for CI.)"
|
|
418
|
+
);
|
|
409
419
|
blank();
|
|
410
420
|
process.stdout.write("Paste token: ");
|
|
411
421
|
const token = await new Promise((resolve3) => {
|
|
@@ -429,13 +439,22 @@ async function tokenLogin() {
|
|
|
429
439
|
printError("No token provided.");
|
|
430
440
|
process.exit(1);
|
|
431
441
|
}
|
|
442
|
+
await validateAndStoreToken(token);
|
|
443
|
+
}
|
|
444
|
+
async function validateAndStoreToken(token) {
|
|
432
445
|
const spinner = ora("Validating token...").start();
|
|
433
446
|
try {
|
|
434
|
-
const result = await api.
|
|
447
|
+
const result = await api.get("/me", {
|
|
448
|
+
headers: { Authorization: `Bearer ${token}` }
|
|
449
|
+
});
|
|
450
|
+
if (!result.user) {
|
|
451
|
+
throw new Error(
|
|
452
|
+
"This token is not a user login (e.g. a runner key). Paste a CLI token instead."
|
|
453
|
+
);
|
|
454
|
+
}
|
|
435
455
|
await storeToken({
|
|
436
456
|
token,
|
|
437
|
-
user: result.user
|
|
438
|
-
expiresAt: result.expires_at
|
|
457
|
+
user: { email: result.user.email }
|
|
439
458
|
});
|
|
440
459
|
spinner.stop();
|
|
441
460
|
printSuccess(`Logged in as ${chalk2.bold(result.user.email)}`);
|
|
@@ -455,9 +474,22 @@ async function login(options) {
|
|
|
455
474
|
}
|
|
456
475
|
|
|
457
476
|
// src/commands/logout.ts
|
|
477
|
+
function describeFailure(failure) {
|
|
478
|
+
if (failure.type === "enumerate") {
|
|
479
|
+
return `could not list stored keychain entries (${failure.error.message})`;
|
|
480
|
+
}
|
|
481
|
+
return `${failure.account} (${failure.error.message})`;
|
|
482
|
+
}
|
|
458
483
|
async function logout(options = {}) {
|
|
459
484
|
if (options.all) {
|
|
460
|
-
await deleteToken({ all: true });
|
|
485
|
+
const result = await deleteToken({ all: true });
|
|
486
|
+
if (result.failures.length > 0) {
|
|
487
|
+
printError(
|
|
488
|
+
`Failed to fully clear your keychain: ${result.failures.map(describeFailure).join("; ")}. Your local credentials file was cleared, but stale keychain entries may remain \u2014 run \`evident logout --all\` again, or remove them manually from your OS keychain / credential manager.`
|
|
489
|
+
);
|
|
490
|
+
process.exitCode = 1;
|
|
491
|
+
return;
|
|
492
|
+
}
|
|
461
493
|
printSuccess("Logged out of all endpoints.");
|
|
462
494
|
return;
|
|
463
495
|
}
|
|
@@ -482,7 +514,9 @@ async function whoami() {
|
|
|
482
514
|
blank();
|
|
483
515
|
console.log(keyValue("Endpoint", apiUrl));
|
|
484
516
|
console.log(keyValue("User", chalk3.bold(credentials2.user.email)));
|
|
485
|
-
|
|
517
|
+
if (credentials2.user.id) {
|
|
518
|
+
console.log(keyValue("User ID", credentials2.user.id));
|
|
519
|
+
}
|
|
486
520
|
if (credentials2.expiresAt) {
|
|
487
521
|
const expiresAt = new Date(credentials2.expiresAt);
|
|
488
522
|
const now = /* @__PURE__ */ new Date();
|
|
@@ -495,14 +529,477 @@ async function whoami() {
|
|
|
495
529
|
console.log(keyValue("Expires", `${daysRemaining} days`));
|
|
496
530
|
}
|
|
497
531
|
}
|
|
498
|
-
blank();
|
|
532
|
+
blank();
|
|
533
|
+
}
|
|
534
|
+
|
|
535
|
+
// src/lib/auth.ts
|
|
536
|
+
async function getAuthCredentials() {
|
|
537
|
+
const runnerKey = process.env.EVIDENT_RUNNER_KEY;
|
|
538
|
+
const agentKey = process.env.EVIDENT_AGENT_KEY;
|
|
539
|
+
if (runnerKey) {
|
|
540
|
+
return {
|
|
541
|
+
token: runnerKey,
|
|
542
|
+
authType: "agent_key",
|
|
543
|
+
keySource: "runner_key",
|
|
544
|
+
notice: agentKey ? "Both EVIDENT_RUNNER_KEY and EVIDENT_AGENT_KEY are set; using EVIDENT_RUNNER_KEY." : void 0
|
|
545
|
+
};
|
|
546
|
+
}
|
|
547
|
+
if (agentKey) {
|
|
548
|
+
return { token: agentKey, authType: "agent_key", keySource: "agent_key" };
|
|
549
|
+
}
|
|
550
|
+
const userToken = process.env.EVIDENT_TOKEN;
|
|
551
|
+
if (userToken) {
|
|
552
|
+
return { token: userToken, authType: "bearer" };
|
|
553
|
+
}
|
|
554
|
+
const keychainCreds = await getToken();
|
|
555
|
+
if (keychainCreds) {
|
|
556
|
+
return {
|
|
557
|
+
token: keychainCreds.token,
|
|
558
|
+
authType: "bearer",
|
|
559
|
+
user: keychainCreds.user
|
|
560
|
+
};
|
|
561
|
+
}
|
|
562
|
+
return null;
|
|
563
|
+
}
|
|
564
|
+
function getAuthHeader(credentials2) {
|
|
565
|
+
if (credentials2.authType === "agent_key") {
|
|
566
|
+
return `SandboxKey ${credentials2.token}`;
|
|
567
|
+
}
|
|
568
|
+
return `Bearer ${credentials2.token}`;
|
|
569
|
+
}
|
|
570
|
+
function isInteractive(jsonOutput) {
|
|
571
|
+
if (jsonOutput) return false;
|
|
572
|
+
if (process.env.CI) return false;
|
|
573
|
+
if (process.env.GITHUB_ACTIONS) return false;
|
|
574
|
+
if (!process.stdin.isTTY) return false;
|
|
575
|
+
return true;
|
|
576
|
+
}
|
|
577
|
+
|
|
578
|
+
// src/commands/agent-lookup.ts
|
|
579
|
+
async function readErrorMessage(response) {
|
|
580
|
+
const text = await response.text().catch(() => "");
|
|
581
|
+
if (!text) return response.statusText || void 0;
|
|
582
|
+
try {
|
|
583
|
+
const data = JSON.parse(text);
|
|
584
|
+
const message = data.message ?? data.error;
|
|
585
|
+
if (typeof message === "string" && message.trim()) {
|
|
586
|
+
return message;
|
|
587
|
+
}
|
|
588
|
+
} catch {
|
|
589
|
+
}
|
|
590
|
+
return text.trim() || response.statusText || void 0;
|
|
591
|
+
}
|
|
592
|
+
function authFailureHint(apiUrl, serverMessage) {
|
|
593
|
+
const reason = serverMessage ? `: ${serverMessage}` : "";
|
|
594
|
+
return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
|
|
595
|
+
}
|
|
596
|
+
async function resolveAgentIdFromKey(authHeader) {
|
|
597
|
+
const apiUrl = getApiUrlConfig();
|
|
598
|
+
try {
|
|
599
|
+
const response = await fetch(`${apiUrl}/me`, {
|
|
600
|
+
headers: { Authorization: authHeader }
|
|
601
|
+
});
|
|
602
|
+
if (response.status === 401) {
|
|
603
|
+
const serverMessage = await readErrorMessage(response);
|
|
604
|
+
return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
605
|
+
}
|
|
606
|
+
if (!response.ok) {
|
|
607
|
+
const serverMessage = await readErrorMessage(response);
|
|
608
|
+
return {
|
|
609
|
+
error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
610
|
+
};
|
|
611
|
+
}
|
|
612
|
+
const data = await response.json();
|
|
613
|
+
if (data.auth_type === "agent_key" && data.agent_id) {
|
|
614
|
+
return { agent_id: data.agent_id };
|
|
615
|
+
}
|
|
616
|
+
return {
|
|
617
|
+
error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --runner explicitly."
|
|
618
|
+
};
|
|
619
|
+
} catch (error2) {
|
|
620
|
+
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
621
|
+
return { error: `Failed to resolve runner from key: ${message}` };
|
|
622
|
+
}
|
|
623
|
+
}
|
|
624
|
+
var BEST_EFFORT_NOTIFY_TIMEOUT_MS = 2e3;
|
|
625
|
+
async function notifyAgentDisconnected(agentId, authHeader) {
|
|
626
|
+
const apiUrl = getApiUrlConfig();
|
|
627
|
+
try {
|
|
628
|
+
const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
|
|
629
|
+
method: "POST",
|
|
630
|
+
headers: { Authorization: authHeader },
|
|
631
|
+
signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
|
|
632
|
+
});
|
|
633
|
+
if (!response.ok) {
|
|
634
|
+
const serverMessage = await readErrorMessage(response);
|
|
635
|
+
return {
|
|
636
|
+
ok: false,
|
|
637
|
+
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
638
|
+
};
|
|
639
|
+
}
|
|
640
|
+
return { ok: true };
|
|
641
|
+
} catch (error2) {
|
|
642
|
+
return { ok: false, error: describeBestEffortError(error2) };
|
|
643
|
+
}
|
|
644
|
+
}
|
|
645
|
+
function describeBestEffortError(error2) {
|
|
646
|
+
const name = error2?.name;
|
|
647
|
+
if (name === "TimeoutError" || name === "AbortError") {
|
|
648
|
+
return `timed out after ${BEST_EFFORT_NOTIFY_TIMEOUT_MS}ms`;
|
|
649
|
+
}
|
|
650
|
+
return error2 instanceof Error ? error2.message : String(error2);
|
|
651
|
+
}
|
|
652
|
+
async function reportMicrovmId(agentId, authHeader, microvmId) {
|
|
653
|
+
try {
|
|
654
|
+
const apiUrl = getApiUrlConfig();
|
|
655
|
+
const response = await fetch(`${apiUrl}/runners/${agentId}/microvm`, {
|
|
656
|
+
method: "POST",
|
|
657
|
+
headers: { Authorization: authHeader, "Content-Type": "application/json" },
|
|
658
|
+
body: JSON.stringify({ microvm_id: microvmId }),
|
|
659
|
+
signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
|
|
660
|
+
});
|
|
661
|
+
if (!response.ok) {
|
|
662
|
+
const serverMessage = await readErrorMessage(response);
|
|
663
|
+
return {
|
|
664
|
+
ok: false,
|
|
665
|
+
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
666
|
+
};
|
|
667
|
+
}
|
|
668
|
+
return { ok: true };
|
|
669
|
+
} catch (error2) {
|
|
670
|
+
return { ok: false, error: describeBestEffortError(error2) };
|
|
671
|
+
}
|
|
672
|
+
}
|
|
673
|
+
function toReportedWindow(window) {
|
|
674
|
+
if (!window) return null;
|
|
675
|
+
return { utilization: window.utilization, resets_at: window.resetsAt };
|
|
676
|
+
}
|
|
677
|
+
async function reportClaudeUsage(agentId, authHeader, snapshot) {
|
|
678
|
+
try {
|
|
679
|
+
const apiUrl = getApiUrlConfig();
|
|
680
|
+
const response = await fetch(`${apiUrl}/runners/${agentId}/claude-usage`, {
|
|
681
|
+
method: "POST",
|
|
682
|
+
headers: { Authorization: authHeader, "Content-Type": "application/json" },
|
|
683
|
+
body: JSON.stringify({
|
|
684
|
+
five_hour: toReportedWindow(snapshot.fiveHour),
|
|
685
|
+
seven_day: toReportedWindow(snapshot.sevenDay)
|
|
686
|
+
}),
|
|
687
|
+
signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
|
|
688
|
+
});
|
|
689
|
+
if (!response.ok) {
|
|
690
|
+
const serverMessage = await readErrorMessage(response);
|
|
691
|
+
return {
|
|
692
|
+
ok: false,
|
|
693
|
+
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
694
|
+
};
|
|
695
|
+
}
|
|
696
|
+
return { ok: true };
|
|
697
|
+
} catch (error2) {
|
|
698
|
+
return { ok: false, error: describeBestEffortError(error2) };
|
|
699
|
+
}
|
|
700
|
+
}
|
|
701
|
+
async function getAgentInfo(agentId, authHeader) {
|
|
702
|
+
const apiUrl = getApiUrlConfig();
|
|
703
|
+
try {
|
|
704
|
+
const response = await fetch(`${apiUrl}/runners/${agentId}`, {
|
|
705
|
+
headers: { Authorization: authHeader }
|
|
706
|
+
});
|
|
707
|
+
if (response.status === 401) {
|
|
708
|
+
const serverMessage = await readErrorMessage(response);
|
|
709
|
+
return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
710
|
+
}
|
|
711
|
+
if (response.status === 403) {
|
|
712
|
+
const serverMessage = await readErrorMessage(response);
|
|
713
|
+
return {
|
|
714
|
+
valid: false,
|
|
715
|
+
error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
|
|
716
|
+
};
|
|
717
|
+
}
|
|
718
|
+
if (response.status === 404) {
|
|
719
|
+
const serverMessage = await readErrorMessage(response);
|
|
720
|
+
return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
|
|
721
|
+
}
|
|
722
|
+
if (!response.ok) {
|
|
723
|
+
const serverMessage = await readErrorMessage(response);
|
|
724
|
+
return {
|
|
725
|
+
valid: false,
|
|
726
|
+
error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
727
|
+
};
|
|
728
|
+
}
|
|
729
|
+
const agent = await response.json();
|
|
730
|
+
if (agent.agent_type !== "local") {
|
|
731
|
+
return {
|
|
732
|
+
valid: false,
|
|
733
|
+
error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
|
|
734
|
+
};
|
|
735
|
+
}
|
|
736
|
+
return { valid: true, agent };
|
|
737
|
+
} catch (error2) {
|
|
738
|
+
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
739
|
+
return { valid: false, error: `Failed to validate runner: ${message}` };
|
|
740
|
+
}
|
|
741
|
+
}
|
|
742
|
+
|
|
743
|
+
// src/commands/status.ts
|
|
744
|
+
var STATUS_TIMEOUT_MS = 1e4;
|
|
745
|
+
function authLabelFor(credentials2) {
|
|
746
|
+
if (credentials2.authType === "agent_key") {
|
|
747
|
+
return credentials2.keySource === "agent_key" ? "runner key (EVIDENT_AGENT_KEY)" : "runner key (EVIDENT_RUNNER_KEY)";
|
|
748
|
+
}
|
|
749
|
+
return "user token";
|
|
750
|
+
}
|
|
751
|
+
function describeFetchError(error2) {
|
|
752
|
+
const name = error2?.name;
|
|
753
|
+
if (name === "TimeoutError" || name === "AbortError") {
|
|
754
|
+
return `timed out after ${STATUS_TIMEOUT_MS}ms waiting for a response`;
|
|
755
|
+
}
|
|
756
|
+
return error2 instanceof Error ? error2.message : String(error2);
|
|
757
|
+
}
|
|
758
|
+
async function checkStatus(jsonMode) {
|
|
759
|
+
const apiUrl = getApiUrlConfig();
|
|
760
|
+
const credentials2 = await getAuthCredentials();
|
|
761
|
+
if (!credentials2) {
|
|
762
|
+
return {
|
|
763
|
+
ok: false,
|
|
764
|
+
endpoint: apiUrl,
|
|
765
|
+
reason: "no_credentials",
|
|
766
|
+
error: "No credentials configured. Set EVIDENT_RUNNER_KEY (or EVIDENT_AGENT_KEY), or EVIDENT_TOKEN, or run `evident login`.",
|
|
767
|
+
exitCode: 1
|
|
768
|
+
};
|
|
769
|
+
}
|
|
770
|
+
if (credentials2.notice && !jsonMode) {
|
|
771
|
+
printWarning(credentials2.notice);
|
|
772
|
+
}
|
|
773
|
+
let response;
|
|
774
|
+
try {
|
|
775
|
+
response = await fetch(`${apiUrl}/me`, {
|
|
776
|
+
headers: { Authorization: getAuthHeader(credentials2) },
|
|
777
|
+
signal: AbortSignal.timeout(STATUS_TIMEOUT_MS)
|
|
778
|
+
});
|
|
779
|
+
} catch (error2) {
|
|
780
|
+
return {
|
|
781
|
+
ok: false,
|
|
782
|
+
endpoint: apiUrl,
|
|
783
|
+
authLabel: authLabelFor(credentials2),
|
|
784
|
+
reason: "unreachable",
|
|
785
|
+
error: `Could not reach ${apiUrl}: ${describeFetchError(error2)}. The credentials were NOT validated.`,
|
|
786
|
+
exitCode: 75
|
|
787
|
+
};
|
|
788
|
+
}
|
|
789
|
+
if (response.status === 401) {
|
|
790
|
+
const serverMessage = await readErrorMessage(response);
|
|
791
|
+
return {
|
|
792
|
+
ok: false,
|
|
793
|
+
endpoint: apiUrl,
|
|
794
|
+
authLabel: authLabelFor(credentials2),
|
|
795
|
+
reason: "unauthorized",
|
|
796
|
+
error: authFailureHint(apiUrl, serverMessage),
|
|
797
|
+
exitCode: 1
|
|
798
|
+
};
|
|
799
|
+
}
|
|
800
|
+
if (response.status >= 500) {
|
|
801
|
+
const serverMessage = await readErrorMessage(response);
|
|
802
|
+
return {
|
|
803
|
+
ok: false,
|
|
804
|
+
endpoint: apiUrl,
|
|
805
|
+
authLabel: authLabelFor(credentials2),
|
|
806
|
+
reason: "unreachable",
|
|
807
|
+
error: `${apiUrl} returned HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}. The credentials were NOT validated.`,
|
|
808
|
+
exitCode: 75
|
|
809
|
+
};
|
|
810
|
+
}
|
|
811
|
+
if (!response.ok) {
|
|
812
|
+
const serverMessage = await readErrorMessage(response);
|
|
813
|
+
return {
|
|
814
|
+
ok: false,
|
|
815
|
+
endpoint: apiUrl,
|
|
816
|
+
authLabel: authLabelFor(credentials2),
|
|
817
|
+
reason: "http_error",
|
|
818
|
+
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`,
|
|
819
|
+
exitCode: 1
|
|
820
|
+
};
|
|
821
|
+
}
|
|
822
|
+
const data = await response.json();
|
|
823
|
+
return {
|
|
824
|
+
ok: true,
|
|
825
|
+
endpoint: apiUrl,
|
|
826
|
+
authType: data.auth_type,
|
|
827
|
+
authLabel: authLabelFor(credentials2),
|
|
828
|
+
runnerId: data.auth_type === "agent_key" ? data.agent_id : void 0,
|
|
829
|
+
reason: "ok",
|
|
830
|
+
exitCode: 0
|
|
831
|
+
};
|
|
832
|
+
}
|
|
833
|
+
function printJson(result) {
|
|
834
|
+
const payload = {
|
|
835
|
+
ok: result.ok,
|
|
836
|
+
endpoint: result.endpoint
|
|
837
|
+
};
|
|
838
|
+
if (result.authType) payload.auth_type = result.authType;
|
|
839
|
+
if (result.runnerId) payload.runner_id = result.runnerId;
|
|
840
|
+
if (result.reason) payload.reason = result.reason;
|
|
841
|
+
if (result.error) payload.error = result.error;
|
|
842
|
+
console.log(JSON.stringify(payload));
|
|
843
|
+
}
|
|
844
|
+
function printHuman(result) {
|
|
845
|
+
blank();
|
|
846
|
+
console.log(keyValue("Endpoint", result.endpoint));
|
|
847
|
+
if (result.ok) {
|
|
848
|
+
console.log(keyValue("Auth", result.authLabel ?? "\u2014"));
|
|
849
|
+
if (result.runnerId) {
|
|
850
|
+
console.log(keyValue("Runner", result.runnerId));
|
|
851
|
+
}
|
|
852
|
+
console.log(keyValue("Status", "OK \u2014 credentials accepted"));
|
|
853
|
+
blank();
|
|
854
|
+
return;
|
|
855
|
+
}
|
|
856
|
+
if (result.authLabel) {
|
|
857
|
+
console.log(keyValue("Auth", result.authLabel));
|
|
858
|
+
}
|
|
859
|
+
blank();
|
|
860
|
+
printError(result.error ?? "Unknown error");
|
|
861
|
+
}
|
|
862
|
+
async function status(options = {}) {
|
|
863
|
+
const result = await checkStatus(Boolean(options.json));
|
|
864
|
+
if (options.json) {
|
|
865
|
+
printJson(result);
|
|
866
|
+
} else {
|
|
867
|
+
printHuman(result);
|
|
868
|
+
}
|
|
869
|
+
process.exit(result.exitCode);
|
|
870
|
+
}
|
|
871
|
+
|
|
872
|
+
// src/lib/claude-usage.ts
|
|
873
|
+
import { execFileSync } from "child_process";
|
|
874
|
+
import { readFileSync } from "fs";
|
|
875
|
+
import { homedir } from "os";
|
|
876
|
+
import { join } from "path";
|
|
877
|
+
var CLAUDE_USAGE_URL = "https://api.anthropic.com/api/oauth/usage";
|
|
878
|
+
var KEYCHAIN_SERVICE = "Claude Code-credentials";
|
|
879
|
+
function parseClaudeCliCredentials(raw) {
|
|
880
|
+
let parsed;
|
|
881
|
+
try {
|
|
882
|
+
parsed = JSON.parse(raw);
|
|
883
|
+
} catch {
|
|
884
|
+
return null;
|
|
885
|
+
}
|
|
886
|
+
const data = parsed.claudeAiOauth ?? parsed;
|
|
887
|
+
const creds = data;
|
|
888
|
+
if (typeof creds.accessToken !== "string" || typeof creds.expiresAt !== "number") {
|
|
889
|
+
return null;
|
|
890
|
+
}
|
|
891
|
+
return { accessToken: creds.accessToken, expiresAt: creds.expiresAt };
|
|
892
|
+
}
|
|
893
|
+
function readClaudeCliCredentials() {
|
|
894
|
+
if (process.platform === "darwin") {
|
|
895
|
+
try {
|
|
896
|
+
const raw = execFileSync(
|
|
897
|
+
"/usr/bin/security",
|
|
898
|
+
["find-generic-password", "-s", KEYCHAIN_SERVICE, "-w"],
|
|
899
|
+
{ encoding: "utf-8", timeout: 2e3, stdio: ["pipe", "pipe", "ignore"] }
|
|
900
|
+
);
|
|
901
|
+
return parseClaudeCliCredentials(raw);
|
|
902
|
+
} catch {
|
|
903
|
+
return null;
|
|
904
|
+
}
|
|
905
|
+
}
|
|
906
|
+
try {
|
|
907
|
+
const raw = readFileSync(join(homedir(), ".claude", ".credentials.json"), "utf-8");
|
|
908
|
+
return parseClaudeCliCredentials(raw);
|
|
909
|
+
} catch {
|
|
910
|
+
return null;
|
|
911
|
+
}
|
|
912
|
+
}
|
|
913
|
+
var ClaudeUsageError = class extends Error {
|
|
914
|
+
constructor(message, reason) {
|
|
915
|
+
super(message);
|
|
916
|
+
this.reason = reason;
|
|
917
|
+
}
|
|
918
|
+
};
|
|
919
|
+
function isLocalCredentialProblem(err) {
|
|
920
|
+
return err instanceof ClaudeUsageError && (err.reason === "no_credentials" || err.reason === "credentials_expired");
|
|
921
|
+
}
|
|
922
|
+
function normalizeResetsAt(value) {
|
|
923
|
+
const ms = Date.parse(value);
|
|
924
|
+
return Number.isNaN(ms) ? null : new Date(ms).toISOString();
|
|
925
|
+
}
|
|
926
|
+
function toWindow(value) {
|
|
927
|
+
if (!value || typeof value !== "object") {
|
|
928
|
+
return null;
|
|
929
|
+
}
|
|
930
|
+
const window = value;
|
|
931
|
+
if (typeof window.utilization !== "number" || typeof window.resets_at !== "string") {
|
|
932
|
+
return null;
|
|
933
|
+
}
|
|
934
|
+
const resetsAt = normalizeResetsAt(window.resets_at);
|
|
935
|
+
if (resetsAt === null) {
|
|
936
|
+
return null;
|
|
937
|
+
}
|
|
938
|
+
return { utilization: window.utilization, resetsAt };
|
|
939
|
+
}
|
|
940
|
+
async function getClaudeUsage() {
|
|
941
|
+
const credentials2 = readClaudeCliCredentials();
|
|
942
|
+
if (!credentials2) {
|
|
943
|
+
throw new ClaudeUsageError(
|
|
944
|
+
"No local Claude Code login found. Run `claude` once to sign in with your Claude subscription.",
|
|
945
|
+
"no_credentials"
|
|
946
|
+
);
|
|
947
|
+
}
|
|
948
|
+
if (credentials2.expiresAt < Date.now()) {
|
|
949
|
+
throw new ClaudeUsageError(
|
|
950
|
+
"Claude Code credentials have expired. Run `claude` to refresh them.",
|
|
951
|
+
"credentials_expired"
|
|
952
|
+
);
|
|
953
|
+
}
|
|
954
|
+
const res = await fetch(CLAUDE_USAGE_URL, {
|
|
955
|
+
headers: {
|
|
956
|
+
Authorization: `Bearer ${credentials2.accessToken}`,
|
|
957
|
+
"Content-Type": "application/json",
|
|
958
|
+
"anthropic-version": "2023-06-01"
|
|
959
|
+
}
|
|
960
|
+
});
|
|
961
|
+
if (!res.ok) {
|
|
962
|
+
throw new ClaudeUsageError(`Claude usage request failed: HTTP ${res.status}`, "request_failed");
|
|
963
|
+
}
|
|
964
|
+
const body = await res.json();
|
|
965
|
+
return {
|
|
966
|
+
fiveHour: toWindow(body.five_hour),
|
|
967
|
+
sevenDay: toWindow(body.seven_day)
|
|
968
|
+
};
|
|
969
|
+
}
|
|
970
|
+
|
|
971
|
+
// src/commands/claude-usage.ts
|
|
972
|
+
function formatWindow(label, window) {
|
|
973
|
+
if (!window) {
|
|
974
|
+
return keyValue(label, "not available for this plan");
|
|
975
|
+
}
|
|
976
|
+
const resetsAt = new Date(window.resetsAt);
|
|
977
|
+
return keyValue(label, `${window.utilization}% used, resets ${resetsAt.toLocaleString()}`);
|
|
978
|
+
}
|
|
979
|
+
async function claudeUsage() {
|
|
980
|
+
try {
|
|
981
|
+
const usage = await getClaudeUsage();
|
|
982
|
+
blank();
|
|
983
|
+
console.log(formatWindow("5-hour session", usage.fiveHour));
|
|
984
|
+
console.log(formatWindow("7-day", usage.sevenDay));
|
|
985
|
+
blank();
|
|
986
|
+
} catch (err) {
|
|
987
|
+
if (err instanceof ClaudeUsageError) {
|
|
988
|
+
printError(err.message);
|
|
989
|
+
process.exit(1);
|
|
990
|
+
}
|
|
991
|
+
throw err;
|
|
992
|
+
}
|
|
499
993
|
}
|
|
500
994
|
|
|
501
995
|
// src/commands/run.ts
|
|
502
|
-
import { homedir as
|
|
503
|
-
import { isAbsolute as isAbsolute2, join as
|
|
996
|
+
import { homedir as homedir3 } from "os";
|
|
997
|
+
import { isAbsolute as isAbsolute2, join as join3, parse, resolve as resolvePath } from "path";
|
|
504
998
|
import chalk6 from "chalk";
|
|
505
999
|
|
|
1000
|
+
// ../../packages/types/src/agents/index.ts
|
|
1001
|
+
var MICROVM_MAX_LIFETIME_MS = 8 * 60 * 6e4;
|
|
1002
|
+
|
|
506
1003
|
// ../../packages/types/src/telemetry/index.ts
|
|
507
1004
|
var TelemetryEventTypes = {
|
|
508
1005
|
// Agent activity events (shown in web UI activity log)
|
|
@@ -510,7 +1007,10 @@ var TelemetryEventTypes = {
|
|
|
510
1007
|
AGENT_DISCONNECTED: "agent.disconnected",
|
|
511
1008
|
AGENT_MESSAGE_PROCESSING: "agent.message_processing",
|
|
512
1009
|
AGENT_MESSAGE_DONE: "agent.message_done",
|
|
513
|
-
AGENT_MESSAGE_FAILED: "agent.message_failed"
|
|
1010
|
+
AGENT_MESSAGE_FAILED: "agent.message_failed",
|
|
1011
|
+
// A `warn`/`error` runner-side log line forwarded server-side for
|
|
1012
|
+
// observability (issue #916) — see `apps/cli/src/lib/runner-activity-telemetry.ts`.
|
|
1013
|
+
RUNNER_ACTIVITY: "runner.activity"
|
|
514
1014
|
};
|
|
515
1015
|
|
|
516
1016
|
// ../../packages/types/src/tunnel/index.ts
|
|
@@ -565,6 +1065,13 @@ var isShuttingDown = false;
|
|
|
565
1065
|
var FLUSH_INTERVAL_MS = 5e3;
|
|
566
1066
|
var MAX_BUFFER_SIZE = 50;
|
|
567
1067
|
var FLUSH_TIMEOUT_MS = 3e3;
|
|
1068
|
+
var authProvider = null;
|
|
1069
|
+
function setTelemetryAuthProvider(provider) {
|
|
1070
|
+
authProvider = provider;
|
|
1071
|
+
}
|
|
1072
|
+
var FLUSH_FAILURE_LOG_INTERVAL_MS = 6e4;
|
|
1073
|
+
var lastFlushFailureLoggedAt = 0;
|
|
1074
|
+
var suppressedFlushFailureCount = 0;
|
|
568
1075
|
function logEvent(eventType, options = {}) {
|
|
569
1076
|
const event = {
|
|
570
1077
|
event_type: eventType,
|
|
@@ -599,9 +1106,16 @@ async function flushEvents() {
|
|
|
599
1106
|
flushTimeout = null;
|
|
600
1107
|
}
|
|
601
1108
|
try {
|
|
602
|
-
const
|
|
603
|
-
|
|
604
|
-
|
|
1109
|
+
const providerContext = authProvider?.();
|
|
1110
|
+
let authHeader;
|
|
1111
|
+
if (providerContext?.authHeader) {
|
|
1112
|
+
authHeader = providerContext.authHeader;
|
|
1113
|
+
} else {
|
|
1114
|
+
const credentials2 = await getToken();
|
|
1115
|
+
if (!credentials2) {
|
|
1116
|
+
return;
|
|
1117
|
+
}
|
|
1118
|
+
authHeader = `Bearer ${credentials2.token}`;
|
|
605
1119
|
}
|
|
606
1120
|
const apiUrl = getApiUrlConfig();
|
|
607
1121
|
const controller = new AbortController();
|
|
@@ -616,7 +1130,7 @@ async function flushEvents() {
|
|
|
616
1130
|
method: "POST",
|
|
617
1131
|
headers: {
|
|
618
1132
|
"Content-Type": "application/json",
|
|
619
|
-
Authorization:
|
|
1133
|
+
Authorization: authHeader
|
|
620
1134
|
},
|
|
621
1135
|
body: JSON.stringify(request),
|
|
622
1136
|
signal: controller.signal
|
|
@@ -628,8 +1142,15 @@ async function flushEvents() {
|
|
|
628
1142
|
clearTimeout(timeout);
|
|
629
1143
|
}
|
|
630
1144
|
} catch (error2) {
|
|
631
|
-
|
|
632
|
-
|
|
1145
|
+
const now = Date.now();
|
|
1146
|
+
if (now - lastFlushFailureLoggedAt >= FLUSH_FAILURE_LOG_INTERVAL_MS) {
|
|
1147
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
1148
|
+
const suffix = suppressedFlushFailureCount > 0 ? ` (${suppressedFlushFailureCount} more suppressed in the last ${FLUSH_FAILURE_LOG_INTERVAL_MS / 1e3}s)` : "";
|
|
1149
|
+
console.error(`Telemetry flush error: ${message}${suffix}`);
|
|
1150
|
+
lastFlushFailureLoggedAt = now;
|
|
1151
|
+
suppressedFlushFailureCount = 0;
|
|
1152
|
+
} else {
|
|
1153
|
+
suppressedFlushFailureCount++;
|
|
633
1154
|
}
|
|
634
1155
|
}
|
|
635
1156
|
}
|
|
@@ -698,47 +1219,67 @@ var EventTypes = {
|
|
|
698
1219
|
DEPRECATED_AGENT_KEY_ENV_USED: "cli.deprecated_agent_key_env_used"
|
|
699
1220
|
};
|
|
700
1221
|
|
|
701
|
-
// src/lib/
|
|
702
|
-
|
|
703
|
-
|
|
704
|
-
|
|
705
|
-
|
|
706
|
-
|
|
707
|
-
|
|
708
|
-
|
|
709
|
-
|
|
710
|
-
|
|
711
|
-
|
|
712
|
-
|
|
713
|
-
if (
|
|
714
|
-
|
|
715
|
-
|
|
716
|
-
|
|
717
|
-
|
|
718
|
-
|
|
1222
|
+
// src/lib/runner-activity-telemetry.ts
|
|
1223
|
+
var FORWARDED_LEVELS = /* @__PURE__ */ new Set(["warn", "error"]);
|
|
1224
|
+
var SEVERITY_BY_LEVEL = {
|
|
1225
|
+
warn: "warning",
|
|
1226
|
+
error: "error"
|
|
1227
|
+
};
|
|
1228
|
+
var MAX_MESSAGE_LENGTH = 500;
|
|
1229
|
+
var TRUNCATION_MARKER = "\u2026";
|
|
1230
|
+
function redact(message) {
|
|
1231
|
+
return message.replace(/esk_[A-Za-z0-9_-]+/g, "esk_***").replace(/ct_[A-Za-z0-9_-]+/g, "ct_***").replace(/https?:\/\/\S+/g, "<url>");
|
|
1232
|
+
}
|
|
1233
|
+
function truncate(message) {
|
|
1234
|
+
if (message.length <= MAX_MESSAGE_LENGTH) return message;
|
|
1235
|
+
return message.slice(0, MAX_MESSAGE_LENGTH - TRUNCATION_MARKER.length) + TRUNCATION_MARKER;
|
|
1236
|
+
}
|
|
1237
|
+
var RATE_LIMIT_WINDOW_MS = 6e4;
|
|
1238
|
+
var RATE_LIMIT_MAX_EVENTS = 30;
|
|
1239
|
+
var windowStartedAt = 0;
|
|
1240
|
+
var windowCount = 0;
|
|
1241
|
+
var windowDroppedCount = 0;
|
|
1242
|
+
function admitUnderRateLimit(now) {
|
|
1243
|
+
if (now - windowStartedAt >= RATE_LIMIT_WINDOW_MS) {
|
|
1244
|
+
if (windowDroppedCount > 0) {
|
|
1245
|
+
console.error(
|
|
1246
|
+
`[runner-activity-telemetry] rate cap reached: dropped ${windowDroppedCount} ${windowDroppedCount === 1 ? "entry" : "entries"} in the last ${RATE_LIMIT_WINDOW_MS / 1e3}s (cap ${RATE_LIMIT_MAX_EVENTS}/min)`
|
|
1247
|
+
);
|
|
1248
|
+
}
|
|
1249
|
+
windowStartedAt = now;
|
|
1250
|
+
windowCount = 0;
|
|
1251
|
+
windowDroppedCount = 0;
|
|
719
1252
|
}
|
|
720
|
-
|
|
721
|
-
|
|
722
|
-
|
|
723
|
-
|
|
724
|
-
|
|
725
|
-
|
|
726
|
-
}
|
|
1253
|
+
if (windowCount >= RATE_LIMIT_MAX_EVENTS) {
|
|
1254
|
+
windowDroppedCount++;
|
|
1255
|
+
if (windowDroppedCount === 1) {
|
|
1256
|
+
console.error(
|
|
1257
|
+
`[runner-activity-telemetry] rate cap reached (${RATE_LIMIT_MAX_EVENTS} per ${RATE_LIMIT_WINDOW_MS / 1e3}s) \u2014 dropping further entries this window`
|
|
1258
|
+
);
|
|
1259
|
+
}
|
|
1260
|
+
return false;
|
|
727
1261
|
}
|
|
728
|
-
|
|
1262
|
+
windowCount++;
|
|
1263
|
+
return true;
|
|
729
1264
|
}
|
|
730
|
-
function
|
|
731
|
-
|
|
732
|
-
|
|
1265
|
+
function forwardRunnerActivity(entry, context) {
|
|
1266
|
+
try {
|
|
1267
|
+
if (!FORWARDED_LEVELS.has(entry.level)) return;
|
|
1268
|
+
if (!context.agentId || !context.authHeader) return;
|
|
1269
|
+
if (!admitUnderRateLimit(Date.now())) return;
|
|
1270
|
+
const rawMessage = entry.error ?? entry.message ?? "";
|
|
1271
|
+
const message = truncate(redact(rawMessage));
|
|
1272
|
+
logEvent(TelemetryEventTypes.RUNNER_ACTIVITY, {
|
|
1273
|
+
severity: SEVERITY_BY_LEVEL[entry.level],
|
|
1274
|
+
message,
|
|
1275
|
+
metadata: { source: "cli.run" },
|
|
1276
|
+
agentId: context.agentId
|
|
1277
|
+
});
|
|
1278
|
+
} catch (err) {
|
|
1279
|
+
console.error(
|
|
1280
|
+
`[runner-activity-telemetry] failed to forward runner activity: ${err instanceof Error ? err.message : String(err)}`
|
|
1281
|
+
);
|
|
733
1282
|
}
|
|
734
|
-
return `Bearer ${credentials2.token}`;
|
|
735
|
-
}
|
|
736
|
-
function isInteractive(jsonOutput) {
|
|
737
|
-
if (jsonOutput) return false;
|
|
738
|
-
if (process.env.CI) return false;
|
|
739
|
-
if (process.env.GITHUB_ACTIONS) return false;
|
|
740
|
-
if (!process.stdin.isTTY) return false;
|
|
741
|
-
return true;
|
|
742
1283
|
}
|
|
743
1284
|
|
|
744
1285
|
// src/lib/opencode/health.ts
|
|
@@ -1532,6 +2073,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
|
|
|
1532
2073
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1533
2074
|
return completedOf(reply) != null && finishOf(reply) === "tool-calls";
|
|
1534
2075
|
}
|
|
2076
|
+
function isB2AbandonmentConfirmed(params) {
|
|
2077
|
+
return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
|
|
2078
|
+
}
|
|
1535
2079
|
function messageError(messages, userMessageId) {
|
|
1536
2080
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1537
2081
|
const error2 = errorOf(reply);
|
|
@@ -1545,6 +2089,42 @@ function messageError(messages, userMessageId) {
|
|
|
1545
2089
|
}
|
|
1546
2090
|
return "The agent run failed.";
|
|
1547
2091
|
}
|
|
2092
|
+
function messageFailure(messages, userMessageId) {
|
|
2093
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
2094
|
+
const error2 = errorOf(reply);
|
|
2095
|
+
if (error2 == null || typeof error2 !== "object") return null;
|
|
2096
|
+
const e = error2;
|
|
2097
|
+
const replyProviderId = reply?.info?.providerID ?? null;
|
|
2098
|
+
const replyModelId = reply?.info?.modelID ?? null;
|
|
2099
|
+
if (e.name === "ProviderAuthError") {
|
|
2100
|
+
const data = e.data;
|
|
2101
|
+
const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
|
|
2102
|
+
return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
|
|
2103
|
+
}
|
|
2104
|
+
if (e.name === "APIError") {
|
|
2105
|
+
const data = e.data;
|
|
2106
|
+
const statusCode = data?.statusCode;
|
|
2107
|
+
if (statusCode === 401 || statusCode === 403) {
|
|
2108
|
+
return {
|
|
2109
|
+
kind: "model_auth",
|
|
2110
|
+
providerId: replyProviderId,
|
|
2111
|
+
modelId: replyModelId,
|
|
2112
|
+
reason: "rejected"
|
|
2113
|
+
};
|
|
2114
|
+
}
|
|
2115
|
+
}
|
|
2116
|
+
return null;
|
|
2117
|
+
}
|
|
2118
|
+
function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
|
|
2119
|
+
if (classified != null) return classified;
|
|
2120
|
+
if (hasConfiguredProvider !== false) return null;
|
|
2121
|
+
return {
|
|
2122
|
+
kind: "model_auth",
|
|
2123
|
+
providerId: replyProviderId,
|
|
2124
|
+
modelId: replyModelId,
|
|
2125
|
+
reason: "missing"
|
|
2126
|
+
};
|
|
2127
|
+
}
|
|
1548
2128
|
function hasRunningAssistantExcept(messages, exceptUserMessageId) {
|
|
1549
2129
|
if (!messages || messages.length === 0) return false;
|
|
1550
2130
|
return messages.some(
|
|
@@ -2073,13 +2653,56 @@ var RunnerConnection = class {
|
|
|
2073
2653
|
}
|
|
2074
2654
|
};
|
|
2075
2655
|
|
|
2656
|
+
// src/lib/tunnel/ready-marker.ts
|
|
2657
|
+
import { writeFileSync } from "fs";
|
|
2658
|
+
function writeTunnelReadyMarker(path, agentId) {
|
|
2659
|
+
try {
|
|
2660
|
+
writeFileSync(path, `${agentId}
|
|
2661
|
+
`);
|
|
2662
|
+
return { ok: true };
|
|
2663
|
+
} catch (error2) {
|
|
2664
|
+
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
2665
|
+
}
|
|
2666
|
+
}
|
|
2667
|
+
|
|
2668
|
+
// src/lib/claude-usage-reporting.ts
|
|
2669
|
+
var VALID_MODES = ["auto", "on", "off"];
|
|
2670
|
+
function resolveClaudeUsageReportingMode(flagValue, env) {
|
|
2671
|
+
const raw = flagValue ?? env.EVIDENT_CLAUDE_USAGE_REPORTING;
|
|
2672
|
+
if (raw === void 0 || raw === "") {
|
|
2673
|
+
return { mode: "auto", warnings: [] };
|
|
2674
|
+
}
|
|
2675
|
+
const normalized = raw.trim().toLowerCase();
|
|
2676
|
+
if (VALID_MODES.includes(normalized)) {
|
|
2677
|
+
return { mode: normalized, warnings: [] };
|
|
2678
|
+
}
|
|
2679
|
+
const source = flagValue !== void 0 ? "--claude-usage-reporting" : "EVIDENT_CLAUDE_USAGE_REPORTING";
|
|
2680
|
+
return {
|
|
2681
|
+
mode: "auto",
|
|
2682
|
+
warnings: [
|
|
2683
|
+
`Ignoring invalid ${source} "${raw}": expected one of ${VALID_MODES.join(", ")}; using auto`
|
|
2684
|
+
]
|
|
2685
|
+
};
|
|
2686
|
+
}
|
|
2687
|
+
var BASE_REPORT_DELAY_MS = 10 * 6e4;
|
|
2688
|
+
var REPORT_DELAY_JITTER_FRACTION = 0.2;
|
|
2689
|
+
function nextReportDelayMs(random = Math.random) {
|
|
2690
|
+
const jitterRangeMs = BASE_REPORT_DELAY_MS * REPORT_DELAY_JITTER_FRACTION;
|
|
2691
|
+
return BASE_REPORT_DELAY_MS - jitterRangeMs + random() * (2 * jitterRangeMs);
|
|
2692
|
+
}
|
|
2693
|
+
var FIRST_REPORT_DELAY_MS = 5e3 + Math.random() * 1e4;
|
|
2694
|
+
var CLAUDE_USAGE_FAILURE_REESCALATION_TICKS = 6;
|
|
2695
|
+
function claudeUsageFailureLogLevel(consecutiveFailures) {
|
|
2696
|
+
return consecutiveFailures === 1 || consecutiveFailures % CLAUDE_USAGE_FAILURE_REESCALATION_TICKS === 0 ? "warn" : "debug";
|
|
2697
|
+
}
|
|
2698
|
+
|
|
2076
2699
|
// src/lib/channels/driver.ts
|
|
2077
|
-
import { homedir } from "os";
|
|
2700
|
+
import { homedir as homedir2 } from "os";
|
|
2078
2701
|
|
|
2079
2702
|
// src/lib/file-push.ts
|
|
2080
2703
|
import { randomUUID } from "crypto";
|
|
2081
2704
|
import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
|
|
2082
|
-
import { basename, dirname as dirname2, isAbsolute, join, relative, resolve as resolve2, sep } from "path";
|
|
2705
|
+
import { basename, dirname as dirname2, isAbsolute, join as join2, relative, resolve as resolve2, sep } from "path";
|
|
2083
2706
|
var FILE_MODE = 384;
|
|
2084
2707
|
var DIRECTORY_MODE = 448;
|
|
2085
2708
|
async function writePushedFile(request) {
|
|
@@ -2112,7 +2735,7 @@ async function writePushedFile(request) {
|
|
|
2112
2735
|
const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
|
|
2113
2736
|
dirname2(candidate)
|
|
2114
2737
|
);
|
|
2115
|
-
const realTarget =
|
|
2738
|
+
const realTarget = join2(existingAncestor, ...missingSegments, basename(candidate));
|
|
2116
2739
|
const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
|
|
2117
2740
|
if (allowedDirectory === null) {
|
|
2118
2741
|
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
@@ -2148,7 +2771,7 @@ function expandAndValidate(requestedPath, homeDir) {
|
|
|
2148
2771
|
if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
|
|
2149
2772
|
return null;
|
|
2150
2773
|
}
|
|
2151
|
-
const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ?
|
|
2774
|
+
const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join2(homeDir, requestedPath.slice(2)) : requestedPath;
|
|
2152
2775
|
if (expanded.split(/[/\\]/).includes("..")) {
|
|
2153
2776
|
return null;
|
|
2154
2777
|
}
|
|
@@ -2221,13 +2844,13 @@ function contains(realDirectory, realTarget) {
|
|
|
2221
2844
|
async function createMissingDirectories(existingAncestor, missingSegments) {
|
|
2222
2845
|
let current = existingAncestor;
|
|
2223
2846
|
for (const segment of missingSegments) {
|
|
2224
|
-
current =
|
|
2847
|
+
current = join2(current, segment);
|
|
2225
2848
|
await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
|
|
2226
2849
|
await chmod(current, DIRECTORY_MODE);
|
|
2227
2850
|
}
|
|
2228
2851
|
}
|
|
2229
2852
|
async function writeAtomically(realTarget, content) {
|
|
2230
|
-
const temporaryPath =
|
|
2853
|
+
const temporaryPath = join2(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
|
|
2231
2854
|
let handle;
|
|
2232
2855
|
try {
|
|
2233
2856
|
handle = await open2(temporaryPath, "wx", FILE_MODE);
|
|
@@ -2394,8 +3017,8 @@ async function applyOne(options, file) {
|
|
|
2394
3017
|
await ack(options, file, "applied");
|
|
2395
3018
|
return true;
|
|
2396
3019
|
}
|
|
2397
|
-
function durableDownloadCode(
|
|
2398
|
-
return
|
|
3020
|
+
function durableDownloadCode(status2) {
|
|
3021
|
+
return status2 === 413 ? "file_too_large" : "write_failed";
|
|
2399
3022
|
}
|
|
2400
3023
|
async function downloadContent(options, file, label) {
|
|
2401
3024
|
try {
|
|
@@ -2428,8 +3051,8 @@ async function downloadContent(options, file, label) {
|
|
|
2428
3051
|
return { ok: false, terminal: false };
|
|
2429
3052
|
}
|
|
2430
3053
|
}
|
|
2431
|
-
async function ack(options, file,
|
|
2432
|
-
const outcome = `${
|
|
3054
|
+
async function ack(options, file, status2, reason) {
|
|
3055
|
+
const outcome = `${status2}${reason ? ` (${reason})` : ""}`;
|
|
2433
3056
|
try {
|
|
2434
3057
|
const res = await options.fetchImpl(
|
|
2435
3058
|
`${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
|
|
@@ -2439,7 +3062,7 @@ async function ack(options, file, status, reason) {
|
|
|
2439
3062
|
Authorization: options.getAuthHeader(),
|
|
2440
3063
|
"Content-Type": "application/json"
|
|
2441
3064
|
},
|
|
2442
|
-
body: JSON.stringify(reason ? { status, reason } : { status })
|
|
3065
|
+
body: JSON.stringify(reason ? { status: status2, reason } : { status: status2 })
|
|
2443
3066
|
}
|
|
2444
3067
|
);
|
|
2445
3068
|
if (!res.ok) {
|
|
@@ -2499,6 +3122,8 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
|
|
|
2499
3122
|
var DEFAULT_STUCK_QUEUED_MS = 6e4;
|
|
2500
3123
|
var HEARTBEAT_MS = 6e4;
|
|
2501
3124
|
var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
|
|
3125
|
+
var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
|
|
3126
|
+
var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
|
|
2502
3127
|
var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
|
|
2503
3128
|
var MAX_SUPERSEDED_CONVERSATIONS = 256;
|
|
2504
3129
|
var ChannelAuthError = class extends Error {
|
|
@@ -2509,10 +3134,10 @@ var ChannelAuthError = class extends Error {
|
|
|
2509
3134
|
};
|
|
2510
3135
|
var ChannelTerminalError = class extends Error {
|
|
2511
3136
|
status;
|
|
2512
|
-
constructor(message,
|
|
3137
|
+
constructor(message, status2) {
|
|
2513
3138
|
super(message);
|
|
2514
3139
|
this.name = "ChannelTerminalError";
|
|
2515
|
-
this.status =
|
|
3140
|
+
this.status = status2;
|
|
2516
3141
|
}
|
|
2517
3142
|
};
|
|
2518
3143
|
function backoffDelay(attempt, policy) {
|
|
@@ -2520,8 +3145,8 @@ function backoffDelay(attempt, policy) {
|
|
|
2520
3145
|
const capped = Math.min(policy.maxDelayMs, exp);
|
|
2521
3146
|
return Math.floor(Math.random() * capped);
|
|
2522
3147
|
}
|
|
2523
|
-
function isRetryableStatus(
|
|
2524
|
-
return
|
|
3148
|
+
function isRetryableStatus(status2) {
|
|
3149
|
+
return status2 === 429 || status2 >= 500 && status2 <= 599;
|
|
2525
3150
|
}
|
|
2526
3151
|
var ChannelDriver = class _ChannelDriver {
|
|
2527
3152
|
agentId;
|
|
@@ -2636,6 +3261,24 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2636
3261
|
* processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
|
|
2637
3262
|
*/
|
|
2638
3263
|
readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
3264
|
+
/**
|
|
3265
|
+
* "Already emitted `redrive_unresolved` for this row" (#965). Mirrors
|
|
3266
|
+
* `readoptPollUnresolvedSignalled`: `resolveRedrive`'s `unresolved` leaf recurs
|
|
3267
|
+
* every ~2s drain until opencode's status becomes readable, but the
|
|
3268
|
+
* server-visible signal is an OUTCOME, so it fires at most once per row. Cleared
|
|
3269
|
+
* on any non-`unresolved` outcome so the set cannot grow beyond the currently
|
|
3270
|
+
* unresolvable rows.
|
|
3271
|
+
*/
|
|
3272
|
+
redriveUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
3273
|
+
/**
|
|
3274
|
+
* First `now()` a `pending` row's re-drive was observed `unresolved` (#965). A
|
|
3275
|
+
* `pending` row is invisible to every cron arm (all require `status =
|
|
3276
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
3277
|
+
* nothing driving it. Once `now - since >= pausedMaxWaitMs`, `resolveRedrive`
|
|
3278
|
+
* takes `dispatch` instead of `unresolved` (reusing the existing knob — see
|
|
3279
|
+
* ADR-0047's own "unreachable ⇒ bounded" rule). Cleared on any other outcome.
|
|
3280
|
+
*/
|
|
3281
|
+
redriveUnresolvedSince = /* @__PURE__ */ new Map();
|
|
2639
3282
|
/**
|
|
2640
3283
|
* "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
|
|
2641
3284
|
* Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
|
|
@@ -2724,23 +3367,23 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2724
3367
|
* and stops opencode.
|
|
2725
3368
|
*/
|
|
2726
3369
|
stopped = false;
|
|
2727
|
-
constructor(
|
|
2728
|
-
this.agentId =
|
|
2729
|
-
this.port =
|
|
2730
|
-
this.apiUrl =
|
|
2731
|
-
this.getAuthHeader =
|
|
2732
|
-
this.conversationFilter =
|
|
2733
|
-
this.retry = { ...DEFAULT_RETRY_POLICY, ...
|
|
2734
|
-
this.log =
|
|
3370
|
+
constructor(config) {
|
|
3371
|
+
this.agentId = config.agentId;
|
|
3372
|
+
this.port = config.port;
|
|
3373
|
+
this.apiUrl = config.apiUrl.replace(/\/$/, "");
|
|
3374
|
+
this.getAuthHeader = config.getAuthHeader;
|
|
3375
|
+
this.conversationFilter = config.conversationFilter ?? null;
|
|
3376
|
+
this.retry = { ...DEFAULT_RETRY_POLICY, ...config.retry };
|
|
3377
|
+
this.log = config.log ?? (() => {
|
|
2735
3378
|
});
|
|
2736
|
-
this.fetchImpl =
|
|
2737
|
-
this.sleep =
|
|
2738
|
-
this.pausedPollIntervalMs =
|
|
2739
|
-
this.pausedMaxWaitMs =
|
|
2740
|
-
this.stuckQueuedMs =
|
|
2741
|
-
this.now =
|
|
2742
|
-
this.fileSyncDirectories =
|
|
2743
|
-
this.homeDir =
|
|
3379
|
+
this.fetchImpl = config.fetchImpl ?? fetch;
|
|
3380
|
+
this.sleep = config.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
|
|
3381
|
+
this.pausedPollIntervalMs = config.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
|
|
3382
|
+
this.pausedMaxWaitMs = config.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
|
|
3383
|
+
this.stuckQueuedMs = config.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
|
|
3384
|
+
this.now = config.now ?? (() => Date.now());
|
|
3385
|
+
this.fileSyncDirectories = config.fileSyncDirectories ?? [];
|
|
3386
|
+
this.homeDir = config.homeDir ?? homedir2();
|
|
2744
3387
|
}
|
|
2745
3388
|
/** The IPv4-loopback base URL for the local `opencode serve`. */
|
|
2746
3389
|
get opencodeBase() {
|
|
@@ -2975,6 +3618,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2975
3618
|
skippedAlreadyDispatched += 1;
|
|
2976
3619
|
continue;
|
|
2977
3620
|
}
|
|
3621
|
+
if (message.opencode_message_id) {
|
|
3622
|
+
const outcome = await this.resolveRedrive(conv, sessionId, message, refusedSessionId);
|
|
3623
|
+
if (outcome !== "dispatch") {
|
|
3624
|
+
break;
|
|
3625
|
+
}
|
|
3626
|
+
}
|
|
2978
3627
|
const options = {
|
|
2979
3628
|
agent: message.opencode_agent ?? void 0,
|
|
2980
3629
|
model: message.opencode_model ?? void 0
|
|
@@ -2983,86 +3632,298 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2983
3632
|
try {
|
|
2984
3633
|
this.log({
|
|
2985
3634
|
level: "info",
|
|
2986
|
-
message: `Dispatching message ${message.id.slice(0, 8)} to OpenCode native queue (session ${sessionId.slice(0, 8)})`,
|
|
2987
|
-
conversation_id: conv.id,
|
|
2988
|
-
message_id: message.id
|
|
2989
|
-
});
|
|
2990
|
-
const sendAttachments = this.buildSendAttachments(conv, message);
|
|
2991
|
-
opencodeMessageId = await this.dispatchLocked(
|
|
2992
|
-
sessionId,
|
|
2993
|
-
() => sendPromptAsync(this.port, sessionId, message.content, options, sendAttachments)
|
|
2994
|
-
);
|
|
2995
|
-
} catch (err) {
|
|
2996
|
-
if (err instanceof ChannelAuthError) throw err;
|
|
2997
|
-
this.dispatched.delete(message.id);
|
|
2998
|
-
const exists = await sessionExists(this.port, sessionId);
|
|
2999
|
-
if (exists === false) {
|
|
3000
|
-
this.sessions.delete(conv.id);
|
|
3001
|
-
this.log({
|
|
3002
|
-
level: "warn",
|
|
3003
|
-
message: `Message ${message.id.slice(0, 8)} dispatch hit a session (${sessionId.slice(0, 8)}) that was deleted mid-dispatch (cleanup race) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick (recreated then, in order). Already-dispatched turns keep their watcher.`,
|
|
3004
|
-
conversation_id: conv.id,
|
|
3005
|
-
message_id: message.id
|
|
3006
|
-
});
|
|
3007
|
-
break;
|
|
3008
|
-
}
|
|
3009
|
-
if (exists === null) {
|
|
3010
|
-
this.log({
|
|
3011
|
-
level: "warn",
|
|
3012
|
-
message: `Message ${message.id.slice(0, 8)} dispatch failed and session (${sessionId.slice(0, 8)}) existence could not be confirmed (opencode momentarily unreachable) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick rather than treating it as a genuine failure.`,
|
|
3013
|
-
conversation_id: conv.id,
|
|
3014
|
-
message_id: message.id
|
|
3015
|
-
});
|
|
3016
|
-
break;
|
|
3017
|
-
}
|
|
3018
|
-
const errorMessage = err instanceof Error ? err.message : String(err);
|
|
3019
|
-
this.sessions.delete(conv.id);
|
|
3020
|
-
this.supersede(conv.id, sessionId);
|
|
3021
|
-
this.log({
|
|
3022
|
-
level: "warn",
|
|
3023
|
-
message: `Abandoning OpenCode session ${sessionId.slice(0, 8)} as the binding for conversation ${conv.id.slice(0, 8)} (it exists but failed to run a turn) \u2014 a fresh session is created on the next tick, whatever the persisted binding says by then.`,
|
|
3635
|
+
message: `Dispatching message ${message.id.slice(0, 8)} to OpenCode native queue (session ${sessionId.slice(0, 8)})`,
|
|
3636
|
+
conversation_id: conv.id,
|
|
3637
|
+
message_id: message.id
|
|
3638
|
+
});
|
|
3639
|
+
const sendAttachments = this.buildSendAttachments(conv, message);
|
|
3640
|
+
opencodeMessageId = await this.dispatchLocked(
|
|
3641
|
+
sessionId,
|
|
3642
|
+
() => sendPromptAsync(this.port, sessionId, message.content, options, sendAttachments)
|
|
3643
|
+
);
|
|
3644
|
+
} catch (err) {
|
|
3645
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3646
|
+
this.dispatched.delete(message.id);
|
|
3647
|
+
const exists = await sessionExists(this.port, sessionId);
|
|
3648
|
+
if (exists === false) {
|
|
3649
|
+
this.sessions.delete(conv.id);
|
|
3650
|
+
this.log({
|
|
3651
|
+
level: "warn",
|
|
3652
|
+
message: `Message ${message.id.slice(0, 8)} dispatch hit a session (${sessionId.slice(0, 8)}) that was deleted mid-dispatch (cleanup race) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick (recreated then, in order). Already-dispatched turns keep their watcher.`,
|
|
3653
|
+
conversation_id: conv.id,
|
|
3654
|
+
message_id: message.id
|
|
3655
|
+
});
|
|
3656
|
+
break;
|
|
3657
|
+
}
|
|
3658
|
+
if (exists === null) {
|
|
3659
|
+
this.log({
|
|
3660
|
+
level: "warn",
|
|
3661
|
+
message: `Message ${message.id.slice(0, 8)} dispatch failed and session (${sessionId.slice(0, 8)}) existence could not be confirmed (opencode momentarily unreachable) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick rather than treating it as a genuine failure.`,
|
|
3662
|
+
conversation_id: conv.id,
|
|
3663
|
+
message_id: message.id
|
|
3664
|
+
});
|
|
3665
|
+
break;
|
|
3666
|
+
}
|
|
3667
|
+
const errorMessage = err instanceof Error ? err.message : String(err);
|
|
3668
|
+
this.sessions.delete(conv.id);
|
|
3669
|
+
this.supersede(conv.id, sessionId);
|
|
3670
|
+
this.log({
|
|
3671
|
+
level: "warn",
|
|
3672
|
+
message: `Abandoning OpenCode session ${sessionId.slice(0, 8)} as the binding for conversation ${conv.id.slice(0, 8)} (it exists but failed to run a turn) \u2014 a fresh session is created on the next tick, whatever the persisted binding says by then.`,
|
|
3673
|
+
conversation_id: conv.id,
|
|
3674
|
+
message_id: message.id
|
|
3675
|
+
});
|
|
3676
|
+
await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
|
|
3677
|
+
this.log({
|
|
3678
|
+
level: "warn",
|
|
3679
|
+
message: `markFailed PATCH for message ${message.id.slice(0, 8)} (conversation ${conv.id.slice(0, 8)}) failed (best-effort, not retried): ${markErr instanceof Error ? markErr.message : String(markErr)}`,
|
|
3680
|
+
conversation_id: conv.id,
|
|
3681
|
+
message_id: message.id
|
|
3682
|
+
});
|
|
3683
|
+
});
|
|
3684
|
+
this.log({
|
|
3685
|
+
level: "error",
|
|
3686
|
+
message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
|
|
3687
|
+
conversation_id: conv.id,
|
|
3688
|
+
message_id: message.id
|
|
3689
|
+
});
|
|
3690
|
+
break;
|
|
3691
|
+
}
|
|
3692
|
+
if (opencodeMessageId === null) {
|
|
3693
|
+
this.log({
|
|
3694
|
+
level: "warn",
|
|
3695
|
+
message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
|
|
3696
|
+
conversation_id: conv.id,
|
|
3697
|
+
message_id: message.id
|
|
3698
|
+
});
|
|
3699
|
+
continue;
|
|
3700
|
+
}
|
|
3701
|
+
this.dispatched.add(message.id);
|
|
3702
|
+
this.registerInFlight(conv, sessionId, message, opencodeMessageId);
|
|
3703
|
+
dispatched += 1;
|
|
3704
|
+
void this.postSignal(conv.id, message.id, "dispatched");
|
|
3705
|
+
}
|
|
3706
|
+
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3707
|
+
this.log({
|
|
3708
|
+
level: "warn",
|
|
3709
|
+
message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
|
|
3710
|
+
conversation_id: conv.id
|
|
3711
|
+
});
|
|
3712
|
+
}
|
|
3713
|
+
this.ensureWatcherRunning(sessionId);
|
|
3714
|
+
return dispatched;
|
|
3715
|
+
}
|
|
3716
|
+
/**
|
|
3717
|
+
* Poll a session's message list for the re-drive fence (#965), via the
|
|
3718
|
+
* INJECTED `fetchImpl` — NOT the imported `getSessionMessages` helper, which
|
|
3719
|
+
* hits the global `fetch` and would bypass the same override every other
|
|
3720
|
+
* opencode poll in this file respects. Mirrors `readoptProcessing`'s own
|
|
3721
|
+
* snapshot fetch (`:3081-3111`). `null` = unreadable (non-OK response,
|
|
3722
|
+
* non-array body, or a network exception) — treated as "can't observe",
|
|
3723
|
+
* never as "confirmed gone".
|
|
3724
|
+
*/
|
|
3725
|
+
async pollSessionMessagesForRedrive(conv, message, sessionId) {
|
|
3726
|
+
try {
|
|
3727
|
+
const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
|
|
3728
|
+
if (!res.ok) {
|
|
3729
|
+
this.log({
|
|
3730
|
+
level: "warn",
|
|
3731
|
+
message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned HTTP ${res.status} \u2014 treating as unreadable this tick`,
|
|
3732
|
+
conversation_id: conv.id,
|
|
3733
|
+
message_id: message.id
|
|
3734
|
+
});
|
|
3735
|
+
return null;
|
|
3736
|
+
}
|
|
3737
|
+
const body = await res.json();
|
|
3738
|
+
if (!Array.isArray(body)) {
|
|
3739
|
+
this.log({
|
|
3740
|
+
level: "warn",
|
|
3741
|
+
message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned a non-array message body \u2014 treating as unreadable this tick`,
|
|
3742
|
+
conversation_id: conv.id,
|
|
3743
|
+
message_id: message.id
|
|
3744
|
+
});
|
|
3745
|
+
return null;
|
|
3746
|
+
}
|
|
3747
|
+
return body;
|
|
3748
|
+
} catch (err) {
|
|
3749
|
+
this.log({
|
|
3750
|
+
level: "warn",
|
|
3751
|
+
message: `Re-drive: failed to poll session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3752
|
+
conversation_id: conv.id,
|
|
3753
|
+
message_id: message.id
|
|
3754
|
+
});
|
|
3755
|
+
return null;
|
|
3756
|
+
}
|
|
3757
|
+
}
|
|
3758
|
+
/**
|
|
3759
|
+
* The re-drive fence for a `pending` row that already carries a stored
|
|
3760
|
+
* `opencode_message_id` (#965) — i.e. it has already been handed to opencode at
|
|
3761
|
+
* least once (see the invariant at `QueuedMessage.opencode_message_id`'s doc).
|
|
3762
|
+
* The lifecycle cron can falsely reclaim a `processing` row back to `pending`
|
|
3763
|
+
* mid-turn (a 5-minute liveness-staleness check racing a still-running turn);
|
|
3764
|
+
* without this fence the drain loop would re-`prompt_async` the SAME turn a
|
|
3765
|
+
* second time against live GitHub state. Mirrors `readoptOne`'s job for the
|
|
3766
|
+
* `processing` re-adopt path, but simpler: no b1/b2 preamble cross-check is
|
|
3767
|
+
* needed here because `refusedSessionId` already handles the one case
|
|
3768
|
+
* (#553 abandoned session) that path exists for.
|
|
3769
|
+
*
|
|
3770
|
+
* Only `ChannelAuthError` propagates; every other failure resolves to
|
|
3771
|
+
* `unresolved` and is retried whole on the next ~2s drain tick.
|
|
3772
|
+
*/
|
|
3773
|
+
async resolveRedrive(conv, sessionId, message, refusedSessionId) {
|
|
3774
|
+
const ocId = message.opencode_message_id ?? null;
|
|
3775
|
+
if (refusedSessionId) {
|
|
3776
|
+
this.clearRedriveUnresolved(message.id);
|
|
3777
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3778
|
+
return "dispatch";
|
|
3779
|
+
}
|
|
3780
|
+
const messages = await this.pollSessionMessagesForRedrive(conv, message, sessionId);
|
|
3781
|
+
if (messages == null || messages.length === 0) {
|
|
3782
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3783
|
+
}
|
|
3784
|
+
const state = messageRunState(messages, ocId ?? "");
|
|
3785
|
+
if (state === "done" || state === "failed") {
|
|
3786
|
+
return this.settleRedrive(conv, sessionId, message, ocId, messages, state);
|
|
3787
|
+
}
|
|
3788
|
+
if (state === "running" || state === "queued") {
|
|
3789
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3790
|
+
if (ongoing === true) {
|
|
3791
|
+
return this.reattachRedrive(conv, sessionId, message, ocId);
|
|
3792
|
+
}
|
|
3793
|
+
if (ongoing === false) {
|
|
3794
|
+
this.clearRedriveUnresolved(message.id);
|
|
3795
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3796
|
+
return "dispatch";
|
|
3797
|
+
}
|
|
3798
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3799
|
+
}
|
|
3800
|
+
this.clearRedriveUnresolved(message.id);
|
|
3801
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3802
|
+
return "dispatch";
|
|
3803
|
+
}
|
|
3804
|
+
/**
|
|
3805
|
+
* The `reattached` outcome (Task 3.3): the prior turn is STILL ONGOING per
|
|
3806
|
+
* opencode's own status map — undo the false reclaim instead of starting a
|
|
3807
|
+
* second turn.
|
|
3808
|
+
*/
|
|
3809
|
+
async reattachRedrive(conv, sessionId, message, ocId) {
|
|
3810
|
+
let anchorMs;
|
|
3811
|
+
const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
|
|
3812
|
+
if (!Number.isNaN(parsed)) {
|
|
3813
|
+
anchorMs = parsed;
|
|
3814
|
+
} else {
|
|
3815
|
+
anchorMs = this.now();
|
|
3816
|
+
this.log({
|
|
3817
|
+
level: "error",
|
|
3818
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} has null/unparseable processing_started_at (${String(message.processing_started_at)}) \u2014 anchoring the watcher's absolute-age ceiling to now (defensive)`,
|
|
3819
|
+
conversation_id: conv.id,
|
|
3820
|
+
message_id: message.id
|
|
3821
|
+
});
|
|
3822
|
+
}
|
|
3823
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3824
|
+
try {
|
|
3825
|
+
await this.markProcessing(conv.id, message.id, sessionId, ocId, title);
|
|
3826
|
+
} catch (err) {
|
|
3827
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3828
|
+
this.log({
|
|
3829
|
+
level: "warn",
|
|
3830
|
+
message: `Re-drive: failed to restore message ${message.id.slice(0, 8)} to processing (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3831
|
+
conversation_id: conv.id,
|
|
3832
|
+
message_id: message.id
|
|
3833
|
+
});
|
|
3834
|
+
return "unresolved";
|
|
3835
|
+
}
|
|
3836
|
+
this.clearRedriveUnresolved(message.id);
|
|
3837
|
+
this.registerReadopted(conv, sessionId, message, ocId ?? "", anchorMs);
|
|
3838
|
+
this.dispatched.add(message.id);
|
|
3839
|
+
this.readopted.add(message.id);
|
|
3840
|
+
this.ensureWatcherRunning(sessionId);
|
|
3841
|
+
const watchedForMs = this.now() - anchorMs;
|
|
3842
|
+
void this.postSignal(conv.id, message.id, "redrive_reattached", {
|
|
3843
|
+
watched_for_ms: watchedForMs
|
|
3844
|
+
});
|
|
3845
|
+
this.log({
|
|
3846
|
+
level: "warn",
|
|
3847
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) was wrongly reclaimed to pending while its turn was still running (watched ${watchedForMs}ms) \u2014 restored to processing instead of re-dispatching`,
|
|
3848
|
+
conversation_id: conv.id,
|
|
3849
|
+
message_id: message.id
|
|
3850
|
+
});
|
|
3851
|
+
return "reattached";
|
|
3852
|
+
}
|
|
3853
|
+
/**
|
|
3854
|
+
* The `settled` outcome (Task 3.2): the prior turn already finished (or
|
|
3855
|
+
* errored) while nobody was watching — deliver/report it instead of re-running.
|
|
3856
|
+
* Mirrors `readoptOne`'s `done`/`failed` branches' error discipline, simplified
|
|
3857
|
+
* (no `doneUndeliverable` park: a terminal PATCH failure here just retries next
|
|
3858
|
+
* drain, same as any other non-auth failure).
|
|
3859
|
+
*/
|
|
3860
|
+
async settleRedrive(conv, sessionId, message, ocId, messages, state) {
|
|
3861
|
+
try {
|
|
3862
|
+
if (state === "done") {
|
|
3863
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3864
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3865
|
+
this.log({
|
|
3866
|
+
level: "info",
|
|
3867
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} completed while its row was wrongly reclaimed to pending \u2014 marking done instead of re-dispatching`,
|
|
3024
3868
|
conversation_id: conv.id,
|
|
3025
3869
|
message_id: message.id
|
|
3026
3870
|
});
|
|
3027
|
-
await this.
|
|
3028
|
-
|
|
3029
|
-
|
|
3030
|
-
|
|
3031
|
-
|
|
3032
|
-
message_id: message.id
|
|
3033
|
-
});
|
|
3034
|
-
});
|
|
3871
|
+
await this.markDone(conv.id, message.id, sessionId, ocId, title, usage);
|
|
3872
|
+
} else {
|
|
3873
|
+
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3874
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3875
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3035
3876
|
this.log({
|
|
3036
3877
|
level: "error",
|
|
3037
|
-
message: `
|
|
3038
|
-
conversation_id: conv.id,
|
|
3039
|
-
message_id: message.id
|
|
3040
|
-
});
|
|
3041
|
-
break;
|
|
3042
|
-
}
|
|
3043
|
-
if (opencodeMessageId === null) {
|
|
3044
|
-
this.log({
|
|
3045
|
-
level: "warn",
|
|
3046
|
-
message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
|
|
3878
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} errored while its row was wrongly reclaimed to pending \u2014 marking failed instead of re-dispatching: ${error2 ?? "(no error text)"}`,
|
|
3047
3879
|
conversation_id: conv.id,
|
|
3048
3880
|
message_id: message.id
|
|
3049
3881
|
});
|
|
3050
|
-
|
|
3882
|
+
await this.markFailed(conv.id, message.id, sessionId, error2, usage, failure);
|
|
3051
3883
|
}
|
|
3052
|
-
|
|
3053
|
-
|
|
3054
|
-
dispatched += 1;
|
|
3055
|
-
void this.postSignal(conv.id, message.id, "dispatched");
|
|
3056
|
-
}
|
|
3057
|
-
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3884
|
+
} catch (err) {
|
|
3885
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3058
3886
|
this.log({
|
|
3059
3887
|
level: "warn",
|
|
3060
|
-
message: `
|
|
3061
|
-
conversation_id: conv.id
|
|
3888
|
+
message: `Re-drive: failed to report message ${message.id.slice(0, 8)} ${state} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3889
|
+
conversation_id: conv.id,
|
|
3890
|
+
message_id: message.id
|
|
3062
3891
|
});
|
|
3892
|
+
return "unresolved";
|
|
3063
3893
|
}
|
|
3064
|
-
this.
|
|
3065
|
-
|
|
3894
|
+
this.clearRedriveUnresolved(message.id);
|
|
3895
|
+
void this.postSignal(conv.id, message.id, "redrive_settled");
|
|
3896
|
+
return "settled";
|
|
3897
|
+
}
|
|
3898
|
+
/**
|
|
3899
|
+
* The bounded `unresolved` outcome (Task 3.4): opencode's state could not be
|
|
3900
|
+
* observed (snapshot unreadable/empty, or `isSessionOngoing` returned `null`).
|
|
3901
|
+
* A `pending` row is invisible to every cron arm (all require `status =
|
|
3902
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
3903
|
+
* nothing driving it — bound it to the existing `pausedMaxWaitMs` window
|
|
3904
|
+
* (reusing the knob, not a new constant) and take `dispatch` once elapsed.
|
|
3905
|
+
*/
|
|
3906
|
+
resolveRedriveUnresolved(conv, message) {
|
|
3907
|
+
const now = this.now();
|
|
3908
|
+
const since = this.redriveUnresolvedSince.get(message.id);
|
|
3909
|
+
if (since !== void 0 && now - since >= this.pausedMaxWaitMs) {
|
|
3910
|
+
this.clearRedriveUnresolved(message.id);
|
|
3911
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3912
|
+
return "dispatch";
|
|
3913
|
+
}
|
|
3914
|
+
if (since === void 0) {
|
|
3915
|
+
this.redriveUnresolvedSince.set(message.id, now);
|
|
3916
|
+
}
|
|
3917
|
+
if (!this.redriveUnresolvedSignalled.has(message.id)) {
|
|
3918
|
+
this.redriveUnresolvedSignalled.add(message.id);
|
|
3919
|
+
void this.postSignal(conv.id, message.id, "redrive_unresolved");
|
|
3920
|
+
}
|
|
3921
|
+
return "unresolved";
|
|
3922
|
+
}
|
|
3923
|
+
/** Clear both `unresolved`-bound trackers for a row (any non-`unresolved` outcome). */
|
|
3924
|
+
clearRedriveUnresolved(messageId) {
|
|
3925
|
+
this.redriveUnresolvedSince.delete(messageId);
|
|
3926
|
+
this.redriveUnresolvedSignalled.delete(messageId);
|
|
3066
3927
|
}
|
|
3067
3928
|
/**
|
|
3068
3929
|
* Record that `sessionId` is no longer a valid binding for `conversationId`
|
|
@@ -3125,7 +3986,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3125
3986
|
const directory = await this.resolveOpenCodeDirectory();
|
|
3126
3987
|
const sessionId = await createOpenCodeSession(this.port, directory);
|
|
3127
3988
|
this.sessions.set(conversationId, sessionId);
|
|
3128
|
-
await this.persistSession(conversationId, sessionId).catch(() => {
|
|
3989
|
+
await this.persistSession(conversationId, sessionId).catch((err) => {
|
|
3990
|
+
this.log({
|
|
3991
|
+
level: "warn",
|
|
3992
|
+
message: `Persisting the OpenCode session binding ${sessionId.slice(0, 8)} for conversation ${conversationId.slice(0, 8)} failed (best-effort, not retried) \u2014 the completion PATCH also carries opencode_session_id, so the binding is repaired when the turn finishes: ${err instanceof Error ? err.message : String(err)}`,
|
|
3993
|
+
conversation_id: conversationId
|
|
3994
|
+
});
|
|
3129
3995
|
});
|
|
3130
3996
|
return sessionId;
|
|
3131
3997
|
}
|
|
@@ -3317,12 +4183,17 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3317
4183
|
stuckReported: false,
|
|
3318
4184
|
lastAliveAt: 0,
|
|
3319
4185
|
aliveInFlight: false,
|
|
4186
|
+
titleSynced: false,
|
|
4187
|
+
titleSyncInFlight: false,
|
|
3320
4188
|
awaitingHumanLatched: false,
|
|
3321
4189
|
pausedOnQuestion: false,
|
|
3322
4190
|
pausedOnPermission: false,
|
|
3323
4191
|
pausedClearConfirmed: false,
|
|
3324
4192
|
pausedInFlight: false,
|
|
3325
|
-
deliveryDeadlineAnchored: false
|
|
4193
|
+
deliveryDeadlineAnchored: false,
|
|
4194
|
+
b2PinnedSinceMs: 0,
|
|
4195
|
+
b2LastDescendantCheckMs: 0,
|
|
4196
|
+
b2AbandonedSignalled: false
|
|
3326
4197
|
});
|
|
3327
4198
|
}
|
|
3328
4199
|
/**
|
|
@@ -3337,9 +4208,10 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3337
4208
|
* opencode reports ACTIVELY `running` is watched to completion (its liveness
|
|
3338
4209
|
* heartbeat keeps the cron off its row), while a re-adopted turn that is paused
|
|
3339
4210
|
* awaiting a human — or queued/unreachable — is still bounded by `deadline` and
|
|
3340
|
-
* handed to the cron.
|
|
3341
|
-
*
|
|
3342
|
-
*
|
|
4211
|
+
* handed to the cron. Real invariant (#965): the cron MAY reclaim a row this
|
|
4212
|
+
* runner still holds; a reclaimed row that already ran is never re-dispatched
|
|
4213
|
+
* while opencode reports its turn ongoing (readopt's own gate here, and the
|
|
4214
|
+
* `pending`-row re-drive fence, `resolveRedrive`). `dispatchedAt` stays `now`
|
|
3343
4215
|
* (only the appear-guard uses it).
|
|
3344
4216
|
*
|
|
3345
4217
|
* `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
|
|
@@ -3390,12 +4262,17 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3390
4262
|
// with no extra `re_adopted` signal needed (folds old WI-6).
|
|
3391
4263
|
lastAliveAt: 0,
|
|
3392
4264
|
aliveInFlight: false,
|
|
4265
|
+
titleSynced: false,
|
|
4266
|
+
titleSyncInFlight: false,
|
|
3393
4267
|
awaitingHumanLatched: false,
|
|
3394
4268
|
pausedOnQuestion: false,
|
|
3395
4269
|
pausedOnPermission: false,
|
|
3396
4270
|
pausedClearConfirmed: false,
|
|
3397
4271
|
pausedInFlight: false,
|
|
3398
|
-
deliveryDeadlineAnchored: false
|
|
4272
|
+
deliveryDeadlineAnchored: false,
|
|
4273
|
+
b2PinnedSinceMs: 0,
|
|
4274
|
+
b2LastDescendantCheckMs: 0,
|
|
4275
|
+
b2AbandonedSignalled: false
|
|
3399
4276
|
});
|
|
3400
4277
|
}
|
|
3401
4278
|
/**
|
|
@@ -3557,58 +4434,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3557
4434
|
}
|
|
3558
4435
|
}
|
|
3559
4436
|
if (state === "done") {
|
|
3560
|
-
this.
|
|
3561
|
-
if (!inFlight.done) {
|
|
3562
|
-
this.log({
|
|
3563
|
-
level: "info",
|
|
3564
|
-
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
3565
|
-
conversation_id: conv.id,
|
|
3566
|
-
message_id: inFlight.evidentMessageId
|
|
3567
|
-
});
|
|
3568
|
-
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
3569
|
-
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
3570
|
-
try {
|
|
3571
|
-
await this.markDone(
|
|
3572
|
-
conv.id,
|
|
3573
|
-
inFlight.evidentMessageId,
|
|
3574
|
-
sessionId,
|
|
3575
|
-
inFlight.opencodeMessageId,
|
|
3576
|
-
title,
|
|
3577
|
-
usage
|
|
3578
|
-
);
|
|
3579
|
-
} catch (err) {
|
|
3580
|
-
if (err instanceof ChannelAuthError) throw err;
|
|
3581
|
-
if (err instanceof ChannelTerminalError) {
|
|
3582
|
-
this.log({
|
|
3583
|
-
level: "warn",
|
|
3584
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
3585
|
-
conversation_id: conv.id,
|
|
3586
|
-
message_id: inFlight.evidentMessageId
|
|
3587
|
-
});
|
|
3588
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3589
|
-
return;
|
|
3590
|
-
}
|
|
3591
|
-
if (this.now() >= inFlight.deadline) {
|
|
3592
|
-
this.log({
|
|
3593
|
-
level: "warn",
|
|
3594
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
|
|
3595
|
-
conversation_id: conv.id,
|
|
3596
|
-
message_id: inFlight.evidentMessageId
|
|
3597
|
-
});
|
|
3598
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3599
|
-
return;
|
|
3600
|
-
}
|
|
3601
|
-
this.log({
|
|
3602
|
-
level: "warn",
|
|
3603
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
3604
|
-
conversation_id: conv.id,
|
|
3605
|
-
message_id: inFlight.evidentMessageId
|
|
3606
|
-
});
|
|
3607
|
-
return;
|
|
3608
|
-
}
|
|
3609
|
-
inFlight.done = true;
|
|
3610
|
-
}
|
|
3611
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4437
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
3612
4438
|
return;
|
|
3613
4439
|
}
|
|
3614
4440
|
if (state === "failed") {
|
|
@@ -3622,8 +4448,16 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3622
4448
|
message_id: inFlight.evidentMessageId
|
|
3623
4449
|
});
|
|
3624
4450
|
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4451
|
+
const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
|
|
3625
4452
|
try {
|
|
3626
|
-
await this.markFailed(
|
|
4453
|
+
await this.markFailed(
|
|
4454
|
+
conv.id,
|
|
4455
|
+
inFlight.evidentMessageId,
|
|
4456
|
+
sessionId,
|
|
4457
|
+
error2,
|
|
4458
|
+
usage,
|
|
4459
|
+
failure
|
|
4460
|
+
);
|
|
3627
4461
|
} catch (err) {
|
|
3628
4462
|
if (err instanceof ChannelAuthError) throw err;
|
|
3629
4463
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -3668,6 +4502,44 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3668
4502
|
});
|
|
3669
4503
|
}
|
|
3670
4504
|
const activelyRunning = state === "running" && !awaitingHuman;
|
|
4505
|
+
const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
|
|
4506
|
+
const snapshotReadable = messages != null && messages.length > 0;
|
|
4507
|
+
if (!pinnedNow) {
|
|
4508
|
+
if (snapshotReadable) {
|
|
4509
|
+
inFlight.b2PinnedSinceMs = 0;
|
|
4510
|
+
inFlight.b2LastDescendantCheckMs = 0;
|
|
4511
|
+
inFlight.b2AbandonedSignalled = false;
|
|
4512
|
+
}
|
|
4513
|
+
} else {
|
|
4514
|
+
if (inFlight.b2AbandonedSignalled) {
|
|
4515
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4516
|
+
return;
|
|
4517
|
+
}
|
|
4518
|
+
if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
|
|
4519
|
+
const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
|
|
4520
|
+
if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
|
|
4521
|
+
inFlight.b2LastDescendantCheckMs = this.now();
|
|
4522
|
+
const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
|
|
4523
|
+
if (isB2AbandonmentConfirmed({
|
|
4524
|
+
pinnedForMs,
|
|
4525
|
+
minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
|
|
4526
|
+
descendantOngoing
|
|
4527
|
+
})) {
|
|
4528
|
+
inFlight.b2AbandonedSignalled = true;
|
|
4529
|
+
this.log({
|
|
4530
|
+
level: "warn",
|
|
4531
|
+
message: `Message ${id.slice(0, 8)} b2-pinned for ${Math.round(pinnedForMs / 1e3)}s with no ongoing descendant sub-agent session (status-map confirmed) \u2014 treating the delegated/tool turn as abandoned, resolving done`,
|
|
4532
|
+
conversation_id: conv.id,
|
|
4533
|
+
message_id: id
|
|
4534
|
+
});
|
|
4535
|
+
void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
|
|
4536
|
+
watched_for_ms: pinnedForMs
|
|
4537
|
+
});
|
|
4538
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4539
|
+
return;
|
|
4540
|
+
}
|
|
4541
|
+
}
|
|
4542
|
+
}
|
|
3671
4543
|
if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
|
|
3672
4544
|
this.log({
|
|
3673
4545
|
level: "warn",
|
|
@@ -3687,6 +4559,18 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3687
4559
|
inFlight.aliveInFlight = false;
|
|
3688
4560
|
if (ok) inFlight.lastAliveAt = this.now();
|
|
3689
4561
|
});
|
|
4562
|
+
if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
|
|
4563
|
+
inFlight.titleSyncInFlight = true;
|
|
4564
|
+
void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
|
|
4565
|
+
if (!title) {
|
|
4566
|
+
inFlight.titleSyncInFlight = false;
|
|
4567
|
+
return;
|
|
4568
|
+
}
|
|
4569
|
+
const ok = await this.patchConversationTitle(conv.id, title);
|
|
4570
|
+
inFlight.titleSyncInFlight = false;
|
|
4571
|
+
if (ok) inFlight.titleSynced = true;
|
|
4572
|
+
});
|
|
4573
|
+
}
|
|
3690
4574
|
}
|
|
3691
4575
|
if (awaitingHuman) {
|
|
3692
4576
|
if (!inFlight.awaitingHumanLatched) {
|
|
@@ -3724,6 +4608,70 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3724
4608
|
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3725
4609
|
}
|
|
3726
4610
|
}
|
|
4611
|
+
/**
|
|
4612
|
+
* Settle a message whose run-state has resolved `'done'` — extracted verbatim
|
|
4613
|
+
* (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
|
|
4614
|
+
* inline `state === 'done'` branch body, so a SECOND caller (the #721
|
|
4615
|
+
* b2-abandonment resolution) can reach the exact same completion behavior
|
|
4616
|
+
* (delivery-deadline anchoring, title resolution, usage extraction, and
|
|
4617
|
+
* `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
|
|
4618
|
+
* and risking the two copies silently drifting apart.
|
|
4619
|
+
*/
|
|
4620
|
+
async settleMessageDone(sessionId, watcher, inFlight, messages) {
|
|
4621
|
+
const conv = watcher.conv;
|
|
4622
|
+
this.anchorDeliveryDeadline(inFlight);
|
|
4623
|
+
if (!inFlight.done) {
|
|
4624
|
+
this.log({
|
|
4625
|
+
level: "info",
|
|
4626
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
4627
|
+
conversation_id: conv.id,
|
|
4628
|
+
message_id: inFlight.evidentMessageId
|
|
4629
|
+
});
|
|
4630
|
+
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
4631
|
+
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4632
|
+
try {
|
|
4633
|
+
await this.markDone(
|
|
4634
|
+
conv.id,
|
|
4635
|
+
inFlight.evidentMessageId,
|
|
4636
|
+
sessionId,
|
|
4637
|
+
inFlight.opencodeMessageId,
|
|
4638
|
+
title,
|
|
4639
|
+
usage
|
|
4640
|
+
);
|
|
4641
|
+
} catch (err) {
|
|
4642
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
4643
|
+
if (err instanceof ChannelTerminalError) {
|
|
4644
|
+
this.log({
|
|
4645
|
+
level: "warn",
|
|
4646
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
4647
|
+
conversation_id: conv.id,
|
|
4648
|
+
message_id: inFlight.evidentMessageId
|
|
4649
|
+
});
|
|
4650
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4651
|
+
return;
|
|
4652
|
+
}
|
|
4653
|
+
if (this.now() >= inFlight.deadline) {
|
|
4654
|
+
this.log({
|
|
4655
|
+
level: "warn",
|
|
4656
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
|
|
4657
|
+
conversation_id: conv.id,
|
|
4658
|
+
message_id: inFlight.evidentMessageId
|
|
4659
|
+
});
|
|
4660
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4661
|
+
return;
|
|
4662
|
+
}
|
|
4663
|
+
this.log({
|
|
4664
|
+
level: "warn",
|
|
4665
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
4666
|
+
conversation_id: conv.id,
|
|
4667
|
+
message_id: inFlight.evidentMessageId
|
|
4668
|
+
});
|
|
4669
|
+
return;
|
|
4670
|
+
}
|
|
4671
|
+
inFlight.done = true;
|
|
4672
|
+
}
|
|
4673
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4674
|
+
}
|
|
3727
4675
|
// Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
|
|
3728
4676
|
/**
|
|
3729
4677
|
* Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
|
|
@@ -3890,6 +4838,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3890
4838
|
if (state === "failed") {
|
|
3891
4839
|
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3892
4840
|
const usage = messageUsage(messages, ocId ?? "");
|
|
4841
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3893
4842
|
this.log({
|
|
3894
4843
|
level: "error",
|
|
3895
4844
|
message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
|
|
@@ -3897,7 +4846,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3897
4846
|
message_id: row.id
|
|
3898
4847
|
});
|
|
3899
4848
|
try {
|
|
3900
|
-
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
|
|
4849
|
+
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
|
|
3901
4850
|
} catch (err) {
|
|
3902
4851
|
if (err instanceof ChannelAuthError) throw err;
|
|
3903
4852
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -4164,7 +5113,8 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4164
5113
|
opencode_model: row.opencode_model,
|
|
4165
5114
|
source_message_id: row.source_message_id,
|
|
4166
5115
|
slack_user_id: row.slack_user_id,
|
|
4167
|
-
attachments: row.attachments ?? null
|
|
5116
|
+
attachments: row.attachments ?? null,
|
|
5117
|
+
opencode_message_id: row.opencode_message_id
|
|
4168
5118
|
};
|
|
4169
5119
|
}
|
|
4170
5120
|
/**
|
|
@@ -4303,6 +5253,47 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4303
5253
|
}
|
|
4304
5254
|
return false;
|
|
4305
5255
|
}
|
|
5256
|
+
/**
|
|
5257
|
+
* Tri-state variant of the upward parentID membership walk (#721), used ONLY
|
|
5258
|
+
* by `isAnyDescendantSessionOngoing`. Walks the SAME cached
|
|
5259
|
+
* `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
|
|
5260
|
+
* `sessionBelongsTo`, which deliberately collapses "confirmed not a
|
|
5261
|
+
* descendant" and "the walk's fetch failed" into the same `false` (safe for
|
|
5262
|
+
* its OTHER callers: interaction attribution and the recovery-path
|
|
5263
|
+
* `isAnyDescendantSessionAlive`, both of which just retry next tick with no
|
|
5264
|
+
* safety consequence either way) — this variant keeps those two outcomes
|
|
5265
|
+
* SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
|
|
5266
|
+
* (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
|
|
5267
|
+
* not ongoing".
|
|
5268
|
+
*
|
|
5269
|
+
* Return contract:
|
|
5270
|
+
* - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
|
|
5271
|
+
* - `false` → the walk reached a definitive, parent-less root session
|
|
5272
|
+
* WITHOUT ever matching `rootSessionId` — `sessionId` is
|
|
5273
|
+
* CONFIRMED NOT a descendant of it.
|
|
5274
|
+
* - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
|
|
5275
|
+
* through the walk (`resolveSessionParent` returned `undefined`),
|
|
5276
|
+
* or the depth cap (32) was hit without a definitive answer (a
|
|
5277
|
+
* pathological/cyclic chain proves nothing either way). NEVER
|
|
5278
|
+
* treat this the same as `false` — see `sessionBelongsTo`'s own
|
|
5279
|
+
* doc comment above for why that collapse is safe THERE but not
|
|
5280
|
+
* here.
|
|
5281
|
+
*
|
|
5282
|
+
* `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
|
|
5283
|
+
* to the live-path descendant check, not a modification of shared code used
|
|
5284
|
+
* by interaction attribution or the recovery path.
|
|
5285
|
+
*/
|
|
5286
|
+
async resolveSessionMembership(sessionId, rootSessionId) {
|
|
5287
|
+
let current = sessionId;
|
|
5288
|
+
for (let depth = 0; current && depth < 32; depth++) {
|
|
5289
|
+
if (current === rootSessionId) return true;
|
|
5290
|
+
const parent = await this.resolveSessionParent(current);
|
|
5291
|
+
if (parent === void 0) return null;
|
|
5292
|
+
if (parent === null) return false;
|
|
5293
|
+
current = parent;
|
|
5294
|
+
}
|
|
5295
|
+
return null;
|
|
5296
|
+
}
|
|
4306
5297
|
/**
|
|
4307
5298
|
* Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
|
|
4308
5299
|
* `null` for a root session (no parent) and `undefined` when opencode is
|
|
@@ -4387,6 +5378,54 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4387
5378
|
}
|
|
4388
5379
|
return null;
|
|
4389
5380
|
}
|
|
5381
|
+
/**
|
|
5382
|
+
* Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
|
|
5383
|
+
* session title onto the conversation via the PLAIN conversation-update
|
|
5384
|
+
* endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
|
|
5385
|
+
* message-status endpoint `markProcessing`/`markDone` use. Deliberately a
|
|
5386
|
+
* separate, lighter call: it carries no `status`, so it cannot re-trigger the
|
|
5387
|
+
* `processing`/`done` transition side effects (Slack notices, activity-log
|
|
5388
|
+
* rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
|
|
5389
|
+
* ever touches `conversations.title`. That route (`routes/conversations.ts`)
|
|
5390
|
+
* skips a title write matching the stored value, so a redundant call with the
|
|
5391
|
+
* same title is a real no-op — it does not bump `updated_at`, which the
|
|
5392
|
+
* conversation list sorts and paginates on. (Note this is a DIFFERENT guard
|
|
5393
|
+
* from `threads.ts`'s "non-empty AND changed" one, which only covers the
|
|
5394
|
+
* message-status PATCH; the non-empty half is enforced here instead, by
|
|
5395
|
+
* `resolveSessionTitle` never returning an empty/placeholder title.)
|
|
5396
|
+
*
|
|
5397
|
+
* Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
|
|
5398
|
+
* is logged and the title is simply retried on the next heartbeat tick (the
|
|
5399
|
+
* caller only latches `titleSynced` on `true`).
|
|
5400
|
+
*/
|
|
5401
|
+
async patchConversationTitle(conversationId, title) {
|
|
5402
|
+
try {
|
|
5403
|
+
const res = await this.fetchImpl(
|
|
5404
|
+
`${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
|
|
5405
|
+
{
|
|
5406
|
+
method: "PATCH",
|
|
5407
|
+
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
5408
|
+
body: JSON.stringify({ title })
|
|
5409
|
+
}
|
|
5410
|
+
);
|
|
5411
|
+
if (!res.ok) {
|
|
5412
|
+
this.log({
|
|
5413
|
+
level: "debug",
|
|
5414
|
+
message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
|
|
5415
|
+
conversation_id: conversationId
|
|
5416
|
+
});
|
|
5417
|
+
return false;
|
|
5418
|
+
}
|
|
5419
|
+
return true;
|
|
5420
|
+
} catch (err) {
|
|
5421
|
+
this.log({
|
|
5422
|
+
level: "debug",
|
|
5423
|
+
message: `Best-effort mid-turn title sync PATCH failed for conversation ${conversationId.slice(0, 8)} (will retry next heartbeat): ${err instanceof Error ? err.message : String(err)}`,
|
|
5424
|
+
conversation_id: conversationId
|
|
5425
|
+
});
|
|
5426
|
+
return false;
|
|
5427
|
+
}
|
|
5428
|
+
}
|
|
4390
5429
|
/**
|
|
4391
5430
|
* DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
|
|
4392
5431
|
* (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
|
|
@@ -4419,31 +5458,109 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4419
5458
|
* longer exists". The inversion lives in the caller; this method just reports
|
|
4420
5459
|
* true/false/null faithfully.
|
|
4421
5460
|
*
|
|
4422
|
-
* VERIFY-BEFORE-DEPEND: we depend ONLY on (a) `parentID` from `GET /session/:id`
|
|
4423
|
-
* (already proven by the existing child-session interaction tests, via
|
|
4424
|
-
* `resolveSessionParent`/`sessionBelongsTo`) and (b) the child's own message-list
|
|
4425
|
-
* terminal state. We do NOT depend on any session-level `busy`/`idle` field —
|
|
4426
|
-
* there is none on `GET /session/:id`; OpenCode's busy state is in-memory
|
|
4427
|
-
* `SessionStatus` only.
|
|
5461
|
+
* VERIFY-BEFORE-DEPEND: we depend ONLY on (a) `parentID` from `GET /session/:id`
|
|
5462
|
+
* (already proven by the existing child-session interaction tests, via
|
|
5463
|
+
* `resolveSessionParent`/`sessionBelongsTo`) and (b) the child's own message-list
|
|
5464
|
+
* terminal state. We do NOT depend on any session-level `busy`/`idle` field —
|
|
5465
|
+
* there is none on `GET /session/:id`; OpenCode's busy state is in-memory
|
|
5466
|
+
* `SessionStatus` only.
|
|
5467
|
+
*/
|
|
5468
|
+
async isAnyDescendantSessionAlive(rootSessionId) {
|
|
5469
|
+
const sessions = await listSessions(this.port);
|
|
5470
|
+
if (!sessions) {
|
|
5471
|
+
this.log({
|
|
5472
|
+
level: "warn",
|
|
5473
|
+
message: `Re-adopt: could not enumerate sessions to cross-check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating child liveness as indeterminate`
|
|
5474
|
+
});
|
|
5475
|
+
return null;
|
|
5476
|
+
}
|
|
5477
|
+
for (const candidate of sessions) {
|
|
5478
|
+
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
5479
|
+
if (!await this.sessionBelongsTo(candidate.id, rootSessionId)) continue;
|
|
5480
|
+
const childMsgs = await getSessionMessages(this.port, candidate.id);
|
|
5481
|
+
if (isSessionActivelyGenerating(childMsgs)) {
|
|
5482
|
+
return true;
|
|
5483
|
+
}
|
|
5484
|
+
}
|
|
5485
|
+
return false;
|
|
5486
|
+
}
|
|
5487
|
+
/**
|
|
5488
|
+
* LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
|
|
5489
|
+
* sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
|
|
5490
|
+
* in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
|
|
5491
|
+
*
|
|
5492
|
+
* Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
|
|
5493
|
+
* cross-check above): that method judges liveness from the child's OWN
|
|
5494
|
+
* TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
|
|
5495
|
+
* on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
|
|
5496
|
+
* path the local opencode server IS running, so its in-memory status map is
|
|
5497
|
+
* live and authoritative — and per ADR-0047 §4a ("the child has its own entry
|
|
5498
|
+
* [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
|
|
5499
|
+
* ENTIRE turn (including any tool call it is itself executing), not a
|
|
5500
|
+
* per-message transcript snapshot. This sidesteps the "child's own tool is
|
|
5501
|
+
* executing, between its step's completion and the next generation step"
|
|
5502
|
+
* transcript gap that a transcript-based check would need a second,
|
|
5503
|
+
* sustained-window bound to guard against — it is simply not derived from
|
|
5504
|
+
* message timestamps at all.
|
|
5505
|
+
*
|
|
5506
|
+
* Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
|
|
5507
|
+
* status, as the recovery path does per §4a)? Because on the LIVE path the
|
|
5508
|
+
* root session can be shared: a SECOND, unrelated user message can land on the
|
|
5509
|
+
* SAME session (issue #721's own root cause) and keep the root `busy` for a
|
|
5510
|
+
* reason that has nothing to do with THIS message's delegation. A `task`
|
|
5511
|
+
* descendant session is spawned for exactly one delegated turn and never
|
|
5512
|
+
* reused, so its OWN status-map entry is unambiguous evidence about that one
|
|
5513
|
+
* delegation — which the root's status is not.
|
|
5514
|
+
*
|
|
5515
|
+
* Why membership is checked via `resolveSessionMembership`, NOT
|
|
5516
|
+
* `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
|
|
5517
|
+
* `GET /session/:id` fetch failure into "not a descendant", which would
|
|
5518
|
+
* silently drop a genuinely-live candidate from consideration on the one
|
|
5519
|
+
* unlucky tick its membership-walk fetch hiccups (#721).
|
|
5520
|
+
* `resolveSessionMembership` keeps that failure mode as a distinct `null`
|
|
5521
|
+
* (indeterminate) so it is folded into THIS method's own `indeterminate` flag
|
|
5522
|
+
* instead.
|
|
5523
|
+
*
|
|
5524
|
+
* Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
|
|
5525
|
+
* - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
|
|
5526
|
+
* - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
|
|
5527
|
+
* confirmed either way (`resolveSessionMembership` never
|
|
5528
|
+
* returned `null`), and every CONFIRMED descendant's status read
|
|
5529
|
+
* succeeded and is not ongoing (includes "no descendant session
|
|
5530
|
+
* exists at all" — e.g. a plain, non-`task` tool call).
|
|
5531
|
+
* - `null` → INDETERMINATE: `listSessions` failed, OR at least one
|
|
5532
|
+
* candidate's MEMBERSHIP could not be confirmed
|
|
5533
|
+
* (`resolveSessionMembership` returned `null` — a fetch failure
|
|
5534
|
+
* or pathological chain partway through the parent walk), OR at
|
|
5535
|
+
* least one CONFIRMED descendant's `isSessionOngoing` read
|
|
5536
|
+
* failed — and no OTHER candidate was already confirmed `true`.
|
|
5537
|
+
* The caller MUST NOT treat `null` the same as `false` here
|
|
5538
|
+
* (unlike the recovery cross-check's contract) — see
|
|
5539
|
+
* `isB2AbandonmentConfirmed`.
|
|
4428
5540
|
*/
|
|
4429
|
-
async
|
|
5541
|
+
async isAnyDescendantSessionOngoing(rootSessionId) {
|
|
4430
5542
|
const sessions = await listSessions(this.port);
|
|
4431
5543
|
if (!sessions) {
|
|
4432
5544
|
this.log({
|
|
4433
5545
|
level: "warn",
|
|
4434
|
-
message: `
|
|
5546
|
+
message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
|
|
4435
5547
|
});
|
|
4436
5548
|
return null;
|
|
4437
5549
|
}
|
|
5550
|
+
let indeterminate = false;
|
|
4438
5551
|
for (const candidate of sessions) {
|
|
4439
5552
|
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
4440
|
-
|
|
4441
|
-
|
|
4442
|
-
|
|
4443
|
-
|
|
5553
|
+
const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
|
|
5554
|
+
if (membership === null) {
|
|
5555
|
+
indeterminate = true;
|
|
5556
|
+
continue;
|
|
4444
5557
|
}
|
|
5558
|
+
if (membership === false) continue;
|
|
5559
|
+
const ongoing = await isSessionOngoing(this.port, candidate.id);
|
|
5560
|
+
if (ongoing === true) return true;
|
|
5561
|
+
if (ongoing === null) indeterminate = true;
|
|
4445
5562
|
}
|
|
4446
|
-
return false;
|
|
5563
|
+
return indeterminate ? null : false;
|
|
4447
5564
|
}
|
|
4448
5565
|
/**
|
|
4449
5566
|
* Cheap decision-telemetry label for a running row's LAST correlated reply
|
|
@@ -4589,11 +5706,11 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4589
5706
|
* `ensureSession` guard, not this suppression, is what makes the self-heal
|
|
4590
5707
|
* stick.
|
|
4591
5708
|
*/
|
|
4592
|
-
sessionIdBody(sessionId, conversationId, messageId,
|
|
5709
|
+
sessionIdBody(sessionId, conversationId, messageId, status2) {
|
|
4593
5710
|
if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
|
|
4594
5711
|
this.log({
|
|
4595
5712
|
level: "debug",
|
|
4596
|
-
message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${
|
|
5713
|
+
message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${status2}' update for message ${messageId.slice(0, 8)} so it is not re-bound to conversation ${conversationId.slice(0, 8)}`,
|
|
4597
5714
|
conversation_id: conversationId,
|
|
4598
5715
|
message_id: messageId
|
|
4599
5716
|
});
|
|
@@ -4708,7 +5825,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4708
5825
|
* exists but is wedged, so the next attempt must get a fresh one
|
|
4709
5826
|
* instead of reusing it — see WI-1's server-side null-clearing PATCH).
|
|
4710
5827
|
*/
|
|
4711
|
-
async markFailed(conversationId, messageId, sessionId, error2, usage) {
|
|
5828
|
+
async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
|
|
4712
5829
|
const body = { status: "failed" };
|
|
4713
5830
|
if (sessionId === null) {
|
|
4714
5831
|
body.opencode_session_id = null;
|
|
@@ -4717,6 +5834,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4717
5834
|
}
|
|
4718
5835
|
if (error2 !== void 0) body.error = error2;
|
|
4719
5836
|
if (usage) Object.assign(body, usage);
|
|
5837
|
+
if (failure) {
|
|
5838
|
+
body.failure_kind = failure.kind;
|
|
5839
|
+
body.failure_provider_id = failure.providerId;
|
|
5840
|
+
body.failure_model_id = failure.modelId;
|
|
5841
|
+
body.failure_reason = failure.reason;
|
|
5842
|
+
}
|
|
4720
5843
|
await this.callWithRetry(
|
|
4721
5844
|
"marking message as failed",
|
|
4722
5845
|
() => this.fetchImpl(
|
|
@@ -4729,6 +5852,29 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4729
5852
|
)
|
|
4730
5853
|
);
|
|
4731
5854
|
}
|
|
5855
|
+
/**
|
|
5856
|
+
* Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
|
|
5857
|
+
*
|
|
5858
|
+
* `messageFailure` alone (structured OpenCode error → `model_auth`) covers
|
|
5859
|
+
* most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
|
|
5860
|
+
* to the P1-2b zero-provider check — one extra loopback call to
|
|
5861
|
+
* `hasAnyConfiguredProvider`, only reached when the structured classifier
|
|
5862
|
+
* couldn't place it. Fails open (never throws): a fallback probe failure
|
|
5863
|
+
* (`null`/indeterminate) leaves the classification `null`, which produces
|
|
5864
|
+
* today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
|
|
5865
|
+
*/
|
|
5866
|
+
async classifyModelAuthFailure(messages, userMessageId) {
|
|
5867
|
+
const classified = messageFailure(messages, userMessageId);
|
|
5868
|
+
if (classified != null) return classified;
|
|
5869
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
5870
|
+
const hasProvider = await hasAnyConfiguredProvider(this.port);
|
|
5871
|
+
return applyZeroProviderFallback(
|
|
5872
|
+
classified,
|
|
5873
|
+
hasProvider,
|
|
5874
|
+
reply?.info?.providerID ?? null,
|
|
5875
|
+
reply?.info?.modelID ?? null
|
|
5876
|
+
);
|
|
5877
|
+
}
|
|
4732
5878
|
/**
|
|
4733
5879
|
* Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
|
|
4734
5880
|
* (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
|
|
@@ -4881,10 +6027,16 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4881
6027
|
import chalk5 from "chalk";
|
|
4882
6028
|
import ora2 from "ora";
|
|
4883
6029
|
import { select as select2 } from "@inquirer/prompts";
|
|
6030
|
+
var INTERACTIVE_START_TIMEOUT_MS = 3e4;
|
|
4884
6031
|
async function ensureOpenCodeRunning(ctx) {
|
|
4885
6032
|
const healthCheck = await checkOpenCodeHealth(ctx.port);
|
|
4886
6033
|
if (healthCheck.healthy) {
|
|
4887
|
-
return {
|
|
6034
|
+
return {
|
|
6035
|
+
port: ctx.port,
|
|
6036
|
+
process: null,
|
|
6037
|
+
version: healthCheck.version ?? null,
|
|
6038
|
+
notReadyReason: null
|
|
6039
|
+
};
|
|
4888
6040
|
}
|
|
4889
6041
|
const runningInstances = await findHealthyOpenCodeInstances();
|
|
4890
6042
|
if (runningInstances.length > 0) {
|
|
@@ -4905,7 +6057,7 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4905
6057
|
console.log(chalk5.yellow("Tip: Run with the correct port:"));
|
|
4906
6058
|
console.log(
|
|
4907
6059
|
chalk5.dim(
|
|
4908
|
-
` ${getCliName()} run --
|
|
6060
|
+
` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
|
|
4909
6061
|
)
|
|
4910
6062
|
);
|
|
4911
6063
|
}
|
|
@@ -4925,14 +6077,22 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4925
6077
|
if (!ctx.interactive) {
|
|
4926
6078
|
ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
|
|
4927
6079
|
const proc = await startOpenCode(ctx.port);
|
|
4928
|
-
const health = await waitForOpenCodeHealth(ctx.port,
|
|
6080
|
+
const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
|
|
4929
6081
|
if (!health.healthy) {
|
|
4930
|
-
|
|
4931
|
-
|
|
4932
|
-
|
|
6082
|
+
return {
|
|
6083
|
+
port: ctx.port,
|
|
6084
|
+
process: proc,
|
|
6085
|
+
version: null,
|
|
6086
|
+
notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
|
|
6087
|
+
};
|
|
4933
6088
|
}
|
|
4934
6089
|
ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
|
|
4935
|
-
return {
|
|
6090
|
+
return {
|
|
6091
|
+
port: ctx.port,
|
|
6092
|
+
process: proc,
|
|
6093
|
+
version: health.version ?? null,
|
|
6094
|
+
notReadyReason: null
|
|
6095
|
+
};
|
|
4936
6096
|
}
|
|
4937
6097
|
let port = ctx.port;
|
|
4938
6098
|
if (isPortInUse(port)) {
|
|
@@ -4985,122 +6145,15 @@ Port ${port} is already in use.`));
|
|
|
4985
6145
|
if (action === "start") {
|
|
4986
6146
|
const spinner = ora2("Starting OpenCode...").start();
|
|
4987
6147
|
const proc = await startOpenCode(port);
|
|
4988
|
-
const health = await waitForOpenCodeHealth(port,
|
|
6148
|
+
const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
|
|
4989
6149
|
if (!health.healthy) {
|
|
4990
6150
|
spinner.fail("Failed to start OpenCode");
|
|
4991
6151
|
throw new Error("OpenCode failed to start");
|
|
4992
6152
|
}
|
|
4993
6153
|
spinner.stop();
|
|
4994
|
-
return { port, process: proc, version: health.version ?? null };
|
|
4995
|
-
}
|
|
4996
|
-
return { port, process: null, version: null };
|
|
4997
|
-
}
|
|
4998
|
-
|
|
4999
|
-
// src/commands/agent-lookup.ts
|
|
5000
|
-
async function readErrorMessage(response) {
|
|
5001
|
-
const text = await response.text().catch(() => "");
|
|
5002
|
-
if (!text) return response.statusText || void 0;
|
|
5003
|
-
try {
|
|
5004
|
-
const data = JSON.parse(text);
|
|
5005
|
-
const message = data.message ?? data.error;
|
|
5006
|
-
if (typeof message === "string" && message.trim()) {
|
|
5007
|
-
return message;
|
|
5008
|
-
}
|
|
5009
|
-
} catch {
|
|
5010
|
-
}
|
|
5011
|
-
return text.trim() || response.statusText || void 0;
|
|
5012
|
-
}
|
|
5013
|
-
function authFailureHint(apiUrl, serverMessage) {
|
|
5014
|
-
const reason = serverMessage ? `: ${serverMessage}` : "";
|
|
5015
|
-
return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
|
|
5016
|
-
}
|
|
5017
|
-
async function resolveAgentIdFromKey(authHeader) {
|
|
5018
|
-
const apiUrl = getApiUrlConfig();
|
|
5019
|
-
try {
|
|
5020
|
-
const response = await fetch(`${apiUrl}/me`, {
|
|
5021
|
-
headers: { Authorization: authHeader }
|
|
5022
|
-
});
|
|
5023
|
-
if (response.status === 401) {
|
|
5024
|
-
const serverMessage = await readErrorMessage(response);
|
|
5025
|
-
return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
5026
|
-
}
|
|
5027
|
-
if (!response.ok) {
|
|
5028
|
-
const serverMessage = await readErrorMessage(response);
|
|
5029
|
-
return {
|
|
5030
|
-
error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
5031
|
-
};
|
|
5032
|
-
}
|
|
5033
|
-
const data = await response.json();
|
|
5034
|
-
if (data.auth_type === "agent_key" && data.agent_id) {
|
|
5035
|
-
return { agent_id: data.agent_id };
|
|
5036
|
-
}
|
|
5037
|
-
return {
|
|
5038
|
-
error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
|
|
5039
|
-
};
|
|
5040
|
-
} catch (error2) {
|
|
5041
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
5042
|
-
return { error: `Failed to resolve runner from key: ${message}` };
|
|
5043
|
-
}
|
|
5044
|
-
}
|
|
5045
|
-
async function notifyAgentDisconnected(agentId, authHeader) {
|
|
5046
|
-
const apiUrl = getApiUrlConfig();
|
|
5047
|
-
try {
|
|
5048
|
-
const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
|
|
5049
|
-
method: "POST",
|
|
5050
|
-
headers: { Authorization: authHeader }
|
|
5051
|
-
});
|
|
5052
|
-
if (!response.ok) {
|
|
5053
|
-
const serverMessage = await readErrorMessage(response);
|
|
5054
|
-
return {
|
|
5055
|
-
ok: false,
|
|
5056
|
-
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
5057
|
-
};
|
|
5058
|
-
}
|
|
5059
|
-
return { ok: true };
|
|
5060
|
-
} catch (error2) {
|
|
5061
|
-
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
5062
|
-
}
|
|
5063
|
-
}
|
|
5064
|
-
async function getAgentInfo(agentId, authHeader) {
|
|
5065
|
-
const apiUrl = getApiUrlConfig();
|
|
5066
|
-
try {
|
|
5067
|
-
const response = await fetch(`${apiUrl}/runners/${agentId}`, {
|
|
5068
|
-
headers: { Authorization: authHeader }
|
|
5069
|
-
});
|
|
5070
|
-
if (response.status === 401) {
|
|
5071
|
-
const serverMessage = await readErrorMessage(response);
|
|
5072
|
-
return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
5073
|
-
}
|
|
5074
|
-
if (response.status === 403) {
|
|
5075
|
-
const serverMessage = await readErrorMessage(response);
|
|
5076
|
-
return {
|
|
5077
|
-
valid: false,
|
|
5078
|
-
error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
|
|
5079
|
-
};
|
|
5080
|
-
}
|
|
5081
|
-
if (response.status === 404) {
|
|
5082
|
-
const serverMessage = await readErrorMessage(response);
|
|
5083
|
-
return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
|
|
5084
|
-
}
|
|
5085
|
-
if (!response.ok) {
|
|
5086
|
-
const serverMessage = await readErrorMessage(response);
|
|
5087
|
-
return {
|
|
5088
|
-
valid: false,
|
|
5089
|
-
error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
5090
|
-
};
|
|
5091
|
-
}
|
|
5092
|
-
const agent = await response.json();
|
|
5093
|
-
if (agent.agent_type !== "local") {
|
|
5094
|
-
return {
|
|
5095
|
-
valid: false,
|
|
5096
|
-
error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
|
|
5097
|
-
};
|
|
5098
|
-
}
|
|
5099
|
-
return { valid: true, agent };
|
|
5100
|
-
} catch (error2) {
|
|
5101
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
5102
|
-
return { valid: false, error: `Failed to validate runner: ${message}` };
|
|
6154
|
+
return { port, process: proc, version: health.version ?? null, notReadyReason: null };
|
|
5103
6155
|
}
|
|
6156
|
+
return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
|
|
5104
6157
|
}
|
|
5105
6158
|
|
|
5106
6159
|
// src/commands/run.ts
|
|
@@ -5108,6 +6161,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
|
|
|
5108
6161
|
var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
|
|
5109
6162
|
var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
|
|
5110
6163
|
var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
|
|
6164
|
+
var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
|
|
5111
6165
|
function resolveLogLevel(options) {
|
|
5112
6166
|
const accepted = Object.keys(LOG_LEVELS);
|
|
5113
6167
|
const validate = (value, source) => {
|
|
@@ -5138,7 +6192,7 @@ function resolveFileSyncDirectories(raw, homeDir) {
|
|
|
5138
6192
|
if (trimmed === "") {
|
|
5139
6193
|
throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
|
|
5140
6194
|
}
|
|
5141
|
-
const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ?
|
|
6195
|
+
const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join3(homeDir, trimmed.slice(2)) : trimmed;
|
|
5142
6196
|
if (!isAbsolute2(expanded)) {
|
|
5143
6197
|
throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
|
|
5144
6198
|
}
|
|
@@ -5159,6 +6213,35 @@ function resolveFileSyncDirectories(raw, homeDir) {
|
|
|
5159
6213
|
}
|
|
5160
6214
|
return directories;
|
|
5161
6215
|
}
|
|
6216
|
+
var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
|
|
6217
|
+
var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
|
|
6218
|
+
var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
|
|
6219
|
+
function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
|
|
6220
|
+
const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
|
|
6221
|
+
let raw;
|
|
6222
|
+
let source;
|
|
6223
|
+
if (options.opencodeStartTimeout !== void 0) {
|
|
6224
|
+
raw = options.opencodeStartTimeout;
|
|
6225
|
+
source = "--opencode-start-timeout";
|
|
6226
|
+
} else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
|
|
6227
|
+
raw = env[OPENCODE_START_TIMEOUT_ENV];
|
|
6228
|
+
source = OPENCODE_START_TIMEOUT_ENV;
|
|
6229
|
+
} else {
|
|
6230
|
+
return { timeoutMs: defaultMs, warnings: [] };
|
|
6231
|
+
}
|
|
6232
|
+
const trimmed = raw.trim();
|
|
6233
|
+
const seconds = Number(trimmed);
|
|
6234
|
+
const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
|
|
6235
|
+
if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
|
|
6236
|
+
return {
|
|
6237
|
+
timeoutMs: defaultMs,
|
|
6238
|
+
warnings: [
|
|
6239
|
+
`Ignoring invalid ${source} "${raw}": expected a positive integer number of seconds (at most ${MAX_OPENCODE_START_TIMEOUT_SECONDS}); using the default ${DEFAULT_OPENCODE_START_TIMEOUT_SECONDS}s`
|
|
6240
|
+
]
|
|
6241
|
+
};
|
|
6242
|
+
}
|
|
6243
|
+
return { timeoutMs: seconds * 1e3, warnings: [] };
|
|
6244
|
+
}
|
|
5162
6245
|
function meetsThreshold(state, level) {
|
|
5163
6246
|
return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
|
|
5164
6247
|
}
|
|
@@ -5180,6 +6263,10 @@ function log2(state, message, level = "info") {
|
|
|
5180
6263
|
function logActivity(state, entry) {
|
|
5181
6264
|
const level = entry.level ?? (entry.type === "error" ? "error" : "info");
|
|
5182
6265
|
if (!meetsThreshold(state, level)) return;
|
|
6266
|
+
forwardRunnerActivity(
|
|
6267
|
+
{ level, message: entry.message, error: entry.error },
|
|
6268
|
+
{ agentId: state.agentId, authHeader: state.authHeader }
|
|
6269
|
+
);
|
|
5183
6270
|
const fullEntry = {
|
|
5184
6271
|
...entry,
|
|
5185
6272
|
level,
|
|
@@ -5279,9 +6366,15 @@ async function handleAuthError(state, error2) {
|
|
|
5279
6366
|
}
|
|
5280
6367
|
async function driveChannels(state, driver) {
|
|
5281
6368
|
let idlePolls = 0;
|
|
6369
|
+
let idleMs = 0;
|
|
6370
|
+
let consecutiveDrainFailures = 0;
|
|
6371
|
+
let unreachableMs = 0;
|
|
5282
6372
|
let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
5283
6373
|
let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
5284
6374
|
while (state.running) {
|
|
6375
|
+
const cycleStartedAtMs = performance.now();
|
|
6376
|
+
let idleThisCycle = false;
|
|
6377
|
+
let unreachableThisCycle = false;
|
|
5285
6378
|
if (state.connection?.reconnecting && state.connection.reconnectPromise) {
|
|
5286
6379
|
logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
|
|
5287
6380
|
if (state.interactive) displayStatus(state);
|
|
@@ -5296,17 +6389,23 @@ async function driveChannels(state, driver) {
|
|
|
5296
6389
|
);
|
|
5297
6390
|
try {
|
|
5298
6391
|
const processed = await driver.drainPending();
|
|
6392
|
+
consecutiveDrainFailures = 0;
|
|
6393
|
+
unreachableMs = 0;
|
|
5299
6394
|
state.messageCount += processed;
|
|
5300
6395
|
const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
|
|
5301
6396
|
lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
5302
6397
|
const appliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
5303
|
-
const
|
|
6398
|
+
const filesApplied = appliedFiles !== lastSeenAppliedFiles;
|
|
6399
|
+
const fileActivity = carriedOverFileSync || filesApplied;
|
|
5304
6400
|
lastSeenAppliedFiles = appliedFiles;
|
|
6401
|
+
if (filesApplied) state.claudeUsageRearm?.();
|
|
5305
6402
|
if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
|
|
5306
6403
|
idlePolls = 0;
|
|
6404
|
+
idleMs = 0;
|
|
5307
6405
|
if (processed > 0 && state.interactive) displayStatus(state);
|
|
5308
6406
|
} else if (state.idleTimeout !== null) {
|
|
5309
6407
|
idlePolls++;
|
|
6408
|
+
idleThisCycle = true;
|
|
5310
6409
|
if (idlePolls === 1) {
|
|
5311
6410
|
logActivity(state, {
|
|
5312
6411
|
type: "info",
|
|
@@ -5330,21 +6429,44 @@ async function driveChannels(state, driver) {
|
|
|
5330
6429
|
const errorMessage = error2 instanceof Error ? error2.message : String(error2);
|
|
5331
6430
|
logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
|
|
5332
6431
|
if (state.interactive) displayStatus(state);
|
|
6432
|
+
if (driver.hasInFlightWatchers()) {
|
|
6433
|
+
consecutiveDrainFailures = 0;
|
|
6434
|
+
unreachableMs = 0;
|
|
6435
|
+
} else if (state.idleTimeout !== null) {
|
|
6436
|
+
consecutiveDrainFailures++;
|
|
6437
|
+
unreachableThisCycle = true;
|
|
6438
|
+
if (consecutiveDrainFailures === 1) {
|
|
6439
|
+
logActivity(state, {
|
|
6440
|
+
type: "info",
|
|
6441
|
+
message: `Cannot reach Evident, will exit if this persists past the idle timeout (timeout: ${state.idleTimeout}s)...`
|
|
6442
|
+
});
|
|
6443
|
+
if (state.interactive) displayStatus(state);
|
|
6444
|
+
}
|
|
6445
|
+
}
|
|
5333
6446
|
}
|
|
5334
6447
|
await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
|
|
5335
|
-
|
|
5336
|
-
|
|
5337
|
-
|
|
5338
|
-
|
|
5339
|
-
|
|
5340
|
-
|
|
5341
|
-
|
|
6448
|
+
const cycleMs = performance.now() - cycleStartedAtMs;
|
|
6449
|
+
if (idleThisCycle) idleMs += cycleMs;
|
|
6450
|
+
if (unreachableThisCycle) unreachableMs += cycleMs;
|
|
6451
|
+
if (state.idleTimeout !== null && consecutiveDrainFailures >= 2 && unreachableMs > state.idleTimeout * 1e3) {
|
|
6452
|
+
logActivity(state, {
|
|
6453
|
+
type: "info",
|
|
6454
|
+
level: "warn",
|
|
6455
|
+
message: `Exiting: could not reach Evident for ${consecutiveDrainFailures} consecutive polls (${Math.round(unreachableMs / 1e3)}s)`
|
|
6456
|
+
});
|
|
6457
|
+
if (state.interactive) displayStatus(state);
|
|
6458
|
+
break;
|
|
6459
|
+
}
|
|
6460
|
+
if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
|
|
6461
|
+
logActivity(state, { type: "info", message: "Idle timeout reached" });
|
|
6462
|
+
if (state.interactive) displayStatus(state);
|
|
6463
|
+
break;
|
|
5342
6464
|
}
|
|
5343
6465
|
}
|
|
5344
6466
|
}
|
|
5345
6467
|
var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
|
|
5346
|
-
async function runSweep(state, driver,
|
|
5347
|
-
const mode = `age=${
|
|
6468
|
+
async function runSweep(state, driver, config) {
|
|
6469
|
+
const mode = `age=${config.maxAgeMs ?? "\u2014"} count=${config.maxCount ?? "\u2014"}`;
|
|
5348
6470
|
try {
|
|
5349
6471
|
const sessions = await listSessions(state.port);
|
|
5350
6472
|
if (sessions === null) {
|
|
@@ -5357,8 +6479,8 @@ async function runSweep(state, driver, config2) {
|
|
|
5357
6479
|
const toDelete = selectSessionsToDelete(
|
|
5358
6480
|
sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
|
|
5359
6481
|
{
|
|
5360
|
-
maxAgeMs:
|
|
5361
|
-
maxCount:
|
|
6482
|
+
maxAgeMs: config.maxAgeMs,
|
|
6483
|
+
maxCount: config.maxCount,
|
|
5362
6484
|
nowMs: Date.now(),
|
|
5363
6485
|
protectedIds: driver.protectedSessionIds()
|
|
5364
6486
|
}
|
|
@@ -5394,7 +6516,7 @@ async function runSweep(state, driver, config2) {
|
|
|
5394
6516
|
}
|
|
5395
6517
|
}
|
|
5396
6518
|
function scheduleSessionCleanup(state, driver, options) {
|
|
5397
|
-
const
|
|
6519
|
+
const config = resolveSessionCleanupConfig(
|
|
5398
6520
|
{
|
|
5399
6521
|
maxAge: options.sessionCleanupMaxAge,
|
|
5400
6522
|
maxCount: options.sessionCleanupMaxCount,
|
|
@@ -5402,21 +6524,129 @@ function scheduleSessionCleanup(state, driver, options) {
|
|
|
5402
6524
|
},
|
|
5403
6525
|
process.env
|
|
5404
6526
|
);
|
|
5405
|
-
for (const warning2 of
|
|
6527
|
+
for (const warning2 of config.warnings) {
|
|
5406
6528
|
logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
|
|
5407
6529
|
}
|
|
5408
|
-
if (!
|
|
6530
|
+
if (!config.enabled) return;
|
|
5409
6531
|
logActivity(state, {
|
|
5410
6532
|
type: "info",
|
|
5411
|
-
message: `Session cleanup enabled (age=${
|
|
6533
|
+
message: `Session cleanup enabled (age=${config.maxAgeMs ?? "\u2014"}, count=${config.maxCount ?? "\u2014"}, interval=${config.intervalMs}ms)`
|
|
5412
6534
|
});
|
|
5413
|
-
const interval = setInterval(() => void runSweep(state, driver,
|
|
6535
|
+
const interval = setInterval(() => void runSweep(state, driver, config), config.intervalMs);
|
|
5414
6536
|
const firstSweep = setTimeout(
|
|
5415
|
-
() => void runSweep(state, driver,
|
|
6537
|
+
() => void runSweep(state, driver, config),
|
|
5416
6538
|
SESSION_CLEANUP_FIRST_SWEEP_MS
|
|
5417
6539
|
);
|
|
5418
6540
|
state.sessionCleanupTimers.push(interval, firstSweep);
|
|
5419
6541
|
}
|
|
6542
|
+
function claudeUsageFailureStreakSuffix(consecutiveFailures) {
|
|
6543
|
+
return consecutiveFailures > 1 ? ` (${consecutiveFailures} consecutive failures)` : "";
|
|
6544
|
+
}
|
|
6545
|
+
function scheduleClaudeUsageReporting(state, options) {
|
|
6546
|
+
const { mode, warnings } = resolveClaudeUsageReportingMode(
|
|
6547
|
+
options.claudeUsageReporting,
|
|
6548
|
+
process.env
|
|
6549
|
+
);
|
|
6550
|
+
for (const warning2 of warnings) {
|
|
6551
|
+
logActivity(state, {
|
|
6552
|
+
type: "info",
|
|
6553
|
+
level: "warn",
|
|
6554
|
+
message: `Claude usage reporting: ${warning2}`
|
|
6555
|
+
});
|
|
6556
|
+
}
|
|
6557
|
+
if (mode === "off") {
|
|
6558
|
+
logActivity(state, {
|
|
6559
|
+
type: "info",
|
|
6560
|
+
level: "debug",
|
|
6561
|
+
message: "Claude usage reporting is off (--claude-usage-reporting off)"
|
|
6562
|
+
});
|
|
6563
|
+
return null;
|
|
6564
|
+
}
|
|
6565
|
+
let consecutiveFailures = 0;
|
|
6566
|
+
let armed = false;
|
|
6567
|
+
let rearmRequested = false;
|
|
6568
|
+
const scheduleNextTick = () => {
|
|
6569
|
+
armed = true;
|
|
6570
|
+
rearmRequested = false;
|
|
6571
|
+
state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
|
|
6572
|
+
};
|
|
6573
|
+
const rearm = () => {
|
|
6574
|
+
if (armed) {
|
|
6575
|
+
rearmRequested = true;
|
|
6576
|
+
return;
|
|
6577
|
+
}
|
|
6578
|
+
rearmRequested = false;
|
|
6579
|
+
armed = true;
|
|
6580
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6581
|
+
};
|
|
6582
|
+
const tick = async (isProbe) => {
|
|
6583
|
+
try {
|
|
6584
|
+
const usage = await getClaudeUsage();
|
|
6585
|
+
const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
|
|
6586
|
+
if (result.ok) {
|
|
6587
|
+
if (consecutiveFailures > 0) {
|
|
6588
|
+
logActivity(state, {
|
|
6589
|
+
type: "info",
|
|
6590
|
+
level: "info",
|
|
6591
|
+
message: "Claude usage reporting recovered"
|
|
6592
|
+
});
|
|
6593
|
+
}
|
|
6594
|
+
consecutiveFailures = 0;
|
|
6595
|
+
logActivity(state, {
|
|
6596
|
+
type: "info",
|
|
6597
|
+
level: "debug",
|
|
6598
|
+
message: "Reported Claude usage to Evident"
|
|
6599
|
+
});
|
|
6600
|
+
} else {
|
|
6601
|
+
consecutiveFailures++;
|
|
6602
|
+
logActivity(state, {
|
|
6603
|
+
type: "info",
|
|
6604
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6605
|
+
message: `Failed to report Claude usage: ${result.error}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6606
|
+
});
|
|
6607
|
+
}
|
|
6608
|
+
scheduleNextTick();
|
|
6609
|
+
} catch (error2) {
|
|
6610
|
+
if (error2 instanceof ClaudeUsageError && isLocalCredentialProblem(error2)) {
|
|
6611
|
+
if (mode === "on") {
|
|
6612
|
+
logActivity(state, {
|
|
6613
|
+
type: "info",
|
|
6614
|
+
level: "warn",
|
|
6615
|
+
message: "Claude usage reporting is forced on but no usable Claude Code login was found \u2014 run `claude` to sign in; reporting will keep retrying"
|
|
6616
|
+
});
|
|
6617
|
+
scheduleNextTick();
|
|
6618
|
+
} else if (isProbe) {
|
|
6619
|
+
logActivity(state, {
|
|
6620
|
+
type: "info",
|
|
6621
|
+
level: "debug",
|
|
6622
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6623
|
+
});
|
|
6624
|
+
armed = false;
|
|
6625
|
+
if (rearmRequested) rearm();
|
|
6626
|
+
} else {
|
|
6627
|
+
logActivity(state, {
|
|
6628
|
+
type: "info",
|
|
6629
|
+
level: "debug",
|
|
6630
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6631
|
+
});
|
|
6632
|
+
scheduleNextTick();
|
|
6633
|
+
}
|
|
6634
|
+
} else {
|
|
6635
|
+
consecutiveFailures++;
|
|
6636
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
6637
|
+
logActivity(state, {
|
|
6638
|
+
type: "info",
|
|
6639
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6640
|
+
message: `Claude usage reporting failed: ${message}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6641
|
+
});
|
|
6642
|
+
scheduleNextTick();
|
|
6643
|
+
}
|
|
6644
|
+
}
|
|
6645
|
+
};
|
|
6646
|
+
armed = true;
|
|
6647
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6648
|
+
return rearm;
|
|
6649
|
+
}
|
|
5420
6650
|
async function notifyOffline(state) {
|
|
5421
6651
|
if (!state.agentId || !state.authHeader) return;
|
|
5422
6652
|
if (!state.connected) {
|
|
@@ -5434,13 +6664,29 @@ async function notifyOffline(state) {
|
|
|
5434
6664
|
if (state.interactive) displayStatus(state);
|
|
5435
6665
|
}
|
|
5436
6666
|
}
|
|
6667
|
+
async function timeShutdownPhase(state, durations, name, run2) {
|
|
6668
|
+
const startedAt = Date.now();
|
|
6669
|
+
try {
|
|
6670
|
+
return await run2();
|
|
6671
|
+
} finally {
|
|
6672
|
+
const elapsedMs = Date.now() - startedAt;
|
|
6673
|
+
durations[name] = elapsedMs;
|
|
6674
|
+
log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
|
|
6675
|
+
}
|
|
6676
|
+
}
|
|
5437
6677
|
async function cleanup(state, opts = {}) {
|
|
6678
|
+
const durations = {};
|
|
5438
6679
|
state.running = false;
|
|
5439
6680
|
for (const timer of state.sessionCleanupTimers) {
|
|
5440
6681
|
clearInterval(timer);
|
|
5441
6682
|
clearTimeout(timer);
|
|
5442
6683
|
}
|
|
5443
6684
|
state.sessionCleanupTimers = [];
|
|
6685
|
+
if (state.claudeUsageTimer) {
|
|
6686
|
+
clearTimeout(state.claudeUsageTimer);
|
|
6687
|
+
state.claudeUsageTimer = null;
|
|
6688
|
+
}
|
|
6689
|
+
state.claudeUsageRearm = null;
|
|
5444
6690
|
if (opts.graceful && state.channelDriver) {
|
|
5445
6691
|
state.channelDriver.stop();
|
|
5446
6692
|
log2(state, "Draining in-flight channel work before shutdown...");
|
|
@@ -5448,7 +6694,13 @@ async function cleanup(state, opts = {}) {
|
|
|
5448
6694
|
logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
|
|
5449
6695
|
displayStatus(state);
|
|
5450
6696
|
}
|
|
5451
|
-
const
|
|
6697
|
+
const driver = state.channelDriver;
|
|
6698
|
+
const settled = await timeShutdownPhase(
|
|
6699
|
+
state,
|
|
6700
|
+
durations,
|
|
6701
|
+
"drain",
|
|
6702
|
+
() => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
|
|
6703
|
+
);
|
|
5452
6704
|
if (!settled) {
|
|
5453
6705
|
logActivity(state, {
|
|
5454
6706
|
type: "info",
|
|
@@ -5457,13 +6709,15 @@ async function cleanup(state, opts = {}) {
|
|
|
5457
6709
|
if (state.interactive) displayStatus(state);
|
|
5458
6710
|
}
|
|
5459
6711
|
}
|
|
5460
|
-
await notifyOffline(state);
|
|
6712
|
+
await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
|
|
5461
6713
|
if (state.connection) {
|
|
5462
|
-
state.connection
|
|
6714
|
+
const connection = state.connection;
|
|
6715
|
+
await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
|
|
5463
6716
|
state.connection = null;
|
|
5464
6717
|
}
|
|
5465
6718
|
if (state.opencodeProcess) {
|
|
5466
|
-
|
|
6719
|
+
const opencodeProcess = state.opencodeProcess;
|
|
6720
|
+
await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
|
|
5467
6721
|
if (state.interactive) {
|
|
5468
6722
|
logActivity(state, { type: "info", message: "Stopped OpenCode process" });
|
|
5469
6723
|
displayStatus(state);
|
|
@@ -5472,6 +6726,7 @@ async function cleanup(state, opts = {}) {
|
|
|
5472
6726
|
}
|
|
5473
6727
|
state.opencodeProcess = null;
|
|
5474
6728
|
}
|
|
6729
|
+
return durations;
|
|
5475
6730
|
}
|
|
5476
6731
|
async function run(options) {
|
|
5477
6732
|
const interactive = isInteractive(options.json);
|
|
@@ -5479,7 +6734,7 @@ async function run(options) {
|
|
|
5479
6734
|
let fileSyncDirectories;
|
|
5480
6735
|
try {
|
|
5481
6736
|
logLevel = resolveLogLevel(options);
|
|
5482
|
-
fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo,
|
|
6737
|
+
fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir3());
|
|
5483
6738
|
} catch (error2) {
|
|
5484
6739
|
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
5485
6740
|
if (options.json) {
|
|
@@ -5512,8 +6767,11 @@ async function run(options) {
|
|
|
5512
6767
|
messageCount: 0,
|
|
5513
6768
|
lastProxiedActivityAt: null,
|
|
5514
6769
|
sessionCleanupTimers: [],
|
|
6770
|
+
claudeUsageTimer: null,
|
|
6771
|
+
claudeUsageRearm: null,
|
|
5515
6772
|
authHeader: ""
|
|
5516
6773
|
};
|
|
6774
|
+
setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
|
|
5517
6775
|
if (fileSyncDirectories.length > 0) {
|
|
5518
6776
|
log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
|
|
5519
6777
|
} else {
|
|
@@ -5542,14 +6800,38 @@ async function run(options) {
|
|
|
5542
6800
|
const handleSignal = async () => {
|
|
5543
6801
|
if (state.shuttingDown) return;
|
|
5544
6802
|
state.shuttingDown = true;
|
|
6803
|
+
const shutdownStartedAt = Date.now();
|
|
5545
6804
|
if (state.interactive) {
|
|
5546
6805
|
logActivity(state, { type: "info", message: "Shutting down..." });
|
|
5547
6806
|
displayStatus(state);
|
|
5548
6807
|
} else {
|
|
5549
6808
|
log2(state, "Shutting down...");
|
|
5550
6809
|
}
|
|
5551
|
-
await cleanup(state, { graceful: true });
|
|
5552
|
-
|
|
6810
|
+
const durations = await cleanup(state, { graceful: true });
|
|
6811
|
+
const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
|
|
6812
|
+
await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
|
|
6813
|
+
let timer;
|
|
6814
|
+
const flushed = shutdownTelemetry().then(
|
|
6815
|
+
() => true,
|
|
6816
|
+
(error2) => {
|
|
6817
|
+
log2(
|
|
6818
|
+
state,
|
|
6819
|
+
`Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
|
|
6820
|
+
"warn"
|
|
6821
|
+
);
|
|
6822
|
+
return true;
|
|
6823
|
+
}
|
|
6824
|
+
);
|
|
6825
|
+
const timedOut = new Promise((resolve3) => {
|
|
6826
|
+
timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
|
|
6827
|
+
});
|
|
6828
|
+
if (!await Promise.race([flushed, timedOut])) {
|
|
6829
|
+
log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
|
|
6830
|
+
}
|
|
6831
|
+
clearTimeout(timer);
|
|
6832
|
+
});
|
|
6833
|
+
const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
|
|
6834
|
+
log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
|
|
5553
6835
|
process.exit(0);
|
|
5554
6836
|
};
|
|
5555
6837
|
process.on("SIGINT", handleSignal);
|
|
@@ -5566,6 +6848,7 @@ async function run(options) {
|
|
|
5566
6848
|
console.log(chalk6.dim("Or run `evident login` for interactive authentication"));
|
|
5567
6849
|
blank();
|
|
5568
6850
|
process.exit(1);
|
|
6851
|
+
return;
|
|
5569
6852
|
}
|
|
5570
6853
|
blank();
|
|
5571
6854
|
console.log(chalk6.yellow("You are not logged in to Evident."));
|
|
@@ -5610,6 +6893,7 @@ async function run(options) {
|
|
|
5610
6893
|
} else {
|
|
5611
6894
|
printError(resolved.error || "Failed to resolve runner ID from key");
|
|
5612
6895
|
process.exit(1);
|
|
6896
|
+
return;
|
|
5613
6897
|
}
|
|
5614
6898
|
} else {
|
|
5615
6899
|
printError(
|
|
@@ -5623,6 +6907,7 @@ async function run(options) {
|
|
|
5623
6907
|
);
|
|
5624
6908
|
blank();
|
|
5625
6909
|
process.exit(1);
|
|
6910
|
+
return;
|
|
5626
6911
|
}
|
|
5627
6912
|
}
|
|
5628
6913
|
telemetry.info(
|
|
@@ -5663,40 +6948,67 @@ async function run(options) {
|
|
|
5663
6948
|
}
|
|
5664
6949
|
spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
|
|
5665
6950
|
state.agentName = validation.agent.name;
|
|
6951
|
+
const microvmId = process.env.MICROVM_ID?.trim();
|
|
6952
|
+
if (microvmId) {
|
|
6953
|
+
const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
|
|
6954
|
+
if (reported.ok) {
|
|
6955
|
+
log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
|
|
6956
|
+
} else {
|
|
6957
|
+
const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
|
|
6958
|
+
log2(state, message, "warn");
|
|
6959
|
+
if (state.interactive && !state.json) {
|
|
6960
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
6961
|
+
}
|
|
6962
|
+
}
|
|
6963
|
+
} else {
|
|
6964
|
+
log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
|
|
6965
|
+
}
|
|
6966
|
+
const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
|
|
6967
|
+
for (const warning2 of opencodeStartTimeoutWarnings) {
|
|
6968
|
+
logActivity(state, { type: "info", level: "warn", message: warning2 });
|
|
6969
|
+
}
|
|
5666
6970
|
const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
|
|
5667
6971
|
try {
|
|
5668
6972
|
const oc = await ensureOpenCodeRunning({
|
|
5669
6973
|
port: state.port,
|
|
5670
6974
|
interactive: state.interactive,
|
|
5671
6975
|
agentId: state.agentId,
|
|
5672
|
-
log: (message) => log2(state, message)
|
|
6976
|
+
log: (message) => log2(state, message),
|
|
6977
|
+
startTimeoutMs: opencodeStartTimeoutMs
|
|
5673
6978
|
});
|
|
5674
6979
|
state.port = oc.port;
|
|
5675
6980
|
state.opencodeProcess = oc.process;
|
|
5676
6981
|
state.opencodeVersion = oc.version;
|
|
5677
|
-
state.opencodeConnected = oc.
|
|
6982
|
+
state.opencodeConnected = oc.notReadyReason === null;
|
|
5678
6983
|
const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
|
|
5679
6984
|
ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
|
|
5680
|
-
|
|
5681
|
-
|
|
5682
|
-
|
|
5683
|
-
|
|
5684
|
-
|
|
6985
|
+
if (!state.interactive && oc.notReadyReason !== null) {
|
|
6986
|
+
const message = `OpenCode is not ready on port ${state.port}: ${oc.notReadyReason}. The runner will still come online, but messages will fail until opencode answers \u2014 raise the wait with --opencode-start-timeout <seconds> (env ${OPENCODE_START_TIMEOUT_ENV}).`;
|
|
6987
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
6988
|
+
} else {
|
|
6989
|
+
const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
|
|
6990
|
+
if (versionWarning) {
|
|
6991
|
+
log2(state, versionWarning, "warn");
|
|
6992
|
+
if (state.interactive && !state.json) {
|
|
6993
|
+
logActivity(state, { type: "info", level: "warn", message: versionWarning });
|
|
6994
|
+
}
|
|
5685
6995
|
}
|
|
5686
|
-
|
|
5687
|
-
|
|
5688
|
-
|
|
5689
|
-
|
|
5690
|
-
|
|
5691
|
-
|
|
5692
|
-
|
|
5693
|
-
|
|
5694
|
-
|
|
5695
|
-
|
|
5696
|
-
|
|
5697
|
-
|
|
5698
|
-
|
|
5699
|
-
|
|
6996
|
+
const noProviderWarning = buildNoProviderWarning(
|
|
6997
|
+
await hasAnyConfiguredProvider(state.port)
|
|
6998
|
+
);
|
|
6999
|
+
if (noProviderWarning) {
|
|
7000
|
+
log2(state, noProviderWarning, "warn");
|
|
7001
|
+
if (state.interactive && !state.json) {
|
|
7002
|
+
logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
|
|
7003
|
+
blank();
|
|
7004
|
+
console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
|
|
7005
|
+
console.log(
|
|
7006
|
+
chalk6.dim(
|
|
7007
|
+
`Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
|
|
7008
|
+
)
|
|
7009
|
+
);
|
|
7010
|
+
blank();
|
|
7011
|
+
}
|
|
5700
7012
|
}
|
|
5701
7013
|
}
|
|
5702
7014
|
} catch (error2) {
|
|
@@ -5714,7 +7026,7 @@ async function run(options) {
|
|
|
5714
7026
|
// #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
|
|
5715
7027
|
// REJECTED with `file_sync_disabled` on the ack, not silently ignored.
|
|
5716
7028
|
fileSyncDirectories,
|
|
5717
|
-
homeDir:
|
|
7029
|
+
homeDir: homedir3(),
|
|
5718
7030
|
log: (entry) => (
|
|
5719
7031
|
// Thread the driver's real level straight through so `debug`/`warn`
|
|
5720
7032
|
// survive the sink filter (they no longer collapse to info). `type`
|
|
@@ -5741,6 +7053,18 @@ async function run(options) {
|
|
|
5741
7053
|
type: "info",
|
|
5742
7054
|
message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
|
|
5743
7055
|
});
|
|
7056
|
+
if (options.tunnelReadyFile) {
|
|
7057
|
+
const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
|
|
7058
|
+
if (marker.ok) {
|
|
7059
|
+
log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
|
|
7060
|
+
} else {
|
|
7061
|
+
log2(
|
|
7062
|
+
state,
|
|
7063
|
+
`Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
|
|
7064
|
+
"error"
|
|
7065
|
+
);
|
|
7066
|
+
}
|
|
7067
|
+
}
|
|
5744
7068
|
emitAgentConnected(state.agentId, {
|
|
5745
7069
|
port: state.port,
|
|
5746
7070
|
cli_version: getCliVersion(),
|
|
@@ -5831,6 +7155,7 @@ async function run(options) {
|
|
|
5831
7155
|
throw error2;
|
|
5832
7156
|
}
|
|
5833
7157
|
scheduleSessionCleanup(state, channelDriver, options);
|
|
7158
|
+
state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
|
|
5834
7159
|
if (!interactive || state.json) {
|
|
5835
7160
|
log2(state, "Driving channel messages...");
|
|
5836
7161
|
}
|
|
@@ -5885,13 +7210,18 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
|
|
|
5885
7210
|
program.command("login").description("Authenticate with Evident").option("--token", "Use token-based authentication (for CI/CD)").option("--no-browser", "Do not open the browser automatically").action(login);
|
|
5886
7211
|
program.command("logout").description("Remove stored credentials for the current endpoint").option("--all", "Remove stored credentials for all endpoints").action((options) => logout({ all: options.all }));
|
|
5887
7212
|
program.command("whoami").description("Show the currently logged in user").action(whoami);
|
|
7213
|
+
program.command("status").description("Check whether the configured credentials can reach Evident").option("--json", "Output in JSON format").action((options) => status({ json: options.json }));
|
|
7214
|
+
program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
|
|
5888
7215
|
program.command("run").description("Connect to Evident and process messages").option("--runner [id]", "Runner ID to connect to (optional when EVIDENT_RUNNER_KEY is set)").option(
|
|
5889
7216
|
"-a, --agent [id]",
|
|
5890
7217
|
"Deprecated alias for --runner (still supported; --runner wins if both are given)"
|
|
5891
7218
|
).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
|
|
5892
7219
|
"--log-level <level>",
|
|
5893
7220
|
"Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
|
|
5894
|
-
).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option(
|
|
7221
|
+
).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option(
|
|
7222
|
+
"--opencode-start-timeout <seconds>",
|
|
7223
|
+
"Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
|
|
7224
|
+
).option("--json", "Output in JSON format").option(
|
|
5895
7225
|
"--session-cleanup-max-age <duration>",
|
|
5896
7226
|
"Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
|
|
5897
7227
|
).option(
|
|
@@ -5900,11 +7230,17 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5900
7230
|
).option(
|
|
5901
7231
|
"--session-cleanup-interval <duration>",
|
|
5902
7232
|
"How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
|
|
7233
|
+
).option(
|
|
7234
|
+
"--claude-usage-reporting <mode>",
|
|
7235
|
+
"Report Claude subscription usage to Evident: auto | on | off (default: auto). Env: EVIDENT_CLAUDE_USAGE_REPORTING"
|
|
5903
7236
|
).option(
|
|
5904
7237
|
"--enable-file-sync-to <dir>",
|
|
5905
7238
|
"Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
|
|
5906
7239
|
(value, previous) => previous.concat([value]),
|
|
5907
7240
|
[]
|
|
7241
|
+
).option(
|
|
7242
|
+
"--tunnel-ready-file <path>",
|
|
7243
|
+
"Path to write once the tunnel is connected (opt-in; unused unless an operator/image sets it)"
|
|
5908
7244
|
).action(
|
|
5909
7245
|
(options) => {
|
|
5910
7246
|
run({
|
|
@@ -5917,14 +7253,21 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5917
7253
|
verbose: options.verbose,
|
|
5918
7254
|
conversation: options.conversation,
|
|
5919
7255
|
idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
|
|
7256
|
+
// Raw string — validation/precedence is single-sourced in run.ts's
|
|
7257
|
+
// resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
|
|
7258
|
+
opencodeStartTimeout: options.opencodeStartTimeout,
|
|
5920
7259
|
json: options.json,
|
|
5921
7260
|
// Raw strings — the resolver in run.ts single-sources parsing (M1).
|
|
5922
7261
|
sessionCleanupMaxAge: options.sessionCleanupMaxAge,
|
|
5923
7262
|
sessionCleanupMaxCount: options.sessionCleanupMaxCount,
|
|
5924
7263
|
sessionCleanupInterval: options.sessionCleanupInterval,
|
|
7264
|
+
// Raw string — the resolver in run.ts single-sources parsing
|
|
7265
|
+
// (resolveClaudeUsageReportingMode).
|
|
7266
|
+
claudeUsageReporting: options.claudeUsageReporting,
|
|
5925
7267
|
// Raw values — expansion/validation is single-sourced in run.ts's
|
|
5926
7268
|
// resolveFileSyncDirectories.
|
|
5927
|
-
enableFileSyncTo: options.enableFileSyncTo
|
|
7269
|
+
enableFileSyncTo: options.enableFileSyncTo,
|
|
7270
|
+
tunnelReadyFile: options.tunnelReadyFile
|
|
5928
7271
|
});
|
|
5929
7272
|
}
|
|
5930
7273
|
);
|