@evident-ai/cli 3.1.1-dev.38348be → 3.1.1-dev.391a05e
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 +1877 -394
- 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
|
}
|
|
@@ -466,43 +498,508 @@ async function logout(options = {}) {
|
|
|
466
498
|
printWarning(`You are not logged in to ${getApiUrlConfig()}.`);
|
|
467
499
|
return;
|
|
468
500
|
}
|
|
469
|
-
await deleteToken();
|
|
470
|
-
printSuccess(`Logged out of ${getApiUrlConfig()}.`);
|
|
501
|
+
await deleteToken();
|
|
502
|
+
printSuccess(`Logged out of ${getApiUrlConfig()}.`);
|
|
503
|
+
}
|
|
504
|
+
|
|
505
|
+
// src/commands/whoami.ts
|
|
506
|
+
import chalk3 from "chalk";
|
|
507
|
+
async function whoami() {
|
|
508
|
+
const apiUrl = getApiUrlConfig();
|
|
509
|
+
const credentials2 = await getToken();
|
|
510
|
+
if (!credentials2) {
|
|
511
|
+
printError(`Not logged in to ${apiUrl}. Run the \`login\` command to authenticate.`);
|
|
512
|
+
process.exit(1);
|
|
513
|
+
}
|
|
514
|
+
blank();
|
|
515
|
+
console.log(keyValue("Endpoint", apiUrl));
|
|
516
|
+
console.log(keyValue("User", chalk3.bold(credentials2.user.email)));
|
|
517
|
+
if (credentials2.user.id) {
|
|
518
|
+
console.log(keyValue("User ID", credentials2.user.id));
|
|
519
|
+
}
|
|
520
|
+
if (credentials2.expiresAt) {
|
|
521
|
+
const expiresAt = new Date(credentials2.expiresAt);
|
|
522
|
+
const now = /* @__PURE__ */ new Date();
|
|
523
|
+
if (expiresAt < now) {
|
|
524
|
+
console.log(keyValue("Status", chalk3.red("Token expired")));
|
|
525
|
+
} else {
|
|
526
|
+
const daysRemaining = Math.ceil(
|
|
527
|
+
(expiresAt.getTime() - now.getTime()) / (1e3 * 60 * 60 * 24)
|
|
528
|
+
);
|
|
529
|
+
console.log(keyValue("Expires", `${daysRemaining} days`));
|
|
530
|
+
}
|
|
531
|
+
}
|
|
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
|
+
};
|
|
471
969
|
}
|
|
472
970
|
|
|
473
|
-
// src/commands/
|
|
474
|
-
|
|
475
|
-
|
|
476
|
-
|
|
477
|
-
const credentials2 = await getToken();
|
|
478
|
-
if (!credentials2) {
|
|
479
|
-
printError(`Not logged in to ${apiUrl}. Run the \`login\` command to authenticate.`);
|
|
480
|
-
process.exit(1);
|
|
971
|
+
// src/commands/claude-usage.ts
|
|
972
|
+
function formatWindow(label, window) {
|
|
973
|
+
if (!window) {
|
|
974
|
+
return keyValue(label, "not available for this plan");
|
|
481
975
|
}
|
|
482
|
-
|
|
483
|
-
|
|
484
|
-
|
|
485
|
-
|
|
486
|
-
|
|
487
|
-
const
|
|
488
|
-
|
|
489
|
-
|
|
490
|
-
|
|
491
|
-
|
|
492
|
-
|
|
493
|
-
|
|
494
|
-
);
|
|
495
|
-
|
|
976
|
+
const resetsAt = new Date(window.resetsAt);
|
|
977
|
+
return keyValue(label, `${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);
|
|
496
990
|
}
|
|
991
|
+
throw err;
|
|
497
992
|
}
|
|
498
|
-
blank();
|
|
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 +2089,57 @@ function messageError(messages, userMessageId) {
|
|
|
1545
2089
|
}
|
|
1546
2090
|
return "The agent run failed.";
|
|
1547
2091
|
}
|
|
2092
|
+
function isAbortedTerminalReply(messages, userMessageId) {
|
|
2093
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
2094
|
+
const error2 = errorOf(reply);
|
|
2095
|
+
if (error2 == null) return false;
|
|
2096
|
+
if (typeof error2 === "string") return error2.trim() === "Aborted";
|
|
2097
|
+
if (typeof error2 === "object") {
|
|
2098
|
+
const e = error2;
|
|
2099
|
+
if (e.name === "MessageAbortedError") return true;
|
|
2100
|
+
if (e.name === "AbortError") return true;
|
|
2101
|
+
const dataMessage = e.data?.message;
|
|
2102
|
+
const rendered = typeof dataMessage === "string" ? dataMessage : typeof e.message === "string" ? e.message : null;
|
|
2103
|
+
return rendered != null && rendered.trim() === "Aborted";
|
|
2104
|
+
}
|
|
2105
|
+
return false;
|
|
2106
|
+
}
|
|
2107
|
+
function messageFailure(messages, userMessageId) {
|
|
2108
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
2109
|
+
const error2 = errorOf(reply);
|
|
2110
|
+
if (error2 == null || typeof error2 !== "object") return null;
|
|
2111
|
+
const e = error2;
|
|
2112
|
+
const replyProviderId = reply?.info?.providerID ?? null;
|
|
2113
|
+
const replyModelId = reply?.info?.modelID ?? null;
|
|
2114
|
+
if (e.name === "ProviderAuthError") {
|
|
2115
|
+
const data = e.data;
|
|
2116
|
+
const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
|
|
2117
|
+
return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
|
|
2118
|
+
}
|
|
2119
|
+
if (e.name === "APIError") {
|
|
2120
|
+
const data = e.data;
|
|
2121
|
+
const statusCode = data?.statusCode;
|
|
2122
|
+
if (statusCode === 401 || statusCode === 403) {
|
|
2123
|
+
return {
|
|
2124
|
+
kind: "model_auth",
|
|
2125
|
+
providerId: replyProviderId,
|
|
2126
|
+
modelId: replyModelId,
|
|
2127
|
+
reason: "rejected"
|
|
2128
|
+
};
|
|
2129
|
+
}
|
|
2130
|
+
}
|
|
2131
|
+
return null;
|
|
2132
|
+
}
|
|
2133
|
+
function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
|
|
2134
|
+
if (classified != null) return classified;
|
|
2135
|
+
if (hasConfiguredProvider !== false) return null;
|
|
2136
|
+
return {
|
|
2137
|
+
kind: "model_auth",
|
|
2138
|
+
providerId: replyProviderId,
|
|
2139
|
+
modelId: replyModelId,
|
|
2140
|
+
reason: "missing"
|
|
2141
|
+
};
|
|
2142
|
+
}
|
|
1548
2143
|
function hasRunningAssistantExcept(messages, exceptUserMessageId) {
|
|
1549
2144
|
if (!messages || messages.length === 0) return false;
|
|
1550
2145
|
return messages.some(
|
|
@@ -2073,13 +2668,56 @@ var RunnerConnection = class {
|
|
|
2073
2668
|
}
|
|
2074
2669
|
};
|
|
2075
2670
|
|
|
2671
|
+
// src/lib/tunnel/ready-marker.ts
|
|
2672
|
+
import { writeFileSync } from "fs";
|
|
2673
|
+
function writeTunnelReadyMarker(path, agentId) {
|
|
2674
|
+
try {
|
|
2675
|
+
writeFileSync(path, `${agentId}
|
|
2676
|
+
`);
|
|
2677
|
+
return { ok: true };
|
|
2678
|
+
} catch (error2) {
|
|
2679
|
+
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
2680
|
+
}
|
|
2681
|
+
}
|
|
2682
|
+
|
|
2683
|
+
// src/lib/claude-usage-reporting.ts
|
|
2684
|
+
var VALID_MODES = ["auto", "on", "off"];
|
|
2685
|
+
function resolveClaudeUsageReportingMode(flagValue, env) {
|
|
2686
|
+
const raw = flagValue ?? env.EVIDENT_CLAUDE_USAGE_REPORTING;
|
|
2687
|
+
if (raw === void 0 || raw === "") {
|
|
2688
|
+
return { mode: "auto", warnings: [] };
|
|
2689
|
+
}
|
|
2690
|
+
const normalized = raw.trim().toLowerCase();
|
|
2691
|
+
if (VALID_MODES.includes(normalized)) {
|
|
2692
|
+
return { mode: normalized, warnings: [] };
|
|
2693
|
+
}
|
|
2694
|
+
const source = flagValue !== void 0 ? "--claude-usage-reporting" : "EVIDENT_CLAUDE_USAGE_REPORTING";
|
|
2695
|
+
return {
|
|
2696
|
+
mode: "auto",
|
|
2697
|
+
warnings: [
|
|
2698
|
+
`Ignoring invalid ${source} "${raw}": expected one of ${VALID_MODES.join(", ")}; using auto`
|
|
2699
|
+
]
|
|
2700
|
+
};
|
|
2701
|
+
}
|
|
2702
|
+
var BASE_REPORT_DELAY_MS = 10 * 6e4;
|
|
2703
|
+
var REPORT_DELAY_JITTER_FRACTION = 0.2;
|
|
2704
|
+
function nextReportDelayMs(random = Math.random) {
|
|
2705
|
+
const jitterRangeMs = BASE_REPORT_DELAY_MS * REPORT_DELAY_JITTER_FRACTION;
|
|
2706
|
+
return BASE_REPORT_DELAY_MS - jitterRangeMs + random() * (2 * jitterRangeMs);
|
|
2707
|
+
}
|
|
2708
|
+
var FIRST_REPORT_DELAY_MS = 5e3 + Math.random() * 1e4;
|
|
2709
|
+
var CLAUDE_USAGE_FAILURE_REESCALATION_TICKS = 6;
|
|
2710
|
+
function claudeUsageFailureLogLevel(consecutiveFailures) {
|
|
2711
|
+
return consecutiveFailures === 1 || consecutiveFailures % CLAUDE_USAGE_FAILURE_REESCALATION_TICKS === 0 ? "warn" : "debug";
|
|
2712
|
+
}
|
|
2713
|
+
|
|
2076
2714
|
// src/lib/channels/driver.ts
|
|
2077
|
-
import { homedir } from "os";
|
|
2715
|
+
import { homedir as homedir2 } from "os";
|
|
2078
2716
|
|
|
2079
2717
|
// src/lib/file-push.ts
|
|
2080
2718
|
import { randomUUID } from "crypto";
|
|
2081
2719
|
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";
|
|
2720
|
+
import { basename, dirname as dirname2, isAbsolute, join as join2, relative, resolve as resolve2, sep } from "path";
|
|
2083
2721
|
var FILE_MODE = 384;
|
|
2084
2722
|
var DIRECTORY_MODE = 448;
|
|
2085
2723
|
async function writePushedFile(request) {
|
|
@@ -2112,7 +2750,7 @@ async function writePushedFile(request) {
|
|
|
2112
2750
|
const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
|
|
2113
2751
|
dirname2(candidate)
|
|
2114
2752
|
);
|
|
2115
|
-
const realTarget =
|
|
2753
|
+
const realTarget = join2(existingAncestor, ...missingSegments, basename(candidate));
|
|
2116
2754
|
const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
|
|
2117
2755
|
if (allowedDirectory === null) {
|
|
2118
2756
|
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
@@ -2148,7 +2786,7 @@ function expandAndValidate(requestedPath, homeDir) {
|
|
|
2148
2786
|
if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
|
|
2149
2787
|
return null;
|
|
2150
2788
|
}
|
|
2151
|
-
const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ?
|
|
2789
|
+
const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join2(homeDir, requestedPath.slice(2)) : requestedPath;
|
|
2152
2790
|
if (expanded.split(/[/\\]/).includes("..")) {
|
|
2153
2791
|
return null;
|
|
2154
2792
|
}
|
|
@@ -2221,13 +2859,13 @@ function contains(realDirectory, realTarget) {
|
|
|
2221
2859
|
async function createMissingDirectories(existingAncestor, missingSegments) {
|
|
2222
2860
|
let current = existingAncestor;
|
|
2223
2861
|
for (const segment of missingSegments) {
|
|
2224
|
-
current =
|
|
2862
|
+
current = join2(current, segment);
|
|
2225
2863
|
await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
|
|
2226
2864
|
await chmod(current, DIRECTORY_MODE);
|
|
2227
2865
|
}
|
|
2228
2866
|
}
|
|
2229
2867
|
async function writeAtomically(realTarget, content) {
|
|
2230
|
-
const temporaryPath =
|
|
2868
|
+
const temporaryPath = join2(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
|
|
2231
2869
|
let handle;
|
|
2232
2870
|
try {
|
|
2233
2871
|
handle = await open2(temporaryPath, "wx", FILE_MODE);
|
|
@@ -2394,8 +3032,8 @@ async function applyOne(options, file) {
|
|
|
2394
3032
|
await ack(options, file, "applied");
|
|
2395
3033
|
return true;
|
|
2396
3034
|
}
|
|
2397
|
-
function durableDownloadCode(
|
|
2398
|
-
return
|
|
3035
|
+
function durableDownloadCode(status2) {
|
|
3036
|
+
return status2 === 413 ? "file_too_large" : "write_failed";
|
|
2399
3037
|
}
|
|
2400
3038
|
async function downloadContent(options, file, label) {
|
|
2401
3039
|
try {
|
|
@@ -2428,8 +3066,8 @@ async function downloadContent(options, file, label) {
|
|
|
2428
3066
|
return { ok: false, terminal: false };
|
|
2429
3067
|
}
|
|
2430
3068
|
}
|
|
2431
|
-
async function ack(options, file,
|
|
2432
|
-
const outcome = `${
|
|
3069
|
+
async function ack(options, file, status2, reason) {
|
|
3070
|
+
const outcome = `${status2}${reason ? ` (${reason})` : ""}`;
|
|
2433
3071
|
try {
|
|
2434
3072
|
const res = await options.fetchImpl(
|
|
2435
3073
|
`${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
|
|
@@ -2439,7 +3077,7 @@ async function ack(options, file, status, reason) {
|
|
|
2439
3077
|
Authorization: options.getAuthHeader(),
|
|
2440
3078
|
"Content-Type": "application/json"
|
|
2441
3079
|
},
|
|
2442
|
-
body: JSON.stringify(reason ? { status, reason } : { status })
|
|
3080
|
+
body: JSON.stringify(reason ? { status: status2, reason } : { status: status2 })
|
|
2443
3081
|
}
|
|
2444
3082
|
);
|
|
2445
3083
|
if (!res.ok) {
|
|
@@ -2499,8 +3137,11 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
|
|
|
2499
3137
|
var DEFAULT_STUCK_QUEUED_MS = 6e4;
|
|
2500
3138
|
var HEARTBEAT_MS = 6e4;
|
|
2501
3139
|
var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
|
|
3140
|
+
var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
|
|
3141
|
+
var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
|
|
2502
3142
|
var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
|
|
2503
3143
|
var MAX_SUPERSEDED_CONVERSATIONS = 256;
|
|
3144
|
+
var MAX_IDENTICAL_REDRIVE_POLL_FAILURES = 5;
|
|
2504
3145
|
var ChannelAuthError = class extends Error {
|
|
2505
3146
|
constructor(message) {
|
|
2506
3147
|
super(message);
|
|
@@ -2509,10 +3150,10 @@ var ChannelAuthError = class extends Error {
|
|
|
2509
3150
|
};
|
|
2510
3151
|
var ChannelTerminalError = class extends Error {
|
|
2511
3152
|
status;
|
|
2512
|
-
constructor(message,
|
|
3153
|
+
constructor(message, status2) {
|
|
2513
3154
|
super(message);
|
|
2514
3155
|
this.name = "ChannelTerminalError";
|
|
2515
|
-
this.status =
|
|
3156
|
+
this.status = status2;
|
|
2516
3157
|
}
|
|
2517
3158
|
};
|
|
2518
3159
|
function backoffDelay(attempt, policy) {
|
|
@@ -2520,8 +3161,12 @@ function backoffDelay(attempt, policy) {
|
|
|
2520
3161
|
const capped = Math.min(policy.maxDelayMs, exp);
|
|
2521
3162
|
return Math.floor(Math.random() * capped);
|
|
2522
3163
|
}
|
|
2523
|
-
function isRetryableStatus(
|
|
2524
|
-
return
|
|
3164
|
+
function isRetryableStatus(status2) {
|
|
3165
|
+
return status2 === 429 || status2 >= 500 && status2 <= 599;
|
|
3166
|
+
}
|
|
3167
|
+
var VOLATILE_BODY_FIELD_PATTERN = /("(?:ref|requestId|request_id|traceId|trace_id)"\s*:\s*)"[^"]*"/gi;
|
|
3168
|
+
function normalizeRedrivePollFailureBody(body) {
|
|
3169
|
+
return body.replace(VOLATILE_BODY_FIELD_PATTERN, '$1"<redacted>"').replace(/\s+/g, " ").trim().slice(0, 200);
|
|
2525
3170
|
}
|
|
2526
3171
|
var ChannelDriver = class _ChannelDriver {
|
|
2527
3172
|
agentId;
|
|
@@ -2636,6 +3281,33 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2636
3281
|
* processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
|
|
2637
3282
|
*/
|
|
2638
3283
|
readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
3284
|
+
/**
|
|
3285
|
+
* "Already emitted `redrive_unresolved` for this row" (#965). Mirrors
|
|
3286
|
+
* `readoptPollUnresolvedSignalled`: `resolveRedrive`'s `unresolved` leaf recurs
|
|
3287
|
+
* every ~2s drain until opencode's status becomes readable, but the
|
|
3288
|
+
* server-visible signal is an OUTCOME, so it fires at most once per row. Cleared
|
|
3289
|
+
* on any non-`unresolved` outcome so the set cannot grow beyond the currently
|
|
3290
|
+
* unresolvable rows.
|
|
3291
|
+
*/
|
|
3292
|
+
redriveUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
3293
|
+
/**
|
|
3294
|
+
* First `now()` a `pending` row's re-drive was observed `unresolved` (#965). A
|
|
3295
|
+
* `pending` row is invisible to every cron arm (all require `status =
|
|
3296
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
3297
|
+
* nothing driving it. Once `now - since >= pausedMaxWaitMs`, `resolveRedrive`
|
|
3298
|
+
* takes `dispatch` instead of `unresolved` (reusing the existing knob — see
|
|
3299
|
+
* ADR-0047's own "unreachable ⇒ bounded" rule). Cleared on any other outcome.
|
|
3300
|
+
*/
|
|
3301
|
+
redriveUnresolvedSince = /* @__PURE__ */ new Map();
|
|
3302
|
+
/**
|
|
3303
|
+
* Consecutive-identical-poll-failure streak for the re-drive fence (#1348),
|
|
3304
|
+
* keyed by Evident **message id** (not session) so `clearRedriveUnresolved`
|
|
3305
|
+
* can drop it with the other two trackers and it cannot leak. `sessionId` is
|
|
3306
|
+
* carried inside the entry, not the key: a session change is a different
|
|
3307
|
+
* situation and resets the streak, which gives the `(sessionId, message.id)`
|
|
3308
|
+
* pairing #1348 asks for without a composite map key.
|
|
3309
|
+
*/
|
|
3310
|
+
redrivePollFailures = /* @__PURE__ */ new Map();
|
|
2639
3311
|
/**
|
|
2640
3312
|
* "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
|
|
2641
3313
|
* Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
|
|
@@ -2724,23 +3396,23 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2724
3396
|
* and stops opencode.
|
|
2725
3397
|
*/
|
|
2726
3398
|
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 =
|
|
3399
|
+
constructor(config) {
|
|
3400
|
+
this.agentId = config.agentId;
|
|
3401
|
+
this.port = config.port;
|
|
3402
|
+
this.apiUrl = config.apiUrl.replace(/\/$/, "");
|
|
3403
|
+
this.getAuthHeader = config.getAuthHeader;
|
|
3404
|
+
this.conversationFilter = config.conversationFilter ?? null;
|
|
3405
|
+
this.retry = { ...DEFAULT_RETRY_POLICY, ...config.retry };
|
|
3406
|
+
this.log = config.log ?? (() => {
|
|
2735
3407
|
});
|
|
2736
|
-
this.fetchImpl =
|
|
2737
|
-
this.sleep =
|
|
2738
|
-
this.pausedPollIntervalMs =
|
|
2739
|
-
this.pausedMaxWaitMs =
|
|
2740
|
-
this.stuckQueuedMs =
|
|
2741
|
-
this.now =
|
|
2742
|
-
this.fileSyncDirectories =
|
|
2743
|
-
this.homeDir =
|
|
3408
|
+
this.fetchImpl = config.fetchImpl ?? fetch;
|
|
3409
|
+
this.sleep = config.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
|
|
3410
|
+
this.pausedPollIntervalMs = config.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
|
|
3411
|
+
this.pausedMaxWaitMs = config.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
|
|
3412
|
+
this.stuckQueuedMs = config.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
|
|
3413
|
+
this.now = config.now ?? (() => Date.now());
|
|
3414
|
+
this.fileSyncDirectories = config.fileSyncDirectories ?? [];
|
|
3415
|
+
this.homeDir = config.homeDir ?? homedir2();
|
|
2744
3416
|
}
|
|
2745
3417
|
/** The IPv4-loopback base URL for the local `opencode serve`. */
|
|
2746
3418
|
get opencodeBase() {
|
|
@@ -2975,6 +3647,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2975
3647
|
skippedAlreadyDispatched += 1;
|
|
2976
3648
|
continue;
|
|
2977
3649
|
}
|
|
3650
|
+
if (message.opencode_message_id) {
|
|
3651
|
+
const outcome = await this.resolveRedrive(conv, sessionId, message, refusedSessionId);
|
|
3652
|
+
if (outcome !== "dispatch") {
|
|
3653
|
+
break;
|
|
3654
|
+
}
|
|
3655
|
+
}
|
|
2978
3656
|
const options = {
|
|
2979
3657
|
agent: message.opencode_agent ?? void 0,
|
|
2980
3658
|
model: message.opencode_model ?? void 0
|
|
@@ -3019,50 +3697,361 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3019
3697
|
this.sessions.delete(conv.id);
|
|
3020
3698
|
this.supersede(conv.id, sessionId);
|
|
3021
3699
|
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.`,
|
|
3700
|
+
level: "warn",
|
|
3701
|
+
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.`,
|
|
3702
|
+
conversation_id: conv.id,
|
|
3703
|
+
message_id: message.id
|
|
3704
|
+
});
|
|
3705
|
+
await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
|
|
3706
|
+
this.log({
|
|
3707
|
+
level: "warn",
|
|
3708
|
+
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)}`,
|
|
3709
|
+
conversation_id: conv.id,
|
|
3710
|
+
message_id: message.id
|
|
3711
|
+
});
|
|
3712
|
+
});
|
|
3713
|
+
this.log({
|
|
3714
|
+
level: "error",
|
|
3715
|
+
message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
|
|
3716
|
+
conversation_id: conv.id,
|
|
3717
|
+
message_id: message.id
|
|
3718
|
+
});
|
|
3719
|
+
break;
|
|
3720
|
+
}
|
|
3721
|
+
if (opencodeMessageId === null) {
|
|
3722
|
+
this.log({
|
|
3723
|
+
level: "warn",
|
|
3724
|
+
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`,
|
|
3725
|
+
conversation_id: conv.id,
|
|
3726
|
+
message_id: message.id
|
|
3727
|
+
});
|
|
3728
|
+
continue;
|
|
3729
|
+
}
|
|
3730
|
+
this.dispatched.add(message.id);
|
|
3731
|
+
this.registerInFlight(conv, sessionId, message, opencodeMessageId);
|
|
3732
|
+
dispatched += 1;
|
|
3733
|
+
void this.postSignal(conv.id, message.id, "dispatched");
|
|
3734
|
+
}
|
|
3735
|
+
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3736
|
+
this.log({
|
|
3737
|
+
level: "warn",
|
|
3738
|
+
message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
|
|
3739
|
+
conversation_id: conv.id
|
|
3740
|
+
});
|
|
3741
|
+
}
|
|
3742
|
+
this.ensureWatcherRunning(sessionId);
|
|
3743
|
+
return dispatched;
|
|
3744
|
+
}
|
|
3745
|
+
/**
|
|
3746
|
+
* Poll a session's message list for the re-drive fence (#965), via the
|
|
3747
|
+
* INJECTED `fetchImpl` — NOT the imported `getSessionMessages` helper, which
|
|
3748
|
+
* hits the global `fetch` and would bypass the same override every other
|
|
3749
|
+
* opencode poll in this file respects. Mirrors `readoptProcessing`'s own
|
|
3750
|
+
* snapshot fetch (`:3081-3111`).
|
|
3751
|
+
*
|
|
3752
|
+
* Returns `{ ok: true, messages }` on a readable snapshot, or
|
|
3753
|
+
* `{ ok: false, signature }` on failure — `signature` is a string that
|
|
3754
|
+
* repeats across attempts for the SAME underlying fault (used by the
|
|
3755
|
+
* consecutive-identical-failure bound, #1348), or `null` for a thrown
|
|
3756
|
+
* exception, which is NOT countable toward that bound (a network blip / an
|
|
3757
|
+
* opencode restart also throws identically every tick, and must keep
|
|
3758
|
+
* retrying unbounded rather than ever being treated as permanent).
|
|
3759
|
+
*/
|
|
3760
|
+
async pollSessionMessagesForRedrive(conv, message, sessionId) {
|
|
3761
|
+
try {
|
|
3762
|
+
const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
|
|
3763
|
+
if (!res.ok) {
|
|
3764
|
+
const rawBody = await res.text();
|
|
3765
|
+
const normalized = normalizeRedrivePollFailureBody(rawBody);
|
|
3766
|
+
this.log({
|
|
3767
|
+
level: "warn",
|
|
3768
|
+
message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned HTTP ${res.status}${normalized ? `: ${normalized}` : ""} \u2014 treating as unreadable this tick`,
|
|
3769
|
+
conversation_id: conv.id,
|
|
3770
|
+
message_id: message.id
|
|
3771
|
+
});
|
|
3772
|
+
return { ok: false, signature: `HTTP ${res.status}${normalized ? `: ${normalized}` : ""}` };
|
|
3773
|
+
}
|
|
3774
|
+
const body = await res.json();
|
|
3775
|
+
if (!Array.isArray(body)) {
|
|
3776
|
+
this.log({
|
|
3777
|
+
level: "warn",
|
|
3778
|
+
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`,
|
|
3779
|
+
conversation_id: conv.id,
|
|
3780
|
+
message_id: message.id
|
|
3781
|
+
});
|
|
3782
|
+
return { ok: false, signature: "non-array message body" };
|
|
3783
|
+
}
|
|
3784
|
+
return { ok: true, messages: body };
|
|
3785
|
+
} catch (err) {
|
|
3786
|
+
this.log({
|
|
3787
|
+
level: "warn",
|
|
3788
|
+
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)}`,
|
|
3789
|
+
conversation_id: conv.id,
|
|
3790
|
+
message_id: message.id
|
|
3791
|
+
});
|
|
3792
|
+
return { ok: false, signature: null };
|
|
3793
|
+
}
|
|
3794
|
+
}
|
|
3795
|
+
/**
|
|
3796
|
+
* The re-drive fence for a `pending` row that already carries a stored
|
|
3797
|
+
* `opencode_message_id` (#965) — i.e. it has already been handed to opencode at
|
|
3798
|
+
* least once (see the invariant at `QueuedMessage.opencode_message_id`'s doc).
|
|
3799
|
+
* The lifecycle cron can falsely reclaim a `processing` row back to `pending`
|
|
3800
|
+
* mid-turn (a 5-minute liveness-staleness check racing a still-running turn);
|
|
3801
|
+
* without this fence the drain loop would re-`prompt_async` the SAME turn a
|
|
3802
|
+
* second time against live GitHub state. Mirrors `readoptOne`'s job for the
|
|
3803
|
+
* `processing` re-adopt path, but simpler: no b1/b2 preamble cross-check is
|
|
3804
|
+
* needed here because `refusedSessionId` already handles the one case
|
|
3805
|
+
* (#553 abandoned session) that path exists for.
|
|
3806
|
+
*
|
|
3807
|
+
* Only `ChannelAuthError` propagates. A poll that fails identically
|
|
3808
|
+
* `MAX_IDENTICAL_REDRIVE_POLL_FAILURES` times in a row reports the message
|
|
3809
|
+
* failed instead of retrying it (#1348) — SEPARATE from, not a replacement
|
|
3810
|
+
* for, `resolveRedriveUnresolved`'s own `pausedMaxWaitMs` bound below. Every
|
|
3811
|
+
* other failure resolves to `unresolved` and is retried whole on the next
|
|
3812
|
+
* ~2s drain tick.
|
|
3813
|
+
*/
|
|
3814
|
+
async resolveRedrive(conv, sessionId, message, refusedSessionId) {
|
|
3815
|
+
const ocId = message.opencode_message_id ?? null;
|
|
3816
|
+
if (refusedSessionId) {
|
|
3817
|
+
this.clearRedriveUnresolved(message.id);
|
|
3818
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3819
|
+
return "dispatch";
|
|
3820
|
+
}
|
|
3821
|
+
const polled = await this.pollSessionMessagesForRedrive(conv, message, sessionId);
|
|
3822
|
+
if (!polled.ok) {
|
|
3823
|
+
const streak = this.recordRedrivePollFailure(message.id, sessionId, polled.signature);
|
|
3824
|
+
if (streak >= MAX_IDENTICAL_REDRIVE_POLL_FAILURES && polled.signature !== null) {
|
|
3825
|
+
return this.failRedrivePollPermanent(conv, sessionId, message, polled.signature, streak);
|
|
3826
|
+
}
|
|
3827
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3828
|
+
}
|
|
3829
|
+
this.redrivePollFailures.delete(message.id);
|
|
3830
|
+
const messages = polled.messages;
|
|
3831
|
+
if (messages.length === 0) {
|
|
3832
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3833
|
+
}
|
|
3834
|
+
const state = messageRunState(messages, ocId ?? "");
|
|
3835
|
+
if (state === "failed" && isAbortedTerminalReply(messages, ocId ?? "")) {
|
|
3836
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3837
|
+
if (ongoing === false) {
|
|
3838
|
+
this.log({
|
|
3839
|
+
level: "info",
|
|
3840
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} carries an abort-shaped terminal error and session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status \u2014 restart orphan, re-dispatching instead of marking it permanently failed`,
|
|
3841
|
+
conversation_id: conv.id,
|
|
3842
|
+
message_id: message.id
|
|
3843
|
+
});
|
|
3844
|
+
this.clearRedriveUnresolved(message.id);
|
|
3845
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3846
|
+
return "dispatch";
|
|
3847
|
+
}
|
|
3848
|
+
}
|
|
3849
|
+
if (state === "done" || state === "failed") {
|
|
3850
|
+
return this.settleRedrive(conv, sessionId, message, ocId, messages, state);
|
|
3851
|
+
}
|
|
3852
|
+
if (state === "running" || state === "queued") {
|
|
3853
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3854
|
+
if (ongoing === true) {
|
|
3855
|
+
return this.reattachRedrive(conv, sessionId, message, ocId);
|
|
3856
|
+
}
|
|
3857
|
+
if (ongoing === false) {
|
|
3858
|
+
this.clearRedriveUnresolved(message.id);
|
|
3859
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3860
|
+
return "dispatch";
|
|
3861
|
+
}
|
|
3862
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3863
|
+
}
|
|
3864
|
+
this.clearRedriveUnresolved(message.id);
|
|
3865
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3866
|
+
return "dispatch";
|
|
3867
|
+
}
|
|
3868
|
+
/**
|
|
3869
|
+
* The `reattached` outcome (Task 3.3): the prior turn is STILL ONGOING per
|
|
3870
|
+
* opencode's own status map — undo the false reclaim instead of starting a
|
|
3871
|
+
* second turn.
|
|
3872
|
+
*/
|
|
3873
|
+
async reattachRedrive(conv, sessionId, message, ocId) {
|
|
3874
|
+
let anchorMs;
|
|
3875
|
+
const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
|
|
3876
|
+
if (!Number.isNaN(parsed)) {
|
|
3877
|
+
anchorMs = parsed;
|
|
3878
|
+
} else {
|
|
3879
|
+
anchorMs = this.now();
|
|
3880
|
+
this.log({
|
|
3881
|
+
level: "error",
|
|
3882
|
+
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)`,
|
|
3883
|
+
conversation_id: conv.id,
|
|
3884
|
+
message_id: message.id
|
|
3885
|
+
});
|
|
3886
|
+
}
|
|
3887
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3888
|
+
try {
|
|
3889
|
+
await this.markProcessing(conv.id, message.id, sessionId, ocId, title);
|
|
3890
|
+
} catch (err) {
|
|
3891
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3892
|
+
this.log({
|
|
3893
|
+
level: "warn",
|
|
3894
|
+
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)}`,
|
|
3895
|
+
conversation_id: conv.id,
|
|
3896
|
+
message_id: message.id
|
|
3897
|
+
});
|
|
3898
|
+
return "unresolved";
|
|
3899
|
+
}
|
|
3900
|
+
this.clearRedriveUnresolved(message.id);
|
|
3901
|
+
this.registerReadopted(conv, sessionId, message, ocId ?? "", anchorMs);
|
|
3902
|
+
this.dispatched.add(message.id);
|
|
3903
|
+
this.readopted.add(message.id);
|
|
3904
|
+
this.ensureWatcherRunning(sessionId);
|
|
3905
|
+
const watchedForMs = this.now() - anchorMs;
|
|
3906
|
+
void this.postSignal(conv.id, message.id, "redrive_reattached", {
|
|
3907
|
+
watched_for_ms: watchedForMs
|
|
3908
|
+
});
|
|
3909
|
+
this.log({
|
|
3910
|
+
level: "warn",
|
|
3911
|
+
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`,
|
|
3912
|
+
conversation_id: conv.id,
|
|
3913
|
+
message_id: message.id
|
|
3914
|
+
});
|
|
3915
|
+
return "reattached";
|
|
3916
|
+
}
|
|
3917
|
+
/**
|
|
3918
|
+
* The `settled` outcome (Task 3.2): the prior turn already finished (or
|
|
3919
|
+
* errored) while nobody was watching — deliver/report it instead of re-running.
|
|
3920
|
+
* Mirrors `readoptOne`'s `done`/`failed` branches' error discipline, simplified
|
|
3921
|
+
* (no `doneUndeliverable` park: a terminal PATCH failure here just retries next
|
|
3922
|
+
* drain, same as any other non-auth failure). The restart-abort carve-out that
|
|
3923
|
+
* keeps the two in step for `failed` lives in the caller (`resolveRedrive`, #1310),
|
|
3924
|
+
* so a row reaching this `failed` branch is a GENUINE failure.
|
|
3925
|
+
*/
|
|
3926
|
+
async settleRedrive(conv, sessionId, message, ocId, messages, state) {
|
|
3927
|
+
try {
|
|
3928
|
+
if (state === "done") {
|
|
3929
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3930
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3931
|
+
this.log({
|
|
3932
|
+
level: "info",
|
|
3933
|
+
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
3934
|
conversation_id: conv.id,
|
|
3025
3935
|
message_id: message.id
|
|
3026
3936
|
});
|
|
3027
|
-
await this.
|
|
3028
|
-
|
|
3029
|
-
|
|
3030
|
-
|
|
3031
|
-
|
|
3032
|
-
message_id: message.id
|
|
3033
|
-
});
|
|
3034
|
-
});
|
|
3937
|
+
await this.markDone(conv.id, message.id, sessionId, ocId, title, usage);
|
|
3938
|
+
} else {
|
|
3939
|
+
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3940
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3941
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3035
3942
|
this.log({
|
|
3036
3943
|
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`,
|
|
3944
|
+
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
3945
|
conversation_id: conv.id,
|
|
3048
3946
|
message_id: message.id
|
|
3049
3947
|
});
|
|
3050
|
-
|
|
3948
|
+
await this.markFailed(conv.id, message.id, sessionId, error2, usage, failure);
|
|
3051
3949
|
}
|
|
3052
|
-
|
|
3053
|
-
|
|
3054
|
-
|
|
3055
|
-
|
|
3950
|
+
} catch (err) {
|
|
3951
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3952
|
+
this.log({
|
|
3953
|
+
level: "warn",
|
|
3954
|
+
message: `Re-drive: failed to report message ${message.id.slice(0, 8)} ${state} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3955
|
+
conversation_id: conv.id,
|
|
3956
|
+
message_id: message.id
|
|
3957
|
+
});
|
|
3958
|
+
return "unresolved";
|
|
3056
3959
|
}
|
|
3057
|
-
|
|
3960
|
+
this.clearRedriveUnresolved(message.id);
|
|
3961
|
+
void this.postSignal(conv.id, message.id, "redrive_settled");
|
|
3962
|
+
return "settled";
|
|
3963
|
+
}
|
|
3964
|
+
/**
|
|
3965
|
+
* The permanent-failure outcome (#1348): the fence's own poll of this session
|
|
3966
|
+
* failed with the SAME opencode-answered signature
|
|
3967
|
+
* `MAX_IDENTICAL_REDRIVE_POLL_FAILURES` times in a row — a transient blip
|
|
3968
|
+
* would have varied or eventually cleared (see `pollSessionMessagesForRedrive`
|
|
3969
|
+
* and `recordRedrivePollFailure`), so this is a durable fault (e.g. #1345's
|
|
3970
|
+
* corrupted opencode session) rather than something worth retrying forever.
|
|
3971
|
+
* Mirrors `settleRedrive`'s error discipline: no `usage`/`failure` args to
|
|
3972
|
+
* `markFailed` (no opencode snapshot to extract them from — this poll never
|
|
3973
|
+
* got a readable one).
|
|
3974
|
+
*/
|
|
3975
|
+
async failRedrivePollPermanent(conv, sessionId, message, signature, streak) {
|
|
3976
|
+
this.log({
|
|
3977
|
+
level: "error",
|
|
3978
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) failed to poll with the identical signature "${signature}" ${streak} times in a row \u2014 reporting the message failed instead of retrying forever`,
|
|
3979
|
+
conversation_id: conv.id,
|
|
3980
|
+
message_id: message.id
|
|
3981
|
+
});
|
|
3982
|
+
try {
|
|
3983
|
+
await this.markFailed(
|
|
3984
|
+
conv.id,
|
|
3985
|
+
message.id,
|
|
3986
|
+
sessionId,
|
|
3987
|
+
`The runner could not read this conversation's state from OpenCode (${signature}). The same failure repeated ${streak} times in a row, so the message was not retried further.`
|
|
3988
|
+
);
|
|
3989
|
+
} catch (err) {
|
|
3990
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3058
3991
|
this.log({
|
|
3059
3992
|
level: "warn",
|
|
3060
|
-
message: `
|
|
3061
|
-
conversation_id: conv.id
|
|
3993
|
+
message: `Re-drive: failed to report message ${message.id.slice(0, 8)} permanently failed (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3994
|
+
conversation_id: conv.id,
|
|
3995
|
+
message_id: message.id
|
|
3062
3996
|
});
|
|
3997
|
+
return "unresolved";
|
|
3063
3998
|
}
|
|
3064
|
-
this.
|
|
3065
|
-
|
|
3999
|
+
this.clearRedriveUnresolved(message.id);
|
|
4000
|
+
void this.postSignal(conv.id, message.id, "redrive_poll_failed");
|
|
4001
|
+
return "settled";
|
|
4002
|
+
}
|
|
4003
|
+
/**
|
|
4004
|
+
* The bounded `unresolved` outcome (Task 3.4): opencode's state could not be
|
|
4005
|
+
* observed (snapshot unreadable/empty, or `isSessionOngoing` returned `null`).
|
|
4006
|
+
* A `pending` row is invisible to every cron arm (all require `status =
|
|
4007
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
4008
|
+
* nothing driving it — bound it to the existing `pausedMaxWaitMs` window
|
|
4009
|
+
* (reusing the knob, not a new constant) and take `dispatch` once elapsed.
|
|
4010
|
+
*/
|
|
4011
|
+
resolveRedriveUnresolved(conv, message) {
|
|
4012
|
+
const now = this.now();
|
|
4013
|
+
const since = this.redriveUnresolvedSince.get(message.id);
|
|
4014
|
+
if (since !== void 0 && now - since >= this.pausedMaxWaitMs) {
|
|
4015
|
+
this.clearRedriveUnresolved(message.id);
|
|
4016
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
4017
|
+
return "dispatch";
|
|
4018
|
+
}
|
|
4019
|
+
if (since === void 0) {
|
|
4020
|
+
this.redriveUnresolvedSince.set(message.id, now);
|
|
4021
|
+
}
|
|
4022
|
+
if (!this.redriveUnresolvedSignalled.has(message.id)) {
|
|
4023
|
+
this.redriveUnresolvedSignalled.add(message.id);
|
|
4024
|
+
void this.postSignal(conv.id, message.id, "redrive_unresolved");
|
|
4025
|
+
}
|
|
4026
|
+
return "unresolved";
|
|
4027
|
+
}
|
|
4028
|
+
/** Clear all three `unresolved`/failure-streak trackers for a row (any non-`unresolved` outcome). */
|
|
4029
|
+
clearRedriveUnresolved(messageId) {
|
|
4030
|
+
this.redriveUnresolvedSince.delete(messageId);
|
|
4031
|
+
this.redriveUnresolvedSignalled.delete(messageId);
|
|
4032
|
+
this.redrivePollFailures.delete(messageId);
|
|
4033
|
+
}
|
|
4034
|
+
/**
|
|
4035
|
+
* Record one poll outcome toward the re-drive fence's consecutive-identical-
|
|
4036
|
+
* failure streak (#1348) and return the resulting count. `signature === null`
|
|
4037
|
+
* (a thrown exception, H1) always clears the streak and returns `0` — it is
|
|
4038
|
+
* never countable. Otherwise the streak continues only when BOTH the session
|
|
4039
|
+
* and the signature match the previous failure; anything else (a different
|
|
4040
|
+
* session, or the same session failing a DIFFERENT way) starts a fresh streak
|
|
4041
|
+
* at `1`.
|
|
4042
|
+
*/
|
|
4043
|
+
recordRedrivePollFailure(messageId, sessionId, signature) {
|
|
4044
|
+
if (signature === null) {
|
|
4045
|
+
this.redrivePollFailures.delete(messageId);
|
|
4046
|
+
return 0;
|
|
4047
|
+
}
|
|
4048
|
+
const existing = this.redrivePollFailures.get(messageId);
|
|
4049
|
+
if (existing && existing.sessionId === sessionId && existing.signature === signature) {
|
|
4050
|
+
existing.count += 1;
|
|
4051
|
+
return existing.count;
|
|
4052
|
+
}
|
|
4053
|
+
this.redrivePollFailures.set(messageId, { sessionId, signature, count: 1 });
|
|
4054
|
+
return 1;
|
|
3066
4055
|
}
|
|
3067
4056
|
/**
|
|
3068
4057
|
* Record that `sessionId` is no longer a valid binding for `conversationId`
|
|
@@ -3125,7 +4114,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3125
4114
|
const directory = await this.resolveOpenCodeDirectory();
|
|
3126
4115
|
const sessionId = await createOpenCodeSession(this.port, directory);
|
|
3127
4116
|
this.sessions.set(conversationId, sessionId);
|
|
3128
|
-
await this.persistSession(conversationId, sessionId).catch(() => {
|
|
4117
|
+
await this.persistSession(conversationId, sessionId).catch((err) => {
|
|
4118
|
+
this.log({
|
|
4119
|
+
level: "warn",
|
|
4120
|
+
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)}`,
|
|
4121
|
+
conversation_id: conversationId
|
|
4122
|
+
});
|
|
3129
4123
|
});
|
|
3130
4124
|
return sessionId;
|
|
3131
4125
|
}
|
|
@@ -3317,12 +4311,17 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3317
4311
|
stuckReported: false,
|
|
3318
4312
|
lastAliveAt: 0,
|
|
3319
4313
|
aliveInFlight: false,
|
|
4314
|
+
titleSynced: false,
|
|
4315
|
+
titleSyncInFlight: false,
|
|
3320
4316
|
awaitingHumanLatched: false,
|
|
3321
4317
|
pausedOnQuestion: false,
|
|
3322
4318
|
pausedOnPermission: false,
|
|
3323
4319
|
pausedClearConfirmed: false,
|
|
3324
4320
|
pausedInFlight: false,
|
|
3325
|
-
deliveryDeadlineAnchored: false
|
|
4321
|
+
deliveryDeadlineAnchored: false,
|
|
4322
|
+
b2PinnedSinceMs: 0,
|
|
4323
|
+
b2LastDescendantCheckMs: 0,
|
|
4324
|
+
b2AbandonedSignalled: false
|
|
3326
4325
|
});
|
|
3327
4326
|
}
|
|
3328
4327
|
/**
|
|
@@ -3337,9 +4336,10 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3337
4336
|
* opencode reports ACTIVELY `running` is watched to completion (its liveness
|
|
3338
4337
|
* heartbeat keeps the cron off its row), while a re-adopted turn that is paused
|
|
3339
4338
|
* awaiting a human — or queued/unreachable — is still bounded by `deadline` and
|
|
3340
|
-
* handed to the cron.
|
|
3341
|
-
*
|
|
3342
|
-
*
|
|
4339
|
+
* handed to the cron. Real invariant (#965): the cron MAY reclaim a row this
|
|
4340
|
+
* runner still holds; a reclaimed row that already ran is never re-dispatched
|
|
4341
|
+
* while opencode reports its turn ongoing (readopt's own gate here, and the
|
|
4342
|
+
* `pending`-row re-drive fence, `resolveRedrive`). `dispatchedAt` stays `now`
|
|
3343
4343
|
* (only the appear-guard uses it).
|
|
3344
4344
|
*
|
|
3345
4345
|
* `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
|
|
@@ -3390,12 +4390,17 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3390
4390
|
// with no extra `re_adopted` signal needed (folds old WI-6).
|
|
3391
4391
|
lastAliveAt: 0,
|
|
3392
4392
|
aliveInFlight: false,
|
|
4393
|
+
titleSynced: false,
|
|
4394
|
+
titleSyncInFlight: false,
|
|
3393
4395
|
awaitingHumanLatched: false,
|
|
3394
4396
|
pausedOnQuestion: false,
|
|
3395
4397
|
pausedOnPermission: false,
|
|
3396
4398
|
pausedClearConfirmed: false,
|
|
3397
4399
|
pausedInFlight: false,
|
|
3398
|
-
deliveryDeadlineAnchored: false
|
|
4400
|
+
deliveryDeadlineAnchored: false,
|
|
4401
|
+
b2PinnedSinceMs: 0,
|
|
4402
|
+
b2LastDescendantCheckMs: 0,
|
|
4403
|
+
b2AbandonedSignalled: false
|
|
3399
4404
|
});
|
|
3400
4405
|
}
|
|
3401
4406
|
/**
|
|
@@ -3557,58 +4562,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3557
4562
|
}
|
|
3558
4563
|
}
|
|
3559
4564
|
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);
|
|
4565
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
3612
4566
|
return;
|
|
3613
4567
|
}
|
|
3614
4568
|
if (state === "failed") {
|
|
@@ -3622,8 +4576,16 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3622
4576
|
message_id: inFlight.evidentMessageId
|
|
3623
4577
|
});
|
|
3624
4578
|
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4579
|
+
const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
|
|
3625
4580
|
try {
|
|
3626
|
-
await this.markFailed(
|
|
4581
|
+
await this.markFailed(
|
|
4582
|
+
conv.id,
|
|
4583
|
+
inFlight.evidentMessageId,
|
|
4584
|
+
sessionId,
|
|
4585
|
+
error2,
|
|
4586
|
+
usage,
|
|
4587
|
+
failure
|
|
4588
|
+
);
|
|
3627
4589
|
} catch (err) {
|
|
3628
4590
|
if (err instanceof ChannelAuthError) throw err;
|
|
3629
4591
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -3668,6 +4630,44 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3668
4630
|
});
|
|
3669
4631
|
}
|
|
3670
4632
|
const activelyRunning = state === "running" && !awaitingHuman;
|
|
4633
|
+
const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
|
|
4634
|
+
const snapshotReadable = messages != null && messages.length > 0;
|
|
4635
|
+
if (!pinnedNow) {
|
|
4636
|
+
if (snapshotReadable) {
|
|
4637
|
+
inFlight.b2PinnedSinceMs = 0;
|
|
4638
|
+
inFlight.b2LastDescendantCheckMs = 0;
|
|
4639
|
+
inFlight.b2AbandonedSignalled = false;
|
|
4640
|
+
}
|
|
4641
|
+
} else {
|
|
4642
|
+
if (inFlight.b2AbandonedSignalled) {
|
|
4643
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4644
|
+
return;
|
|
4645
|
+
}
|
|
4646
|
+
if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
|
|
4647
|
+
const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
|
|
4648
|
+
if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
|
|
4649
|
+
inFlight.b2LastDescendantCheckMs = this.now();
|
|
4650
|
+
const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
|
|
4651
|
+
if (isB2AbandonmentConfirmed({
|
|
4652
|
+
pinnedForMs,
|
|
4653
|
+
minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
|
|
4654
|
+
descendantOngoing
|
|
4655
|
+
})) {
|
|
4656
|
+
inFlight.b2AbandonedSignalled = true;
|
|
4657
|
+
this.log({
|
|
4658
|
+
level: "warn",
|
|
4659
|
+
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`,
|
|
4660
|
+
conversation_id: conv.id,
|
|
4661
|
+
message_id: id
|
|
4662
|
+
});
|
|
4663
|
+
void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
|
|
4664
|
+
watched_for_ms: pinnedForMs
|
|
4665
|
+
});
|
|
4666
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4667
|
+
return;
|
|
4668
|
+
}
|
|
4669
|
+
}
|
|
4670
|
+
}
|
|
3671
4671
|
if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
|
|
3672
4672
|
this.log({
|
|
3673
4673
|
level: "warn",
|
|
@@ -3687,6 +4687,18 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3687
4687
|
inFlight.aliveInFlight = false;
|
|
3688
4688
|
if (ok) inFlight.lastAliveAt = this.now();
|
|
3689
4689
|
});
|
|
4690
|
+
if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
|
|
4691
|
+
inFlight.titleSyncInFlight = true;
|
|
4692
|
+
void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
|
|
4693
|
+
if (!title) {
|
|
4694
|
+
inFlight.titleSyncInFlight = false;
|
|
4695
|
+
return;
|
|
4696
|
+
}
|
|
4697
|
+
const ok = await this.patchConversationTitle(conv.id, title);
|
|
4698
|
+
inFlight.titleSyncInFlight = false;
|
|
4699
|
+
if (ok) inFlight.titleSynced = true;
|
|
4700
|
+
});
|
|
4701
|
+
}
|
|
3690
4702
|
}
|
|
3691
4703
|
if (awaitingHuman) {
|
|
3692
4704
|
if (!inFlight.awaitingHumanLatched) {
|
|
@@ -3724,6 +4736,70 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3724
4736
|
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3725
4737
|
}
|
|
3726
4738
|
}
|
|
4739
|
+
/**
|
|
4740
|
+
* Settle a message whose run-state has resolved `'done'` — extracted verbatim
|
|
4741
|
+
* (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
|
|
4742
|
+
* inline `state === 'done'` branch body, so a SECOND caller (the #721
|
|
4743
|
+
* b2-abandonment resolution) can reach the exact same completion behavior
|
|
4744
|
+
* (delivery-deadline anchoring, title resolution, usage extraction, and
|
|
4745
|
+
* `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
|
|
4746
|
+
* and risking the two copies silently drifting apart.
|
|
4747
|
+
*/
|
|
4748
|
+
async settleMessageDone(sessionId, watcher, inFlight, messages) {
|
|
4749
|
+
const conv = watcher.conv;
|
|
4750
|
+
this.anchorDeliveryDeadline(inFlight);
|
|
4751
|
+
if (!inFlight.done) {
|
|
4752
|
+
this.log({
|
|
4753
|
+
level: "info",
|
|
4754
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
4755
|
+
conversation_id: conv.id,
|
|
4756
|
+
message_id: inFlight.evidentMessageId
|
|
4757
|
+
});
|
|
4758
|
+
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
4759
|
+
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4760
|
+
try {
|
|
4761
|
+
await this.markDone(
|
|
4762
|
+
conv.id,
|
|
4763
|
+
inFlight.evidentMessageId,
|
|
4764
|
+
sessionId,
|
|
4765
|
+
inFlight.opencodeMessageId,
|
|
4766
|
+
title,
|
|
4767
|
+
usage
|
|
4768
|
+
);
|
|
4769
|
+
} catch (err) {
|
|
4770
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
4771
|
+
if (err instanceof ChannelTerminalError) {
|
|
4772
|
+
this.log({
|
|
4773
|
+
level: "warn",
|
|
4774
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
4775
|
+
conversation_id: conv.id,
|
|
4776
|
+
message_id: inFlight.evidentMessageId
|
|
4777
|
+
});
|
|
4778
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4779
|
+
return;
|
|
4780
|
+
}
|
|
4781
|
+
if (this.now() >= inFlight.deadline) {
|
|
4782
|
+
this.log({
|
|
4783
|
+
level: "warn",
|
|
4784
|
+
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)}`,
|
|
4785
|
+
conversation_id: conv.id,
|
|
4786
|
+
message_id: inFlight.evidentMessageId
|
|
4787
|
+
});
|
|
4788
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4789
|
+
return;
|
|
4790
|
+
}
|
|
4791
|
+
this.log({
|
|
4792
|
+
level: "warn",
|
|
4793
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
4794
|
+
conversation_id: conv.id,
|
|
4795
|
+
message_id: inFlight.evidentMessageId
|
|
4796
|
+
});
|
|
4797
|
+
return;
|
|
4798
|
+
}
|
|
4799
|
+
inFlight.done = true;
|
|
4800
|
+
}
|
|
4801
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4802
|
+
}
|
|
3727
4803
|
// Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
|
|
3728
4804
|
/**
|
|
3729
4805
|
* Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
|
|
@@ -3823,7 +4899,10 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3823
4899
|
* re-dispatched (at most once, see `forceReadoptRun`):
|
|
3824
4900
|
* - `done` → `markDone` now (guarded like the watcher's done branch);
|
|
3825
4901
|
* - `failed` → `markFailed` with the surfaced error (issue #182), so an
|
|
3826
|
-
* errored turn is reported failed on restart, NOT re-dispatched
|
|
4902
|
+
* errored turn is reported failed on restart, NOT re-dispatched —
|
|
4903
|
+
* EXCEPT a restart-ABORTED turn under a not-ongoing session,
|
|
4904
|
+
* which is a restart orphan wearing a terminal error and is
|
|
4905
|
+
* re-dispatched instead (issue #1310, see the branch below);
|
|
3827
4906
|
* - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
|
|
3828
4907
|
* tracking the stored id so the reply correlates by it;
|
|
3829
4908
|
* - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
|
|
@@ -3887,9 +4966,19 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3887
4966
|
void this.postSignal(row.conversation_id, row.id, "readopt_done");
|
|
3888
4967
|
return;
|
|
3889
4968
|
}
|
|
3890
|
-
|
|
4969
|
+
const restartAborted = state === "failed" && sessionOngoing === false && isAbortedTerminalReply(messages, ocId ?? "");
|
|
4970
|
+
if (restartAborted) {
|
|
4971
|
+
this.log({
|
|
4972
|
+
level: "info",
|
|
4973
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} carries an abort-shaped terminal error and session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status \u2014 restart orphan, re-dispatching instead of marking it permanently failed`,
|
|
4974
|
+
conversation_id: row.conversation_id,
|
|
4975
|
+
message_id: row.id
|
|
4976
|
+
});
|
|
4977
|
+
}
|
|
4978
|
+
if (state === "failed" && !restartAborted) {
|
|
3891
4979
|
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3892
4980
|
const usage = messageUsage(messages, ocId ?? "");
|
|
4981
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3893
4982
|
this.log({
|
|
3894
4983
|
level: "error",
|
|
3895
4984
|
message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
|
|
@@ -3897,7 +4986,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3897
4986
|
message_id: row.id
|
|
3898
4987
|
});
|
|
3899
4988
|
try {
|
|
3900
|
-
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
|
|
4989
|
+
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
|
|
3901
4990
|
} catch (err) {
|
|
3902
4991
|
if (err instanceof ChannelAuthError) throw err;
|
|
3903
4992
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -4164,7 +5253,8 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4164
5253
|
opencode_model: row.opencode_model,
|
|
4165
5254
|
source_message_id: row.source_message_id,
|
|
4166
5255
|
slack_user_id: row.slack_user_id,
|
|
4167
|
-
attachments: row.attachments ?? null
|
|
5256
|
+
attachments: row.attachments ?? null,
|
|
5257
|
+
opencode_message_id: row.opencode_message_id
|
|
4168
5258
|
};
|
|
4169
5259
|
}
|
|
4170
5260
|
/**
|
|
@@ -4303,6 +5393,47 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4303
5393
|
}
|
|
4304
5394
|
return false;
|
|
4305
5395
|
}
|
|
5396
|
+
/**
|
|
5397
|
+
* Tri-state variant of the upward parentID membership walk (#721), used ONLY
|
|
5398
|
+
* by `isAnyDescendantSessionOngoing`. Walks the SAME cached
|
|
5399
|
+
* `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
|
|
5400
|
+
* `sessionBelongsTo`, which deliberately collapses "confirmed not a
|
|
5401
|
+
* descendant" and "the walk's fetch failed" into the same `false` (safe for
|
|
5402
|
+
* its OTHER callers: interaction attribution and the recovery-path
|
|
5403
|
+
* `isAnyDescendantSessionAlive`, both of which just retry next tick with no
|
|
5404
|
+
* safety consequence either way) — this variant keeps those two outcomes
|
|
5405
|
+
* SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
|
|
5406
|
+
* (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
|
|
5407
|
+
* not ongoing".
|
|
5408
|
+
*
|
|
5409
|
+
* Return contract:
|
|
5410
|
+
* - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
|
|
5411
|
+
* - `false` → the walk reached a definitive, parent-less root session
|
|
5412
|
+
* WITHOUT ever matching `rootSessionId` — `sessionId` is
|
|
5413
|
+
* CONFIRMED NOT a descendant of it.
|
|
5414
|
+
* - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
|
|
5415
|
+
* through the walk (`resolveSessionParent` returned `undefined`),
|
|
5416
|
+
* or the depth cap (32) was hit without a definitive answer (a
|
|
5417
|
+
* pathological/cyclic chain proves nothing either way). NEVER
|
|
5418
|
+
* treat this the same as `false` — see `sessionBelongsTo`'s own
|
|
5419
|
+
* doc comment above for why that collapse is safe THERE but not
|
|
5420
|
+
* here.
|
|
5421
|
+
*
|
|
5422
|
+
* `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
|
|
5423
|
+
* to the live-path descendant check, not a modification of shared code used
|
|
5424
|
+
* by interaction attribution or the recovery path.
|
|
5425
|
+
*/
|
|
5426
|
+
async resolveSessionMembership(sessionId, rootSessionId) {
|
|
5427
|
+
let current = sessionId;
|
|
5428
|
+
for (let depth = 0; current && depth < 32; depth++) {
|
|
5429
|
+
if (current === rootSessionId) return true;
|
|
5430
|
+
const parent = await this.resolveSessionParent(current);
|
|
5431
|
+
if (parent === void 0) return null;
|
|
5432
|
+
if (parent === null) return false;
|
|
5433
|
+
current = parent;
|
|
5434
|
+
}
|
|
5435
|
+
return null;
|
|
5436
|
+
}
|
|
4306
5437
|
/**
|
|
4307
5438
|
* Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
|
|
4308
5439
|
* `null` for a root session (no parent) and `undefined` when opencode is
|
|
@@ -4387,6 +5518,54 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4387
5518
|
}
|
|
4388
5519
|
return null;
|
|
4389
5520
|
}
|
|
5521
|
+
/**
|
|
5522
|
+
* Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
|
|
5523
|
+
* session title onto the conversation via the PLAIN conversation-update
|
|
5524
|
+
* endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
|
|
5525
|
+
* message-status endpoint `markProcessing`/`markDone` use. Deliberately a
|
|
5526
|
+
* separate, lighter call: it carries no `status`, so it cannot re-trigger the
|
|
5527
|
+
* `processing`/`done` transition side effects (Slack notices, activity-log
|
|
5528
|
+
* rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
|
|
5529
|
+
* ever touches `conversations.title`. That route (`routes/conversations.ts`)
|
|
5530
|
+
* skips a title write matching the stored value, so a redundant call with the
|
|
5531
|
+
* same title is a real no-op — it does not bump `updated_at`, which the
|
|
5532
|
+
* conversation list sorts and paginates on. (Note this is a DIFFERENT guard
|
|
5533
|
+
* from `threads.ts`'s "non-empty AND changed" one, which only covers the
|
|
5534
|
+
* message-status PATCH; the non-empty half is enforced here instead, by
|
|
5535
|
+
* `resolveSessionTitle` never returning an empty/placeholder title.)
|
|
5536
|
+
*
|
|
5537
|
+
* Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
|
|
5538
|
+
* is logged and the title is simply retried on the next heartbeat tick (the
|
|
5539
|
+
* caller only latches `titleSynced` on `true`).
|
|
5540
|
+
*/
|
|
5541
|
+
async patchConversationTitle(conversationId, title) {
|
|
5542
|
+
try {
|
|
5543
|
+
const res = await this.fetchImpl(
|
|
5544
|
+
`${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
|
|
5545
|
+
{
|
|
5546
|
+
method: "PATCH",
|
|
5547
|
+
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
5548
|
+
body: JSON.stringify({ title })
|
|
5549
|
+
}
|
|
5550
|
+
);
|
|
5551
|
+
if (!res.ok) {
|
|
5552
|
+
this.log({
|
|
5553
|
+
level: "debug",
|
|
5554
|
+
message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
|
|
5555
|
+
conversation_id: conversationId
|
|
5556
|
+
});
|
|
5557
|
+
return false;
|
|
5558
|
+
}
|
|
5559
|
+
return true;
|
|
5560
|
+
} catch (err) {
|
|
5561
|
+
this.log({
|
|
5562
|
+
level: "debug",
|
|
5563
|
+
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)}`,
|
|
5564
|
+
conversation_id: conversationId
|
|
5565
|
+
});
|
|
5566
|
+
return false;
|
|
5567
|
+
}
|
|
5568
|
+
}
|
|
4390
5569
|
/**
|
|
4391
5570
|
* DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
|
|
4392
5571
|
* (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
|
|
@@ -4419,31 +5598,109 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4419
5598
|
* longer exists". The inversion lives in the caller; this method just reports
|
|
4420
5599
|
* true/false/null faithfully.
|
|
4421
5600
|
*
|
|
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.
|
|
5601
|
+
* VERIFY-BEFORE-DEPEND: we depend ONLY on (a) `parentID` from `GET /session/:id`
|
|
5602
|
+
* (already proven by the existing child-session interaction tests, via
|
|
5603
|
+
* `resolveSessionParent`/`sessionBelongsTo`) and (b) the child's own message-list
|
|
5604
|
+
* terminal state. We do NOT depend on any session-level `busy`/`idle` field —
|
|
5605
|
+
* there is none on `GET /session/:id`; OpenCode's busy state is in-memory
|
|
5606
|
+
* `SessionStatus` only.
|
|
5607
|
+
*/
|
|
5608
|
+
async isAnyDescendantSessionAlive(rootSessionId) {
|
|
5609
|
+
const sessions = await listSessions(this.port);
|
|
5610
|
+
if (!sessions) {
|
|
5611
|
+
this.log({
|
|
5612
|
+
level: "warn",
|
|
5613
|
+
message: `Re-adopt: could not enumerate sessions to cross-check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating child liveness as indeterminate`
|
|
5614
|
+
});
|
|
5615
|
+
return null;
|
|
5616
|
+
}
|
|
5617
|
+
for (const candidate of sessions) {
|
|
5618
|
+
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
5619
|
+
if (!await this.sessionBelongsTo(candidate.id, rootSessionId)) continue;
|
|
5620
|
+
const childMsgs = await getSessionMessages(this.port, candidate.id);
|
|
5621
|
+
if (isSessionActivelyGenerating(childMsgs)) {
|
|
5622
|
+
return true;
|
|
5623
|
+
}
|
|
5624
|
+
}
|
|
5625
|
+
return false;
|
|
5626
|
+
}
|
|
5627
|
+
/**
|
|
5628
|
+
* LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
|
|
5629
|
+
* sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
|
|
5630
|
+
* in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
|
|
5631
|
+
*
|
|
5632
|
+
* Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
|
|
5633
|
+
* cross-check above): that method judges liveness from the child's OWN
|
|
5634
|
+
* TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
|
|
5635
|
+
* on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
|
|
5636
|
+
* path the local opencode server IS running, so its in-memory status map is
|
|
5637
|
+
* live and authoritative — and per ADR-0047 §4a ("the child has its own entry
|
|
5638
|
+
* [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
|
|
5639
|
+
* ENTIRE turn (including any tool call it is itself executing), not a
|
|
5640
|
+
* per-message transcript snapshot. This sidesteps the "child's own tool is
|
|
5641
|
+
* executing, between its step's completion and the next generation step"
|
|
5642
|
+
* transcript gap that a transcript-based check would need a second,
|
|
5643
|
+
* sustained-window bound to guard against — it is simply not derived from
|
|
5644
|
+
* message timestamps at all.
|
|
5645
|
+
*
|
|
5646
|
+
* Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
|
|
5647
|
+
* status, as the recovery path does per §4a)? Because on the LIVE path the
|
|
5648
|
+
* root session can be shared: a SECOND, unrelated user message can land on the
|
|
5649
|
+
* SAME session (issue #721's own root cause) and keep the root `busy` for a
|
|
5650
|
+
* reason that has nothing to do with THIS message's delegation. A `task`
|
|
5651
|
+
* descendant session is spawned for exactly one delegated turn and never
|
|
5652
|
+
* reused, so its OWN status-map entry is unambiguous evidence about that one
|
|
5653
|
+
* delegation — which the root's status is not.
|
|
5654
|
+
*
|
|
5655
|
+
* Why membership is checked via `resolveSessionMembership`, NOT
|
|
5656
|
+
* `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
|
|
5657
|
+
* `GET /session/:id` fetch failure into "not a descendant", which would
|
|
5658
|
+
* silently drop a genuinely-live candidate from consideration on the one
|
|
5659
|
+
* unlucky tick its membership-walk fetch hiccups (#721).
|
|
5660
|
+
* `resolveSessionMembership` keeps that failure mode as a distinct `null`
|
|
5661
|
+
* (indeterminate) so it is folded into THIS method's own `indeterminate` flag
|
|
5662
|
+
* instead.
|
|
5663
|
+
*
|
|
5664
|
+
* Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
|
|
5665
|
+
* - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
|
|
5666
|
+
* - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
|
|
5667
|
+
* confirmed either way (`resolveSessionMembership` never
|
|
5668
|
+
* returned `null`), and every CONFIRMED descendant's status read
|
|
5669
|
+
* succeeded and is not ongoing (includes "no descendant session
|
|
5670
|
+
* exists at all" — e.g. a plain, non-`task` tool call).
|
|
5671
|
+
* - `null` → INDETERMINATE: `listSessions` failed, OR at least one
|
|
5672
|
+
* candidate's MEMBERSHIP could not be confirmed
|
|
5673
|
+
* (`resolveSessionMembership` returned `null` — a fetch failure
|
|
5674
|
+
* or pathological chain partway through the parent walk), OR at
|
|
5675
|
+
* least one CONFIRMED descendant's `isSessionOngoing` read
|
|
5676
|
+
* failed — and no OTHER candidate was already confirmed `true`.
|
|
5677
|
+
* The caller MUST NOT treat `null` the same as `false` here
|
|
5678
|
+
* (unlike the recovery cross-check's contract) — see
|
|
5679
|
+
* `isB2AbandonmentConfirmed`.
|
|
4428
5680
|
*/
|
|
4429
|
-
async
|
|
5681
|
+
async isAnyDescendantSessionOngoing(rootSessionId) {
|
|
4430
5682
|
const sessions = await listSessions(this.port);
|
|
4431
5683
|
if (!sessions) {
|
|
4432
5684
|
this.log({
|
|
4433
5685
|
level: "warn",
|
|
4434
|
-
message: `
|
|
5686
|
+
message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
|
|
4435
5687
|
});
|
|
4436
5688
|
return null;
|
|
4437
5689
|
}
|
|
5690
|
+
let indeterminate = false;
|
|
4438
5691
|
for (const candidate of sessions) {
|
|
4439
5692
|
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
4440
|
-
|
|
4441
|
-
|
|
4442
|
-
|
|
4443
|
-
|
|
5693
|
+
const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
|
|
5694
|
+
if (membership === null) {
|
|
5695
|
+
indeterminate = true;
|
|
5696
|
+
continue;
|
|
4444
5697
|
}
|
|
5698
|
+
if (membership === false) continue;
|
|
5699
|
+
const ongoing = await isSessionOngoing(this.port, candidate.id);
|
|
5700
|
+
if (ongoing === true) return true;
|
|
5701
|
+
if (ongoing === null) indeterminate = true;
|
|
4445
5702
|
}
|
|
4446
|
-
return false;
|
|
5703
|
+
return indeterminate ? null : false;
|
|
4447
5704
|
}
|
|
4448
5705
|
/**
|
|
4449
5706
|
* Cheap decision-telemetry label for a running row's LAST correlated reply
|
|
@@ -4589,11 +5846,11 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4589
5846
|
* `ensureSession` guard, not this suppression, is what makes the self-heal
|
|
4590
5847
|
* stick.
|
|
4591
5848
|
*/
|
|
4592
|
-
sessionIdBody(sessionId, conversationId, messageId,
|
|
5849
|
+
sessionIdBody(sessionId, conversationId, messageId, status2) {
|
|
4593
5850
|
if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
|
|
4594
5851
|
this.log({
|
|
4595
5852
|
level: "debug",
|
|
4596
|
-
message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${
|
|
5853
|
+
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
5854
|
conversation_id: conversationId,
|
|
4598
5855
|
message_id: messageId
|
|
4599
5856
|
});
|
|
@@ -4708,7 +5965,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4708
5965
|
* exists but is wedged, so the next attempt must get a fresh one
|
|
4709
5966
|
* instead of reusing it — see WI-1's server-side null-clearing PATCH).
|
|
4710
5967
|
*/
|
|
4711
|
-
async markFailed(conversationId, messageId, sessionId, error2, usage) {
|
|
5968
|
+
async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
|
|
4712
5969
|
const body = { status: "failed" };
|
|
4713
5970
|
if (sessionId === null) {
|
|
4714
5971
|
body.opencode_session_id = null;
|
|
@@ -4717,6 +5974,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4717
5974
|
}
|
|
4718
5975
|
if (error2 !== void 0) body.error = error2;
|
|
4719
5976
|
if (usage) Object.assign(body, usage);
|
|
5977
|
+
if (failure) {
|
|
5978
|
+
body.failure_kind = failure.kind;
|
|
5979
|
+
body.failure_provider_id = failure.providerId;
|
|
5980
|
+
body.failure_model_id = failure.modelId;
|
|
5981
|
+
body.failure_reason = failure.reason;
|
|
5982
|
+
}
|
|
4720
5983
|
await this.callWithRetry(
|
|
4721
5984
|
"marking message as failed",
|
|
4722
5985
|
() => this.fetchImpl(
|
|
@@ -4729,6 +5992,29 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4729
5992
|
)
|
|
4730
5993
|
);
|
|
4731
5994
|
}
|
|
5995
|
+
/**
|
|
5996
|
+
* Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
|
|
5997
|
+
*
|
|
5998
|
+
* `messageFailure` alone (structured OpenCode error → `model_auth`) covers
|
|
5999
|
+
* most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
|
|
6000
|
+
* to the P1-2b zero-provider check — one extra loopback call to
|
|
6001
|
+
* `hasAnyConfiguredProvider`, only reached when the structured classifier
|
|
6002
|
+
* couldn't place it. Fails open (never throws): a fallback probe failure
|
|
6003
|
+
* (`null`/indeterminate) leaves the classification `null`, which produces
|
|
6004
|
+
* today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
|
|
6005
|
+
*/
|
|
6006
|
+
async classifyModelAuthFailure(messages, userMessageId) {
|
|
6007
|
+
const classified = messageFailure(messages, userMessageId);
|
|
6008
|
+
if (classified != null) return classified;
|
|
6009
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
6010
|
+
const hasProvider = await hasAnyConfiguredProvider(this.port);
|
|
6011
|
+
return applyZeroProviderFallback(
|
|
6012
|
+
classified,
|
|
6013
|
+
hasProvider,
|
|
6014
|
+
reply?.info?.providerID ?? null,
|
|
6015
|
+
reply?.info?.modelID ?? null
|
|
6016
|
+
);
|
|
6017
|
+
}
|
|
4732
6018
|
/**
|
|
4733
6019
|
* Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
|
|
4734
6020
|
* (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
|
|
@@ -4881,10 +6167,16 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4881
6167
|
import chalk5 from "chalk";
|
|
4882
6168
|
import ora2 from "ora";
|
|
4883
6169
|
import { select as select2 } from "@inquirer/prompts";
|
|
6170
|
+
var INTERACTIVE_START_TIMEOUT_MS = 3e4;
|
|
4884
6171
|
async function ensureOpenCodeRunning(ctx) {
|
|
4885
6172
|
const healthCheck = await checkOpenCodeHealth(ctx.port);
|
|
4886
6173
|
if (healthCheck.healthy) {
|
|
4887
|
-
return {
|
|
6174
|
+
return {
|
|
6175
|
+
port: ctx.port,
|
|
6176
|
+
process: null,
|
|
6177
|
+
version: healthCheck.version ?? null,
|
|
6178
|
+
notReadyReason: null
|
|
6179
|
+
};
|
|
4888
6180
|
}
|
|
4889
6181
|
const runningInstances = await findHealthyOpenCodeInstances();
|
|
4890
6182
|
if (runningInstances.length > 0) {
|
|
@@ -4905,7 +6197,7 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4905
6197
|
console.log(chalk5.yellow("Tip: Run with the correct port:"));
|
|
4906
6198
|
console.log(
|
|
4907
6199
|
chalk5.dim(
|
|
4908
|
-
` ${getCliName()} run --
|
|
6200
|
+
` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
|
|
4909
6201
|
)
|
|
4910
6202
|
);
|
|
4911
6203
|
}
|
|
@@ -4925,14 +6217,22 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4925
6217
|
if (!ctx.interactive) {
|
|
4926
6218
|
ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
|
|
4927
6219
|
const proc = await startOpenCode(ctx.port);
|
|
4928
|
-
const health = await waitForOpenCodeHealth(ctx.port,
|
|
6220
|
+
const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
|
|
4929
6221
|
if (!health.healthy) {
|
|
4930
|
-
|
|
4931
|
-
|
|
4932
|
-
|
|
6222
|
+
return {
|
|
6223
|
+
port: ctx.port,
|
|
6224
|
+
process: proc,
|
|
6225
|
+
version: null,
|
|
6226
|
+
notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
|
|
6227
|
+
};
|
|
4933
6228
|
}
|
|
4934
6229
|
ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
|
|
4935
|
-
return {
|
|
6230
|
+
return {
|
|
6231
|
+
port: ctx.port,
|
|
6232
|
+
process: proc,
|
|
6233
|
+
version: health.version ?? null,
|
|
6234
|
+
notReadyReason: null
|
|
6235
|
+
};
|
|
4936
6236
|
}
|
|
4937
6237
|
let port = ctx.port;
|
|
4938
6238
|
if (isPortInUse(port)) {
|
|
@@ -4985,122 +6285,15 @@ Port ${port} is already in use.`));
|
|
|
4985
6285
|
if (action === "start") {
|
|
4986
6286
|
const spinner = ora2("Starting OpenCode...").start();
|
|
4987
6287
|
const proc = await startOpenCode(port);
|
|
4988
|
-
const health = await waitForOpenCodeHealth(port,
|
|
6288
|
+
const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
|
|
4989
6289
|
if (!health.healthy) {
|
|
4990
6290
|
spinner.fail("Failed to start OpenCode");
|
|
4991
6291
|
throw new Error("OpenCode failed to start");
|
|
4992
6292
|
}
|
|
4993
6293
|
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: ${message}` };
|
|
6294
|
+
return { port, process: proc, version: health.version ?? null, notReadyReason: null };
|
|
5103
6295
|
}
|
|
6296
|
+
return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
|
|
5104
6297
|
}
|
|
5105
6298
|
|
|
5106
6299
|
// src/commands/run.ts
|
|
@@ -5108,6 +6301,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
|
|
|
5108
6301
|
var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
|
|
5109
6302
|
var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
|
|
5110
6303
|
var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
|
|
6304
|
+
var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
|
|
5111
6305
|
function resolveLogLevel(options) {
|
|
5112
6306
|
const accepted = Object.keys(LOG_LEVELS);
|
|
5113
6307
|
const validate = (value, source) => {
|
|
@@ -5138,7 +6332,7 @@ function resolveFileSyncDirectories(raw, homeDir) {
|
|
|
5138
6332
|
if (trimmed === "") {
|
|
5139
6333
|
throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
|
|
5140
6334
|
}
|
|
5141
|
-
const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ?
|
|
6335
|
+
const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join3(homeDir, trimmed.slice(2)) : trimmed;
|
|
5142
6336
|
if (!isAbsolute2(expanded)) {
|
|
5143
6337
|
throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
|
|
5144
6338
|
}
|
|
@@ -5159,6 +6353,35 @@ function resolveFileSyncDirectories(raw, homeDir) {
|
|
|
5159
6353
|
}
|
|
5160
6354
|
return directories;
|
|
5161
6355
|
}
|
|
6356
|
+
var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
|
|
6357
|
+
var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
|
|
6358
|
+
var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
|
|
6359
|
+
function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
|
|
6360
|
+
const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
|
|
6361
|
+
let raw;
|
|
6362
|
+
let source;
|
|
6363
|
+
if (options.opencodeStartTimeout !== void 0) {
|
|
6364
|
+
raw = options.opencodeStartTimeout;
|
|
6365
|
+
source = "--opencode-start-timeout";
|
|
6366
|
+
} else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
|
|
6367
|
+
raw = env[OPENCODE_START_TIMEOUT_ENV];
|
|
6368
|
+
source = OPENCODE_START_TIMEOUT_ENV;
|
|
6369
|
+
} else {
|
|
6370
|
+
return { timeoutMs: defaultMs, warnings: [] };
|
|
6371
|
+
}
|
|
6372
|
+
const trimmed = raw.trim();
|
|
6373
|
+
const seconds = Number(trimmed);
|
|
6374
|
+
const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
|
|
6375
|
+
if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
|
|
6376
|
+
return {
|
|
6377
|
+
timeoutMs: defaultMs,
|
|
6378
|
+
warnings: [
|
|
6379
|
+
`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`
|
|
6380
|
+
]
|
|
6381
|
+
};
|
|
6382
|
+
}
|
|
6383
|
+
return { timeoutMs: seconds * 1e3, warnings: [] };
|
|
6384
|
+
}
|
|
5162
6385
|
function meetsThreshold(state, level) {
|
|
5163
6386
|
return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
|
|
5164
6387
|
}
|
|
@@ -5180,6 +6403,10 @@ function log2(state, message, level = "info") {
|
|
|
5180
6403
|
function logActivity(state, entry) {
|
|
5181
6404
|
const level = entry.level ?? (entry.type === "error" ? "error" : "info");
|
|
5182
6405
|
if (!meetsThreshold(state, level)) return;
|
|
6406
|
+
forwardRunnerActivity(
|
|
6407
|
+
{ level, message: entry.message, error: entry.error },
|
|
6408
|
+
{ agentId: state.agentId, authHeader: state.authHeader }
|
|
6409
|
+
);
|
|
5183
6410
|
const fullEntry = {
|
|
5184
6411
|
...entry,
|
|
5185
6412
|
level,
|
|
@@ -5279,9 +6506,15 @@ async function handleAuthError(state, error2) {
|
|
|
5279
6506
|
}
|
|
5280
6507
|
async function driveChannels(state, driver) {
|
|
5281
6508
|
let idlePolls = 0;
|
|
6509
|
+
let idleMs = 0;
|
|
6510
|
+
let consecutiveDrainFailures = 0;
|
|
6511
|
+
let unreachableMs = 0;
|
|
5282
6512
|
let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
5283
6513
|
let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
5284
6514
|
while (state.running) {
|
|
6515
|
+
const cycleStartedAtMs = performance.now();
|
|
6516
|
+
let idleThisCycle = false;
|
|
6517
|
+
let unreachableThisCycle = false;
|
|
5285
6518
|
if (state.connection?.reconnecting && state.connection.reconnectPromise) {
|
|
5286
6519
|
logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
|
|
5287
6520
|
if (state.interactive) displayStatus(state);
|
|
@@ -5296,17 +6529,23 @@ async function driveChannels(state, driver) {
|
|
|
5296
6529
|
);
|
|
5297
6530
|
try {
|
|
5298
6531
|
const processed = await driver.drainPending();
|
|
6532
|
+
consecutiveDrainFailures = 0;
|
|
6533
|
+
unreachableMs = 0;
|
|
5299
6534
|
state.messageCount += processed;
|
|
5300
6535
|
const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
|
|
5301
6536
|
lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
5302
6537
|
const appliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
5303
|
-
const
|
|
6538
|
+
const filesApplied = appliedFiles !== lastSeenAppliedFiles;
|
|
6539
|
+
const fileActivity = carriedOverFileSync || filesApplied;
|
|
5304
6540
|
lastSeenAppliedFiles = appliedFiles;
|
|
6541
|
+
if (filesApplied) state.claudeUsageRearm?.();
|
|
5305
6542
|
if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
|
|
5306
6543
|
idlePolls = 0;
|
|
6544
|
+
idleMs = 0;
|
|
5307
6545
|
if (processed > 0 && state.interactive) displayStatus(state);
|
|
5308
6546
|
} else if (state.idleTimeout !== null) {
|
|
5309
6547
|
idlePolls++;
|
|
6548
|
+
idleThisCycle = true;
|
|
5310
6549
|
if (idlePolls === 1) {
|
|
5311
6550
|
logActivity(state, {
|
|
5312
6551
|
type: "info",
|
|
@@ -5330,21 +6569,44 @@ async function driveChannels(state, driver) {
|
|
|
5330
6569
|
const errorMessage = error2 instanceof Error ? error2.message : String(error2);
|
|
5331
6570
|
logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
|
|
5332
6571
|
if (state.interactive) displayStatus(state);
|
|
6572
|
+
if (driver.hasInFlightWatchers()) {
|
|
6573
|
+
consecutiveDrainFailures = 0;
|
|
6574
|
+
unreachableMs = 0;
|
|
6575
|
+
} else if (state.idleTimeout !== null) {
|
|
6576
|
+
consecutiveDrainFailures++;
|
|
6577
|
+
unreachableThisCycle = true;
|
|
6578
|
+
if (consecutiveDrainFailures === 1) {
|
|
6579
|
+
logActivity(state, {
|
|
6580
|
+
type: "info",
|
|
6581
|
+
message: `Cannot reach Evident, will exit if this persists past the idle timeout (timeout: ${state.idleTimeout}s)...`
|
|
6582
|
+
});
|
|
6583
|
+
if (state.interactive) displayStatus(state);
|
|
6584
|
+
}
|
|
6585
|
+
}
|
|
5333
6586
|
}
|
|
5334
6587
|
await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
|
|
5335
|
-
|
|
5336
|
-
|
|
5337
|
-
|
|
5338
|
-
|
|
5339
|
-
|
|
5340
|
-
|
|
5341
|
-
|
|
6588
|
+
const cycleMs = performance.now() - cycleStartedAtMs;
|
|
6589
|
+
if (idleThisCycle) idleMs += cycleMs;
|
|
6590
|
+
if (unreachableThisCycle) unreachableMs += cycleMs;
|
|
6591
|
+
if (state.idleTimeout !== null && consecutiveDrainFailures >= 2 && unreachableMs > state.idleTimeout * 1e3) {
|
|
6592
|
+
logActivity(state, {
|
|
6593
|
+
type: "info",
|
|
6594
|
+
level: "warn",
|
|
6595
|
+
message: `Exiting: could not reach Evident for ${consecutiveDrainFailures} consecutive polls (${Math.round(unreachableMs / 1e3)}s)`
|
|
6596
|
+
});
|
|
6597
|
+
if (state.interactive) displayStatus(state);
|
|
6598
|
+
break;
|
|
6599
|
+
}
|
|
6600
|
+
if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
|
|
6601
|
+
logActivity(state, { type: "info", message: "Idle timeout reached" });
|
|
6602
|
+
if (state.interactive) displayStatus(state);
|
|
6603
|
+
break;
|
|
5342
6604
|
}
|
|
5343
6605
|
}
|
|
5344
6606
|
}
|
|
5345
6607
|
var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
|
|
5346
|
-
async function runSweep(state, driver,
|
|
5347
|
-
const mode = `age=${
|
|
6608
|
+
async function runSweep(state, driver, config) {
|
|
6609
|
+
const mode = `age=${config.maxAgeMs ?? "\u2014"} count=${config.maxCount ?? "\u2014"}`;
|
|
5348
6610
|
try {
|
|
5349
6611
|
const sessions = await listSessions(state.port);
|
|
5350
6612
|
if (sessions === null) {
|
|
@@ -5357,8 +6619,8 @@ async function runSweep(state, driver, config2) {
|
|
|
5357
6619
|
const toDelete = selectSessionsToDelete(
|
|
5358
6620
|
sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
|
|
5359
6621
|
{
|
|
5360
|
-
maxAgeMs:
|
|
5361
|
-
maxCount:
|
|
6622
|
+
maxAgeMs: config.maxAgeMs,
|
|
6623
|
+
maxCount: config.maxCount,
|
|
5362
6624
|
nowMs: Date.now(),
|
|
5363
6625
|
protectedIds: driver.protectedSessionIds()
|
|
5364
6626
|
}
|
|
@@ -5394,7 +6656,7 @@ async function runSweep(state, driver, config2) {
|
|
|
5394
6656
|
}
|
|
5395
6657
|
}
|
|
5396
6658
|
function scheduleSessionCleanup(state, driver, options) {
|
|
5397
|
-
const
|
|
6659
|
+
const config = resolveSessionCleanupConfig(
|
|
5398
6660
|
{
|
|
5399
6661
|
maxAge: options.sessionCleanupMaxAge,
|
|
5400
6662
|
maxCount: options.sessionCleanupMaxCount,
|
|
@@ -5402,21 +6664,129 @@ function scheduleSessionCleanup(state, driver, options) {
|
|
|
5402
6664
|
},
|
|
5403
6665
|
process.env
|
|
5404
6666
|
);
|
|
5405
|
-
for (const warning2 of
|
|
6667
|
+
for (const warning2 of config.warnings) {
|
|
5406
6668
|
logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
|
|
5407
6669
|
}
|
|
5408
|
-
if (!
|
|
6670
|
+
if (!config.enabled) return;
|
|
5409
6671
|
logActivity(state, {
|
|
5410
6672
|
type: "info",
|
|
5411
|
-
message: `Session cleanup enabled (age=${
|
|
6673
|
+
message: `Session cleanup enabled (age=${config.maxAgeMs ?? "\u2014"}, count=${config.maxCount ?? "\u2014"}, interval=${config.intervalMs}ms)`
|
|
5412
6674
|
});
|
|
5413
|
-
const interval = setInterval(() => void runSweep(state, driver,
|
|
6675
|
+
const interval = setInterval(() => void runSweep(state, driver, config), config.intervalMs);
|
|
5414
6676
|
const firstSweep = setTimeout(
|
|
5415
|
-
() => void runSweep(state, driver,
|
|
6677
|
+
() => void runSweep(state, driver, config),
|
|
5416
6678
|
SESSION_CLEANUP_FIRST_SWEEP_MS
|
|
5417
6679
|
);
|
|
5418
6680
|
state.sessionCleanupTimers.push(interval, firstSweep);
|
|
5419
6681
|
}
|
|
6682
|
+
function claudeUsageFailureStreakSuffix(consecutiveFailures) {
|
|
6683
|
+
return consecutiveFailures > 1 ? ` (${consecutiveFailures} consecutive failures)` : "";
|
|
6684
|
+
}
|
|
6685
|
+
function scheduleClaudeUsageReporting(state, options) {
|
|
6686
|
+
const { mode, warnings } = resolveClaudeUsageReportingMode(
|
|
6687
|
+
options.claudeUsageReporting,
|
|
6688
|
+
process.env
|
|
6689
|
+
);
|
|
6690
|
+
for (const warning2 of warnings) {
|
|
6691
|
+
logActivity(state, {
|
|
6692
|
+
type: "info",
|
|
6693
|
+
level: "warn",
|
|
6694
|
+
message: `Claude usage reporting: ${warning2}`
|
|
6695
|
+
});
|
|
6696
|
+
}
|
|
6697
|
+
if (mode === "off") {
|
|
6698
|
+
logActivity(state, {
|
|
6699
|
+
type: "info",
|
|
6700
|
+
level: "debug",
|
|
6701
|
+
message: "Claude usage reporting is off (--claude-usage-reporting off)"
|
|
6702
|
+
});
|
|
6703
|
+
return null;
|
|
6704
|
+
}
|
|
6705
|
+
let consecutiveFailures = 0;
|
|
6706
|
+
let armed = false;
|
|
6707
|
+
let rearmRequested = false;
|
|
6708
|
+
const scheduleNextTick = () => {
|
|
6709
|
+
armed = true;
|
|
6710
|
+
rearmRequested = false;
|
|
6711
|
+
state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
|
|
6712
|
+
};
|
|
6713
|
+
const rearm = () => {
|
|
6714
|
+
if (armed) {
|
|
6715
|
+
rearmRequested = true;
|
|
6716
|
+
return;
|
|
6717
|
+
}
|
|
6718
|
+
rearmRequested = false;
|
|
6719
|
+
armed = true;
|
|
6720
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6721
|
+
};
|
|
6722
|
+
const tick = async (isProbe) => {
|
|
6723
|
+
try {
|
|
6724
|
+
const usage = await getClaudeUsage();
|
|
6725
|
+
const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
|
|
6726
|
+
if (result.ok) {
|
|
6727
|
+
if (consecutiveFailures > 0) {
|
|
6728
|
+
logActivity(state, {
|
|
6729
|
+
type: "info",
|
|
6730
|
+
level: "info",
|
|
6731
|
+
message: "Claude usage reporting recovered"
|
|
6732
|
+
});
|
|
6733
|
+
}
|
|
6734
|
+
consecutiveFailures = 0;
|
|
6735
|
+
logActivity(state, {
|
|
6736
|
+
type: "info",
|
|
6737
|
+
level: "debug",
|
|
6738
|
+
message: "Reported Claude usage to Evident"
|
|
6739
|
+
});
|
|
6740
|
+
} else {
|
|
6741
|
+
consecutiveFailures++;
|
|
6742
|
+
logActivity(state, {
|
|
6743
|
+
type: "info",
|
|
6744
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6745
|
+
message: `Failed to report Claude usage: ${result.error}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6746
|
+
});
|
|
6747
|
+
}
|
|
6748
|
+
scheduleNextTick();
|
|
6749
|
+
} catch (error2) {
|
|
6750
|
+
if (error2 instanceof ClaudeUsageError && isLocalCredentialProblem(error2)) {
|
|
6751
|
+
if (mode === "on") {
|
|
6752
|
+
logActivity(state, {
|
|
6753
|
+
type: "info",
|
|
6754
|
+
level: "warn",
|
|
6755
|
+
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"
|
|
6756
|
+
});
|
|
6757
|
+
scheduleNextTick();
|
|
6758
|
+
} else if (isProbe) {
|
|
6759
|
+
logActivity(state, {
|
|
6760
|
+
type: "info",
|
|
6761
|
+
level: "debug",
|
|
6762
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6763
|
+
});
|
|
6764
|
+
armed = false;
|
|
6765
|
+
if (rearmRequested) rearm();
|
|
6766
|
+
} else {
|
|
6767
|
+
logActivity(state, {
|
|
6768
|
+
type: "info",
|
|
6769
|
+
level: "debug",
|
|
6770
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6771
|
+
});
|
|
6772
|
+
scheduleNextTick();
|
|
6773
|
+
}
|
|
6774
|
+
} else {
|
|
6775
|
+
consecutiveFailures++;
|
|
6776
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
6777
|
+
logActivity(state, {
|
|
6778
|
+
type: "info",
|
|
6779
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6780
|
+
message: `Claude usage reporting failed: ${message}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6781
|
+
});
|
|
6782
|
+
scheduleNextTick();
|
|
6783
|
+
}
|
|
6784
|
+
}
|
|
6785
|
+
};
|
|
6786
|
+
armed = true;
|
|
6787
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6788
|
+
return rearm;
|
|
6789
|
+
}
|
|
5420
6790
|
async function notifyOffline(state) {
|
|
5421
6791
|
if (!state.agentId || !state.authHeader) return;
|
|
5422
6792
|
if (!state.connected) {
|
|
@@ -5434,13 +6804,29 @@ async function notifyOffline(state) {
|
|
|
5434
6804
|
if (state.interactive) displayStatus(state);
|
|
5435
6805
|
}
|
|
5436
6806
|
}
|
|
6807
|
+
async function timeShutdownPhase(state, durations, name, run2) {
|
|
6808
|
+
const startedAt = Date.now();
|
|
6809
|
+
try {
|
|
6810
|
+
return await run2();
|
|
6811
|
+
} finally {
|
|
6812
|
+
const elapsedMs = Date.now() - startedAt;
|
|
6813
|
+
durations[name] = elapsedMs;
|
|
6814
|
+
log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
|
|
6815
|
+
}
|
|
6816
|
+
}
|
|
5437
6817
|
async function cleanup(state, opts = {}) {
|
|
6818
|
+
const durations = {};
|
|
5438
6819
|
state.running = false;
|
|
5439
6820
|
for (const timer of state.sessionCleanupTimers) {
|
|
5440
6821
|
clearInterval(timer);
|
|
5441
6822
|
clearTimeout(timer);
|
|
5442
6823
|
}
|
|
5443
6824
|
state.sessionCleanupTimers = [];
|
|
6825
|
+
if (state.claudeUsageTimer) {
|
|
6826
|
+
clearTimeout(state.claudeUsageTimer);
|
|
6827
|
+
state.claudeUsageTimer = null;
|
|
6828
|
+
}
|
|
6829
|
+
state.claudeUsageRearm = null;
|
|
5444
6830
|
if (opts.graceful && state.channelDriver) {
|
|
5445
6831
|
state.channelDriver.stop();
|
|
5446
6832
|
log2(state, "Draining in-flight channel work before shutdown...");
|
|
@@ -5448,7 +6834,13 @@ async function cleanup(state, opts = {}) {
|
|
|
5448
6834
|
logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
|
|
5449
6835
|
displayStatus(state);
|
|
5450
6836
|
}
|
|
5451
|
-
const
|
|
6837
|
+
const driver = state.channelDriver;
|
|
6838
|
+
const settled = await timeShutdownPhase(
|
|
6839
|
+
state,
|
|
6840
|
+
durations,
|
|
6841
|
+
"drain",
|
|
6842
|
+
() => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
|
|
6843
|
+
);
|
|
5452
6844
|
if (!settled) {
|
|
5453
6845
|
logActivity(state, {
|
|
5454
6846
|
type: "info",
|
|
@@ -5457,13 +6849,15 @@ async function cleanup(state, opts = {}) {
|
|
|
5457
6849
|
if (state.interactive) displayStatus(state);
|
|
5458
6850
|
}
|
|
5459
6851
|
}
|
|
5460
|
-
await notifyOffline(state);
|
|
6852
|
+
await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
|
|
5461
6853
|
if (state.connection) {
|
|
5462
|
-
state.connection
|
|
6854
|
+
const connection = state.connection;
|
|
6855
|
+
await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
|
|
5463
6856
|
state.connection = null;
|
|
5464
6857
|
}
|
|
5465
6858
|
if (state.opencodeProcess) {
|
|
5466
|
-
|
|
6859
|
+
const opencodeProcess = state.opencodeProcess;
|
|
6860
|
+
await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
|
|
5467
6861
|
if (state.interactive) {
|
|
5468
6862
|
logActivity(state, { type: "info", message: "Stopped OpenCode process" });
|
|
5469
6863
|
displayStatus(state);
|
|
@@ -5472,6 +6866,7 @@ async function cleanup(state, opts = {}) {
|
|
|
5472
6866
|
}
|
|
5473
6867
|
state.opencodeProcess = null;
|
|
5474
6868
|
}
|
|
6869
|
+
return durations;
|
|
5475
6870
|
}
|
|
5476
6871
|
async function run(options) {
|
|
5477
6872
|
const interactive = isInteractive(options.json);
|
|
@@ -5479,7 +6874,7 @@ async function run(options) {
|
|
|
5479
6874
|
let fileSyncDirectories;
|
|
5480
6875
|
try {
|
|
5481
6876
|
logLevel = resolveLogLevel(options);
|
|
5482
|
-
fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo,
|
|
6877
|
+
fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir3());
|
|
5483
6878
|
} catch (error2) {
|
|
5484
6879
|
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
5485
6880
|
if (options.json) {
|
|
@@ -5512,8 +6907,11 @@ async function run(options) {
|
|
|
5512
6907
|
messageCount: 0,
|
|
5513
6908
|
lastProxiedActivityAt: null,
|
|
5514
6909
|
sessionCleanupTimers: [],
|
|
6910
|
+
claudeUsageTimer: null,
|
|
6911
|
+
claudeUsageRearm: null,
|
|
5515
6912
|
authHeader: ""
|
|
5516
6913
|
};
|
|
6914
|
+
setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
|
|
5517
6915
|
if (fileSyncDirectories.length > 0) {
|
|
5518
6916
|
log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
|
|
5519
6917
|
} else {
|
|
@@ -5542,14 +6940,38 @@ async function run(options) {
|
|
|
5542
6940
|
const handleSignal = async () => {
|
|
5543
6941
|
if (state.shuttingDown) return;
|
|
5544
6942
|
state.shuttingDown = true;
|
|
6943
|
+
const shutdownStartedAt = Date.now();
|
|
5545
6944
|
if (state.interactive) {
|
|
5546
6945
|
logActivity(state, { type: "info", message: "Shutting down..." });
|
|
5547
6946
|
displayStatus(state);
|
|
5548
6947
|
} else {
|
|
5549
6948
|
log2(state, "Shutting down...");
|
|
5550
6949
|
}
|
|
5551
|
-
await cleanup(state, { graceful: true });
|
|
5552
|
-
|
|
6950
|
+
const durations = await cleanup(state, { graceful: true });
|
|
6951
|
+
const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
|
|
6952
|
+
await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
|
|
6953
|
+
let timer;
|
|
6954
|
+
const flushed = shutdownTelemetry().then(
|
|
6955
|
+
() => true,
|
|
6956
|
+
(error2) => {
|
|
6957
|
+
log2(
|
|
6958
|
+
state,
|
|
6959
|
+
`Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
|
|
6960
|
+
"warn"
|
|
6961
|
+
);
|
|
6962
|
+
return true;
|
|
6963
|
+
}
|
|
6964
|
+
);
|
|
6965
|
+
const timedOut = new Promise((resolve3) => {
|
|
6966
|
+
timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
|
|
6967
|
+
});
|
|
6968
|
+
if (!await Promise.race([flushed, timedOut])) {
|
|
6969
|
+
log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
|
|
6970
|
+
}
|
|
6971
|
+
clearTimeout(timer);
|
|
6972
|
+
});
|
|
6973
|
+
const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
|
|
6974
|
+
log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
|
|
5553
6975
|
process.exit(0);
|
|
5554
6976
|
};
|
|
5555
6977
|
process.on("SIGINT", handleSignal);
|
|
@@ -5566,6 +6988,7 @@ async function run(options) {
|
|
|
5566
6988
|
console.log(chalk6.dim("Or run `evident login` for interactive authentication"));
|
|
5567
6989
|
blank();
|
|
5568
6990
|
process.exit(1);
|
|
6991
|
+
return;
|
|
5569
6992
|
}
|
|
5570
6993
|
blank();
|
|
5571
6994
|
console.log(chalk6.yellow("You are not logged in to Evident."));
|
|
@@ -5610,6 +7033,7 @@ async function run(options) {
|
|
|
5610
7033
|
} else {
|
|
5611
7034
|
printError(resolved.error || "Failed to resolve runner ID from key");
|
|
5612
7035
|
process.exit(1);
|
|
7036
|
+
return;
|
|
5613
7037
|
}
|
|
5614
7038
|
} else {
|
|
5615
7039
|
printError(
|
|
@@ -5623,6 +7047,7 @@ async function run(options) {
|
|
|
5623
7047
|
);
|
|
5624
7048
|
blank();
|
|
5625
7049
|
process.exit(1);
|
|
7050
|
+
return;
|
|
5626
7051
|
}
|
|
5627
7052
|
}
|
|
5628
7053
|
telemetry.info(
|
|
@@ -5663,40 +7088,67 @@ async function run(options) {
|
|
|
5663
7088
|
}
|
|
5664
7089
|
spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
|
|
5665
7090
|
state.agentName = validation.agent.name;
|
|
7091
|
+
const microvmId = process.env.MICROVM_ID?.trim();
|
|
7092
|
+
if (microvmId) {
|
|
7093
|
+
const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
|
|
7094
|
+
if (reported.ok) {
|
|
7095
|
+
log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
|
|
7096
|
+
} else {
|
|
7097
|
+
const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
|
|
7098
|
+
log2(state, message, "warn");
|
|
7099
|
+
if (state.interactive && !state.json) {
|
|
7100
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
7101
|
+
}
|
|
7102
|
+
}
|
|
7103
|
+
} else {
|
|
7104
|
+
log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
|
|
7105
|
+
}
|
|
7106
|
+
const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
|
|
7107
|
+
for (const warning2 of opencodeStartTimeoutWarnings) {
|
|
7108
|
+
logActivity(state, { type: "info", level: "warn", message: warning2 });
|
|
7109
|
+
}
|
|
5666
7110
|
const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
|
|
5667
7111
|
try {
|
|
5668
7112
|
const oc = await ensureOpenCodeRunning({
|
|
5669
7113
|
port: state.port,
|
|
5670
7114
|
interactive: state.interactive,
|
|
5671
7115
|
agentId: state.agentId,
|
|
5672
|
-
log: (message) => log2(state, message)
|
|
7116
|
+
log: (message) => log2(state, message),
|
|
7117
|
+
startTimeoutMs: opencodeStartTimeoutMs
|
|
5673
7118
|
});
|
|
5674
7119
|
state.port = oc.port;
|
|
5675
7120
|
state.opencodeProcess = oc.process;
|
|
5676
7121
|
state.opencodeVersion = oc.version;
|
|
5677
|
-
state.opencodeConnected = oc.
|
|
7122
|
+
state.opencodeConnected = oc.notReadyReason === null;
|
|
5678
7123
|
const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
|
|
5679
7124
|
ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
|
|
5680
|
-
|
|
5681
|
-
|
|
5682
|
-
|
|
5683
|
-
|
|
5684
|
-
|
|
7125
|
+
if (!state.interactive && oc.notReadyReason !== null) {
|
|
7126
|
+
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}).`;
|
|
7127
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
7128
|
+
} else {
|
|
7129
|
+
const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
|
|
7130
|
+
if (versionWarning) {
|
|
7131
|
+
log2(state, versionWarning, "warn");
|
|
7132
|
+
if (state.interactive && !state.json) {
|
|
7133
|
+
logActivity(state, { type: "info", level: "warn", message: versionWarning });
|
|
7134
|
+
}
|
|
5685
7135
|
}
|
|
5686
|
-
|
|
5687
|
-
|
|
5688
|
-
|
|
5689
|
-
|
|
5690
|
-
|
|
5691
|
-
|
|
5692
|
-
|
|
5693
|
-
|
|
5694
|
-
|
|
5695
|
-
|
|
5696
|
-
|
|
5697
|
-
|
|
5698
|
-
|
|
5699
|
-
|
|
7136
|
+
const noProviderWarning = buildNoProviderWarning(
|
|
7137
|
+
await hasAnyConfiguredProvider(state.port)
|
|
7138
|
+
);
|
|
7139
|
+
if (noProviderWarning) {
|
|
7140
|
+
log2(state, noProviderWarning, "warn");
|
|
7141
|
+
if (state.interactive && !state.json) {
|
|
7142
|
+
logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
|
|
7143
|
+
blank();
|
|
7144
|
+
console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
|
|
7145
|
+
console.log(
|
|
7146
|
+
chalk6.dim(
|
|
7147
|
+
`Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
|
|
7148
|
+
)
|
|
7149
|
+
);
|
|
7150
|
+
blank();
|
|
7151
|
+
}
|
|
5700
7152
|
}
|
|
5701
7153
|
}
|
|
5702
7154
|
} catch (error2) {
|
|
@@ -5714,7 +7166,7 @@ async function run(options) {
|
|
|
5714
7166
|
// #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
|
|
5715
7167
|
// REJECTED with `file_sync_disabled` on the ack, not silently ignored.
|
|
5716
7168
|
fileSyncDirectories,
|
|
5717
|
-
homeDir:
|
|
7169
|
+
homeDir: homedir3(),
|
|
5718
7170
|
log: (entry) => (
|
|
5719
7171
|
// Thread the driver's real level straight through so `debug`/`warn`
|
|
5720
7172
|
// survive the sink filter (they no longer collapse to info). `type`
|
|
@@ -5741,6 +7193,18 @@ async function run(options) {
|
|
|
5741
7193
|
type: "info",
|
|
5742
7194
|
message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
|
|
5743
7195
|
});
|
|
7196
|
+
if (options.tunnelReadyFile) {
|
|
7197
|
+
const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
|
|
7198
|
+
if (marker.ok) {
|
|
7199
|
+
log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
|
|
7200
|
+
} else {
|
|
7201
|
+
log2(
|
|
7202
|
+
state,
|
|
7203
|
+
`Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
|
|
7204
|
+
"error"
|
|
7205
|
+
);
|
|
7206
|
+
}
|
|
7207
|
+
}
|
|
5744
7208
|
emitAgentConnected(state.agentId, {
|
|
5745
7209
|
port: state.port,
|
|
5746
7210
|
cli_version: getCliVersion(),
|
|
@@ -5831,6 +7295,7 @@ async function run(options) {
|
|
|
5831
7295
|
throw error2;
|
|
5832
7296
|
}
|
|
5833
7297
|
scheduleSessionCleanup(state, channelDriver, options);
|
|
7298
|
+
state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
|
|
5834
7299
|
if (!interactive || state.json) {
|
|
5835
7300
|
log2(state, "Driving channel messages...");
|
|
5836
7301
|
}
|
|
@@ -5885,13 +7350,18 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
|
|
|
5885
7350
|
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
7351
|
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
7352
|
program.command("whoami").description("Show the currently logged in user").action(whoami);
|
|
7353
|
+
program.command("status").description("Check whether the configured credentials can reach Evident").option("--json", "Output in JSON format").action((options) => status({ json: options.json }));
|
|
7354
|
+
program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
|
|
5888
7355
|
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
7356
|
"-a, --agent [id]",
|
|
5890
7357
|
"Deprecated alias for --runner (still supported; --runner wins if both are given)"
|
|
5891
7358
|
).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
|
|
5892
7359
|
"--log-level <level>",
|
|
5893
7360
|
"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(
|
|
7361
|
+
).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(
|
|
7362
|
+
"--opencode-start-timeout <seconds>",
|
|
7363
|
+
"Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
|
|
7364
|
+
).option("--json", "Output in JSON format").option(
|
|
5895
7365
|
"--session-cleanup-max-age <duration>",
|
|
5896
7366
|
"Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
|
|
5897
7367
|
).option(
|
|
@@ -5900,11 +7370,17 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5900
7370
|
).option(
|
|
5901
7371
|
"--session-cleanup-interval <duration>",
|
|
5902
7372
|
"How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
|
|
7373
|
+
).option(
|
|
7374
|
+
"--claude-usage-reporting <mode>",
|
|
7375
|
+
"Report Claude subscription usage to Evident: auto | on | off (default: auto). Env: EVIDENT_CLAUDE_USAGE_REPORTING"
|
|
5903
7376
|
).option(
|
|
5904
7377
|
"--enable-file-sync-to <dir>",
|
|
5905
7378
|
"Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
|
|
5906
7379
|
(value, previous) => previous.concat([value]),
|
|
5907
7380
|
[]
|
|
7381
|
+
).option(
|
|
7382
|
+
"--tunnel-ready-file <path>",
|
|
7383
|
+
"Path to write once the tunnel is connected (opt-in; unused unless an operator/image sets it)"
|
|
5908
7384
|
).action(
|
|
5909
7385
|
(options) => {
|
|
5910
7386
|
run({
|
|
@@ -5917,14 +7393,21 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5917
7393
|
verbose: options.verbose,
|
|
5918
7394
|
conversation: options.conversation,
|
|
5919
7395
|
idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
|
|
7396
|
+
// Raw string — validation/precedence is single-sourced in run.ts's
|
|
7397
|
+
// resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
|
|
7398
|
+
opencodeStartTimeout: options.opencodeStartTimeout,
|
|
5920
7399
|
json: options.json,
|
|
5921
7400
|
// Raw strings — the resolver in run.ts single-sources parsing (M1).
|
|
5922
7401
|
sessionCleanupMaxAge: options.sessionCleanupMaxAge,
|
|
5923
7402
|
sessionCleanupMaxCount: options.sessionCleanupMaxCount,
|
|
5924
7403
|
sessionCleanupInterval: options.sessionCleanupInterval,
|
|
7404
|
+
// Raw string — the resolver in run.ts single-sources parsing
|
|
7405
|
+
// (resolveClaudeUsageReportingMode).
|
|
7406
|
+
claudeUsageReporting: options.claudeUsageReporting,
|
|
5925
7407
|
// Raw values — expansion/validation is single-sourced in run.ts's
|
|
5926
7408
|
// resolveFileSyncDirectories.
|
|
5927
|
-
enableFileSyncTo: options.enableFileSyncTo
|
|
7409
|
+
enableFileSyncTo: options.enableFileSyncTo,
|
|
7410
|
+
tunnelReadyFile: options.tunnelReadyFile
|
|
5928
7411
|
});
|
|
5929
7412
|
}
|
|
5930
7413
|
);
|