@evident-ai/cli 3.1.1-dev.97fd633 → 3.1.1-dev.9deb35f
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 +1926 -384
- 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 +3281,58 @@ 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();
|
|
3311
|
+
/**
|
|
3312
|
+
* "Already emitted `redrive_outcome_unreported` for THIS (message, outcome)
|
|
3313
|
+
* streak" (Class B, #1340: the runner DECIDED reattach/settle/fail_permanent
|
|
3314
|
+
* but its own PATCH to record it failed — distinct from Class A's
|
|
3315
|
+
* `redrive_poll_failed`, where opencode itself can't be observed). Keyed by
|
|
3316
|
+
* message id, valued by the outcome currently failing to report, so a
|
|
3317
|
+
* change of outcome starts a fresh signal. Cleared by
|
|
3318
|
+
* `clearRedriveUnresolved` the instant either PATCH succeeds.
|
|
3319
|
+
*/
|
|
3320
|
+
redriveOutcomeUnreportedSignalled = /* @__PURE__ */ new Map();
|
|
3321
|
+
/**
|
|
3322
|
+
* Consecutive-UNCONFIRMED-dispatch streak for a `pending` row with NO stored
|
|
3323
|
+
* `opencode_message_id` yet — i.e. one that has never even reached the
|
|
3324
|
+
* re-drive fence above. `sendPromptAsync`'s POST may 2xx, but its own
|
|
3325
|
+
* read-back retries can never confirm the assigned id when the session's
|
|
3326
|
+
* message list is PERMANENTLY unreadable (e.g. a corrupted local opencode
|
|
3327
|
+
* SQLite DB, #1345/#1348's exact fault, just hit BEFORE the row is ever
|
|
3328
|
+
* dispatched instead of after). Unlike an already-dispatched row, THIS row has
|
|
3329
|
+
* no other safety net at all: the lifecycle cron only reclaims `status =
|
|
3330
|
+
* 'processing'` rows, and a row stuck here never reaches `processing`. Keyed
|
|
3331
|
+
* by message id, carrying `sessionId` so a session change (a fresh one bound
|
|
3332
|
+
* after abandonment) starts a new streak rather than inheriting the old
|
|
3333
|
+
* session's count — same shape as `redrivePollFailures` above.
|
|
3334
|
+
*/
|
|
3335
|
+
unconfirmedDispatchFailures = /* @__PURE__ */ new Map();
|
|
2639
3336
|
/**
|
|
2640
3337
|
* "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
|
|
2641
3338
|
* Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
|
|
@@ -2724,23 +3421,23 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2724
3421
|
* and stops opencode.
|
|
2725
3422
|
*/
|
|
2726
3423
|
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 =
|
|
3424
|
+
constructor(config) {
|
|
3425
|
+
this.agentId = config.agentId;
|
|
3426
|
+
this.port = config.port;
|
|
3427
|
+
this.apiUrl = config.apiUrl.replace(/\/$/, "");
|
|
3428
|
+
this.getAuthHeader = config.getAuthHeader;
|
|
3429
|
+
this.conversationFilter = config.conversationFilter ?? null;
|
|
3430
|
+
this.retry = { ...DEFAULT_RETRY_POLICY, ...config.retry };
|
|
3431
|
+
this.log = config.log ?? (() => {
|
|
2735
3432
|
});
|
|
2736
|
-
this.fetchImpl =
|
|
2737
|
-
this.sleep =
|
|
2738
|
-
this.pausedPollIntervalMs =
|
|
2739
|
-
this.pausedMaxWaitMs =
|
|
2740
|
-
this.stuckQueuedMs =
|
|
2741
|
-
this.now =
|
|
2742
|
-
this.fileSyncDirectories =
|
|
2743
|
-
this.homeDir =
|
|
3433
|
+
this.fetchImpl = config.fetchImpl ?? fetch;
|
|
3434
|
+
this.sleep = config.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
|
|
3435
|
+
this.pausedPollIntervalMs = config.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
|
|
3436
|
+
this.pausedMaxWaitMs = config.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
|
|
3437
|
+
this.stuckQueuedMs = config.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
|
|
3438
|
+
this.now = config.now ?? (() => Date.now());
|
|
3439
|
+
this.fileSyncDirectories = config.fileSyncDirectories ?? [];
|
|
3440
|
+
this.homeDir = config.homeDir ?? homedir2();
|
|
2744
3441
|
}
|
|
2745
3442
|
/** The IPv4-loopback base URL for the local `opencode serve`. */
|
|
2746
3443
|
get opencodeBase() {
|
|
@@ -2975,6 +3672,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2975
3672
|
skippedAlreadyDispatched += 1;
|
|
2976
3673
|
continue;
|
|
2977
3674
|
}
|
|
3675
|
+
if (message.opencode_message_id) {
|
|
3676
|
+
const outcome = await this.resolveRedrive(conv, sessionId, message, refusedSessionId);
|
|
3677
|
+
if (outcome !== "dispatch") {
|
|
3678
|
+
break;
|
|
3679
|
+
}
|
|
3680
|
+
}
|
|
2978
3681
|
const options = {
|
|
2979
3682
|
agent: message.opencode_agent ?? void 0,
|
|
2980
3683
|
model: message.opencode_model ?? void 0
|
|
@@ -3019,50 +3722,422 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3019
3722
|
this.sessions.delete(conv.id);
|
|
3020
3723
|
this.supersede(conv.id, sessionId);
|
|
3021
3724
|
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.`,
|
|
3725
|
+
level: "warn",
|
|
3726
|
+
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.`,
|
|
3727
|
+
conversation_id: conv.id,
|
|
3728
|
+
message_id: message.id
|
|
3729
|
+
});
|
|
3730
|
+
await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
|
|
3731
|
+
this.log({
|
|
3732
|
+
level: "warn",
|
|
3733
|
+
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)}`,
|
|
3734
|
+
conversation_id: conv.id,
|
|
3735
|
+
message_id: message.id
|
|
3736
|
+
});
|
|
3737
|
+
});
|
|
3738
|
+
this.log({
|
|
3739
|
+
level: "error",
|
|
3740
|
+
message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
|
|
3741
|
+
conversation_id: conv.id,
|
|
3742
|
+
message_id: message.id
|
|
3743
|
+
});
|
|
3744
|
+
break;
|
|
3745
|
+
}
|
|
3746
|
+
if (opencodeMessageId === null) {
|
|
3747
|
+
const streak = this.recordUnconfirmedDispatch(message.id, sessionId);
|
|
3748
|
+
if (streak < MAX_IDENTICAL_REDRIVE_POLL_FAILURES) {
|
|
3749
|
+
this.log({
|
|
3750
|
+
level: "warn",
|
|
3751
|
+
message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back (${streak}/${MAX_IDENTICAL_REDRIVE_POLL_FAILURES}) \u2014 leaving un-tracked to retry next tick`,
|
|
3752
|
+
conversation_id: conv.id,
|
|
3753
|
+
message_id: message.id
|
|
3754
|
+
});
|
|
3755
|
+
continue;
|
|
3756
|
+
}
|
|
3757
|
+
this.unconfirmedDispatchFailures.delete(message.id);
|
|
3758
|
+
this.sessions.delete(conv.id);
|
|
3759
|
+
this.supersede(conv.id, sessionId);
|
|
3760
|
+
const errorMessage = `OpenCode accepted this message ${streak} times in a row but never confirmed its assigned id (session ${sessionId.slice(0, 8)}'s message list could not be read back) \u2014 the session was abandoned; a fresh one is used for further messages.`;
|
|
3761
|
+
this.log({
|
|
3762
|
+
level: "error",
|
|
3763
|
+
message: errorMessage,
|
|
3764
|
+
conversation_id: conv.id,
|
|
3765
|
+
message_id: message.id
|
|
3766
|
+
});
|
|
3767
|
+
await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
|
|
3768
|
+
this.log({
|
|
3769
|
+
level: "warn",
|
|
3770
|
+
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)}`,
|
|
3771
|
+
conversation_id: conv.id,
|
|
3772
|
+
message_id: message.id
|
|
3773
|
+
});
|
|
3774
|
+
});
|
|
3775
|
+
break;
|
|
3776
|
+
}
|
|
3777
|
+
this.unconfirmedDispatchFailures.delete(message.id);
|
|
3778
|
+
this.dispatched.add(message.id);
|
|
3779
|
+
this.registerInFlight(conv, sessionId, message, opencodeMessageId);
|
|
3780
|
+
dispatched += 1;
|
|
3781
|
+
void this.postSignal(conv.id, message.id, "dispatched");
|
|
3782
|
+
}
|
|
3783
|
+
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3784
|
+
this.log({
|
|
3785
|
+
level: "warn",
|
|
3786
|
+
message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
|
|
3787
|
+
conversation_id: conv.id
|
|
3788
|
+
});
|
|
3789
|
+
}
|
|
3790
|
+
this.ensureWatcherRunning(sessionId);
|
|
3791
|
+
return dispatched;
|
|
3792
|
+
}
|
|
3793
|
+
/**
|
|
3794
|
+
* Poll a session's message list for the re-drive fence (#965), via the
|
|
3795
|
+
* INJECTED `fetchImpl` — NOT the imported `getSessionMessages` helper, which
|
|
3796
|
+
* hits the global `fetch` and would bypass the same override every other
|
|
3797
|
+
* opencode poll in this file respects. Mirrors `readoptProcessing`'s own
|
|
3798
|
+
* snapshot fetch (`:3081-3111`).
|
|
3799
|
+
*
|
|
3800
|
+
* Returns `{ ok: true, messages }` on a readable snapshot, or
|
|
3801
|
+
* `{ ok: false, signature }` on failure — `signature` is a string that
|
|
3802
|
+
* repeats across attempts for the SAME underlying fault (used by the
|
|
3803
|
+
* consecutive-identical-failure bound, #1348), or `null` for a thrown
|
|
3804
|
+
* exception, which is NOT countable toward that bound (a network blip / an
|
|
3805
|
+
* opencode restart also throws identically every tick, and must keep
|
|
3806
|
+
* retrying unbounded rather than ever being treated as permanent).
|
|
3807
|
+
*/
|
|
3808
|
+
async pollSessionMessagesForRedrive(conv, message, sessionId) {
|
|
3809
|
+
try {
|
|
3810
|
+
const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
|
|
3811
|
+
if (!res.ok) {
|
|
3812
|
+
const rawBody = await res.text();
|
|
3813
|
+
const normalized = normalizeRedrivePollFailureBody(rawBody);
|
|
3814
|
+
this.log({
|
|
3815
|
+
level: "warn",
|
|
3816
|
+
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`,
|
|
3817
|
+
conversation_id: conv.id,
|
|
3818
|
+
message_id: message.id
|
|
3819
|
+
});
|
|
3820
|
+
return { ok: false, signature: `HTTP ${res.status}${normalized ? `: ${normalized}` : ""}` };
|
|
3821
|
+
}
|
|
3822
|
+
const body = await res.json();
|
|
3823
|
+
if (!Array.isArray(body)) {
|
|
3824
|
+
this.log({
|
|
3825
|
+
level: "warn",
|
|
3826
|
+
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`,
|
|
3827
|
+
conversation_id: conv.id,
|
|
3828
|
+
message_id: message.id
|
|
3829
|
+
});
|
|
3830
|
+
return { ok: false, signature: "non-array message body" };
|
|
3831
|
+
}
|
|
3832
|
+
return { ok: true, messages: body };
|
|
3833
|
+
} catch (err) {
|
|
3834
|
+
this.log({
|
|
3835
|
+
level: "warn",
|
|
3836
|
+
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)}`,
|
|
3837
|
+
conversation_id: conv.id,
|
|
3838
|
+
message_id: message.id
|
|
3839
|
+
});
|
|
3840
|
+
return { ok: false, signature: null };
|
|
3841
|
+
}
|
|
3842
|
+
}
|
|
3843
|
+
/**
|
|
3844
|
+
* The re-drive fence for a `pending` row that already carries a stored
|
|
3845
|
+
* `opencode_message_id` (#965) — i.e. it has already been handed to opencode at
|
|
3846
|
+
* least once (see the invariant at `QueuedMessage.opencode_message_id`'s doc).
|
|
3847
|
+
* The lifecycle cron can falsely reclaim a `processing` row back to `pending`
|
|
3848
|
+
* mid-turn (a 5-minute liveness-staleness check racing a still-running turn);
|
|
3849
|
+
* without this fence the drain loop would re-`prompt_async` the SAME turn a
|
|
3850
|
+
* second time against live GitHub state. Mirrors `readoptOne`'s job for the
|
|
3851
|
+
* `processing` re-adopt path, but simpler: no b1/b2 preamble cross-check is
|
|
3852
|
+
* needed here because `refusedSessionId` already handles the one case
|
|
3853
|
+
* (#553 abandoned session) that path exists for.
|
|
3854
|
+
*
|
|
3855
|
+
* Only `ChannelAuthError` propagates. A poll that fails identically
|
|
3856
|
+
* `MAX_IDENTICAL_REDRIVE_POLL_FAILURES` times in a row reports the message
|
|
3857
|
+
* failed instead of retrying it (#1348) — SEPARATE from, not a replacement
|
|
3858
|
+
* for, `resolveRedriveUnresolved`'s own `pausedMaxWaitMs` bound below. Every
|
|
3859
|
+
* other failure resolves to `unresolved` and is retried whole on the next
|
|
3860
|
+
* ~2s drain tick.
|
|
3861
|
+
*/
|
|
3862
|
+
async resolveRedrive(conv, sessionId, message, refusedSessionId) {
|
|
3863
|
+
const ocId = message.opencode_message_id ?? null;
|
|
3864
|
+
if (refusedSessionId) {
|
|
3865
|
+
this.clearRedriveUnresolved(message.id);
|
|
3866
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3867
|
+
return "dispatch";
|
|
3868
|
+
}
|
|
3869
|
+
const polled = await this.pollSessionMessagesForRedrive(conv, message, sessionId);
|
|
3870
|
+
if (!polled.ok) {
|
|
3871
|
+
const streak = this.recordRedrivePollFailure(message.id, sessionId, polled.signature);
|
|
3872
|
+
if (streak >= MAX_IDENTICAL_REDRIVE_POLL_FAILURES && polled.signature !== null) {
|
|
3873
|
+
return this.failRedrivePollPermanent(conv, sessionId, message, polled.signature, streak);
|
|
3874
|
+
}
|
|
3875
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3876
|
+
}
|
|
3877
|
+
this.redrivePollFailures.delete(message.id);
|
|
3878
|
+
const messages = polled.messages;
|
|
3879
|
+
if (messages.length === 0) {
|
|
3880
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3881
|
+
}
|
|
3882
|
+
const state = messageRunState(messages, ocId ?? "");
|
|
3883
|
+
if (state === "failed" && isAbortedTerminalReply(messages, ocId ?? "")) {
|
|
3884
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3885
|
+
if (ongoing === false) {
|
|
3886
|
+
this.log({
|
|
3887
|
+
level: "info",
|
|
3888
|
+
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`,
|
|
3889
|
+
conversation_id: conv.id,
|
|
3890
|
+
message_id: message.id
|
|
3891
|
+
});
|
|
3892
|
+
this.clearRedriveUnresolved(message.id);
|
|
3893
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3894
|
+
return "dispatch";
|
|
3895
|
+
}
|
|
3896
|
+
}
|
|
3897
|
+
if (state === "done" || state === "failed") {
|
|
3898
|
+
return this.settleRedrive(conv, sessionId, message, ocId, messages, state);
|
|
3899
|
+
}
|
|
3900
|
+
if (state === "running" || state === "queued") {
|
|
3901
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3902
|
+
if (ongoing === true) {
|
|
3903
|
+
return this.reattachRedrive(conv, sessionId, message, ocId);
|
|
3904
|
+
}
|
|
3905
|
+
if (ongoing === false) {
|
|
3906
|
+
this.clearRedriveUnresolved(message.id);
|
|
3907
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3908
|
+
return "dispatch";
|
|
3909
|
+
}
|
|
3910
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3911
|
+
}
|
|
3912
|
+
this.clearRedriveUnresolved(message.id);
|
|
3913
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3914
|
+
return "dispatch";
|
|
3915
|
+
}
|
|
3916
|
+
/**
|
|
3917
|
+
* The `reattached` outcome (Task 3.3): the prior turn is STILL ONGOING per
|
|
3918
|
+
* opencode's own status map — undo the false reclaim instead of starting a
|
|
3919
|
+
* second turn.
|
|
3920
|
+
*/
|
|
3921
|
+
async reattachRedrive(conv, sessionId, message, ocId) {
|
|
3922
|
+
let anchorMs;
|
|
3923
|
+
const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
|
|
3924
|
+
if (!Number.isNaN(parsed)) {
|
|
3925
|
+
anchorMs = parsed;
|
|
3926
|
+
} else {
|
|
3927
|
+
anchorMs = this.now();
|
|
3928
|
+
this.log({
|
|
3929
|
+
level: "error",
|
|
3930
|
+
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)`,
|
|
3931
|
+
conversation_id: conv.id,
|
|
3932
|
+
message_id: message.id
|
|
3933
|
+
});
|
|
3934
|
+
}
|
|
3935
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3936
|
+
try {
|
|
3937
|
+
await this.markProcessing(conv.id, message.id, sessionId, ocId, title);
|
|
3938
|
+
} catch (err) {
|
|
3939
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3940
|
+
this.log({
|
|
3941
|
+
level: "warn",
|
|
3942
|
+
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)}`,
|
|
3943
|
+
conversation_id: conv.id,
|
|
3944
|
+
message_id: message.id
|
|
3945
|
+
});
|
|
3946
|
+
this.signalRedriveOutcomeUnreported(conv, message, "reattach");
|
|
3947
|
+
return "unresolved";
|
|
3948
|
+
}
|
|
3949
|
+
this.clearRedriveUnresolved(message.id);
|
|
3950
|
+
this.registerReadopted(conv, sessionId, message, ocId ?? "", anchorMs);
|
|
3951
|
+
this.dispatched.add(message.id);
|
|
3952
|
+
this.readopted.add(message.id);
|
|
3953
|
+
this.ensureWatcherRunning(sessionId);
|
|
3954
|
+
const watchedForMs = this.now() - anchorMs;
|
|
3955
|
+
void this.postSignal(conv.id, message.id, "redrive_reattached", {
|
|
3956
|
+
watched_for_ms: watchedForMs
|
|
3957
|
+
});
|
|
3958
|
+
this.log({
|
|
3959
|
+
level: "warn",
|
|
3960
|
+
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`,
|
|
3961
|
+
conversation_id: conv.id,
|
|
3962
|
+
message_id: message.id
|
|
3963
|
+
});
|
|
3964
|
+
return "reattached";
|
|
3965
|
+
}
|
|
3966
|
+
/**
|
|
3967
|
+
* The `settled` outcome (Task 3.2): the prior turn already finished (or
|
|
3968
|
+
* errored) while nobody was watching — deliver/report it instead of re-running.
|
|
3969
|
+
* Mirrors `readoptOne`'s `done`/`failed` branches' error discipline, simplified
|
|
3970
|
+
* (no `doneUndeliverable` park: a terminal PATCH failure here just retries next
|
|
3971
|
+
* drain, same as any other non-auth failure). The restart-abort carve-out that
|
|
3972
|
+
* keeps the two in step for `failed` lives in the caller (`resolveRedrive`, #1310),
|
|
3973
|
+
* so a row reaching this `failed` branch is a GENUINE failure.
|
|
3974
|
+
*/
|
|
3975
|
+
async settleRedrive(conv, sessionId, message, ocId, messages, state) {
|
|
3976
|
+
try {
|
|
3977
|
+
if (state === "done") {
|
|
3978
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3979
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3980
|
+
this.log({
|
|
3981
|
+
level: "info",
|
|
3982
|
+
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
3983
|
conversation_id: conv.id,
|
|
3025
3984
|
message_id: message.id
|
|
3026
3985
|
});
|
|
3027
|
-
await this.
|
|
3028
|
-
|
|
3029
|
-
|
|
3030
|
-
|
|
3031
|
-
|
|
3032
|
-
message_id: message.id
|
|
3033
|
-
});
|
|
3034
|
-
});
|
|
3986
|
+
await this.markDone(conv.id, message.id, sessionId, ocId, title, usage);
|
|
3987
|
+
} else {
|
|
3988
|
+
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3989
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3990
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3035
3991
|
this.log({
|
|
3036
3992
|
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`,
|
|
3993
|
+
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
3994
|
conversation_id: conv.id,
|
|
3048
3995
|
message_id: message.id
|
|
3049
3996
|
});
|
|
3050
|
-
|
|
3997
|
+
await this.markFailed(conv.id, message.id, sessionId, error2, usage, failure);
|
|
3051
3998
|
}
|
|
3052
|
-
|
|
3053
|
-
|
|
3054
|
-
|
|
3055
|
-
|
|
3999
|
+
} catch (err) {
|
|
4000
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
4001
|
+
this.log({
|
|
4002
|
+
level: "warn",
|
|
4003
|
+
message: `Re-drive: failed to report message ${message.id.slice(0, 8)} ${state} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
4004
|
+
conversation_id: conv.id,
|
|
4005
|
+
message_id: message.id
|
|
4006
|
+
});
|
|
4007
|
+
this.signalRedriveOutcomeUnreported(conv, message, "settle");
|
|
4008
|
+
return "unresolved";
|
|
3056
4009
|
}
|
|
3057
|
-
|
|
4010
|
+
this.clearRedriveUnresolved(message.id);
|
|
4011
|
+
void this.postSignal(conv.id, message.id, "redrive_settled");
|
|
4012
|
+
return "settled";
|
|
4013
|
+
}
|
|
4014
|
+
/**
|
|
4015
|
+
* The permanent-failure outcome (#1348): the fence's own poll of this session
|
|
4016
|
+
* failed with the SAME opencode-answered signature
|
|
4017
|
+
* `MAX_IDENTICAL_REDRIVE_POLL_FAILURES` times in a row — a transient blip
|
|
4018
|
+
* would have varied or eventually cleared (see `pollSessionMessagesForRedrive`
|
|
4019
|
+
* and `recordRedrivePollFailure`), so this is a durable fault (e.g. #1345's
|
|
4020
|
+
* corrupted opencode session) rather than something worth retrying forever.
|
|
4021
|
+
* Mirrors `settleRedrive`'s error discipline: no `usage`/`failure` args to
|
|
4022
|
+
* `markFailed` (no opencode snapshot to extract them from — this poll never
|
|
4023
|
+
* got a readable one).
|
|
4024
|
+
*/
|
|
4025
|
+
async failRedrivePollPermanent(conv, sessionId, message, signature, streak) {
|
|
4026
|
+
this.log({
|
|
4027
|
+
level: "error",
|
|
4028
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) failed to poll with the identical signature "${signature}" ${streak} times in a row \u2014 reporting the message failed instead of retrying forever`,
|
|
4029
|
+
conversation_id: conv.id,
|
|
4030
|
+
message_id: message.id
|
|
4031
|
+
});
|
|
4032
|
+
try {
|
|
4033
|
+
await this.markFailed(
|
|
4034
|
+
conv.id,
|
|
4035
|
+
message.id,
|
|
4036
|
+
sessionId,
|
|
4037
|
+
`The runner could not read this conversation's state from OpenCode (${signature}). The same failure repeated ${streak} times in a row, so the message was not retried further.`
|
|
4038
|
+
);
|
|
4039
|
+
} catch (err) {
|
|
4040
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3058
4041
|
this.log({
|
|
3059
4042
|
level: "warn",
|
|
3060
|
-
message: `
|
|
3061
|
-
conversation_id: conv.id
|
|
4043
|
+
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)}`,
|
|
4044
|
+
conversation_id: conv.id,
|
|
4045
|
+
message_id: message.id
|
|
3062
4046
|
});
|
|
4047
|
+
this.signalRedriveOutcomeUnreported(conv, message, "fail_permanent");
|
|
4048
|
+
return "unresolved";
|
|
3063
4049
|
}
|
|
3064
|
-
this.
|
|
3065
|
-
|
|
4050
|
+
this.clearRedriveUnresolved(message.id);
|
|
4051
|
+
void this.postSignal(conv.id, message.id, "redrive_poll_failed");
|
|
4052
|
+
return "settled";
|
|
4053
|
+
}
|
|
4054
|
+
/**
|
|
4055
|
+
* The bounded `unresolved` outcome (Task 3.4): opencode's state could not be
|
|
4056
|
+
* observed (snapshot unreadable/empty, or `isSessionOngoing` returned `null`).
|
|
4057
|
+
* A `pending` row is invisible to every cron arm (all require `status =
|
|
4058
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
4059
|
+
* nothing driving it — bound it to the existing `pausedMaxWaitMs` window
|
|
4060
|
+
* (reusing the knob, not a new constant) and take `dispatch` once elapsed.
|
|
4061
|
+
*/
|
|
4062
|
+
resolveRedriveUnresolved(conv, message) {
|
|
4063
|
+
const now = this.now();
|
|
4064
|
+
const since = this.redriveUnresolvedSince.get(message.id);
|
|
4065
|
+
if (since !== void 0 && now - since >= this.pausedMaxWaitMs) {
|
|
4066
|
+
this.clearRedriveUnresolved(message.id);
|
|
4067
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
4068
|
+
return "dispatch";
|
|
4069
|
+
}
|
|
4070
|
+
if (since === void 0) {
|
|
4071
|
+
this.redriveUnresolvedSince.set(message.id, now);
|
|
4072
|
+
}
|
|
4073
|
+
if (!this.redriveUnresolvedSignalled.has(message.id)) {
|
|
4074
|
+
this.redriveUnresolvedSignalled.add(message.id);
|
|
4075
|
+
void this.postSignal(conv.id, message.id, "redrive_unresolved");
|
|
4076
|
+
}
|
|
4077
|
+
return "unresolved";
|
|
4078
|
+
}
|
|
4079
|
+
/** Clear all `unresolved`/failure-streak trackers for a row (any non-`unresolved` outcome). */
|
|
4080
|
+
clearRedriveUnresolved(messageId) {
|
|
4081
|
+
this.redriveUnresolvedSince.delete(messageId);
|
|
4082
|
+
this.redriveUnresolvedSignalled.delete(messageId);
|
|
4083
|
+
this.redrivePollFailures.delete(messageId);
|
|
4084
|
+
this.redriveOutcomeUnreportedSignalled.delete(messageId);
|
|
4085
|
+
}
|
|
4086
|
+
/**
|
|
4087
|
+
* Class B (#1340): the runner DECIDED an outcome (reattach/settle/fail_permanent)
|
|
4088
|
+
* but its own PATCH to record it failed. Deliberately NOT bounded like
|
|
4089
|
+
* `resolveRedriveUnresolved`, for two load-bearing reasons: (1) the only
|
|
4090
|
+
* remedy is a PATCH to the server — the very call that is failing, so a bound
|
|
4091
|
+
* would act by retrying it; (2) this class is self-clearing by construction
|
|
4092
|
+
* (the next successful PATCH resolves it via `clearRedriveUnresolved`),
|
|
4093
|
+
* unlike #1340's deterministic, permanent shape. Fires at most once per
|
|
4094
|
+
* (message, outcome) streak.
|
|
4095
|
+
*/
|
|
4096
|
+
signalRedriveOutcomeUnreported(conv, message, outcome) {
|
|
4097
|
+
if (this.redriveOutcomeUnreportedSignalled.get(message.id) === outcome) return;
|
|
4098
|
+
this.redriveOutcomeUnreportedSignalled.set(message.id, outcome);
|
|
4099
|
+
void this.postSignal(conv.id, message.id, "redrive_outcome_unreported", {
|
|
4100
|
+
attempted_outcome: outcome
|
|
4101
|
+
});
|
|
4102
|
+
}
|
|
4103
|
+
/**
|
|
4104
|
+
* Record one poll outcome toward the re-drive fence's consecutive-identical-
|
|
4105
|
+
* failure streak (#1348) and return the resulting count. `signature === null`
|
|
4106
|
+
* (a thrown exception, H1) always clears the streak and returns `0` — it is
|
|
4107
|
+
* never countable. Otherwise the streak continues only when BOTH the session
|
|
4108
|
+
* and the signature match the previous failure; anything else (a different
|
|
4109
|
+
* session, or the same session failing a DIFFERENT way) starts a fresh streak
|
|
4110
|
+
* at `1`.
|
|
4111
|
+
*/
|
|
4112
|
+
recordRedrivePollFailure(messageId, sessionId, signature) {
|
|
4113
|
+
if (signature === null) {
|
|
4114
|
+
this.redrivePollFailures.delete(messageId);
|
|
4115
|
+
return 0;
|
|
4116
|
+
}
|
|
4117
|
+
const existing = this.redrivePollFailures.get(messageId);
|
|
4118
|
+
if (existing && existing.sessionId === sessionId && existing.signature === signature) {
|
|
4119
|
+
existing.count += 1;
|
|
4120
|
+
return existing.count;
|
|
4121
|
+
}
|
|
4122
|
+
this.redrivePollFailures.set(messageId, { sessionId, signature, count: 1 });
|
|
4123
|
+
return 1;
|
|
4124
|
+
}
|
|
4125
|
+
/**
|
|
4126
|
+
* Record one UNCONFIRMED-dispatch outcome (a `pending` row with no stored
|
|
4127
|
+
* `opencode_message_id` whose `sendPromptAsync` returned `null`) toward the
|
|
4128
|
+
* bound in `processConversation`'s dispatch loop, and return the resulting
|
|
4129
|
+
* count. Mirrors `recordRedrivePollFailure`'s session-scoping: a session
|
|
4130
|
+
* change starts a fresh streak at `1` rather than inheriting the old one's
|
|
4131
|
+
* count, since a new session is a genuinely different attempt.
|
|
4132
|
+
*/
|
|
4133
|
+
recordUnconfirmedDispatch(messageId, sessionId) {
|
|
4134
|
+
const existing = this.unconfirmedDispatchFailures.get(messageId);
|
|
4135
|
+
if (existing && existing.sessionId === sessionId) {
|
|
4136
|
+
existing.count += 1;
|
|
4137
|
+
return existing.count;
|
|
4138
|
+
}
|
|
4139
|
+
this.unconfirmedDispatchFailures.set(messageId, { sessionId, count: 1 });
|
|
4140
|
+
return 1;
|
|
3066
4141
|
}
|
|
3067
4142
|
/**
|
|
3068
4143
|
* Record that `sessionId` is no longer a valid binding for `conversationId`
|
|
@@ -3125,7 +4200,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3125
4200
|
const directory = await this.resolveOpenCodeDirectory();
|
|
3126
4201
|
const sessionId = await createOpenCodeSession(this.port, directory);
|
|
3127
4202
|
this.sessions.set(conversationId, sessionId);
|
|
3128
|
-
await this.persistSession(conversationId, sessionId).catch(() => {
|
|
4203
|
+
await this.persistSession(conversationId, sessionId).catch((err) => {
|
|
4204
|
+
this.log({
|
|
4205
|
+
level: "warn",
|
|
4206
|
+
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)}`,
|
|
4207
|
+
conversation_id: conversationId
|
|
4208
|
+
});
|
|
3129
4209
|
});
|
|
3130
4210
|
return sessionId;
|
|
3131
4211
|
}
|
|
@@ -3317,12 +4397,17 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3317
4397
|
stuckReported: false,
|
|
3318
4398
|
lastAliveAt: 0,
|
|
3319
4399
|
aliveInFlight: false,
|
|
4400
|
+
titleSynced: false,
|
|
4401
|
+
titleSyncInFlight: false,
|
|
3320
4402
|
awaitingHumanLatched: false,
|
|
3321
4403
|
pausedOnQuestion: false,
|
|
3322
4404
|
pausedOnPermission: false,
|
|
3323
4405
|
pausedClearConfirmed: false,
|
|
3324
4406
|
pausedInFlight: false,
|
|
3325
|
-
deliveryDeadlineAnchored: false
|
|
4407
|
+
deliveryDeadlineAnchored: false,
|
|
4408
|
+
b2PinnedSinceMs: 0,
|
|
4409
|
+
b2LastDescendantCheckMs: 0,
|
|
4410
|
+
b2AbandonedSignalled: false
|
|
3326
4411
|
});
|
|
3327
4412
|
}
|
|
3328
4413
|
/**
|
|
@@ -3337,9 +4422,10 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3337
4422
|
* opencode reports ACTIVELY `running` is watched to completion (its liveness
|
|
3338
4423
|
* heartbeat keeps the cron off its row), while a re-adopted turn that is paused
|
|
3339
4424
|
* awaiting a human — or queued/unreachable — is still bounded by `deadline` and
|
|
3340
|
-
* handed to the cron.
|
|
3341
|
-
*
|
|
3342
|
-
*
|
|
4425
|
+
* handed to the cron. Real invariant (#965): the cron MAY reclaim a row this
|
|
4426
|
+
* runner still holds; a reclaimed row that already ran is never re-dispatched
|
|
4427
|
+
* while opencode reports its turn ongoing (readopt's own gate here, and the
|
|
4428
|
+
* `pending`-row re-drive fence, `resolveRedrive`). `dispatchedAt` stays `now`
|
|
3343
4429
|
* (only the appear-guard uses it).
|
|
3344
4430
|
*
|
|
3345
4431
|
* `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
|
|
@@ -3390,12 +4476,17 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3390
4476
|
// with no extra `re_adopted` signal needed (folds old WI-6).
|
|
3391
4477
|
lastAliveAt: 0,
|
|
3392
4478
|
aliveInFlight: false,
|
|
4479
|
+
titleSynced: false,
|
|
4480
|
+
titleSyncInFlight: false,
|
|
3393
4481
|
awaitingHumanLatched: false,
|
|
3394
4482
|
pausedOnQuestion: false,
|
|
3395
4483
|
pausedOnPermission: false,
|
|
3396
4484
|
pausedClearConfirmed: false,
|
|
3397
4485
|
pausedInFlight: false,
|
|
3398
|
-
deliveryDeadlineAnchored: false
|
|
4486
|
+
deliveryDeadlineAnchored: false,
|
|
4487
|
+
b2PinnedSinceMs: 0,
|
|
4488
|
+
b2LastDescendantCheckMs: 0,
|
|
4489
|
+
b2AbandonedSignalled: false
|
|
3399
4490
|
});
|
|
3400
4491
|
}
|
|
3401
4492
|
/**
|
|
@@ -3557,58 +4648,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3557
4648
|
}
|
|
3558
4649
|
}
|
|
3559
4650
|
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);
|
|
4651
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
3612
4652
|
return;
|
|
3613
4653
|
}
|
|
3614
4654
|
if (state === "failed") {
|
|
@@ -3622,8 +4662,16 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3622
4662
|
message_id: inFlight.evidentMessageId
|
|
3623
4663
|
});
|
|
3624
4664
|
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4665
|
+
const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
|
|
3625
4666
|
try {
|
|
3626
|
-
await this.markFailed(
|
|
4667
|
+
await this.markFailed(
|
|
4668
|
+
conv.id,
|
|
4669
|
+
inFlight.evidentMessageId,
|
|
4670
|
+
sessionId,
|
|
4671
|
+
error2,
|
|
4672
|
+
usage,
|
|
4673
|
+
failure
|
|
4674
|
+
);
|
|
3627
4675
|
} catch (err) {
|
|
3628
4676
|
if (err instanceof ChannelAuthError) throw err;
|
|
3629
4677
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -3668,6 +4716,44 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3668
4716
|
});
|
|
3669
4717
|
}
|
|
3670
4718
|
const activelyRunning = state === "running" && !awaitingHuman;
|
|
4719
|
+
const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
|
|
4720
|
+
const snapshotReadable = messages != null && messages.length > 0;
|
|
4721
|
+
if (!pinnedNow) {
|
|
4722
|
+
if (snapshotReadable) {
|
|
4723
|
+
inFlight.b2PinnedSinceMs = 0;
|
|
4724
|
+
inFlight.b2LastDescendantCheckMs = 0;
|
|
4725
|
+
inFlight.b2AbandonedSignalled = false;
|
|
4726
|
+
}
|
|
4727
|
+
} else {
|
|
4728
|
+
if (inFlight.b2AbandonedSignalled) {
|
|
4729
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4730
|
+
return;
|
|
4731
|
+
}
|
|
4732
|
+
if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
|
|
4733
|
+
const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
|
|
4734
|
+
if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
|
|
4735
|
+
inFlight.b2LastDescendantCheckMs = this.now();
|
|
4736
|
+
const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
|
|
4737
|
+
if (isB2AbandonmentConfirmed({
|
|
4738
|
+
pinnedForMs,
|
|
4739
|
+
minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
|
|
4740
|
+
descendantOngoing
|
|
4741
|
+
})) {
|
|
4742
|
+
inFlight.b2AbandonedSignalled = true;
|
|
4743
|
+
this.log({
|
|
4744
|
+
level: "warn",
|
|
4745
|
+
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`,
|
|
4746
|
+
conversation_id: conv.id,
|
|
4747
|
+
message_id: id
|
|
4748
|
+
});
|
|
4749
|
+
void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
|
|
4750
|
+
watched_for_ms: pinnedForMs
|
|
4751
|
+
});
|
|
4752
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4753
|
+
return;
|
|
4754
|
+
}
|
|
4755
|
+
}
|
|
4756
|
+
}
|
|
3671
4757
|
if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
|
|
3672
4758
|
this.log({
|
|
3673
4759
|
level: "warn",
|
|
@@ -3687,6 +4773,18 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3687
4773
|
inFlight.aliveInFlight = false;
|
|
3688
4774
|
if (ok) inFlight.lastAliveAt = this.now();
|
|
3689
4775
|
});
|
|
4776
|
+
if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
|
|
4777
|
+
inFlight.titleSyncInFlight = true;
|
|
4778
|
+
void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
|
|
4779
|
+
if (!title) {
|
|
4780
|
+
inFlight.titleSyncInFlight = false;
|
|
4781
|
+
return;
|
|
4782
|
+
}
|
|
4783
|
+
const ok = await this.patchConversationTitle(conv.id, title);
|
|
4784
|
+
inFlight.titleSyncInFlight = false;
|
|
4785
|
+
if (ok) inFlight.titleSynced = true;
|
|
4786
|
+
});
|
|
4787
|
+
}
|
|
3690
4788
|
}
|
|
3691
4789
|
if (awaitingHuman) {
|
|
3692
4790
|
if (!inFlight.awaitingHumanLatched) {
|
|
@@ -3724,6 +4822,70 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3724
4822
|
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3725
4823
|
}
|
|
3726
4824
|
}
|
|
4825
|
+
/**
|
|
4826
|
+
* Settle a message whose run-state has resolved `'done'` — extracted verbatim
|
|
4827
|
+
* (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
|
|
4828
|
+
* inline `state === 'done'` branch body, so a SECOND caller (the #721
|
|
4829
|
+
* b2-abandonment resolution) can reach the exact same completion behavior
|
|
4830
|
+
* (delivery-deadline anchoring, title resolution, usage extraction, and
|
|
4831
|
+
* `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
|
|
4832
|
+
* and risking the two copies silently drifting apart.
|
|
4833
|
+
*/
|
|
4834
|
+
async settleMessageDone(sessionId, watcher, inFlight, messages) {
|
|
4835
|
+
const conv = watcher.conv;
|
|
4836
|
+
this.anchorDeliveryDeadline(inFlight);
|
|
4837
|
+
if (!inFlight.done) {
|
|
4838
|
+
this.log({
|
|
4839
|
+
level: "info",
|
|
4840
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
4841
|
+
conversation_id: conv.id,
|
|
4842
|
+
message_id: inFlight.evidentMessageId
|
|
4843
|
+
});
|
|
4844
|
+
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
4845
|
+
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4846
|
+
try {
|
|
4847
|
+
await this.markDone(
|
|
4848
|
+
conv.id,
|
|
4849
|
+
inFlight.evidentMessageId,
|
|
4850
|
+
sessionId,
|
|
4851
|
+
inFlight.opencodeMessageId,
|
|
4852
|
+
title,
|
|
4853
|
+
usage
|
|
4854
|
+
);
|
|
4855
|
+
} catch (err) {
|
|
4856
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
4857
|
+
if (err instanceof ChannelTerminalError) {
|
|
4858
|
+
this.log({
|
|
4859
|
+
level: "warn",
|
|
4860
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
4861
|
+
conversation_id: conv.id,
|
|
4862
|
+
message_id: inFlight.evidentMessageId
|
|
4863
|
+
});
|
|
4864
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4865
|
+
return;
|
|
4866
|
+
}
|
|
4867
|
+
if (this.now() >= inFlight.deadline) {
|
|
4868
|
+
this.log({
|
|
4869
|
+
level: "warn",
|
|
4870
|
+
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)}`,
|
|
4871
|
+
conversation_id: conv.id,
|
|
4872
|
+
message_id: inFlight.evidentMessageId
|
|
4873
|
+
});
|
|
4874
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4875
|
+
return;
|
|
4876
|
+
}
|
|
4877
|
+
this.log({
|
|
4878
|
+
level: "warn",
|
|
4879
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
4880
|
+
conversation_id: conv.id,
|
|
4881
|
+
message_id: inFlight.evidentMessageId
|
|
4882
|
+
});
|
|
4883
|
+
return;
|
|
4884
|
+
}
|
|
4885
|
+
inFlight.done = true;
|
|
4886
|
+
}
|
|
4887
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4888
|
+
}
|
|
3727
4889
|
// Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
|
|
3728
4890
|
/**
|
|
3729
4891
|
* Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
|
|
@@ -3823,7 +4985,10 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3823
4985
|
* re-dispatched (at most once, see `forceReadoptRun`):
|
|
3824
4986
|
* - `done` → `markDone` now (guarded like the watcher's done branch);
|
|
3825
4987
|
* - `failed` → `markFailed` with the surfaced error (issue #182), so an
|
|
3826
|
-
* errored turn is reported failed on restart, NOT re-dispatched
|
|
4988
|
+
* errored turn is reported failed on restart, NOT re-dispatched —
|
|
4989
|
+
* EXCEPT a restart-ABORTED turn under a not-ongoing session,
|
|
4990
|
+
* which is a restart orphan wearing a terminal error and is
|
|
4991
|
+
* re-dispatched instead (issue #1310, see the branch below);
|
|
3827
4992
|
* - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
|
|
3828
4993
|
* tracking the stored id so the reply correlates by it;
|
|
3829
4994
|
* - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
|
|
@@ -3887,9 +5052,19 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3887
5052
|
void this.postSignal(row.conversation_id, row.id, "readopt_done");
|
|
3888
5053
|
return;
|
|
3889
5054
|
}
|
|
3890
|
-
|
|
5055
|
+
const restartAborted = state === "failed" && sessionOngoing === false && isAbortedTerminalReply(messages, ocId ?? "");
|
|
5056
|
+
if (restartAborted) {
|
|
5057
|
+
this.log({
|
|
5058
|
+
level: "info",
|
|
5059
|
+
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`,
|
|
5060
|
+
conversation_id: row.conversation_id,
|
|
5061
|
+
message_id: row.id
|
|
5062
|
+
});
|
|
5063
|
+
}
|
|
5064
|
+
if (state === "failed" && !restartAborted) {
|
|
3891
5065
|
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3892
5066
|
const usage = messageUsage(messages, ocId ?? "");
|
|
5067
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3893
5068
|
this.log({
|
|
3894
5069
|
level: "error",
|
|
3895
5070
|
message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
|
|
@@ -3897,7 +5072,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3897
5072
|
message_id: row.id
|
|
3898
5073
|
});
|
|
3899
5074
|
try {
|
|
3900
|
-
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
|
|
5075
|
+
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
|
|
3901
5076
|
} catch (err) {
|
|
3902
5077
|
if (err instanceof ChannelAuthError) throw err;
|
|
3903
5078
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -4100,15 +5275,39 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4100
5275
|
}
|
|
4101
5276
|
if (ocId === null) {
|
|
4102
5277
|
this.awaitingReadopt.delete(row.id);
|
|
5278
|
+
const streak = this.recordUnconfirmedDispatch(row.id, sessionId);
|
|
5279
|
+
if (streak >= MAX_IDENTICAL_REDRIVE_POLL_FAILURES) {
|
|
5280
|
+
this.unconfirmedDispatchFailures.delete(row.id);
|
|
5281
|
+
this.sessions.delete(readoptConv.id);
|
|
5282
|
+
this.supersede(readoptConv.id, sessionId);
|
|
5283
|
+
const errorMessage = `OpenCode accepted this re-dispatched message ${streak} times in a row but never confirmed its assigned id (session ${sessionId.slice(0, 8)}'s message list could not be read back) \u2014 the session was abandoned; a fresh one is used for further messages.`;
|
|
5284
|
+
this.log({
|
|
5285
|
+
level: "error",
|
|
5286
|
+
message: errorMessage,
|
|
5287
|
+
conversation_id: row.conversation_id,
|
|
5288
|
+
message_id: row.id
|
|
5289
|
+
});
|
|
5290
|
+
await this.markFailed(row.conversation_id, row.id, null, errorMessage).catch((markErr) => {
|
|
5291
|
+
this.log({
|
|
5292
|
+
level: "warn",
|
|
5293
|
+
message: `markFailed PATCH for message ${row.id.slice(0, 8)} (conversation ${row.conversation_id.slice(0, 8)}) failed (best-effort, not retried): ${markErr instanceof Error ? markErr.message : String(markErr)}`,
|
|
5294
|
+
conversation_id: row.conversation_id,
|
|
5295
|
+
message_id: row.id
|
|
5296
|
+
});
|
|
5297
|
+
});
|
|
5298
|
+
void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
|
|
5299
|
+
return;
|
|
5300
|
+
}
|
|
4103
5301
|
this.log({
|
|
4104
5302
|
level: "warn",
|
|
4105
|
-
message: `Re-adopt: message ${row.id.slice(0, 8)} re-dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next drain`,
|
|
5303
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} re-dispatched but its opencode id could not be read back (${streak}/${MAX_IDENTICAL_REDRIVE_POLL_FAILURES}) \u2014 leaving un-tracked to retry next drain`,
|
|
4106
5304
|
conversation_id: row.conversation_id,
|
|
4107
5305
|
message_id: row.id
|
|
4108
5306
|
});
|
|
4109
5307
|
void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
|
|
4110
5308
|
return;
|
|
4111
5309
|
}
|
|
5310
|
+
this.unconfirmedDispatchFailures.delete(row.id);
|
|
4112
5311
|
this.registerReadopted(readoptConv, sessionId, readoptMessage, ocId, this.processedAtMs(row));
|
|
4113
5312
|
this.dispatched.add(row.id);
|
|
4114
5313
|
this.readopted.add(row.id);
|
|
@@ -4164,7 +5363,8 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4164
5363
|
opencode_model: row.opencode_model,
|
|
4165
5364
|
source_message_id: row.source_message_id,
|
|
4166
5365
|
slack_user_id: row.slack_user_id,
|
|
4167
|
-
attachments: row.attachments ?? null
|
|
5366
|
+
attachments: row.attachments ?? null,
|
|
5367
|
+
opencode_message_id: row.opencode_message_id
|
|
4168
5368
|
};
|
|
4169
5369
|
}
|
|
4170
5370
|
/**
|
|
@@ -4303,6 +5503,47 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4303
5503
|
}
|
|
4304
5504
|
return false;
|
|
4305
5505
|
}
|
|
5506
|
+
/**
|
|
5507
|
+
* Tri-state variant of the upward parentID membership walk (#721), used ONLY
|
|
5508
|
+
* by `isAnyDescendantSessionOngoing`. Walks the SAME cached
|
|
5509
|
+
* `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
|
|
5510
|
+
* `sessionBelongsTo`, which deliberately collapses "confirmed not a
|
|
5511
|
+
* descendant" and "the walk's fetch failed" into the same `false` (safe for
|
|
5512
|
+
* its OTHER callers: interaction attribution and the recovery-path
|
|
5513
|
+
* `isAnyDescendantSessionAlive`, both of which just retry next tick with no
|
|
5514
|
+
* safety consequence either way) — this variant keeps those two outcomes
|
|
5515
|
+
* SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
|
|
5516
|
+
* (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
|
|
5517
|
+
* not ongoing".
|
|
5518
|
+
*
|
|
5519
|
+
* Return contract:
|
|
5520
|
+
* - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
|
|
5521
|
+
* - `false` → the walk reached a definitive, parent-less root session
|
|
5522
|
+
* WITHOUT ever matching `rootSessionId` — `sessionId` is
|
|
5523
|
+
* CONFIRMED NOT a descendant of it.
|
|
5524
|
+
* - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
|
|
5525
|
+
* through the walk (`resolveSessionParent` returned `undefined`),
|
|
5526
|
+
* or the depth cap (32) was hit without a definitive answer (a
|
|
5527
|
+
* pathological/cyclic chain proves nothing either way). NEVER
|
|
5528
|
+
* treat this the same as `false` — see `sessionBelongsTo`'s own
|
|
5529
|
+
* doc comment above for why that collapse is safe THERE but not
|
|
5530
|
+
* here.
|
|
5531
|
+
*
|
|
5532
|
+
* `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
|
|
5533
|
+
* to the live-path descendant check, not a modification of shared code used
|
|
5534
|
+
* by interaction attribution or the recovery path.
|
|
5535
|
+
*/
|
|
5536
|
+
async resolveSessionMembership(sessionId, rootSessionId) {
|
|
5537
|
+
let current = sessionId;
|
|
5538
|
+
for (let depth = 0; current && depth < 32; depth++) {
|
|
5539
|
+
if (current === rootSessionId) return true;
|
|
5540
|
+
const parent = await this.resolveSessionParent(current);
|
|
5541
|
+
if (parent === void 0) return null;
|
|
5542
|
+
if (parent === null) return false;
|
|
5543
|
+
current = parent;
|
|
5544
|
+
}
|
|
5545
|
+
return null;
|
|
5546
|
+
}
|
|
4306
5547
|
/**
|
|
4307
5548
|
* Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
|
|
4308
5549
|
* `null` for a root session (no parent) and `undefined` when opencode is
|
|
@@ -4381,11 +5622,59 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4381
5622
|
} catch (err) {
|
|
4382
5623
|
this.log({
|
|
4383
5624
|
level: "debug",
|
|
4384
|
-
message: `Best-effort session title fetch failed for session ${sessionId.slice(0, 8)} (agent ${this.agentId.slice(0, 8)}) \u2014 omitting title: ${err instanceof Error ? err.message : String(err)}`,
|
|
5625
|
+
message: `Best-effort session title fetch failed for session ${sessionId.slice(0, 8)} (agent ${this.agentId.slice(0, 8)}) \u2014 omitting title: ${err instanceof Error ? err.message : String(err)}`,
|
|
5626
|
+
conversation_id: conversationId
|
|
5627
|
+
});
|
|
5628
|
+
}
|
|
5629
|
+
return null;
|
|
5630
|
+
}
|
|
5631
|
+
/**
|
|
5632
|
+
* Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
|
|
5633
|
+
* session title onto the conversation via the PLAIN conversation-update
|
|
5634
|
+
* endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
|
|
5635
|
+
* message-status endpoint `markProcessing`/`markDone` use. Deliberately a
|
|
5636
|
+
* separate, lighter call: it carries no `status`, so it cannot re-trigger the
|
|
5637
|
+
* `processing`/`done` transition side effects (Slack notices, activity-log
|
|
5638
|
+
* rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
|
|
5639
|
+
* ever touches `conversations.title`. That route (`routes/conversations.ts`)
|
|
5640
|
+
* skips a title write matching the stored value, so a redundant call with the
|
|
5641
|
+
* same title is a real no-op — it does not bump `updated_at`, which the
|
|
5642
|
+
* conversation list sorts and paginates on. (Note this is a DIFFERENT guard
|
|
5643
|
+
* from `threads.ts`'s "non-empty AND changed" one, which only covers the
|
|
5644
|
+
* message-status PATCH; the non-empty half is enforced here instead, by
|
|
5645
|
+
* `resolveSessionTitle` never returning an empty/placeholder title.)
|
|
5646
|
+
*
|
|
5647
|
+
* Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
|
|
5648
|
+
* is logged and the title is simply retried on the next heartbeat tick (the
|
|
5649
|
+
* caller only latches `titleSynced` on `true`).
|
|
5650
|
+
*/
|
|
5651
|
+
async patchConversationTitle(conversationId, title) {
|
|
5652
|
+
try {
|
|
5653
|
+
const res = await this.fetchImpl(
|
|
5654
|
+
`${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
|
|
5655
|
+
{
|
|
5656
|
+
method: "PATCH",
|
|
5657
|
+
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
5658
|
+
body: JSON.stringify({ title })
|
|
5659
|
+
}
|
|
5660
|
+
);
|
|
5661
|
+
if (!res.ok) {
|
|
5662
|
+
this.log({
|
|
5663
|
+
level: "debug",
|
|
5664
|
+
message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
|
|
5665
|
+
conversation_id: conversationId
|
|
5666
|
+
});
|
|
5667
|
+
return false;
|
|
5668
|
+
}
|
|
5669
|
+
return true;
|
|
5670
|
+
} catch (err) {
|
|
5671
|
+
this.log({
|
|
5672
|
+
level: "debug",
|
|
5673
|
+
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)}`,
|
|
4385
5674
|
conversation_id: conversationId
|
|
4386
5675
|
});
|
|
5676
|
+
return false;
|
|
4387
5677
|
}
|
|
4388
|
-
return null;
|
|
4389
5678
|
}
|
|
4390
5679
|
/**
|
|
4391
5680
|
* DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
|
|
@@ -4445,6 +5734,84 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4445
5734
|
}
|
|
4446
5735
|
return false;
|
|
4447
5736
|
}
|
|
5737
|
+
/**
|
|
5738
|
+
* LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
|
|
5739
|
+
* sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
|
|
5740
|
+
* in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
|
|
5741
|
+
*
|
|
5742
|
+
* Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
|
|
5743
|
+
* cross-check above): that method judges liveness from the child's OWN
|
|
5744
|
+
* TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
|
|
5745
|
+
* on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
|
|
5746
|
+
* path the local opencode server IS running, so its in-memory status map is
|
|
5747
|
+
* live and authoritative — and per ADR-0047 §4a ("the child has its own entry
|
|
5748
|
+
* [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
|
|
5749
|
+
* ENTIRE turn (including any tool call it is itself executing), not a
|
|
5750
|
+
* per-message transcript snapshot. This sidesteps the "child's own tool is
|
|
5751
|
+
* executing, between its step's completion and the next generation step"
|
|
5752
|
+
* transcript gap that a transcript-based check would need a second,
|
|
5753
|
+
* sustained-window bound to guard against — it is simply not derived from
|
|
5754
|
+
* message timestamps at all.
|
|
5755
|
+
*
|
|
5756
|
+
* Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
|
|
5757
|
+
* status, as the recovery path does per §4a)? Because on the LIVE path the
|
|
5758
|
+
* root session can be shared: a SECOND, unrelated user message can land on the
|
|
5759
|
+
* SAME session (issue #721's own root cause) and keep the root `busy` for a
|
|
5760
|
+
* reason that has nothing to do with THIS message's delegation. A `task`
|
|
5761
|
+
* descendant session is spawned for exactly one delegated turn and never
|
|
5762
|
+
* reused, so its OWN status-map entry is unambiguous evidence about that one
|
|
5763
|
+
* delegation — which the root's status is not.
|
|
5764
|
+
*
|
|
5765
|
+
* Why membership is checked via `resolveSessionMembership`, NOT
|
|
5766
|
+
* `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
|
|
5767
|
+
* `GET /session/:id` fetch failure into "not a descendant", which would
|
|
5768
|
+
* silently drop a genuinely-live candidate from consideration on the one
|
|
5769
|
+
* unlucky tick its membership-walk fetch hiccups (#721).
|
|
5770
|
+
* `resolveSessionMembership` keeps that failure mode as a distinct `null`
|
|
5771
|
+
* (indeterminate) so it is folded into THIS method's own `indeterminate` flag
|
|
5772
|
+
* instead.
|
|
5773
|
+
*
|
|
5774
|
+
* Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
|
|
5775
|
+
* - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
|
|
5776
|
+
* - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
|
|
5777
|
+
* confirmed either way (`resolveSessionMembership` never
|
|
5778
|
+
* returned `null`), and every CONFIRMED descendant's status read
|
|
5779
|
+
* succeeded and is not ongoing (includes "no descendant session
|
|
5780
|
+
* exists at all" — e.g. a plain, non-`task` tool call).
|
|
5781
|
+
* - `null` → INDETERMINATE: `listSessions` failed, OR at least one
|
|
5782
|
+
* candidate's MEMBERSHIP could not be confirmed
|
|
5783
|
+
* (`resolveSessionMembership` returned `null` — a fetch failure
|
|
5784
|
+
* or pathological chain partway through the parent walk), OR at
|
|
5785
|
+
* least one CONFIRMED descendant's `isSessionOngoing` read
|
|
5786
|
+
* failed — and no OTHER candidate was already confirmed `true`.
|
|
5787
|
+
* The caller MUST NOT treat `null` the same as `false` here
|
|
5788
|
+
* (unlike the recovery cross-check's contract) — see
|
|
5789
|
+
* `isB2AbandonmentConfirmed`.
|
|
5790
|
+
*/
|
|
5791
|
+
async isAnyDescendantSessionOngoing(rootSessionId) {
|
|
5792
|
+
const sessions = await listSessions(this.port);
|
|
5793
|
+
if (!sessions) {
|
|
5794
|
+
this.log({
|
|
5795
|
+
level: "warn",
|
|
5796
|
+
message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
|
|
5797
|
+
});
|
|
5798
|
+
return null;
|
|
5799
|
+
}
|
|
5800
|
+
let indeterminate = false;
|
|
5801
|
+
for (const candidate of sessions) {
|
|
5802
|
+
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
5803
|
+
const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
|
|
5804
|
+
if (membership === null) {
|
|
5805
|
+
indeterminate = true;
|
|
5806
|
+
continue;
|
|
5807
|
+
}
|
|
5808
|
+
if (membership === false) continue;
|
|
5809
|
+
const ongoing = await isSessionOngoing(this.port, candidate.id);
|
|
5810
|
+
if (ongoing === true) return true;
|
|
5811
|
+
if (ongoing === null) indeterminate = true;
|
|
5812
|
+
}
|
|
5813
|
+
return indeterminate ? null : false;
|
|
5814
|
+
}
|
|
4448
5815
|
/**
|
|
4449
5816
|
* Cheap decision-telemetry label for a running row's LAST correlated reply
|
|
4450
5817
|
* (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
|
|
@@ -4589,11 +5956,11 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4589
5956
|
* `ensureSession` guard, not this suppression, is what makes the self-heal
|
|
4590
5957
|
* stick.
|
|
4591
5958
|
*/
|
|
4592
|
-
sessionIdBody(sessionId, conversationId, messageId,
|
|
5959
|
+
sessionIdBody(sessionId, conversationId, messageId, status2) {
|
|
4593
5960
|
if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
|
|
4594
5961
|
this.log({
|
|
4595
5962
|
level: "debug",
|
|
4596
|
-
message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${
|
|
5963
|
+
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
5964
|
conversation_id: conversationId,
|
|
4598
5965
|
message_id: messageId
|
|
4599
5966
|
});
|
|
@@ -4708,7 +6075,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4708
6075
|
* exists but is wedged, so the next attempt must get a fresh one
|
|
4709
6076
|
* instead of reusing it — see WI-1's server-side null-clearing PATCH).
|
|
4710
6077
|
*/
|
|
4711
|
-
async markFailed(conversationId, messageId, sessionId, error2, usage) {
|
|
6078
|
+
async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
|
|
4712
6079
|
const body = { status: "failed" };
|
|
4713
6080
|
if (sessionId === null) {
|
|
4714
6081
|
body.opencode_session_id = null;
|
|
@@ -4717,6 +6084,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4717
6084
|
}
|
|
4718
6085
|
if (error2 !== void 0) body.error = error2;
|
|
4719
6086
|
if (usage) Object.assign(body, usage);
|
|
6087
|
+
if (failure) {
|
|
6088
|
+
body.failure_kind = failure.kind;
|
|
6089
|
+
body.failure_provider_id = failure.providerId;
|
|
6090
|
+
body.failure_model_id = failure.modelId;
|
|
6091
|
+
body.failure_reason = failure.reason;
|
|
6092
|
+
}
|
|
4720
6093
|
await this.callWithRetry(
|
|
4721
6094
|
"marking message as failed",
|
|
4722
6095
|
() => this.fetchImpl(
|
|
@@ -4729,6 +6102,29 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4729
6102
|
)
|
|
4730
6103
|
);
|
|
4731
6104
|
}
|
|
6105
|
+
/**
|
|
6106
|
+
* Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
|
|
6107
|
+
*
|
|
6108
|
+
* `messageFailure` alone (structured OpenCode error → `model_auth`) covers
|
|
6109
|
+
* most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
|
|
6110
|
+
* to the P1-2b zero-provider check — one extra loopback call to
|
|
6111
|
+
* `hasAnyConfiguredProvider`, only reached when the structured classifier
|
|
6112
|
+
* couldn't place it. Fails open (never throws): a fallback probe failure
|
|
6113
|
+
* (`null`/indeterminate) leaves the classification `null`, which produces
|
|
6114
|
+
* today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
|
|
6115
|
+
*/
|
|
6116
|
+
async classifyModelAuthFailure(messages, userMessageId) {
|
|
6117
|
+
const classified = messageFailure(messages, userMessageId);
|
|
6118
|
+
if (classified != null) return classified;
|
|
6119
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
6120
|
+
const hasProvider = await hasAnyConfiguredProvider(this.port);
|
|
6121
|
+
return applyZeroProviderFallback(
|
|
6122
|
+
classified,
|
|
6123
|
+
hasProvider,
|
|
6124
|
+
reply?.info?.providerID ?? null,
|
|
6125
|
+
reply?.info?.modelID ?? null
|
|
6126
|
+
);
|
|
6127
|
+
}
|
|
4732
6128
|
/**
|
|
4733
6129
|
* Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
|
|
4734
6130
|
* (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
|
|
@@ -4881,10 +6277,16 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4881
6277
|
import chalk5 from "chalk";
|
|
4882
6278
|
import ora2 from "ora";
|
|
4883
6279
|
import { select as select2 } from "@inquirer/prompts";
|
|
6280
|
+
var INTERACTIVE_START_TIMEOUT_MS = 3e4;
|
|
4884
6281
|
async function ensureOpenCodeRunning(ctx) {
|
|
4885
6282
|
const healthCheck = await checkOpenCodeHealth(ctx.port);
|
|
4886
6283
|
if (healthCheck.healthy) {
|
|
4887
|
-
return {
|
|
6284
|
+
return {
|
|
6285
|
+
port: ctx.port,
|
|
6286
|
+
process: null,
|
|
6287
|
+
version: healthCheck.version ?? null,
|
|
6288
|
+
notReadyReason: null
|
|
6289
|
+
};
|
|
4888
6290
|
}
|
|
4889
6291
|
const runningInstances = await findHealthyOpenCodeInstances();
|
|
4890
6292
|
if (runningInstances.length > 0) {
|
|
@@ -4905,7 +6307,7 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4905
6307
|
console.log(chalk5.yellow("Tip: Run with the correct port:"));
|
|
4906
6308
|
console.log(
|
|
4907
6309
|
chalk5.dim(
|
|
4908
|
-
` ${getCliName()} run --
|
|
6310
|
+
` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
|
|
4909
6311
|
)
|
|
4910
6312
|
);
|
|
4911
6313
|
}
|
|
@@ -4925,14 +6327,22 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4925
6327
|
if (!ctx.interactive) {
|
|
4926
6328
|
ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
|
|
4927
6329
|
const proc = await startOpenCode(ctx.port);
|
|
4928
|
-
const health = await waitForOpenCodeHealth(ctx.port,
|
|
6330
|
+
const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
|
|
4929
6331
|
if (!health.healthy) {
|
|
4930
|
-
|
|
4931
|
-
|
|
4932
|
-
|
|
6332
|
+
return {
|
|
6333
|
+
port: ctx.port,
|
|
6334
|
+
process: proc,
|
|
6335
|
+
version: null,
|
|
6336
|
+
notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
|
|
6337
|
+
};
|
|
4933
6338
|
}
|
|
4934
6339
|
ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
|
|
4935
|
-
return {
|
|
6340
|
+
return {
|
|
6341
|
+
port: ctx.port,
|
|
6342
|
+
process: proc,
|
|
6343
|
+
version: health.version ?? null,
|
|
6344
|
+
notReadyReason: null
|
|
6345
|
+
};
|
|
4936
6346
|
}
|
|
4937
6347
|
let port = ctx.port;
|
|
4938
6348
|
if (isPortInUse(port)) {
|
|
@@ -4985,128 +6395,15 @@ Port ${port} is already in use.`));
|
|
|
4985
6395
|
if (action === "start") {
|
|
4986
6396
|
const spinner = ora2("Starting OpenCode...").start();
|
|
4987
6397
|
const proc = await startOpenCode(port);
|
|
4988
|
-
const health = await waitForOpenCodeHealth(port,
|
|
6398
|
+
const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
|
|
4989
6399
|
if (!health.healthy) {
|
|
4990
6400
|
spinner.fail("Failed to start OpenCode");
|
|
4991
6401
|
throw new Error("OpenCode failed to start");
|
|
4992
6402
|
}
|
|
4993
6403
|
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: ${message}` };
|
|
5043
|
-
}
|
|
5044
|
-
}
|
|
5045
|
-
var OFFLINE_NOTIFY_TIMEOUT_MS = 2e3;
|
|
5046
|
-
async function notifyAgentDisconnected(agentId, authHeader) {
|
|
5047
|
-
const apiUrl = getApiUrlConfig();
|
|
5048
|
-
try {
|
|
5049
|
-
const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
|
|
5050
|
-
method: "POST",
|
|
5051
|
-
headers: { Authorization: authHeader },
|
|
5052
|
-
signal: AbortSignal.timeout(OFFLINE_NOTIFY_TIMEOUT_MS)
|
|
5053
|
-
});
|
|
5054
|
-
if (!response.ok) {
|
|
5055
|
-
const serverMessage = await readErrorMessage(response);
|
|
5056
|
-
return {
|
|
5057
|
-
ok: false,
|
|
5058
|
-
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
5059
|
-
};
|
|
5060
|
-
}
|
|
5061
|
-
return { ok: true };
|
|
5062
|
-
} catch (error2) {
|
|
5063
|
-
const name = error2?.name;
|
|
5064
|
-
if (name === "TimeoutError" || name === "AbortError") {
|
|
5065
|
-
return { ok: false, error: `timed out after ${OFFLINE_NOTIFY_TIMEOUT_MS}ms` };
|
|
5066
|
-
}
|
|
5067
|
-
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
5068
|
-
}
|
|
5069
|
-
}
|
|
5070
|
-
async function getAgentInfo(agentId, authHeader) {
|
|
5071
|
-
const apiUrl = getApiUrlConfig();
|
|
5072
|
-
try {
|
|
5073
|
-
const response = await fetch(`${apiUrl}/runners/${agentId}`, {
|
|
5074
|
-
headers: { Authorization: authHeader }
|
|
5075
|
-
});
|
|
5076
|
-
if (response.status === 401) {
|
|
5077
|
-
const serverMessage = await readErrorMessage(response);
|
|
5078
|
-
return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
5079
|
-
}
|
|
5080
|
-
if (response.status === 403) {
|
|
5081
|
-
const serverMessage = await readErrorMessage(response);
|
|
5082
|
-
return {
|
|
5083
|
-
valid: false,
|
|
5084
|
-
error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
|
|
5085
|
-
};
|
|
5086
|
-
}
|
|
5087
|
-
if (response.status === 404) {
|
|
5088
|
-
const serverMessage = await readErrorMessage(response);
|
|
5089
|
-
return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
|
|
5090
|
-
}
|
|
5091
|
-
if (!response.ok) {
|
|
5092
|
-
const serverMessage = await readErrorMessage(response);
|
|
5093
|
-
return {
|
|
5094
|
-
valid: false,
|
|
5095
|
-
error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
5096
|
-
};
|
|
5097
|
-
}
|
|
5098
|
-
const agent = await response.json();
|
|
5099
|
-
if (agent.agent_type !== "local") {
|
|
5100
|
-
return {
|
|
5101
|
-
valid: false,
|
|
5102
|
-
error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
|
|
5103
|
-
};
|
|
5104
|
-
}
|
|
5105
|
-
return { valid: true, agent };
|
|
5106
|
-
} catch (error2) {
|
|
5107
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
5108
|
-
return { valid: false, error: `Failed to validate runner: ${message}` };
|
|
6404
|
+
return { port, process: proc, version: health.version ?? null, notReadyReason: null };
|
|
5109
6405
|
}
|
|
6406
|
+
return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
|
|
5110
6407
|
}
|
|
5111
6408
|
|
|
5112
6409
|
// src/commands/run.ts
|
|
@@ -5145,7 +6442,7 @@ function resolveFileSyncDirectories(raw, homeDir) {
|
|
|
5145
6442
|
if (trimmed === "") {
|
|
5146
6443
|
throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
|
|
5147
6444
|
}
|
|
5148
|
-
const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ?
|
|
6445
|
+
const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join3(homeDir, trimmed.slice(2)) : trimmed;
|
|
5149
6446
|
if (!isAbsolute2(expanded)) {
|
|
5150
6447
|
throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
|
|
5151
6448
|
}
|
|
@@ -5166,6 +6463,35 @@ function resolveFileSyncDirectories(raw, homeDir) {
|
|
|
5166
6463
|
}
|
|
5167
6464
|
return directories;
|
|
5168
6465
|
}
|
|
6466
|
+
var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
|
|
6467
|
+
var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
|
|
6468
|
+
var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
|
|
6469
|
+
function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
|
|
6470
|
+
const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
|
|
6471
|
+
let raw;
|
|
6472
|
+
let source;
|
|
6473
|
+
if (options.opencodeStartTimeout !== void 0) {
|
|
6474
|
+
raw = options.opencodeStartTimeout;
|
|
6475
|
+
source = "--opencode-start-timeout";
|
|
6476
|
+
} else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
|
|
6477
|
+
raw = env[OPENCODE_START_TIMEOUT_ENV];
|
|
6478
|
+
source = OPENCODE_START_TIMEOUT_ENV;
|
|
6479
|
+
} else {
|
|
6480
|
+
return { timeoutMs: defaultMs, warnings: [] };
|
|
6481
|
+
}
|
|
6482
|
+
const trimmed = raw.trim();
|
|
6483
|
+
const seconds = Number(trimmed);
|
|
6484
|
+
const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
|
|
6485
|
+
if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
|
|
6486
|
+
return {
|
|
6487
|
+
timeoutMs: defaultMs,
|
|
6488
|
+
warnings: [
|
|
6489
|
+
`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`
|
|
6490
|
+
]
|
|
6491
|
+
};
|
|
6492
|
+
}
|
|
6493
|
+
return { timeoutMs: seconds * 1e3, warnings: [] };
|
|
6494
|
+
}
|
|
5169
6495
|
function meetsThreshold(state, level) {
|
|
5170
6496
|
return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
|
|
5171
6497
|
}
|
|
@@ -5187,6 +6513,10 @@ function log2(state, message, level = "info") {
|
|
|
5187
6513
|
function logActivity(state, entry) {
|
|
5188
6514
|
const level = entry.level ?? (entry.type === "error" ? "error" : "info");
|
|
5189
6515
|
if (!meetsThreshold(state, level)) return;
|
|
6516
|
+
forwardRunnerActivity(
|
|
6517
|
+
{ level, message: entry.message, error: entry.error },
|
|
6518
|
+
{ agentId: state.agentId, authHeader: state.authHeader }
|
|
6519
|
+
);
|
|
5190
6520
|
const fullEntry = {
|
|
5191
6521
|
...entry,
|
|
5192
6522
|
level,
|
|
@@ -5286,9 +6616,15 @@ async function handleAuthError(state, error2) {
|
|
|
5286
6616
|
}
|
|
5287
6617
|
async function driveChannels(state, driver) {
|
|
5288
6618
|
let idlePolls = 0;
|
|
6619
|
+
let idleMs = 0;
|
|
6620
|
+
let consecutiveDrainFailures = 0;
|
|
6621
|
+
let unreachableMs = 0;
|
|
5289
6622
|
let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
5290
6623
|
let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
5291
6624
|
while (state.running) {
|
|
6625
|
+
const cycleStartedAtMs = performance.now();
|
|
6626
|
+
let idleThisCycle = false;
|
|
6627
|
+
let unreachableThisCycle = false;
|
|
5292
6628
|
if (state.connection?.reconnecting && state.connection.reconnectPromise) {
|
|
5293
6629
|
logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
|
|
5294
6630
|
if (state.interactive) displayStatus(state);
|
|
@@ -5303,17 +6639,23 @@ async function driveChannels(state, driver) {
|
|
|
5303
6639
|
);
|
|
5304
6640
|
try {
|
|
5305
6641
|
const processed = await driver.drainPending();
|
|
6642
|
+
consecutiveDrainFailures = 0;
|
|
6643
|
+
unreachableMs = 0;
|
|
5306
6644
|
state.messageCount += processed;
|
|
5307
6645
|
const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
|
|
5308
6646
|
lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
5309
6647
|
const appliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
5310
|
-
const
|
|
6648
|
+
const filesApplied = appliedFiles !== lastSeenAppliedFiles;
|
|
6649
|
+
const fileActivity = carriedOverFileSync || filesApplied;
|
|
5311
6650
|
lastSeenAppliedFiles = appliedFiles;
|
|
6651
|
+
if (filesApplied) state.claudeUsageRearm?.();
|
|
5312
6652
|
if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
|
|
5313
6653
|
idlePolls = 0;
|
|
6654
|
+
idleMs = 0;
|
|
5314
6655
|
if (processed > 0 && state.interactive) displayStatus(state);
|
|
5315
6656
|
} else if (state.idleTimeout !== null) {
|
|
5316
6657
|
idlePolls++;
|
|
6658
|
+
idleThisCycle = true;
|
|
5317
6659
|
if (idlePolls === 1) {
|
|
5318
6660
|
logActivity(state, {
|
|
5319
6661
|
type: "info",
|
|
@@ -5337,21 +6679,44 @@ async function driveChannels(state, driver) {
|
|
|
5337
6679
|
const errorMessage = error2 instanceof Error ? error2.message : String(error2);
|
|
5338
6680
|
logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
|
|
5339
6681
|
if (state.interactive) displayStatus(state);
|
|
6682
|
+
if (driver.hasInFlightWatchers()) {
|
|
6683
|
+
consecutiveDrainFailures = 0;
|
|
6684
|
+
unreachableMs = 0;
|
|
6685
|
+
} else if (state.idleTimeout !== null) {
|
|
6686
|
+
consecutiveDrainFailures++;
|
|
6687
|
+
unreachableThisCycle = true;
|
|
6688
|
+
if (consecutiveDrainFailures === 1) {
|
|
6689
|
+
logActivity(state, {
|
|
6690
|
+
type: "info",
|
|
6691
|
+
message: `Cannot reach Evident, will exit if this persists past the idle timeout (timeout: ${state.idleTimeout}s)...`
|
|
6692
|
+
});
|
|
6693
|
+
if (state.interactive) displayStatus(state);
|
|
6694
|
+
}
|
|
6695
|
+
}
|
|
5340
6696
|
}
|
|
5341
6697
|
await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
|
|
5342
|
-
|
|
5343
|
-
|
|
5344
|
-
|
|
5345
|
-
|
|
5346
|
-
|
|
5347
|
-
|
|
5348
|
-
|
|
6698
|
+
const cycleMs = performance.now() - cycleStartedAtMs;
|
|
6699
|
+
if (idleThisCycle) idleMs += cycleMs;
|
|
6700
|
+
if (unreachableThisCycle) unreachableMs += cycleMs;
|
|
6701
|
+
if (state.idleTimeout !== null && consecutiveDrainFailures >= 2 && unreachableMs > state.idleTimeout * 1e3) {
|
|
6702
|
+
logActivity(state, {
|
|
6703
|
+
type: "info",
|
|
6704
|
+
level: "warn",
|
|
6705
|
+
message: `Exiting: could not reach Evident for ${consecutiveDrainFailures} consecutive polls (${Math.round(unreachableMs / 1e3)}s)`
|
|
6706
|
+
});
|
|
6707
|
+
if (state.interactive) displayStatus(state);
|
|
6708
|
+
break;
|
|
6709
|
+
}
|
|
6710
|
+
if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
|
|
6711
|
+
logActivity(state, { type: "info", message: "Idle timeout reached" });
|
|
6712
|
+
if (state.interactive) displayStatus(state);
|
|
6713
|
+
break;
|
|
5349
6714
|
}
|
|
5350
6715
|
}
|
|
5351
6716
|
}
|
|
5352
6717
|
var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
|
|
5353
|
-
async function runSweep(state, driver,
|
|
5354
|
-
const mode = `age=${
|
|
6718
|
+
async function runSweep(state, driver, config) {
|
|
6719
|
+
const mode = `age=${config.maxAgeMs ?? "\u2014"} count=${config.maxCount ?? "\u2014"}`;
|
|
5355
6720
|
try {
|
|
5356
6721
|
const sessions = await listSessions(state.port);
|
|
5357
6722
|
if (sessions === null) {
|
|
@@ -5364,8 +6729,8 @@ async function runSweep(state, driver, config2) {
|
|
|
5364
6729
|
const toDelete = selectSessionsToDelete(
|
|
5365
6730
|
sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
|
|
5366
6731
|
{
|
|
5367
|
-
maxAgeMs:
|
|
5368
|
-
maxCount:
|
|
6732
|
+
maxAgeMs: config.maxAgeMs,
|
|
6733
|
+
maxCount: config.maxCount,
|
|
5369
6734
|
nowMs: Date.now(),
|
|
5370
6735
|
protectedIds: driver.protectedSessionIds()
|
|
5371
6736
|
}
|
|
@@ -5401,7 +6766,7 @@ async function runSweep(state, driver, config2) {
|
|
|
5401
6766
|
}
|
|
5402
6767
|
}
|
|
5403
6768
|
function scheduleSessionCleanup(state, driver, options) {
|
|
5404
|
-
const
|
|
6769
|
+
const config = resolveSessionCleanupConfig(
|
|
5405
6770
|
{
|
|
5406
6771
|
maxAge: options.sessionCleanupMaxAge,
|
|
5407
6772
|
maxCount: options.sessionCleanupMaxCount,
|
|
@@ -5409,21 +6774,129 @@ function scheduleSessionCleanup(state, driver, options) {
|
|
|
5409
6774
|
},
|
|
5410
6775
|
process.env
|
|
5411
6776
|
);
|
|
5412
|
-
for (const warning2 of
|
|
6777
|
+
for (const warning2 of config.warnings) {
|
|
5413
6778
|
logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
|
|
5414
6779
|
}
|
|
5415
|
-
if (!
|
|
6780
|
+
if (!config.enabled) return;
|
|
5416
6781
|
logActivity(state, {
|
|
5417
6782
|
type: "info",
|
|
5418
|
-
message: `Session cleanup enabled (age=${
|
|
6783
|
+
message: `Session cleanup enabled (age=${config.maxAgeMs ?? "\u2014"}, count=${config.maxCount ?? "\u2014"}, interval=${config.intervalMs}ms)`
|
|
5419
6784
|
});
|
|
5420
|
-
const interval = setInterval(() => void runSweep(state, driver,
|
|
6785
|
+
const interval = setInterval(() => void runSweep(state, driver, config), config.intervalMs);
|
|
5421
6786
|
const firstSweep = setTimeout(
|
|
5422
|
-
() => void runSweep(state, driver,
|
|
6787
|
+
() => void runSweep(state, driver, config),
|
|
5423
6788
|
SESSION_CLEANUP_FIRST_SWEEP_MS
|
|
5424
6789
|
);
|
|
5425
6790
|
state.sessionCleanupTimers.push(interval, firstSweep);
|
|
5426
6791
|
}
|
|
6792
|
+
function claudeUsageFailureStreakSuffix(consecutiveFailures) {
|
|
6793
|
+
return consecutiveFailures > 1 ? ` (${consecutiveFailures} consecutive failures)` : "";
|
|
6794
|
+
}
|
|
6795
|
+
function scheduleClaudeUsageReporting(state, options) {
|
|
6796
|
+
const { mode, warnings } = resolveClaudeUsageReportingMode(
|
|
6797
|
+
options.claudeUsageReporting,
|
|
6798
|
+
process.env
|
|
6799
|
+
);
|
|
6800
|
+
for (const warning2 of warnings) {
|
|
6801
|
+
logActivity(state, {
|
|
6802
|
+
type: "info",
|
|
6803
|
+
level: "warn",
|
|
6804
|
+
message: `Claude usage reporting: ${warning2}`
|
|
6805
|
+
});
|
|
6806
|
+
}
|
|
6807
|
+
if (mode === "off") {
|
|
6808
|
+
logActivity(state, {
|
|
6809
|
+
type: "info",
|
|
6810
|
+
level: "debug",
|
|
6811
|
+
message: "Claude usage reporting is off (--claude-usage-reporting off)"
|
|
6812
|
+
});
|
|
6813
|
+
return null;
|
|
6814
|
+
}
|
|
6815
|
+
let consecutiveFailures = 0;
|
|
6816
|
+
let armed = false;
|
|
6817
|
+
let rearmRequested = false;
|
|
6818
|
+
const scheduleNextTick = () => {
|
|
6819
|
+
armed = true;
|
|
6820
|
+
rearmRequested = false;
|
|
6821
|
+
state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
|
|
6822
|
+
};
|
|
6823
|
+
const rearm = () => {
|
|
6824
|
+
if (armed) {
|
|
6825
|
+
rearmRequested = true;
|
|
6826
|
+
return;
|
|
6827
|
+
}
|
|
6828
|
+
rearmRequested = false;
|
|
6829
|
+
armed = true;
|
|
6830
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6831
|
+
};
|
|
6832
|
+
const tick = async (isProbe) => {
|
|
6833
|
+
try {
|
|
6834
|
+
const usage = await getClaudeUsage();
|
|
6835
|
+
const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
|
|
6836
|
+
if (result.ok) {
|
|
6837
|
+
if (consecutiveFailures > 0) {
|
|
6838
|
+
logActivity(state, {
|
|
6839
|
+
type: "info",
|
|
6840
|
+
level: "info",
|
|
6841
|
+
message: "Claude usage reporting recovered"
|
|
6842
|
+
});
|
|
6843
|
+
}
|
|
6844
|
+
consecutiveFailures = 0;
|
|
6845
|
+
logActivity(state, {
|
|
6846
|
+
type: "info",
|
|
6847
|
+
level: "debug",
|
|
6848
|
+
message: "Reported Claude usage to Evident"
|
|
6849
|
+
});
|
|
6850
|
+
} else {
|
|
6851
|
+
consecutiveFailures++;
|
|
6852
|
+
logActivity(state, {
|
|
6853
|
+
type: "info",
|
|
6854
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6855
|
+
message: `Failed to report Claude usage: ${result.error}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6856
|
+
});
|
|
6857
|
+
}
|
|
6858
|
+
scheduleNextTick();
|
|
6859
|
+
} catch (error2) {
|
|
6860
|
+
if (error2 instanceof ClaudeUsageError && isLocalCredentialProblem(error2)) {
|
|
6861
|
+
if (mode === "on") {
|
|
6862
|
+
logActivity(state, {
|
|
6863
|
+
type: "info",
|
|
6864
|
+
level: "warn",
|
|
6865
|
+
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"
|
|
6866
|
+
});
|
|
6867
|
+
scheduleNextTick();
|
|
6868
|
+
} else if (isProbe) {
|
|
6869
|
+
logActivity(state, {
|
|
6870
|
+
type: "info",
|
|
6871
|
+
level: "debug",
|
|
6872
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6873
|
+
});
|
|
6874
|
+
armed = false;
|
|
6875
|
+
if (rearmRequested) rearm();
|
|
6876
|
+
} else {
|
|
6877
|
+
logActivity(state, {
|
|
6878
|
+
type: "info",
|
|
6879
|
+
level: "debug",
|
|
6880
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6881
|
+
});
|
|
6882
|
+
scheduleNextTick();
|
|
6883
|
+
}
|
|
6884
|
+
} else {
|
|
6885
|
+
consecutiveFailures++;
|
|
6886
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
6887
|
+
logActivity(state, {
|
|
6888
|
+
type: "info",
|
|
6889
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6890
|
+
message: `Claude usage reporting failed: ${message}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6891
|
+
});
|
|
6892
|
+
scheduleNextTick();
|
|
6893
|
+
}
|
|
6894
|
+
}
|
|
6895
|
+
};
|
|
6896
|
+
armed = true;
|
|
6897
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6898
|
+
return rearm;
|
|
6899
|
+
}
|
|
5427
6900
|
async function notifyOffline(state) {
|
|
5428
6901
|
if (!state.agentId || !state.authHeader) return;
|
|
5429
6902
|
if (!state.connected) {
|
|
@@ -5459,6 +6932,11 @@ async function cleanup(state, opts = {}) {
|
|
|
5459
6932
|
clearTimeout(timer);
|
|
5460
6933
|
}
|
|
5461
6934
|
state.sessionCleanupTimers = [];
|
|
6935
|
+
if (state.claudeUsageTimer) {
|
|
6936
|
+
clearTimeout(state.claudeUsageTimer);
|
|
6937
|
+
state.claudeUsageTimer = null;
|
|
6938
|
+
}
|
|
6939
|
+
state.claudeUsageRearm = null;
|
|
5462
6940
|
if (opts.graceful && state.channelDriver) {
|
|
5463
6941
|
state.channelDriver.stop();
|
|
5464
6942
|
log2(state, "Draining in-flight channel work before shutdown...");
|
|
@@ -5506,7 +6984,7 @@ async function run(options) {
|
|
|
5506
6984
|
let fileSyncDirectories;
|
|
5507
6985
|
try {
|
|
5508
6986
|
logLevel = resolveLogLevel(options);
|
|
5509
|
-
fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo,
|
|
6987
|
+
fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir3());
|
|
5510
6988
|
} catch (error2) {
|
|
5511
6989
|
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
5512
6990
|
if (options.json) {
|
|
@@ -5539,8 +7017,11 @@ async function run(options) {
|
|
|
5539
7017
|
messageCount: 0,
|
|
5540
7018
|
lastProxiedActivityAt: null,
|
|
5541
7019
|
sessionCleanupTimers: [],
|
|
7020
|
+
claudeUsageTimer: null,
|
|
7021
|
+
claudeUsageRearm: null,
|
|
5542
7022
|
authHeader: ""
|
|
5543
7023
|
};
|
|
7024
|
+
setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
|
|
5544
7025
|
if (fileSyncDirectories.length > 0) {
|
|
5545
7026
|
log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
|
|
5546
7027
|
} else {
|
|
@@ -5617,6 +7098,7 @@ async function run(options) {
|
|
|
5617
7098
|
console.log(chalk6.dim("Or run `evident login` for interactive authentication"));
|
|
5618
7099
|
blank();
|
|
5619
7100
|
process.exit(1);
|
|
7101
|
+
return;
|
|
5620
7102
|
}
|
|
5621
7103
|
blank();
|
|
5622
7104
|
console.log(chalk6.yellow("You are not logged in to Evident."));
|
|
@@ -5661,6 +7143,7 @@ async function run(options) {
|
|
|
5661
7143
|
} else {
|
|
5662
7144
|
printError(resolved.error || "Failed to resolve runner ID from key");
|
|
5663
7145
|
process.exit(1);
|
|
7146
|
+
return;
|
|
5664
7147
|
}
|
|
5665
7148
|
} else {
|
|
5666
7149
|
printError(
|
|
@@ -5674,6 +7157,7 @@ async function run(options) {
|
|
|
5674
7157
|
);
|
|
5675
7158
|
blank();
|
|
5676
7159
|
process.exit(1);
|
|
7160
|
+
return;
|
|
5677
7161
|
}
|
|
5678
7162
|
}
|
|
5679
7163
|
telemetry.info(
|
|
@@ -5714,40 +7198,67 @@ async function run(options) {
|
|
|
5714
7198
|
}
|
|
5715
7199
|
spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
|
|
5716
7200
|
state.agentName = validation.agent.name;
|
|
7201
|
+
const microvmId = process.env.MICROVM_ID?.trim();
|
|
7202
|
+
if (microvmId) {
|
|
7203
|
+
const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
|
|
7204
|
+
if (reported.ok) {
|
|
7205
|
+
log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
|
|
7206
|
+
} else {
|
|
7207
|
+
const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
|
|
7208
|
+
log2(state, message, "warn");
|
|
7209
|
+
if (state.interactive && !state.json) {
|
|
7210
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
7211
|
+
}
|
|
7212
|
+
}
|
|
7213
|
+
} else {
|
|
7214
|
+
log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
|
|
7215
|
+
}
|
|
7216
|
+
const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
|
|
7217
|
+
for (const warning2 of opencodeStartTimeoutWarnings) {
|
|
7218
|
+
logActivity(state, { type: "info", level: "warn", message: warning2 });
|
|
7219
|
+
}
|
|
5717
7220
|
const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
|
|
5718
7221
|
try {
|
|
5719
7222
|
const oc = await ensureOpenCodeRunning({
|
|
5720
7223
|
port: state.port,
|
|
5721
7224
|
interactive: state.interactive,
|
|
5722
7225
|
agentId: state.agentId,
|
|
5723
|
-
log: (message) => log2(state, message)
|
|
7226
|
+
log: (message) => log2(state, message),
|
|
7227
|
+
startTimeoutMs: opencodeStartTimeoutMs
|
|
5724
7228
|
});
|
|
5725
7229
|
state.port = oc.port;
|
|
5726
7230
|
state.opencodeProcess = oc.process;
|
|
5727
7231
|
state.opencodeVersion = oc.version;
|
|
5728
|
-
state.opencodeConnected = oc.
|
|
7232
|
+
state.opencodeConnected = oc.notReadyReason === null;
|
|
5729
7233
|
const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
|
|
5730
7234
|
ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
|
|
5731
|
-
|
|
5732
|
-
|
|
5733
|
-
|
|
5734
|
-
|
|
5735
|
-
|
|
7235
|
+
if (!state.interactive && oc.notReadyReason !== null) {
|
|
7236
|
+
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}).`;
|
|
7237
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
7238
|
+
} else {
|
|
7239
|
+
const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
|
|
7240
|
+
if (versionWarning) {
|
|
7241
|
+
log2(state, versionWarning, "warn");
|
|
7242
|
+
if (state.interactive && !state.json) {
|
|
7243
|
+
logActivity(state, { type: "info", level: "warn", message: versionWarning });
|
|
7244
|
+
}
|
|
5736
7245
|
}
|
|
5737
|
-
|
|
5738
|
-
|
|
5739
|
-
|
|
5740
|
-
|
|
5741
|
-
|
|
5742
|
-
|
|
5743
|
-
|
|
5744
|
-
|
|
5745
|
-
|
|
5746
|
-
|
|
5747
|
-
|
|
5748
|
-
|
|
5749
|
-
|
|
5750
|
-
|
|
7246
|
+
const noProviderWarning = buildNoProviderWarning(
|
|
7247
|
+
await hasAnyConfiguredProvider(state.port)
|
|
7248
|
+
);
|
|
7249
|
+
if (noProviderWarning) {
|
|
7250
|
+
log2(state, noProviderWarning, "warn");
|
|
7251
|
+
if (state.interactive && !state.json) {
|
|
7252
|
+
logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
|
|
7253
|
+
blank();
|
|
7254
|
+
console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
|
|
7255
|
+
console.log(
|
|
7256
|
+
chalk6.dim(
|
|
7257
|
+
`Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
|
|
7258
|
+
)
|
|
7259
|
+
);
|
|
7260
|
+
blank();
|
|
7261
|
+
}
|
|
5751
7262
|
}
|
|
5752
7263
|
}
|
|
5753
7264
|
} catch (error2) {
|
|
@@ -5765,7 +7276,7 @@ async function run(options) {
|
|
|
5765
7276
|
// #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
|
|
5766
7277
|
// REJECTED with `file_sync_disabled` on the ack, not silently ignored.
|
|
5767
7278
|
fileSyncDirectories,
|
|
5768
|
-
homeDir:
|
|
7279
|
+
homeDir: homedir3(),
|
|
5769
7280
|
log: (entry) => (
|
|
5770
7281
|
// Thread the driver's real level straight through so `debug`/`warn`
|
|
5771
7282
|
// survive the sink filter (they no longer collapse to info). `type`
|
|
@@ -5792,6 +7303,18 @@ async function run(options) {
|
|
|
5792
7303
|
type: "info",
|
|
5793
7304
|
message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
|
|
5794
7305
|
});
|
|
7306
|
+
if (options.tunnelReadyFile) {
|
|
7307
|
+
const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
|
|
7308
|
+
if (marker.ok) {
|
|
7309
|
+
log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
|
|
7310
|
+
} else {
|
|
7311
|
+
log2(
|
|
7312
|
+
state,
|
|
7313
|
+
`Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
|
|
7314
|
+
"error"
|
|
7315
|
+
);
|
|
7316
|
+
}
|
|
7317
|
+
}
|
|
5795
7318
|
emitAgentConnected(state.agentId, {
|
|
5796
7319
|
port: state.port,
|
|
5797
7320
|
cli_version: getCliVersion(),
|
|
@@ -5882,6 +7405,7 @@ async function run(options) {
|
|
|
5882
7405
|
throw error2;
|
|
5883
7406
|
}
|
|
5884
7407
|
scheduleSessionCleanup(state, channelDriver, options);
|
|
7408
|
+
state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
|
|
5885
7409
|
if (!interactive || state.json) {
|
|
5886
7410
|
log2(state, "Driving channel messages...");
|
|
5887
7411
|
}
|
|
@@ -5936,13 +7460,18 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
|
|
|
5936
7460
|
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);
|
|
5937
7461
|
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 }));
|
|
5938
7462
|
program.command("whoami").description("Show the currently logged in user").action(whoami);
|
|
7463
|
+
program.command("status").description("Check whether the configured credentials can reach Evident").option("--json", "Output in JSON format").action((options) => status({ json: options.json }));
|
|
7464
|
+
program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
|
|
5939
7465
|
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(
|
|
5940
7466
|
"-a, --agent [id]",
|
|
5941
7467
|
"Deprecated alias for --runner (still supported; --runner wins if both are given)"
|
|
5942
7468
|
).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
|
|
5943
7469
|
"--log-level <level>",
|
|
5944
7470
|
"Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
|
|
5945
|
-
).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(
|
|
7471
|
+
).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(
|
|
7472
|
+
"--opencode-start-timeout <seconds>",
|
|
7473
|
+
"Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
|
|
7474
|
+
).option("--json", "Output in JSON format").option(
|
|
5946
7475
|
"--session-cleanup-max-age <duration>",
|
|
5947
7476
|
"Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
|
|
5948
7477
|
).option(
|
|
@@ -5951,11 +7480,17 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5951
7480
|
).option(
|
|
5952
7481
|
"--session-cleanup-interval <duration>",
|
|
5953
7482
|
"How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
|
|
7483
|
+
).option(
|
|
7484
|
+
"--claude-usage-reporting <mode>",
|
|
7485
|
+
"Report Claude subscription usage to Evident: auto | on | off (default: auto). Env: EVIDENT_CLAUDE_USAGE_REPORTING"
|
|
5954
7486
|
).option(
|
|
5955
7487
|
"--enable-file-sync-to <dir>",
|
|
5956
7488
|
"Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
|
|
5957
7489
|
(value, previous) => previous.concat([value]),
|
|
5958
7490
|
[]
|
|
7491
|
+
).option(
|
|
7492
|
+
"--tunnel-ready-file <path>",
|
|
7493
|
+
"Path to write once the tunnel is connected (opt-in; unused unless an operator/image sets it)"
|
|
5959
7494
|
).action(
|
|
5960
7495
|
(options) => {
|
|
5961
7496
|
run({
|
|
@@ -5968,14 +7503,21 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5968
7503
|
verbose: options.verbose,
|
|
5969
7504
|
conversation: options.conversation,
|
|
5970
7505
|
idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
|
|
7506
|
+
// Raw string — validation/precedence is single-sourced in run.ts's
|
|
7507
|
+
// resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
|
|
7508
|
+
opencodeStartTimeout: options.opencodeStartTimeout,
|
|
5971
7509
|
json: options.json,
|
|
5972
7510
|
// Raw strings — the resolver in run.ts single-sources parsing (M1).
|
|
5973
7511
|
sessionCleanupMaxAge: options.sessionCleanupMaxAge,
|
|
5974
7512
|
sessionCleanupMaxCount: options.sessionCleanupMaxCount,
|
|
5975
7513
|
sessionCleanupInterval: options.sessionCleanupInterval,
|
|
7514
|
+
// Raw string — the resolver in run.ts single-sources parsing
|
|
7515
|
+
// (resolveClaudeUsageReportingMode).
|
|
7516
|
+
claudeUsageReporting: options.claudeUsageReporting,
|
|
5976
7517
|
// Raw values — expansion/validation is single-sourced in run.ts's
|
|
5977
7518
|
// resolveFileSyncDirectories.
|
|
5978
|
-
enableFileSyncTo: options.enableFileSyncTo
|
|
7519
|
+
enableFileSyncTo: options.enableFileSyncTo,
|
|
7520
|
+
tunnelReadyFile: options.tunnelReadyFile
|
|
5979
7521
|
});
|
|
5980
7522
|
}
|
|
5981
7523
|
);
|