@evident-ai/cli 3.1.1-dev.3ec6e7f → 3.1.1-dev.43b2e9a
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 +1778 -404
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
package/dist/index.js
CHANGED
|
@@ -11,8 +11,8 @@ import chalk2 from "chalk";
|
|
|
11
11
|
|
|
12
12
|
// src/lib/config.ts
|
|
13
13
|
import Conf from "conf";
|
|
14
|
-
import {
|
|
15
|
-
import {
|
|
14
|
+
import { chmodSync, existsSync, statSync } from "fs";
|
|
15
|
+
import { dirname } from "path";
|
|
16
16
|
var PRODUCTION_API_URL = "https://api.production.evident.run/v1";
|
|
17
17
|
var PRODUCTION_TUNNEL_URL = "wss://tunnel.production.evident.run";
|
|
18
18
|
var defaults = {
|
|
@@ -38,17 +38,39 @@ 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: "",
|
|
49
44
|
configName: "credentials",
|
|
50
|
-
defaults: {}
|
|
45
|
+
defaults: {},
|
|
46
|
+
configFileMode: 384
|
|
51
47
|
});
|
|
48
|
+
var CREDENTIALS_FILE_MODE = 384;
|
|
49
|
+
var CREDENTIALS_DIR_MODE = 448;
|
|
50
|
+
var permissionWarningEmitted = false;
|
|
51
|
+
function hardenCredentialsPermissions() {
|
|
52
|
+
if (process.platform === "win32") {
|
|
53
|
+
return;
|
|
54
|
+
}
|
|
55
|
+
const file = credentials.path;
|
|
56
|
+
for (const [path, mode] of [
|
|
57
|
+
[file, CREDENTIALS_FILE_MODE],
|
|
58
|
+
[dirname(file), CREDENTIALS_DIR_MODE]
|
|
59
|
+
]) {
|
|
60
|
+
try {
|
|
61
|
+
if (existsSync(path) && (statSync(path).mode & 511) !== mode) {
|
|
62
|
+
chmodSync(path, mode);
|
|
63
|
+
}
|
|
64
|
+
} catch (err) {
|
|
65
|
+
if (!permissionWarningEmitted) {
|
|
66
|
+
permissionWarningEmitted = true;
|
|
67
|
+
console.error(
|
|
68
|
+
`[config] could not restrict permissions on ${path}; the credentials file may be readable by other users on this machine: ${err instanceof Error ? err.message : String(err)}`
|
|
69
|
+
);
|
|
70
|
+
}
|
|
71
|
+
}
|
|
72
|
+
}
|
|
73
|
+
}
|
|
52
74
|
function getApiUrlConfig() {
|
|
53
75
|
return getApiUrl();
|
|
54
76
|
}
|
|
@@ -59,6 +81,7 @@ function credentialsKey() {
|
|
|
59
81
|
return getApiUrl();
|
|
60
82
|
}
|
|
61
83
|
function getCredentials() {
|
|
84
|
+
hardenCredentialsPermissions();
|
|
62
85
|
const byEndpoint = credentials.get("byEndpoint") ?? {};
|
|
63
86
|
return byEndpoint[credentialsKey()] ?? {};
|
|
64
87
|
}
|
|
@@ -70,14 +93,17 @@ function setCredentials(creds) {
|
|
|
70
93
|
expiresAt: creds.expiresAt
|
|
71
94
|
};
|
|
72
95
|
credentials.set("byEndpoint", byEndpoint);
|
|
96
|
+
hardenCredentialsPermissions();
|
|
73
97
|
}
|
|
74
98
|
function clearCredentials() {
|
|
75
99
|
const byEndpoint = credentials.get("byEndpoint") ?? {};
|
|
76
100
|
delete byEndpoint[credentialsKey()];
|
|
77
101
|
credentials.set("byEndpoint", byEndpoint);
|
|
102
|
+
hardenCredentialsPermissions();
|
|
78
103
|
}
|
|
79
104
|
function clearAllCredentials() {
|
|
80
105
|
credentials.clear();
|
|
106
|
+
hardenCredentialsPermissions();
|
|
81
107
|
}
|
|
82
108
|
function getCliName() {
|
|
83
109
|
const argv1 = process.argv[1] || "";
|
|
@@ -236,16 +262,28 @@ async function getToken() {
|
|
|
236
262
|
}
|
|
237
263
|
return null;
|
|
238
264
|
}
|
|
265
|
+
function toError(err) {
|
|
266
|
+
return err instanceof Error ? err : new Error(String(err));
|
|
267
|
+
}
|
|
239
268
|
async function deleteToken(options = {}) {
|
|
240
269
|
const keytar = await getKeytar();
|
|
270
|
+
const failures = [];
|
|
241
271
|
if (keytar) {
|
|
242
272
|
if (options.all) {
|
|
243
|
-
|
|
273
|
+
let accounts = [];
|
|
274
|
+
try {
|
|
275
|
+
accounts = await keytar.findCredentials(SERVICE_NAME);
|
|
276
|
+
} catch (err) {
|
|
277
|
+
failures.push({ type: "enumerate", error: toError(err) });
|
|
278
|
+
}
|
|
244
279
|
await Promise.all(
|
|
245
|
-
|
|
246
|
-
|
|
247
|
-
|
|
248
|
-
|
|
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
|
+
})
|
|
249
287
|
);
|
|
250
288
|
} else {
|
|
251
289
|
await keytar.deletePassword(SERVICE_NAME, keychainAccount());
|
|
@@ -256,6 +294,7 @@ async function deleteToken(options = {}) {
|
|
|
256
294
|
} else {
|
|
257
295
|
clearCredentials();
|
|
258
296
|
}
|
|
297
|
+
return { failures };
|
|
259
298
|
}
|
|
260
299
|
|
|
261
300
|
// src/utils/ui.ts
|
|
@@ -373,8 +412,10 @@ async function deviceFlowLogin(options) {
|
|
|
373
412
|
}
|
|
374
413
|
async function tokenLogin() {
|
|
375
414
|
console.log("Token login mode.");
|
|
376
|
-
console.log("
|
|
377
|
-
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
|
+
);
|
|
378
419
|
blank();
|
|
379
420
|
process.stdout.write("Paste token: ");
|
|
380
421
|
const token = await new Promise((resolve3) => {
|
|
@@ -398,13 +439,22 @@ async function tokenLogin() {
|
|
|
398
439
|
printError("No token provided.");
|
|
399
440
|
process.exit(1);
|
|
400
441
|
}
|
|
442
|
+
await validateAndStoreToken(token);
|
|
443
|
+
}
|
|
444
|
+
async function validateAndStoreToken(token) {
|
|
401
445
|
const spinner = ora("Validating token...").start();
|
|
402
446
|
try {
|
|
403
|
-
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
|
+
}
|
|
404
455
|
await storeToken({
|
|
405
456
|
token,
|
|
406
|
-
user: result.user
|
|
407
|
-
expiresAt: result.expires_at
|
|
457
|
+
user: { email: result.user.email }
|
|
408
458
|
});
|
|
409
459
|
spinner.stop();
|
|
410
460
|
printSuccess(`Logged in as ${chalk2.bold(result.user.email)}`);
|
|
@@ -424,9 +474,22 @@ async function login(options) {
|
|
|
424
474
|
}
|
|
425
475
|
|
|
426
476
|
// src/commands/logout.ts
|
|
477
|
+
function describeFailure(failure) {
|
|
478
|
+
if (failure.type === "enumerate") {
|
|
479
|
+
return `could not list stored keychain entries (${failure.error.message})`;
|
|
480
|
+
}
|
|
481
|
+
return `${failure.account} (${failure.error.message})`;
|
|
482
|
+
}
|
|
427
483
|
async function logout(options = {}) {
|
|
428
484
|
if (options.all) {
|
|
429
|
-
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: ${result.failures.map(describeFailure).join("; ")}. Your local credentials file was cleared, but stale keychain entries may remain \u2014 run \`evident logout --all\` again, or remove them manually from your OS keychain / credential manager.`
|
|
489
|
+
);
|
|
490
|
+
process.exitCode = 1;
|
|
491
|
+
return;
|
|
492
|
+
}
|
|
430
493
|
printSuccess("Logged out of all endpoints.");
|
|
431
494
|
return;
|
|
432
495
|
}
|
|
@@ -451,7 +514,9 @@ async function whoami() {
|
|
|
451
514
|
blank();
|
|
452
515
|
console.log(keyValue("Endpoint", apiUrl));
|
|
453
516
|
console.log(keyValue("User", chalk3.bold(credentials2.user.email)));
|
|
454
|
-
|
|
517
|
+
if (credentials2.user.id) {
|
|
518
|
+
console.log(keyValue("User ID", credentials2.user.id));
|
|
519
|
+
}
|
|
455
520
|
if (credentials2.expiresAt) {
|
|
456
521
|
const expiresAt = new Date(credentials2.expiresAt);
|
|
457
522
|
const now = /* @__PURE__ */ new Date();
|
|
@@ -464,7 +529,467 @@ async function whoami() {
|
|
|
464
529
|
console.log(keyValue("Expires", `${daysRemaining} days`));
|
|
465
530
|
}
|
|
466
531
|
}
|
|
467
|
-
blank();
|
|
532
|
+
blank();
|
|
533
|
+
}
|
|
534
|
+
|
|
535
|
+
// src/lib/auth.ts
|
|
536
|
+
async function getAuthCredentials() {
|
|
537
|
+
const runnerKey = process.env.EVIDENT_RUNNER_KEY;
|
|
538
|
+
const agentKey = process.env.EVIDENT_AGENT_KEY;
|
|
539
|
+
if (runnerKey) {
|
|
540
|
+
return {
|
|
541
|
+
token: runnerKey,
|
|
542
|
+
authType: "agent_key",
|
|
543
|
+
keySource: "runner_key",
|
|
544
|
+
notice: agentKey ? "Both EVIDENT_RUNNER_KEY and EVIDENT_AGENT_KEY are set; using EVIDENT_RUNNER_KEY." : void 0
|
|
545
|
+
};
|
|
546
|
+
}
|
|
547
|
+
if (agentKey) {
|
|
548
|
+
return { token: agentKey, authType: "agent_key", keySource: "agent_key" };
|
|
549
|
+
}
|
|
550
|
+
const userToken = process.env.EVIDENT_TOKEN;
|
|
551
|
+
if (userToken) {
|
|
552
|
+
return { token: userToken, authType: "bearer" };
|
|
553
|
+
}
|
|
554
|
+
const keychainCreds = await getToken();
|
|
555
|
+
if (keychainCreds) {
|
|
556
|
+
return {
|
|
557
|
+
token: keychainCreds.token,
|
|
558
|
+
authType: "bearer",
|
|
559
|
+
user: keychainCreds.user
|
|
560
|
+
};
|
|
561
|
+
}
|
|
562
|
+
return null;
|
|
563
|
+
}
|
|
564
|
+
function getAuthHeader(credentials2) {
|
|
565
|
+
if (credentials2.authType === "agent_key") {
|
|
566
|
+
return `SandboxKey ${credentials2.token}`;
|
|
567
|
+
}
|
|
568
|
+
return `Bearer ${credentials2.token}`;
|
|
569
|
+
}
|
|
570
|
+
function isInteractive(jsonOutput) {
|
|
571
|
+
if (jsonOutput) return false;
|
|
572
|
+
if (process.env.CI) return false;
|
|
573
|
+
if (process.env.GITHUB_ACTIONS) return false;
|
|
574
|
+
if (!process.stdin.isTTY) return false;
|
|
575
|
+
return true;
|
|
576
|
+
}
|
|
577
|
+
|
|
578
|
+
// src/commands/agent-lookup.ts
|
|
579
|
+
async function readErrorMessage(response) {
|
|
580
|
+
const text = await response.text().catch(() => "");
|
|
581
|
+
if (!text) return response.statusText || void 0;
|
|
582
|
+
try {
|
|
583
|
+
const data = JSON.parse(text);
|
|
584
|
+
const message = data.message ?? data.error;
|
|
585
|
+
if (typeof message === "string" && message.trim()) {
|
|
586
|
+
return message;
|
|
587
|
+
}
|
|
588
|
+
} catch {
|
|
589
|
+
}
|
|
590
|
+
return text.trim() || response.statusText || void 0;
|
|
591
|
+
}
|
|
592
|
+
function authFailureHint(apiUrl, serverMessage) {
|
|
593
|
+
const reason = serverMessage ? `: ${serverMessage}` : "";
|
|
594
|
+
return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
|
|
595
|
+
}
|
|
596
|
+
async function resolveAgentIdFromKey(authHeader) {
|
|
597
|
+
const apiUrl = getApiUrlConfig();
|
|
598
|
+
try {
|
|
599
|
+
const response = await fetch(`${apiUrl}/me`, {
|
|
600
|
+
headers: { Authorization: authHeader }
|
|
601
|
+
});
|
|
602
|
+
if (response.status === 401) {
|
|
603
|
+
const serverMessage = await readErrorMessage(response);
|
|
604
|
+
return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
605
|
+
}
|
|
606
|
+
if (!response.ok) {
|
|
607
|
+
const serverMessage = await readErrorMessage(response);
|
|
608
|
+
return {
|
|
609
|
+
error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
610
|
+
};
|
|
611
|
+
}
|
|
612
|
+
const data = await response.json();
|
|
613
|
+
if (data.auth_type === "agent_key" && data.agent_id) {
|
|
614
|
+
return { agent_id: data.agent_id };
|
|
615
|
+
}
|
|
616
|
+
return {
|
|
617
|
+
error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --runner explicitly."
|
|
618
|
+
};
|
|
619
|
+
} catch (error2) {
|
|
620
|
+
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
621
|
+
return { error: `Failed to resolve runner from key: ${message}` };
|
|
622
|
+
}
|
|
623
|
+
}
|
|
624
|
+
var BEST_EFFORT_NOTIFY_TIMEOUT_MS = 2e3;
|
|
625
|
+
async function notifyAgentDisconnected(agentId, authHeader) {
|
|
626
|
+
const apiUrl = getApiUrlConfig();
|
|
627
|
+
try {
|
|
628
|
+
const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
|
|
629
|
+
method: "POST",
|
|
630
|
+
headers: { Authorization: authHeader },
|
|
631
|
+
signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
|
|
632
|
+
});
|
|
633
|
+
if (!response.ok) {
|
|
634
|
+
const serverMessage = await readErrorMessage(response);
|
|
635
|
+
return {
|
|
636
|
+
ok: false,
|
|
637
|
+
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
638
|
+
};
|
|
639
|
+
}
|
|
640
|
+
return { ok: true };
|
|
641
|
+
} catch (error2) {
|
|
642
|
+
return { ok: false, error: describeBestEffortError(error2) };
|
|
643
|
+
}
|
|
644
|
+
}
|
|
645
|
+
function describeBestEffortError(error2) {
|
|
646
|
+
const name = error2?.name;
|
|
647
|
+
if (name === "TimeoutError" || name === "AbortError") {
|
|
648
|
+
return `timed out after ${BEST_EFFORT_NOTIFY_TIMEOUT_MS}ms`;
|
|
649
|
+
}
|
|
650
|
+
return error2 instanceof Error ? error2.message : String(error2);
|
|
651
|
+
}
|
|
652
|
+
async function reportMicrovmId(agentId, authHeader, microvmId) {
|
|
653
|
+
try {
|
|
654
|
+
const apiUrl = getApiUrlConfig();
|
|
655
|
+
const response = await fetch(`${apiUrl}/runners/${agentId}/microvm`, {
|
|
656
|
+
method: "POST",
|
|
657
|
+
headers: { Authorization: authHeader, "Content-Type": "application/json" },
|
|
658
|
+
body: JSON.stringify({ microvm_id: microvmId }),
|
|
659
|
+
signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
|
|
660
|
+
});
|
|
661
|
+
if (!response.ok) {
|
|
662
|
+
const serverMessage = await readErrorMessage(response);
|
|
663
|
+
return {
|
|
664
|
+
ok: false,
|
|
665
|
+
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
666
|
+
};
|
|
667
|
+
}
|
|
668
|
+
return { ok: true };
|
|
669
|
+
} catch (error2) {
|
|
670
|
+
return { ok: false, error: describeBestEffortError(error2) };
|
|
671
|
+
}
|
|
672
|
+
}
|
|
673
|
+
function toReportedWindow(window) {
|
|
674
|
+
if (!window) return null;
|
|
675
|
+
return { utilization: window.utilization, resets_at: window.resetsAt };
|
|
676
|
+
}
|
|
677
|
+
async function reportClaudeUsage(agentId, authHeader, snapshot) {
|
|
678
|
+
try {
|
|
679
|
+
const apiUrl = getApiUrlConfig();
|
|
680
|
+
const response = await fetch(`${apiUrl}/runners/${agentId}/claude-usage`, {
|
|
681
|
+
method: "POST",
|
|
682
|
+
headers: { Authorization: authHeader, "Content-Type": "application/json" },
|
|
683
|
+
body: JSON.stringify({
|
|
684
|
+
five_hour: toReportedWindow(snapshot.fiveHour),
|
|
685
|
+
seven_day: toReportedWindow(snapshot.sevenDay)
|
|
686
|
+
}),
|
|
687
|
+
signal: AbortSignal.timeout(BEST_EFFORT_NOTIFY_TIMEOUT_MS)
|
|
688
|
+
});
|
|
689
|
+
if (!response.ok) {
|
|
690
|
+
const serverMessage = await readErrorMessage(response);
|
|
691
|
+
return {
|
|
692
|
+
ok: false,
|
|
693
|
+
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
694
|
+
};
|
|
695
|
+
}
|
|
696
|
+
return { ok: true };
|
|
697
|
+
} catch (error2) {
|
|
698
|
+
return { ok: false, error: describeBestEffortError(error2) };
|
|
699
|
+
}
|
|
700
|
+
}
|
|
701
|
+
async function getAgentInfo(agentId, authHeader) {
|
|
702
|
+
const apiUrl = getApiUrlConfig();
|
|
703
|
+
try {
|
|
704
|
+
const response = await fetch(`${apiUrl}/runners/${agentId}`, {
|
|
705
|
+
headers: { Authorization: authHeader }
|
|
706
|
+
});
|
|
707
|
+
if (response.status === 401) {
|
|
708
|
+
const serverMessage = await readErrorMessage(response);
|
|
709
|
+
return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
710
|
+
}
|
|
711
|
+
if (response.status === 403) {
|
|
712
|
+
const serverMessage = await readErrorMessage(response);
|
|
713
|
+
return {
|
|
714
|
+
valid: false,
|
|
715
|
+
error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
|
|
716
|
+
};
|
|
717
|
+
}
|
|
718
|
+
if (response.status === 404) {
|
|
719
|
+
const serverMessage = await readErrorMessage(response);
|
|
720
|
+
return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
|
|
721
|
+
}
|
|
722
|
+
if (!response.ok) {
|
|
723
|
+
const serverMessage = await readErrorMessage(response);
|
|
724
|
+
return {
|
|
725
|
+
valid: false,
|
|
726
|
+
error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
727
|
+
};
|
|
728
|
+
}
|
|
729
|
+
const agent = await response.json();
|
|
730
|
+
if (agent.agent_type !== "local") {
|
|
731
|
+
return {
|
|
732
|
+
valid: false,
|
|
733
|
+
error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
|
|
734
|
+
};
|
|
735
|
+
}
|
|
736
|
+
return { valid: true, agent };
|
|
737
|
+
} catch (error2) {
|
|
738
|
+
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
739
|
+
return { valid: false, error: `Failed to validate runner: ${message}` };
|
|
740
|
+
}
|
|
741
|
+
}
|
|
742
|
+
|
|
743
|
+
// src/commands/status.ts
|
|
744
|
+
var STATUS_TIMEOUT_MS = 1e4;
|
|
745
|
+
function authLabelFor(credentials2) {
|
|
746
|
+
if (credentials2.authType === "agent_key") {
|
|
747
|
+
return credentials2.keySource === "agent_key" ? "runner key (EVIDENT_AGENT_KEY)" : "runner key (EVIDENT_RUNNER_KEY)";
|
|
748
|
+
}
|
|
749
|
+
return "user token";
|
|
750
|
+
}
|
|
751
|
+
function describeFetchError(error2) {
|
|
752
|
+
const name = error2?.name;
|
|
753
|
+
if (name === "TimeoutError" || name === "AbortError") {
|
|
754
|
+
return `timed out after ${STATUS_TIMEOUT_MS}ms waiting for a response`;
|
|
755
|
+
}
|
|
756
|
+
return error2 instanceof Error ? error2.message : String(error2);
|
|
757
|
+
}
|
|
758
|
+
async function checkStatus(jsonMode) {
|
|
759
|
+
const apiUrl = getApiUrlConfig();
|
|
760
|
+
const credentials2 = await getAuthCredentials();
|
|
761
|
+
if (!credentials2) {
|
|
762
|
+
return {
|
|
763
|
+
ok: false,
|
|
764
|
+
endpoint: apiUrl,
|
|
765
|
+
reason: "no_credentials",
|
|
766
|
+
error: "No credentials configured. Set EVIDENT_RUNNER_KEY (or EVIDENT_AGENT_KEY), or EVIDENT_TOKEN, or run `evident login`.",
|
|
767
|
+
exitCode: 1
|
|
768
|
+
};
|
|
769
|
+
}
|
|
770
|
+
if (credentials2.notice && !jsonMode) {
|
|
771
|
+
printWarning(credentials2.notice);
|
|
772
|
+
}
|
|
773
|
+
let response;
|
|
774
|
+
try {
|
|
775
|
+
response = await fetch(`${apiUrl}/me`, {
|
|
776
|
+
headers: { Authorization: getAuthHeader(credentials2) },
|
|
777
|
+
signal: AbortSignal.timeout(STATUS_TIMEOUT_MS)
|
|
778
|
+
});
|
|
779
|
+
} catch (error2) {
|
|
780
|
+
return {
|
|
781
|
+
ok: false,
|
|
782
|
+
endpoint: apiUrl,
|
|
783
|
+
authLabel: authLabelFor(credentials2),
|
|
784
|
+
reason: "unreachable",
|
|
785
|
+
error: `Could not reach ${apiUrl}: ${describeFetchError(error2)}. The credentials were NOT validated.`,
|
|
786
|
+
exitCode: 75
|
|
787
|
+
};
|
|
788
|
+
}
|
|
789
|
+
if (response.status === 401) {
|
|
790
|
+
const serverMessage = await readErrorMessage(response);
|
|
791
|
+
return {
|
|
792
|
+
ok: false,
|
|
793
|
+
endpoint: apiUrl,
|
|
794
|
+
authLabel: authLabelFor(credentials2),
|
|
795
|
+
reason: "unauthorized",
|
|
796
|
+
error: authFailureHint(apiUrl, serverMessage),
|
|
797
|
+
exitCode: 1
|
|
798
|
+
};
|
|
799
|
+
}
|
|
800
|
+
if (response.status >= 500) {
|
|
801
|
+
const serverMessage = await readErrorMessage(response);
|
|
802
|
+
return {
|
|
803
|
+
ok: false,
|
|
804
|
+
endpoint: apiUrl,
|
|
805
|
+
authLabel: authLabelFor(credentials2),
|
|
806
|
+
reason: "unreachable",
|
|
807
|
+
error: `${apiUrl} returned HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}. The credentials were NOT validated.`,
|
|
808
|
+
exitCode: 75
|
|
809
|
+
};
|
|
810
|
+
}
|
|
811
|
+
if (!response.ok) {
|
|
812
|
+
const serverMessage = await readErrorMessage(response);
|
|
813
|
+
return {
|
|
814
|
+
ok: false,
|
|
815
|
+
endpoint: apiUrl,
|
|
816
|
+
authLabel: authLabelFor(credentials2),
|
|
817
|
+
reason: "http_error",
|
|
818
|
+
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`,
|
|
819
|
+
exitCode: 1
|
|
820
|
+
};
|
|
821
|
+
}
|
|
822
|
+
const data = await response.json();
|
|
823
|
+
return {
|
|
824
|
+
ok: true,
|
|
825
|
+
endpoint: apiUrl,
|
|
826
|
+
authType: data.auth_type,
|
|
827
|
+
authLabel: authLabelFor(credentials2),
|
|
828
|
+
runnerId: data.auth_type === "agent_key" ? data.agent_id : void 0,
|
|
829
|
+
reason: "ok",
|
|
830
|
+
exitCode: 0
|
|
831
|
+
};
|
|
832
|
+
}
|
|
833
|
+
function printJson(result) {
|
|
834
|
+
const payload = {
|
|
835
|
+
ok: result.ok,
|
|
836
|
+
endpoint: result.endpoint
|
|
837
|
+
};
|
|
838
|
+
if (result.authType) payload.auth_type = result.authType;
|
|
839
|
+
if (result.runnerId) payload.runner_id = result.runnerId;
|
|
840
|
+
if (result.reason) payload.reason = result.reason;
|
|
841
|
+
if (result.error) payload.error = result.error;
|
|
842
|
+
console.log(JSON.stringify(payload));
|
|
843
|
+
}
|
|
844
|
+
function printHuman(result) {
|
|
845
|
+
blank();
|
|
846
|
+
console.log(keyValue("Endpoint", result.endpoint));
|
|
847
|
+
if (result.ok) {
|
|
848
|
+
console.log(keyValue("Auth", result.authLabel ?? "\u2014"));
|
|
849
|
+
if (result.runnerId) {
|
|
850
|
+
console.log(keyValue("Runner", result.runnerId));
|
|
851
|
+
}
|
|
852
|
+
console.log(keyValue("Status", "OK \u2014 credentials accepted"));
|
|
853
|
+
blank();
|
|
854
|
+
return;
|
|
855
|
+
}
|
|
856
|
+
if (result.authLabel) {
|
|
857
|
+
console.log(keyValue("Auth", result.authLabel));
|
|
858
|
+
}
|
|
859
|
+
blank();
|
|
860
|
+
printError(result.error ?? "Unknown error");
|
|
861
|
+
}
|
|
862
|
+
async function status(options = {}) {
|
|
863
|
+
const result = await checkStatus(Boolean(options.json));
|
|
864
|
+
if (options.json) {
|
|
865
|
+
printJson(result);
|
|
866
|
+
} else {
|
|
867
|
+
printHuman(result);
|
|
868
|
+
}
|
|
869
|
+
process.exit(result.exitCode);
|
|
870
|
+
}
|
|
871
|
+
|
|
872
|
+
// src/lib/claude-usage.ts
|
|
873
|
+
import { execFileSync } from "child_process";
|
|
874
|
+
import { readFileSync } from "fs";
|
|
875
|
+
import { homedir } from "os";
|
|
876
|
+
import { join } from "path";
|
|
877
|
+
var CLAUDE_USAGE_URL = "https://api.anthropic.com/api/oauth/usage";
|
|
878
|
+
var KEYCHAIN_SERVICE = "Claude Code-credentials";
|
|
879
|
+
function parseClaudeCliCredentials(raw) {
|
|
880
|
+
let parsed;
|
|
881
|
+
try {
|
|
882
|
+
parsed = JSON.parse(raw);
|
|
883
|
+
} catch {
|
|
884
|
+
return null;
|
|
885
|
+
}
|
|
886
|
+
const data = parsed.claudeAiOauth ?? parsed;
|
|
887
|
+
const creds = data;
|
|
888
|
+
if (typeof creds.accessToken !== "string" || typeof creds.expiresAt !== "number") {
|
|
889
|
+
return null;
|
|
890
|
+
}
|
|
891
|
+
return { accessToken: creds.accessToken, expiresAt: creds.expiresAt };
|
|
892
|
+
}
|
|
893
|
+
function readClaudeCliCredentials() {
|
|
894
|
+
if (process.platform === "darwin") {
|
|
895
|
+
try {
|
|
896
|
+
const raw = execFileSync(
|
|
897
|
+
"/usr/bin/security",
|
|
898
|
+
["find-generic-password", "-s", KEYCHAIN_SERVICE, "-w"],
|
|
899
|
+
{ encoding: "utf-8", timeout: 2e3, stdio: ["pipe", "pipe", "ignore"] }
|
|
900
|
+
);
|
|
901
|
+
return parseClaudeCliCredentials(raw);
|
|
902
|
+
} catch {
|
|
903
|
+
return null;
|
|
904
|
+
}
|
|
905
|
+
}
|
|
906
|
+
try {
|
|
907
|
+
const raw = readFileSync(join(homedir(), ".claude", ".credentials.json"), "utf-8");
|
|
908
|
+
return parseClaudeCliCredentials(raw);
|
|
909
|
+
} catch {
|
|
910
|
+
return null;
|
|
911
|
+
}
|
|
912
|
+
}
|
|
913
|
+
var ClaudeUsageError = class extends Error {
|
|
914
|
+
constructor(message, reason) {
|
|
915
|
+
super(message);
|
|
916
|
+
this.reason = reason;
|
|
917
|
+
}
|
|
918
|
+
};
|
|
919
|
+
function isLocalCredentialProblem(err) {
|
|
920
|
+
return err instanceof ClaudeUsageError && (err.reason === "no_credentials" || err.reason === "credentials_expired");
|
|
921
|
+
}
|
|
922
|
+
function normalizeResetsAt(value) {
|
|
923
|
+
const ms = Date.parse(value);
|
|
924
|
+
return Number.isNaN(ms) ? null : new Date(ms).toISOString();
|
|
925
|
+
}
|
|
926
|
+
function toWindow(value) {
|
|
927
|
+
if (!value || typeof value !== "object") {
|
|
928
|
+
return null;
|
|
929
|
+
}
|
|
930
|
+
const window = value;
|
|
931
|
+
if (typeof window.utilization !== "number" || typeof window.resets_at !== "string") {
|
|
932
|
+
return null;
|
|
933
|
+
}
|
|
934
|
+
const resetsAt = normalizeResetsAt(window.resets_at);
|
|
935
|
+
if (resetsAt === null) {
|
|
936
|
+
return null;
|
|
937
|
+
}
|
|
938
|
+
return { utilization: window.utilization, resetsAt };
|
|
939
|
+
}
|
|
940
|
+
async function getClaudeUsage() {
|
|
941
|
+
const credentials2 = readClaudeCliCredentials();
|
|
942
|
+
if (!credentials2) {
|
|
943
|
+
throw new ClaudeUsageError(
|
|
944
|
+
"No local Claude Code login found. Run `claude` once to sign in with your Claude subscription.",
|
|
945
|
+
"no_credentials"
|
|
946
|
+
);
|
|
947
|
+
}
|
|
948
|
+
if (credentials2.expiresAt < Date.now()) {
|
|
949
|
+
throw new ClaudeUsageError(
|
|
950
|
+
"Claude Code credentials have expired. Run `claude` to refresh them.",
|
|
951
|
+
"credentials_expired"
|
|
952
|
+
);
|
|
953
|
+
}
|
|
954
|
+
const res = await fetch(CLAUDE_USAGE_URL, {
|
|
955
|
+
headers: {
|
|
956
|
+
Authorization: `Bearer ${credentials2.accessToken}`,
|
|
957
|
+
"Content-Type": "application/json",
|
|
958
|
+
"anthropic-version": "2023-06-01"
|
|
959
|
+
}
|
|
960
|
+
});
|
|
961
|
+
if (!res.ok) {
|
|
962
|
+
throw new ClaudeUsageError(`Claude usage request failed: HTTP ${res.status}`, "request_failed");
|
|
963
|
+
}
|
|
964
|
+
const body = await res.json();
|
|
965
|
+
return {
|
|
966
|
+
fiveHour: toWindow(body.five_hour),
|
|
967
|
+
sevenDay: toWindow(body.seven_day)
|
|
968
|
+
};
|
|
969
|
+
}
|
|
970
|
+
|
|
971
|
+
// src/commands/claude-usage.ts
|
|
972
|
+
function formatWindow(label, window) {
|
|
973
|
+
if (!window) {
|
|
974
|
+
return keyValue(label, "not available for this plan");
|
|
975
|
+
}
|
|
976
|
+
const resetsAt = new Date(window.resetsAt);
|
|
977
|
+
return keyValue(label, `${window.utilization}% used, resets ${resetsAt.toLocaleString()}`);
|
|
978
|
+
}
|
|
979
|
+
async function claudeUsage() {
|
|
980
|
+
try {
|
|
981
|
+
const usage = await getClaudeUsage();
|
|
982
|
+
blank();
|
|
983
|
+
console.log(formatWindow("5-hour session", usage.fiveHour));
|
|
984
|
+
console.log(formatWindow("7-day", usage.sevenDay));
|
|
985
|
+
blank();
|
|
986
|
+
} catch (err) {
|
|
987
|
+
if (err instanceof ClaudeUsageError) {
|
|
988
|
+
printError(err.message);
|
|
989
|
+
process.exit(1);
|
|
990
|
+
}
|
|
991
|
+
throw err;
|
|
992
|
+
}
|
|
468
993
|
}
|
|
469
994
|
|
|
470
995
|
// src/commands/run.ts
|
|
@@ -472,6 +997,9 @@ import { homedir as homedir3 } from "os";
|
|
|
472
997
|
import { isAbsolute as isAbsolute2, join as join3, parse, resolve as resolvePath } from "path";
|
|
473
998
|
import chalk6 from "chalk";
|
|
474
999
|
|
|
1000
|
+
// ../../packages/types/src/agents/index.ts
|
|
1001
|
+
var MICROVM_MAX_LIFETIME_MS = 8 * 60 * 6e4;
|
|
1002
|
+
|
|
475
1003
|
// ../../packages/types/src/telemetry/index.ts
|
|
476
1004
|
var TelemetryEventTypes = {
|
|
477
1005
|
// Agent activity events (shown in web UI activity log)
|
|
@@ -479,7 +1007,10 @@ var TelemetryEventTypes = {
|
|
|
479
1007
|
AGENT_DISCONNECTED: "agent.disconnected",
|
|
480
1008
|
AGENT_MESSAGE_PROCESSING: "agent.message_processing",
|
|
481
1009
|
AGENT_MESSAGE_DONE: "agent.message_done",
|
|
482
|
-
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"
|
|
483
1014
|
};
|
|
484
1015
|
|
|
485
1016
|
// ../../packages/types/src/tunnel/index.ts
|
|
@@ -534,6 +1065,13 @@ var isShuttingDown = false;
|
|
|
534
1065
|
var FLUSH_INTERVAL_MS = 5e3;
|
|
535
1066
|
var MAX_BUFFER_SIZE = 50;
|
|
536
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;
|
|
537
1075
|
function logEvent(eventType, options = {}) {
|
|
538
1076
|
const event = {
|
|
539
1077
|
event_type: eventType,
|
|
@@ -568,9 +1106,16 @@ async function flushEvents() {
|
|
|
568
1106
|
flushTimeout = null;
|
|
569
1107
|
}
|
|
570
1108
|
try {
|
|
571
|
-
const
|
|
572
|
-
|
|
573
|
-
|
|
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 ${credentials2.token}`;
|
|
574
1119
|
}
|
|
575
1120
|
const apiUrl = getApiUrlConfig();
|
|
576
1121
|
const controller = new AbortController();
|
|
@@ -585,7 +1130,7 @@ async function flushEvents() {
|
|
|
585
1130
|
method: "POST",
|
|
586
1131
|
headers: {
|
|
587
1132
|
"Content-Type": "application/json",
|
|
588
|
-
Authorization:
|
|
1133
|
+
Authorization: authHeader
|
|
589
1134
|
},
|
|
590
1135
|
body: JSON.stringify(request),
|
|
591
1136
|
signal: controller.signal
|
|
@@ -597,8 +1142,15 @@ async function flushEvents() {
|
|
|
597
1142
|
clearTimeout(timeout);
|
|
598
1143
|
}
|
|
599
1144
|
} catch (error2) {
|
|
600
|
-
|
|
601
|
-
|
|
1145
|
+
const now = Date.now();
|
|
1146
|
+
if (now - lastFlushFailureLoggedAt >= FLUSH_FAILURE_LOG_INTERVAL_MS) {
|
|
1147
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
1148
|
+
const suffix = suppressedFlushFailureCount > 0 ? ` (${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++;
|
|
602
1154
|
}
|
|
603
1155
|
}
|
|
604
1156
|
}
|
|
@@ -667,47 +1219,67 @@ var EventTypes = {
|
|
|
667
1219
|
DEPRECATED_AGENT_KEY_ENV_USED: "cli.deprecated_agent_key_env_used"
|
|
668
1220
|
};
|
|
669
1221
|
|
|
670
|
-
// src/lib/
|
|
671
|
-
|
|
672
|
-
|
|
673
|
-
|
|
674
|
-
|
|
675
|
-
|
|
676
|
-
|
|
677
|
-
|
|
678
|
-
|
|
679
|
-
|
|
680
|
-
|
|
681
|
-
|
|
682
|
-
if (
|
|
683
|
-
|
|
684
|
-
|
|
685
|
-
|
|
686
|
-
|
|
687
|
-
|
|
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 ${RATE_LIMIT_MAX_EVENTS}/min)`
|
|
1247
|
+
);
|
|
1248
|
+
}
|
|
1249
|
+
windowStartedAt = now;
|
|
1250
|
+
windowCount = 0;
|
|
1251
|
+
windowDroppedCount = 0;
|
|
688
1252
|
}
|
|
689
|
-
|
|
690
|
-
|
|
691
|
-
|
|
692
|
-
|
|
693
|
-
|
|
694
|
-
|
|
695
|
-
}
|
|
1253
|
+
if (windowCount >= RATE_LIMIT_MAX_EVENTS) {
|
|
1254
|
+
windowDroppedCount++;
|
|
1255
|
+
if (windowDroppedCount === 1) {
|
|
1256
|
+
console.error(
|
|
1257
|
+
`[runner-activity-telemetry] rate cap reached (${RATE_LIMIT_MAX_EVENTS} per ${RATE_LIMIT_WINDOW_MS / 1e3}s) \u2014 dropping further entries this window`
|
|
1258
|
+
);
|
|
1259
|
+
}
|
|
1260
|
+
return false;
|
|
696
1261
|
}
|
|
697
|
-
|
|
1262
|
+
windowCount++;
|
|
1263
|
+
return true;
|
|
698
1264
|
}
|
|
699
|
-
function
|
|
700
|
-
|
|
701
|
-
|
|
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
|
+
);
|
|
702
1282
|
}
|
|
703
|
-
return `Bearer ${credentials2.token}`;
|
|
704
|
-
}
|
|
705
|
-
function isInteractive(jsonOutput) {
|
|
706
|
-
if (jsonOutput) return false;
|
|
707
|
-
if (process.env.CI) return false;
|
|
708
|
-
if (process.env.GITHUB_ACTIONS) return false;
|
|
709
|
-
if (!process.stdin.isTTY) return false;
|
|
710
|
-
return true;
|
|
711
1283
|
}
|
|
712
1284
|
|
|
713
1285
|
// src/lib/opencode/health.ts
|
|
@@ -1501,6 +2073,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
|
|
|
1501
2073
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1502
2074
|
return completedOf(reply) != null && finishOf(reply) === "tool-calls";
|
|
1503
2075
|
}
|
|
2076
|
+
function isB2AbandonmentConfirmed(params) {
|
|
2077
|
+
return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
|
|
2078
|
+
}
|
|
1504
2079
|
function messageError(messages, userMessageId) {
|
|
1505
2080
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1506
2081
|
const error2 = errorOf(reply);
|
|
@@ -1514,6 +2089,42 @@ function messageError(messages, userMessageId) {
|
|
|
1514
2089
|
}
|
|
1515
2090
|
return "The agent run failed.";
|
|
1516
2091
|
}
|
|
2092
|
+
function messageFailure(messages, userMessageId) {
|
|
2093
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
2094
|
+
const error2 = errorOf(reply);
|
|
2095
|
+
if (error2 == null || typeof error2 !== "object") return null;
|
|
2096
|
+
const e = error2;
|
|
2097
|
+
const replyProviderId = reply?.info?.providerID ?? null;
|
|
2098
|
+
const replyModelId = reply?.info?.modelID ?? null;
|
|
2099
|
+
if (e.name === "ProviderAuthError") {
|
|
2100
|
+
const data = e.data;
|
|
2101
|
+
const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
|
|
2102
|
+
return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
|
|
2103
|
+
}
|
|
2104
|
+
if (e.name === "APIError") {
|
|
2105
|
+
const data = e.data;
|
|
2106
|
+
const statusCode = data?.statusCode;
|
|
2107
|
+
if (statusCode === 401 || statusCode === 403) {
|
|
2108
|
+
return {
|
|
2109
|
+
kind: "model_auth",
|
|
2110
|
+
providerId: replyProviderId,
|
|
2111
|
+
modelId: replyModelId,
|
|
2112
|
+
reason: "rejected"
|
|
2113
|
+
};
|
|
2114
|
+
}
|
|
2115
|
+
}
|
|
2116
|
+
return null;
|
|
2117
|
+
}
|
|
2118
|
+
function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
|
|
2119
|
+
if (classified != null) return classified;
|
|
2120
|
+
if (hasConfiguredProvider !== false) return null;
|
|
2121
|
+
return {
|
|
2122
|
+
kind: "model_auth",
|
|
2123
|
+
providerId: replyProviderId,
|
|
2124
|
+
modelId: replyModelId,
|
|
2125
|
+
reason: "missing"
|
|
2126
|
+
};
|
|
2127
|
+
}
|
|
1517
2128
|
function hasRunningAssistantExcept(messages, exceptUserMessageId) {
|
|
1518
2129
|
if (!messages || messages.length === 0) return false;
|
|
1519
2130
|
return messages.some(
|
|
@@ -2042,13 +2653,56 @@ var RunnerConnection = class {
|
|
|
2042
2653
|
}
|
|
2043
2654
|
};
|
|
2044
2655
|
|
|
2656
|
+
// src/lib/tunnel/ready-marker.ts
|
|
2657
|
+
import { writeFileSync } from "fs";
|
|
2658
|
+
function writeTunnelReadyMarker(path, agentId) {
|
|
2659
|
+
try {
|
|
2660
|
+
writeFileSync(path, `${agentId}
|
|
2661
|
+
`);
|
|
2662
|
+
return { ok: true };
|
|
2663
|
+
} catch (error2) {
|
|
2664
|
+
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
2665
|
+
}
|
|
2666
|
+
}
|
|
2667
|
+
|
|
2668
|
+
// src/lib/claude-usage-reporting.ts
|
|
2669
|
+
var VALID_MODES = ["auto", "on", "off"];
|
|
2670
|
+
function resolveClaudeUsageReportingMode(flagValue, env) {
|
|
2671
|
+
const raw = flagValue ?? env.EVIDENT_CLAUDE_USAGE_REPORTING;
|
|
2672
|
+
if (raw === void 0 || raw === "") {
|
|
2673
|
+
return { mode: "auto", warnings: [] };
|
|
2674
|
+
}
|
|
2675
|
+
const normalized = raw.trim().toLowerCase();
|
|
2676
|
+
if (VALID_MODES.includes(normalized)) {
|
|
2677
|
+
return { mode: normalized, warnings: [] };
|
|
2678
|
+
}
|
|
2679
|
+
const source = flagValue !== void 0 ? "--claude-usage-reporting" : "EVIDENT_CLAUDE_USAGE_REPORTING";
|
|
2680
|
+
return {
|
|
2681
|
+
mode: "auto",
|
|
2682
|
+
warnings: [
|
|
2683
|
+
`Ignoring invalid ${source} "${raw}": expected one of ${VALID_MODES.join(", ")}; using auto`
|
|
2684
|
+
]
|
|
2685
|
+
};
|
|
2686
|
+
}
|
|
2687
|
+
var BASE_REPORT_DELAY_MS = 10 * 6e4;
|
|
2688
|
+
var REPORT_DELAY_JITTER_FRACTION = 0.2;
|
|
2689
|
+
function nextReportDelayMs(random = Math.random) {
|
|
2690
|
+
const jitterRangeMs = BASE_REPORT_DELAY_MS * REPORT_DELAY_JITTER_FRACTION;
|
|
2691
|
+
return BASE_REPORT_DELAY_MS - jitterRangeMs + random() * (2 * jitterRangeMs);
|
|
2692
|
+
}
|
|
2693
|
+
var FIRST_REPORT_DELAY_MS = 5e3 + Math.random() * 1e4;
|
|
2694
|
+
var CLAUDE_USAGE_FAILURE_REESCALATION_TICKS = 6;
|
|
2695
|
+
function claudeUsageFailureLogLevel(consecutiveFailures) {
|
|
2696
|
+
return consecutiveFailures === 1 || consecutiveFailures % CLAUDE_USAGE_FAILURE_REESCALATION_TICKS === 0 ? "warn" : "debug";
|
|
2697
|
+
}
|
|
2698
|
+
|
|
2045
2699
|
// src/lib/channels/driver.ts
|
|
2046
2700
|
import { homedir as homedir2 } from "os";
|
|
2047
2701
|
|
|
2048
2702
|
// src/lib/file-push.ts
|
|
2049
2703
|
import { randomUUID } from "crypto";
|
|
2050
2704
|
import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
|
|
2051
|
-
import { basename, dirname, isAbsolute, join as join2, relative, resolve as resolve2, sep } from "path";
|
|
2705
|
+
import { basename, dirname as dirname2, isAbsolute, join as join2, relative, resolve as resolve2, sep } from "path";
|
|
2052
2706
|
var FILE_MODE = 384;
|
|
2053
2707
|
var DIRECTORY_MODE = 448;
|
|
2054
2708
|
async function writePushedFile(request) {
|
|
@@ -2079,7 +2733,7 @@ async function writePushedFile(request) {
|
|
|
2079
2733
|
}
|
|
2080
2734
|
try {
|
|
2081
2735
|
const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
|
|
2082
|
-
|
|
2736
|
+
dirname2(candidate)
|
|
2083
2737
|
);
|
|
2084
2738
|
const realTarget = join2(existingAncestor, ...missingSegments, basename(candidate));
|
|
2085
2739
|
const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
|
|
@@ -2091,8 +2745,8 @@ async function writePushedFile(request) {
|
|
|
2091
2745
|
}
|
|
2092
2746
|
if (missingSegments.length > 0) {
|
|
2093
2747
|
await createMissingDirectories(existingAncestor, missingSegments);
|
|
2094
|
-
const realParent = await realpath(
|
|
2095
|
-
if (realParent !==
|
|
2748
|
+
const realParent = await realpath(dirname2(realTarget));
|
|
2749
|
+
if (realParent !== dirname2(realTarget) || !contains(allowedDirectory, realTarget)) {
|
|
2096
2750
|
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
2097
2751
|
path: realTarget,
|
|
2098
2752
|
bytes,
|
|
@@ -2135,7 +2789,7 @@ async function resolveNearestExistingAncestor(directory) {
|
|
|
2135
2789
|
try {
|
|
2136
2790
|
return { existingAncestor: await realpath(current), missingSegments };
|
|
2137
2791
|
} catch (err) {
|
|
2138
|
-
const parent =
|
|
2792
|
+
const parent = dirname2(current);
|
|
2139
2793
|
if (err.code !== "ENOENT" || parent === current) {
|
|
2140
2794
|
throw err;
|
|
2141
2795
|
}
|
|
@@ -2196,7 +2850,7 @@ async function createMissingDirectories(existingAncestor, missingSegments) {
|
|
|
2196
2850
|
}
|
|
2197
2851
|
}
|
|
2198
2852
|
async function writeAtomically(realTarget, content) {
|
|
2199
|
-
const temporaryPath = join2(
|
|
2853
|
+
const temporaryPath = join2(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
|
|
2200
2854
|
let handle;
|
|
2201
2855
|
try {
|
|
2202
2856
|
handle = await open2(temporaryPath, "wx", FILE_MODE);
|
|
@@ -2363,8 +3017,8 @@ async function applyOne(options, file) {
|
|
|
2363
3017
|
await ack(options, file, "applied");
|
|
2364
3018
|
return true;
|
|
2365
3019
|
}
|
|
2366
|
-
function durableDownloadCode(
|
|
2367
|
-
return
|
|
3020
|
+
function durableDownloadCode(status2) {
|
|
3021
|
+
return status2 === 413 ? "file_too_large" : "write_failed";
|
|
2368
3022
|
}
|
|
2369
3023
|
async function downloadContent(options, file, label) {
|
|
2370
3024
|
try {
|
|
@@ -2397,8 +3051,8 @@ async function downloadContent(options, file, label) {
|
|
|
2397
3051
|
return { ok: false, terminal: false };
|
|
2398
3052
|
}
|
|
2399
3053
|
}
|
|
2400
|
-
async function ack(options, file,
|
|
2401
|
-
const outcome = `${
|
|
3054
|
+
async function ack(options, file, status2, reason) {
|
|
3055
|
+
const outcome = `${status2}${reason ? ` (${reason})` : ""}`;
|
|
2402
3056
|
try {
|
|
2403
3057
|
const res = await options.fetchImpl(
|
|
2404
3058
|
`${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
|
|
@@ -2408,7 +3062,7 @@ async function ack(options, file, status, reason) {
|
|
|
2408
3062
|
Authorization: options.getAuthHeader(),
|
|
2409
3063
|
"Content-Type": "application/json"
|
|
2410
3064
|
},
|
|
2411
|
-
body: JSON.stringify(reason ? { status, reason } : { status })
|
|
3065
|
+
body: JSON.stringify(reason ? { status: status2, reason } : { status: status2 })
|
|
2412
3066
|
}
|
|
2413
3067
|
);
|
|
2414
3068
|
if (!res.ok) {
|
|
@@ -2468,6 +3122,8 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
|
|
|
2468
3122
|
var DEFAULT_STUCK_QUEUED_MS = 6e4;
|
|
2469
3123
|
var HEARTBEAT_MS = 6e4;
|
|
2470
3124
|
var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
|
|
3125
|
+
var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
|
|
3126
|
+
var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
|
|
2471
3127
|
var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
|
|
2472
3128
|
var MAX_SUPERSEDED_CONVERSATIONS = 256;
|
|
2473
3129
|
var ChannelAuthError = class extends Error {
|
|
@@ -2478,10 +3134,10 @@ var ChannelAuthError = class extends Error {
|
|
|
2478
3134
|
};
|
|
2479
3135
|
var ChannelTerminalError = class extends Error {
|
|
2480
3136
|
status;
|
|
2481
|
-
constructor(message,
|
|
3137
|
+
constructor(message, status2) {
|
|
2482
3138
|
super(message);
|
|
2483
3139
|
this.name = "ChannelTerminalError";
|
|
2484
|
-
this.status =
|
|
3140
|
+
this.status = status2;
|
|
2485
3141
|
}
|
|
2486
3142
|
};
|
|
2487
3143
|
function backoffDelay(attempt, policy) {
|
|
@@ -2489,8 +3145,8 @@ function backoffDelay(attempt, policy) {
|
|
|
2489
3145
|
const capped = Math.min(policy.maxDelayMs, exp);
|
|
2490
3146
|
return Math.floor(Math.random() * capped);
|
|
2491
3147
|
}
|
|
2492
|
-
function isRetryableStatus(
|
|
2493
|
-
return
|
|
3148
|
+
function isRetryableStatus(status2) {
|
|
3149
|
+
return status2 === 429 || status2 >= 500 && status2 <= 599;
|
|
2494
3150
|
}
|
|
2495
3151
|
var ChannelDriver = class _ChannelDriver {
|
|
2496
3152
|
agentId;
|
|
@@ -2605,6 +3261,24 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2605
3261
|
* processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
|
|
2606
3262
|
*/
|
|
2607
3263
|
readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
3264
|
+
/**
|
|
3265
|
+
* "Already emitted `redrive_unresolved` for this row" (#965). Mirrors
|
|
3266
|
+
* `readoptPollUnresolvedSignalled`: `resolveRedrive`'s `unresolved` leaf recurs
|
|
3267
|
+
* every ~2s drain until opencode's status becomes readable, but the
|
|
3268
|
+
* server-visible signal is an OUTCOME, so it fires at most once per row. Cleared
|
|
3269
|
+
* on any non-`unresolved` outcome so the set cannot grow beyond the currently
|
|
3270
|
+
* unresolvable rows.
|
|
3271
|
+
*/
|
|
3272
|
+
redriveUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
3273
|
+
/**
|
|
3274
|
+
* First `now()` a `pending` row's re-drive was observed `unresolved` (#965). A
|
|
3275
|
+
* `pending` row is invisible to every cron arm (all require `status =
|
|
3276
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
3277
|
+
* nothing driving it. Once `now - since >= pausedMaxWaitMs`, `resolveRedrive`
|
|
3278
|
+
* takes `dispatch` instead of `unresolved` (reusing the existing knob — see
|
|
3279
|
+
* ADR-0047's own "unreachable ⇒ bounded" rule). Cleared on any other outcome.
|
|
3280
|
+
*/
|
|
3281
|
+
redriveUnresolvedSince = /* @__PURE__ */ new Map();
|
|
2608
3282
|
/**
|
|
2609
3283
|
* "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
|
|
2610
3284
|
* Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
|
|
@@ -2693,23 +3367,23 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2693
3367
|
* and stops opencode.
|
|
2694
3368
|
*/
|
|
2695
3369
|
stopped = false;
|
|
2696
|
-
constructor(
|
|
2697
|
-
this.agentId =
|
|
2698
|
-
this.port =
|
|
2699
|
-
this.apiUrl =
|
|
2700
|
-
this.getAuthHeader =
|
|
2701
|
-
this.conversationFilter =
|
|
2702
|
-
this.retry = { ...DEFAULT_RETRY_POLICY, ...
|
|
2703
|
-
this.log =
|
|
3370
|
+
constructor(config) {
|
|
3371
|
+
this.agentId = config.agentId;
|
|
3372
|
+
this.port = config.port;
|
|
3373
|
+
this.apiUrl = config.apiUrl.replace(/\/$/, "");
|
|
3374
|
+
this.getAuthHeader = config.getAuthHeader;
|
|
3375
|
+
this.conversationFilter = config.conversationFilter ?? null;
|
|
3376
|
+
this.retry = { ...DEFAULT_RETRY_POLICY, ...config.retry };
|
|
3377
|
+
this.log = config.log ?? (() => {
|
|
2704
3378
|
});
|
|
2705
|
-
this.fetchImpl =
|
|
2706
|
-
this.sleep =
|
|
2707
|
-
this.pausedPollIntervalMs =
|
|
2708
|
-
this.pausedMaxWaitMs =
|
|
2709
|
-
this.stuckQueuedMs =
|
|
2710
|
-
this.now =
|
|
2711
|
-
this.fileSyncDirectories =
|
|
2712
|
-
this.homeDir =
|
|
3379
|
+
this.fetchImpl = config.fetchImpl ?? fetch;
|
|
3380
|
+
this.sleep = config.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
|
|
3381
|
+
this.pausedPollIntervalMs = config.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
|
|
3382
|
+
this.pausedMaxWaitMs = config.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
|
|
3383
|
+
this.stuckQueuedMs = config.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
|
|
3384
|
+
this.now = config.now ?? (() => Date.now());
|
|
3385
|
+
this.fileSyncDirectories = config.fileSyncDirectories ?? [];
|
|
3386
|
+
this.homeDir = config.homeDir ?? homedir2();
|
|
2713
3387
|
}
|
|
2714
3388
|
/** The IPv4-loopback base URL for the local `opencode serve`. */
|
|
2715
3389
|
get opencodeBase() {
|
|
@@ -2944,6 +3618,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2944
3618
|
skippedAlreadyDispatched += 1;
|
|
2945
3619
|
continue;
|
|
2946
3620
|
}
|
|
3621
|
+
if (message.opencode_message_id) {
|
|
3622
|
+
const outcome = await this.resolveRedrive(conv, sessionId, message, refusedSessionId);
|
|
3623
|
+
if (outcome !== "dispatch") {
|
|
3624
|
+
break;
|
|
3625
|
+
}
|
|
3626
|
+
}
|
|
2947
3627
|
const options = {
|
|
2948
3628
|
agent: message.opencode_agent ?? void 0,
|
|
2949
3629
|
model: message.opencode_model ?? void 0
|
|
@@ -2952,86 +3632,298 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2952
3632
|
try {
|
|
2953
3633
|
this.log({
|
|
2954
3634
|
level: "info",
|
|
2955
|
-
message: `Dispatching message ${message.id.slice(0, 8)} to OpenCode native queue (session ${sessionId.slice(0, 8)})`,
|
|
2956
|
-
conversation_id: conv.id,
|
|
2957
|
-
message_id: message.id
|
|
2958
|
-
});
|
|
2959
|
-
const sendAttachments = this.buildSendAttachments(conv, message);
|
|
2960
|
-
opencodeMessageId = await this.dispatchLocked(
|
|
2961
|
-
sessionId,
|
|
2962
|
-
() => sendPromptAsync(this.port, sessionId, message.content, options, sendAttachments)
|
|
2963
|
-
);
|
|
2964
|
-
} catch (err) {
|
|
2965
|
-
if (err instanceof ChannelAuthError) throw err;
|
|
2966
|
-
this.dispatched.delete(message.id);
|
|
2967
|
-
const exists = await sessionExists(this.port, sessionId);
|
|
2968
|
-
if (exists === false) {
|
|
2969
|
-
this.sessions.delete(conv.id);
|
|
2970
|
-
this.log({
|
|
2971
|
-
level: "warn",
|
|
2972
|
-
message: `Message ${message.id.slice(0, 8)} dispatch hit a session (${sessionId.slice(0, 8)}) that was deleted mid-dispatch (cleanup race) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick (recreated then, in order). Already-dispatched turns keep their watcher.`,
|
|
2973
|
-
conversation_id: conv.id,
|
|
2974
|
-
message_id: message.id
|
|
2975
|
-
});
|
|
2976
|
-
break;
|
|
2977
|
-
}
|
|
2978
|
-
if (exists === null) {
|
|
2979
|
-
this.log({
|
|
2980
|
-
level: "warn",
|
|
2981
|
-
message: `Message ${message.id.slice(0, 8)} dispatch failed and session (${sessionId.slice(0, 8)}) existence could not be confirmed (opencode momentarily unreachable) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick rather than treating it as a genuine failure.`,
|
|
2982
|
-
conversation_id: conv.id,
|
|
2983
|
-
message_id: message.id
|
|
2984
|
-
});
|
|
2985
|
-
break;
|
|
2986
|
-
}
|
|
2987
|
-
const errorMessage = err instanceof Error ? err.message : String(err);
|
|
2988
|
-
this.sessions.delete(conv.id);
|
|
2989
|
-
this.supersede(conv.id, sessionId);
|
|
2990
|
-
this.log({
|
|
2991
|
-
level: "warn",
|
|
2992
|
-
message: `Abandoning OpenCode session ${sessionId.slice(0, 8)} as the binding for conversation ${conv.id.slice(0, 8)} (it exists but failed to run a turn) \u2014 a fresh session is created on the next tick, whatever the persisted binding says by then.`,
|
|
3635
|
+
message: `Dispatching message ${message.id.slice(0, 8)} to OpenCode native queue (session ${sessionId.slice(0, 8)})`,
|
|
3636
|
+
conversation_id: conv.id,
|
|
3637
|
+
message_id: message.id
|
|
3638
|
+
});
|
|
3639
|
+
const sendAttachments = this.buildSendAttachments(conv, message);
|
|
3640
|
+
opencodeMessageId = await this.dispatchLocked(
|
|
3641
|
+
sessionId,
|
|
3642
|
+
() => sendPromptAsync(this.port, sessionId, message.content, options, sendAttachments)
|
|
3643
|
+
);
|
|
3644
|
+
} catch (err) {
|
|
3645
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3646
|
+
this.dispatched.delete(message.id);
|
|
3647
|
+
const exists = await sessionExists(this.port, sessionId);
|
|
3648
|
+
if (exists === false) {
|
|
3649
|
+
this.sessions.delete(conv.id);
|
|
3650
|
+
this.log({
|
|
3651
|
+
level: "warn",
|
|
3652
|
+
message: `Message ${message.id.slice(0, 8)} dispatch hit a session (${sessionId.slice(0, 8)}) that was deleted mid-dispatch (cleanup race) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick (recreated then, in order). Already-dispatched turns keep their watcher.`,
|
|
3653
|
+
conversation_id: conv.id,
|
|
3654
|
+
message_id: message.id
|
|
3655
|
+
});
|
|
3656
|
+
break;
|
|
3657
|
+
}
|
|
3658
|
+
if (exists === null) {
|
|
3659
|
+
this.log({
|
|
3660
|
+
level: "warn",
|
|
3661
|
+
message: `Message ${message.id.slice(0, 8)} dispatch failed and session (${sessionId.slice(0, 8)}) existence could not be confirmed (opencode momentarily unreachable) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick rather than treating it as a genuine failure.`,
|
|
3662
|
+
conversation_id: conv.id,
|
|
3663
|
+
message_id: message.id
|
|
3664
|
+
});
|
|
3665
|
+
break;
|
|
3666
|
+
}
|
|
3667
|
+
const errorMessage = err instanceof Error ? err.message : String(err);
|
|
3668
|
+
this.sessions.delete(conv.id);
|
|
3669
|
+
this.supersede(conv.id, sessionId);
|
|
3670
|
+
this.log({
|
|
3671
|
+
level: "warn",
|
|
3672
|
+
message: `Abandoning OpenCode session ${sessionId.slice(0, 8)} as the binding for conversation ${conv.id.slice(0, 8)} (it exists but failed to run a turn) \u2014 a fresh session is created on the next tick, whatever the persisted binding says by then.`,
|
|
3673
|
+
conversation_id: conv.id,
|
|
3674
|
+
message_id: message.id
|
|
3675
|
+
});
|
|
3676
|
+
await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
|
|
3677
|
+
this.log({
|
|
3678
|
+
level: "warn",
|
|
3679
|
+
message: `markFailed PATCH for message ${message.id.slice(0, 8)} (conversation ${conv.id.slice(0, 8)}) failed (best-effort, not retried): ${markErr instanceof Error ? markErr.message : String(markErr)}`,
|
|
3680
|
+
conversation_id: conv.id,
|
|
3681
|
+
message_id: message.id
|
|
3682
|
+
});
|
|
3683
|
+
});
|
|
3684
|
+
this.log({
|
|
3685
|
+
level: "error",
|
|
3686
|
+
message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
|
|
3687
|
+
conversation_id: conv.id,
|
|
3688
|
+
message_id: message.id
|
|
3689
|
+
});
|
|
3690
|
+
break;
|
|
3691
|
+
}
|
|
3692
|
+
if (opencodeMessageId === null) {
|
|
3693
|
+
this.log({
|
|
3694
|
+
level: "warn",
|
|
3695
|
+
message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
|
|
3696
|
+
conversation_id: conv.id,
|
|
3697
|
+
message_id: message.id
|
|
3698
|
+
});
|
|
3699
|
+
continue;
|
|
3700
|
+
}
|
|
3701
|
+
this.dispatched.add(message.id);
|
|
3702
|
+
this.registerInFlight(conv, sessionId, message, opencodeMessageId);
|
|
3703
|
+
dispatched += 1;
|
|
3704
|
+
void this.postSignal(conv.id, message.id, "dispatched");
|
|
3705
|
+
}
|
|
3706
|
+
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3707
|
+
this.log({
|
|
3708
|
+
level: "warn",
|
|
3709
|
+
message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
|
|
3710
|
+
conversation_id: conv.id
|
|
3711
|
+
});
|
|
3712
|
+
}
|
|
3713
|
+
this.ensureWatcherRunning(sessionId);
|
|
3714
|
+
return dispatched;
|
|
3715
|
+
}
|
|
3716
|
+
/**
|
|
3717
|
+
* Poll a session's message list for the re-drive fence (#965), via the
|
|
3718
|
+
* INJECTED `fetchImpl` — NOT the imported `getSessionMessages` helper, which
|
|
3719
|
+
* hits the global `fetch` and would bypass the same override every other
|
|
3720
|
+
* opencode poll in this file respects. Mirrors `readoptProcessing`'s own
|
|
3721
|
+
* snapshot fetch (`:3081-3111`). `null` = unreadable (non-OK response,
|
|
3722
|
+
* non-array body, or a network exception) — treated as "can't observe",
|
|
3723
|
+
* never as "confirmed gone".
|
|
3724
|
+
*/
|
|
3725
|
+
async pollSessionMessagesForRedrive(conv, message, sessionId) {
|
|
3726
|
+
try {
|
|
3727
|
+
const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
|
|
3728
|
+
if (!res.ok) {
|
|
3729
|
+
this.log({
|
|
3730
|
+
level: "warn",
|
|
3731
|
+
message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned HTTP ${res.status} \u2014 treating as unreadable this tick`,
|
|
3732
|
+
conversation_id: conv.id,
|
|
3733
|
+
message_id: message.id
|
|
3734
|
+
});
|
|
3735
|
+
return null;
|
|
3736
|
+
}
|
|
3737
|
+
const body = await res.json();
|
|
3738
|
+
if (!Array.isArray(body)) {
|
|
3739
|
+
this.log({
|
|
3740
|
+
level: "warn",
|
|
3741
|
+
message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned a non-array message body \u2014 treating as unreadable this tick`,
|
|
3742
|
+
conversation_id: conv.id,
|
|
3743
|
+
message_id: message.id
|
|
3744
|
+
});
|
|
3745
|
+
return null;
|
|
3746
|
+
}
|
|
3747
|
+
return body;
|
|
3748
|
+
} catch (err) {
|
|
3749
|
+
this.log({
|
|
3750
|
+
level: "warn",
|
|
3751
|
+
message: `Re-drive: failed to poll session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3752
|
+
conversation_id: conv.id,
|
|
3753
|
+
message_id: message.id
|
|
3754
|
+
});
|
|
3755
|
+
return null;
|
|
3756
|
+
}
|
|
3757
|
+
}
|
|
3758
|
+
/**
|
|
3759
|
+
* The re-drive fence for a `pending` row that already carries a stored
|
|
3760
|
+
* `opencode_message_id` (#965) — i.e. it has already been handed to opencode at
|
|
3761
|
+
* least once (see the invariant at `QueuedMessage.opencode_message_id`'s doc).
|
|
3762
|
+
* The lifecycle cron can falsely reclaim a `processing` row back to `pending`
|
|
3763
|
+
* mid-turn (a 5-minute liveness-staleness check racing a still-running turn);
|
|
3764
|
+
* without this fence the drain loop would re-`prompt_async` the SAME turn a
|
|
3765
|
+
* second time against live GitHub state. Mirrors `readoptOne`'s job for the
|
|
3766
|
+
* `processing` re-adopt path, but simpler: no b1/b2 preamble cross-check is
|
|
3767
|
+
* needed here because `refusedSessionId` already handles the one case
|
|
3768
|
+
* (#553 abandoned session) that path exists for.
|
|
3769
|
+
*
|
|
3770
|
+
* Only `ChannelAuthError` propagates; every other failure resolves to
|
|
3771
|
+
* `unresolved` and is retried whole on the next ~2s drain tick.
|
|
3772
|
+
*/
|
|
3773
|
+
async resolveRedrive(conv, sessionId, message, refusedSessionId) {
|
|
3774
|
+
const ocId = message.opencode_message_id ?? null;
|
|
3775
|
+
if (refusedSessionId) {
|
|
3776
|
+
this.clearRedriveUnresolved(message.id);
|
|
3777
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3778
|
+
return "dispatch";
|
|
3779
|
+
}
|
|
3780
|
+
const messages = await this.pollSessionMessagesForRedrive(conv, message, sessionId);
|
|
3781
|
+
if (messages == null || messages.length === 0) {
|
|
3782
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3783
|
+
}
|
|
3784
|
+
const state = messageRunState(messages, ocId ?? "");
|
|
3785
|
+
if (state === "done" || state === "failed") {
|
|
3786
|
+
return this.settleRedrive(conv, sessionId, message, ocId, messages, state);
|
|
3787
|
+
}
|
|
3788
|
+
if (state === "running" || state === "queued") {
|
|
3789
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3790
|
+
if (ongoing === true) {
|
|
3791
|
+
return this.reattachRedrive(conv, sessionId, message, ocId);
|
|
3792
|
+
}
|
|
3793
|
+
if (ongoing === false) {
|
|
3794
|
+
this.clearRedriveUnresolved(message.id);
|
|
3795
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3796
|
+
return "dispatch";
|
|
3797
|
+
}
|
|
3798
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3799
|
+
}
|
|
3800
|
+
this.clearRedriveUnresolved(message.id);
|
|
3801
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3802
|
+
return "dispatch";
|
|
3803
|
+
}
|
|
3804
|
+
/**
|
|
3805
|
+
* The `reattached` outcome (Task 3.3): the prior turn is STILL ONGOING per
|
|
3806
|
+
* opencode's own status map — undo the false reclaim instead of starting a
|
|
3807
|
+
* second turn.
|
|
3808
|
+
*/
|
|
3809
|
+
async reattachRedrive(conv, sessionId, message, ocId) {
|
|
3810
|
+
let anchorMs;
|
|
3811
|
+
const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
|
|
3812
|
+
if (!Number.isNaN(parsed)) {
|
|
3813
|
+
anchorMs = parsed;
|
|
3814
|
+
} else {
|
|
3815
|
+
anchorMs = this.now();
|
|
3816
|
+
this.log({
|
|
3817
|
+
level: "error",
|
|
3818
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} has null/unparseable processing_started_at (${String(message.processing_started_at)}) \u2014 anchoring the watcher's absolute-age ceiling to now (defensive)`,
|
|
3819
|
+
conversation_id: conv.id,
|
|
3820
|
+
message_id: message.id
|
|
3821
|
+
});
|
|
3822
|
+
}
|
|
3823
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3824
|
+
try {
|
|
3825
|
+
await this.markProcessing(conv.id, message.id, sessionId, ocId, title);
|
|
3826
|
+
} catch (err) {
|
|
3827
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3828
|
+
this.log({
|
|
3829
|
+
level: "warn",
|
|
3830
|
+
message: `Re-drive: failed to restore message ${message.id.slice(0, 8)} to processing (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3831
|
+
conversation_id: conv.id,
|
|
3832
|
+
message_id: message.id
|
|
3833
|
+
});
|
|
3834
|
+
return "unresolved";
|
|
3835
|
+
}
|
|
3836
|
+
this.clearRedriveUnresolved(message.id);
|
|
3837
|
+
this.registerReadopted(conv, sessionId, message, ocId ?? "", anchorMs);
|
|
3838
|
+
this.dispatched.add(message.id);
|
|
3839
|
+
this.readopted.add(message.id);
|
|
3840
|
+
this.ensureWatcherRunning(sessionId);
|
|
3841
|
+
const watchedForMs = this.now() - anchorMs;
|
|
3842
|
+
void this.postSignal(conv.id, message.id, "redrive_reattached", {
|
|
3843
|
+
watched_for_ms: watchedForMs
|
|
3844
|
+
});
|
|
3845
|
+
this.log({
|
|
3846
|
+
level: "warn",
|
|
3847
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) was wrongly reclaimed to pending while its turn was still running (watched ${watchedForMs}ms) \u2014 restored to processing instead of re-dispatching`,
|
|
3848
|
+
conversation_id: conv.id,
|
|
3849
|
+
message_id: message.id
|
|
3850
|
+
});
|
|
3851
|
+
return "reattached";
|
|
3852
|
+
}
|
|
3853
|
+
/**
|
|
3854
|
+
* The `settled` outcome (Task 3.2): the prior turn already finished (or
|
|
3855
|
+
* errored) while nobody was watching — deliver/report it instead of re-running.
|
|
3856
|
+
* Mirrors `readoptOne`'s `done`/`failed` branches' error discipline, simplified
|
|
3857
|
+
* (no `doneUndeliverable` park: a terminal PATCH failure here just retries next
|
|
3858
|
+
* drain, same as any other non-auth failure).
|
|
3859
|
+
*/
|
|
3860
|
+
async settleRedrive(conv, sessionId, message, ocId, messages, state) {
|
|
3861
|
+
try {
|
|
3862
|
+
if (state === "done") {
|
|
3863
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3864
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3865
|
+
this.log({
|
|
3866
|
+
level: "info",
|
|
3867
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} completed while its row was wrongly reclaimed to pending \u2014 marking done instead of re-dispatching`,
|
|
2993
3868
|
conversation_id: conv.id,
|
|
2994
3869
|
message_id: message.id
|
|
2995
3870
|
});
|
|
2996
|
-
await this.
|
|
2997
|
-
|
|
2998
|
-
|
|
2999
|
-
|
|
3000
|
-
|
|
3001
|
-
message_id: message.id
|
|
3002
|
-
});
|
|
3003
|
-
});
|
|
3871
|
+
await this.markDone(conv.id, message.id, sessionId, ocId, title, usage);
|
|
3872
|
+
} else {
|
|
3873
|
+
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3874
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3875
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3004
3876
|
this.log({
|
|
3005
3877
|
level: "error",
|
|
3006
|
-
message: `
|
|
3007
|
-
conversation_id: conv.id,
|
|
3008
|
-
message_id: message.id
|
|
3009
|
-
});
|
|
3010
|
-
break;
|
|
3011
|
-
}
|
|
3012
|
-
if (opencodeMessageId === null) {
|
|
3013
|
-
this.log({
|
|
3014
|
-
level: "warn",
|
|
3015
|
-
message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
|
|
3878
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} errored while its row was wrongly reclaimed to pending \u2014 marking failed instead of re-dispatching: ${error2 ?? "(no error text)"}`,
|
|
3016
3879
|
conversation_id: conv.id,
|
|
3017
3880
|
message_id: message.id
|
|
3018
3881
|
});
|
|
3019
|
-
|
|
3882
|
+
await this.markFailed(conv.id, message.id, sessionId, error2, usage, failure);
|
|
3020
3883
|
}
|
|
3021
|
-
|
|
3022
|
-
|
|
3023
|
-
dispatched += 1;
|
|
3024
|
-
void this.postSignal(conv.id, message.id, "dispatched");
|
|
3025
|
-
}
|
|
3026
|
-
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3884
|
+
} catch (err) {
|
|
3885
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3027
3886
|
this.log({
|
|
3028
3887
|
level: "warn",
|
|
3029
|
-
message: `
|
|
3030
|
-
conversation_id: conv.id
|
|
3888
|
+
message: `Re-drive: failed to report message ${message.id.slice(0, 8)} ${state} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3889
|
+
conversation_id: conv.id,
|
|
3890
|
+
message_id: message.id
|
|
3031
3891
|
});
|
|
3892
|
+
return "unresolved";
|
|
3032
3893
|
}
|
|
3033
|
-
this.
|
|
3034
|
-
|
|
3894
|
+
this.clearRedriveUnresolved(message.id);
|
|
3895
|
+
void this.postSignal(conv.id, message.id, "redrive_settled");
|
|
3896
|
+
return "settled";
|
|
3897
|
+
}
|
|
3898
|
+
/**
|
|
3899
|
+
* The bounded `unresolved` outcome (Task 3.4): opencode's state could not be
|
|
3900
|
+
* observed (snapshot unreadable/empty, or `isSessionOngoing` returned `null`).
|
|
3901
|
+
* A `pending` row is invisible to every cron arm (all require `status =
|
|
3902
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
3903
|
+
* nothing driving it — bound it to the existing `pausedMaxWaitMs` window
|
|
3904
|
+
* (reusing the knob, not a new constant) and take `dispatch` once elapsed.
|
|
3905
|
+
*/
|
|
3906
|
+
resolveRedriveUnresolved(conv, message) {
|
|
3907
|
+
const now = this.now();
|
|
3908
|
+
const since = this.redriveUnresolvedSince.get(message.id);
|
|
3909
|
+
if (since !== void 0 && now - since >= this.pausedMaxWaitMs) {
|
|
3910
|
+
this.clearRedriveUnresolved(message.id);
|
|
3911
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3912
|
+
return "dispatch";
|
|
3913
|
+
}
|
|
3914
|
+
if (since === void 0) {
|
|
3915
|
+
this.redriveUnresolvedSince.set(message.id, now);
|
|
3916
|
+
}
|
|
3917
|
+
if (!this.redriveUnresolvedSignalled.has(message.id)) {
|
|
3918
|
+
this.redriveUnresolvedSignalled.add(message.id);
|
|
3919
|
+
void this.postSignal(conv.id, message.id, "redrive_unresolved");
|
|
3920
|
+
}
|
|
3921
|
+
return "unresolved";
|
|
3922
|
+
}
|
|
3923
|
+
/** Clear both `unresolved`-bound trackers for a row (any non-`unresolved` outcome). */
|
|
3924
|
+
clearRedriveUnresolved(messageId) {
|
|
3925
|
+
this.redriveUnresolvedSince.delete(messageId);
|
|
3926
|
+
this.redriveUnresolvedSignalled.delete(messageId);
|
|
3035
3927
|
}
|
|
3036
3928
|
/**
|
|
3037
3929
|
* Record that `sessionId` is no longer a valid binding for `conversationId`
|
|
@@ -3094,7 +3986,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3094
3986
|
const directory = await this.resolveOpenCodeDirectory();
|
|
3095
3987
|
const sessionId = await createOpenCodeSession(this.port, directory);
|
|
3096
3988
|
this.sessions.set(conversationId, sessionId);
|
|
3097
|
-
await this.persistSession(conversationId, sessionId).catch(() => {
|
|
3989
|
+
await this.persistSession(conversationId, sessionId).catch((err) => {
|
|
3990
|
+
this.log({
|
|
3991
|
+
level: "warn",
|
|
3992
|
+
message: `Persisting the OpenCode session binding ${sessionId.slice(0, 8)} for conversation ${conversationId.slice(0, 8)} failed (best-effort, not retried) \u2014 the completion PATCH also carries opencode_session_id, so the binding is repaired when the turn finishes: ${err instanceof Error ? err.message : String(err)}`,
|
|
3993
|
+
conversation_id: conversationId
|
|
3994
|
+
});
|
|
3098
3995
|
});
|
|
3099
3996
|
return sessionId;
|
|
3100
3997
|
}
|
|
@@ -3286,12 +4183,17 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3286
4183
|
stuckReported: false,
|
|
3287
4184
|
lastAliveAt: 0,
|
|
3288
4185
|
aliveInFlight: false,
|
|
4186
|
+
titleSynced: false,
|
|
4187
|
+
titleSyncInFlight: false,
|
|
3289
4188
|
awaitingHumanLatched: false,
|
|
3290
4189
|
pausedOnQuestion: false,
|
|
3291
4190
|
pausedOnPermission: false,
|
|
3292
4191
|
pausedClearConfirmed: false,
|
|
3293
4192
|
pausedInFlight: false,
|
|
3294
|
-
deliveryDeadlineAnchored: false
|
|
4193
|
+
deliveryDeadlineAnchored: false,
|
|
4194
|
+
b2PinnedSinceMs: 0,
|
|
4195
|
+
b2LastDescendantCheckMs: 0,
|
|
4196
|
+
b2AbandonedSignalled: false
|
|
3295
4197
|
});
|
|
3296
4198
|
}
|
|
3297
4199
|
/**
|
|
@@ -3306,9 +4208,10 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3306
4208
|
* opencode reports ACTIVELY `running` is watched to completion (its liveness
|
|
3307
4209
|
* heartbeat keeps the cron off its row), while a re-adopted turn that is paused
|
|
3308
4210
|
* awaiting a human — or queued/unreachable — is still bounded by `deadline` and
|
|
3309
|
-
* handed to the cron.
|
|
3310
|
-
*
|
|
3311
|
-
*
|
|
4211
|
+
* handed to the cron. Real invariant (#965): the cron MAY reclaim a row this
|
|
4212
|
+
* runner still holds; a reclaimed row that already ran is never re-dispatched
|
|
4213
|
+
* while opencode reports its turn ongoing (readopt's own gate here, and the
|
|
4214
|
+
* `pending`-row re-drive fence, `resolveRedrive`). `dispatchedAt` stays `now`
|
|
3312
4215
|
* (only the appear-guard uses it).
|
|
3313
4216
|
*
|
|
3314
4217
|
* `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
|
|
@@ -3359,12 +4262,17 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3359
4262
|
// with no extra `re_adopted` signal needed (folds old WI-6).
|
|
3360
4263
|
lastAliveAt: 0,
|
|
3361
4264
|
aliveInFlight: false,
|
|
4265
|
+
titleSynced: false,
|
|
4266
|
+
titleSyncInFlight: false,
|
|
3362
4267
|
awaitingHumanLatched: false,
|
|
3363
4268
|
pausedOnQuestion: false,
|
|
3364
4269
|
pausedOnPermission: false,
|
|
3365
4270
|
pausedClearConfirmed: false,
|
|
3366
4271
|
pausedInFlight: false,
|
|
3367
|
-
deliveryDeadlineAnchored: false
|
|
4272
|
+
deliveryDeadlineAnchored: false,
|
|
4273
|
+
b2PinnedSinceMs: 0,
|
|
4274
|
+
b2LastDescendantCheckMs: 0,
|
|
4275
|
+
b2AbandonedSignalled: false
|
|
3368
4276
|
});
|
|
3369
4277
|
}
|
|
3370
4278
|
/**
|
|
@@ -3526,58 +4434,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3526
4434
|
}
|
|
3527
4435
|
}
|
|
3528
4436
|
if (state === "done") {
|
|
3529
|
-
this.
|
|
3530
|
-
if (!inFlight.done) {
|
|
3531
|
-
this.log({
|
|
3532
|
-
level: "info",
|
|
3533
|
-
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
3534
|
-
conversation_id: conv.id,
|
|
3535
|
-
message_id: inFlight.evidentMessageId
|
|
3536
|
-
});
|
|
3537
|
-
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
3538
|
-
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
3539
|
-
try {
|
|
3540
|
-
await this.markDone(
|
|
3541
|
-
conv.id,
|
|
3542
|
-
inFlight.evidentMessageId,
|
|
3543
|
-
sessionId,
|
|
3544
|
-
inFlight.opencodeMessageId,
|
|
3545
|
-
title,
|
|
3546
|
-
usage
|
|
3547
|
-
);
|
|
3548
|
-
} catch (err) {
|
|
3549
|
-
if (err instanceof ChannelAuthError) throw err;
|
|
3550
|
-
if (err instanceof ChannelTerminalError) {
|
|
3551
|
-
this.log({
|
|
3552
|
-
level: "warn",
|
|
3553
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
3554
|
-
conversation_id: conv.id,
|
|
3555
|
-
message_id: inFlight.evidentMessageId
|
|
3556
|
-
});
|
|
3557
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3558
|
-
return;
|
|
3559
|
-
}
|
|
3560
|
-
if (this.now() >= inFlight.deadline) {
|
|
3561
|
-
this.log({
|
|
3562
|
-
level: "warn",
|
|
3563
|
-
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)}`,
|
|
3564
|
-
conversation_id: conv.id,
|
|
3565
|
-
message_id: inFlight.evidentMessageId
|
|
3566
|
-
});
|
|
3567
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3568
|
-
return;
|
|
3569
|
-
}
|
|
3570
|
-
this.log({
|
|
3571
|
-
level: "warn",
|
|
3572
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
3573
|
-
conversation_id: conv.id,
|
|
3574
|
-
message_id: inFlight.evidentMessageId
|
|
3575
|
-
});
|
|
3576
|
-
return;
|
|
3577
|
-
}
|
|
3578
|
-
inFlight.done = true;
|
|
3579
|
-
}
|
|
3580
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4437
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
3581
4438
|
return;
|
|
3582
4439
|
}
|
|
3583
4440
|
if (state === "failed") {
|
|
@@ -3591,8 +4448,16 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3591
4448
|
message_id: inFlight.evidentMessageId
|
|
3592
4449
|
});
|
|
3593
4450
|
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4451
|
+
const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
|
|
3594
4452
|
try {
|
|
3595
|
-
await this.markFailed(
|
|
4453
|
+
await this.markFailed(
|
|
4454
|
+
conv.id,
|
|
4455
|
+
inFlight.evidentMessageId,
|
|
4456
|
+
sessionId,
|
|
4457
|
+
error2,
|
|
4458
|
+
usage,
|
|
4459
|
+
failure
|
|
4460
|
+
);
|
|
3596
4461
|
} catch (err) {
|
|
3597
4462
|
if (err instanceof ChannelAuthError) throw err;
|
|
3598
4463
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -3637,6 +4502,44 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3637
4502
|
});
|
|
3638
4503
|
}
|
|
3639
4504
|
const activelyRunning = state === "running" && !awaitingHuman;
|
|
4505
|
+
const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
|
|
4506
|
+
const snapshotReadable = messages != null && messages.length > 0;
|
|
4507
|
+
if (!pinnedNow) {
|
|
4508
|
+
if (snapshotReadable) {
|
|
4509
|
+
inFlight.b2PinnedSinceMs = 0;
|
|
4510
|
+
inFlight.b2LastDescendantCheckMs = 0;
|
|
4511
|
+
inFlight.b2AbandonedSignalled = false;
|
|
4512
|
+
}
|
|
4513
|
+
} else {
|
|
4514
|
+
if (inFlight.b2AbandonedSignalled) {
|
|
4515
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4516
|
+
return;
|
|
4517
|
+
}
|
|
4518
|
+
if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
|
|
4519
|
+
const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
|
|
4520
|
+
if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
|
|
4521
|
+
inFlight.b2LastDescendantCheckMs = this.now();
|
|
4522
|
+
const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
|
|
4523
|
+
if (isB2AbandonmentConfirmed({
|
|
4524
|
+
pinnedForMs,
|
|
4525
|
+
minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
|
|
4526
|
+
descendantOngoing
|
|
4527
|
+
})) {
|
|
4528
|
+
inFlight.b2AbandonedSignalled = true;
|
|
4529
|
+
this.log({
|
|
4530
|
+
level: "warn",
|
|
4531
|
+
message: `Message ${id.slice(0, 8)} b2-pinned for ${Math.round(pinnedForMs / 1e3)}s with no ongoing descendant sub-agent session (status-map confirmed) \u2014 treating the delegated/tool turn as abandoned, resolving done`,
|
|
4532
|
+
conversation_id: conv.id,
|
|
4533
|
+
message_id: id
|
|
4534
|
+
});
|
|
4535
|
+
void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
|
|
4536
|
+
watched_for_ms: pinnedForMs
|
|
4537
|
+
});
|
|
4538
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4539
|
+
return;
|
|
4540
|
+
}
|
|
4541
|
+
}
|
|
4542
|
+
}
|
|
3640
4543
|
if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
|
|
3641
4544
|
this.log({
|
|
3642
4545
|
level: "warn",
|
|
@@ -3656,6 +4559,18 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3656
4559
|
inFlight.aliveInFlight = false;
|
|
3657
4560
|
if (ok) inFlight.lastAliveAt = this.now();
|
|
3658
4561
|
});
|
|
4562
|
+
if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
|
|
4563
|
+
inFlight.titleSyncInFlight = true;
|
|
4564
|
+
void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
|
|
4565
|
+
if (!title) {
|
|
4566
|
+
inFlight.titleSyncInFlight = false;
|
|
4567
|
+
return;
|
|
4568
|
+
}
|
|
4569
|
+
const ok = await this.patchConversationTitle(conv.id, title);
|
|
4570
|
+
inFlight.titleSyncInFlight = false;
|
|
4571
|
+
if (ok) inFlight.titleSynced = true;
|
|
4572
|
+
});
|
|
4573
|
+
}
|
|
3659
4574
|
}
|
|
3660
4575
|
if (awaitingHuman) {
|
|
3661
4576
|
if (!inFlight.awaitingHumanLatched) {
|
|
@@ -3693,6 +4608,70 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3693
4608
|
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3694
4609
|
}
|
|
3695
4610
|
}
|
|
4611
|
+
/**
|
|
4612
|
+
* Settle a message whose run-state has resolved `'done'` — extracted verbatim
|
|
4613
|
+
* (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
|
|
4614
|
+
* inline `state === 'done'` branch body, so a SECOND caller (the #721
|
|
4615
|
+
* b2-abandonment resolution) can reach the exact same completion behavior
|
|
4616
|
+
* (delivery-deadline anchoring, title resolution, usage extraction, and
|
|
4617
|
+
* `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
|
|
4618
|
+
* and risking the two copies silently drifting apart.
|
|
4619
|
+
*/
|
|
4620
|
+
async settleMessageDone(sessionId, watcher, inFlight, messages) {
|
|
4621
|
+
const conv = watcher.conv;
|
|
4622
|
+
this.anchorDeliveryDeadline(inFlight);
|
|
4623
|
+
if (!inFlight.done) {
|
|
4624
|
+
this.log({
|
|
4625
|
+
level: "info",
|
|
4626
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
4627
|
+
conversation_id: conv.id,
|
|
4628
|
+
message_id: inFlight.evidentMessageId
|
|
4629
|
+
});
|
|
4630
|
+
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
4631
|
+
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4632
|
+
try {
|
|
4633
|
+
await this.markDone(
|
|
4634
|
+
conv.id,
|
|
4635
|
+
inFlight.evidentMessageId,
|
|
4636
|
+
sessionId,
|
|
4637
|
+
inFlight.opencodeMessageId,
|
|
4638
|
+
title,
|
|
4639
|
+
usage
|
|
4640
|
+
);
|
|
4641
|
+
} catch (err) {
|
|
4642
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
4643
|
+
if (err instanceof ChannelTerminalError) {
|
|
4644
|
+
this.log({
|
|
4645
|
+
level: "warn",
|
|
4646
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
4647
|
+
conversation_id: conv.id,
|
|
4648
|
+
message_id: inFlight.evidentMessageId
|
|
4649
|
+
});
|
|
4650
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4651
|
+
return;
|
|
4652
|
+
}
|
|
4653
|
+
if (this.now() >= inFlight.deadline) {
|
|
4654
|
+
this.log({
|
|
4655
|
+
level: "warn",
|
|
4656
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
|
|
4657
|
+
conversation_id: conv.id,
|
|
4658
|
+
message_id: inFlight.evidentMessageId
|
|
4659
|
+
});
|
|
4660
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4661
|
+
return;
|
|
4662
|
+
}
|
|
4663
|
+
this.log({
|
|
4664
|
+
level: "warn",
|
|
4665
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
4666
|
+
conversation_id: conv.id,
|
|
4667
|
+
message_id: inFlight.evidentMessageId
|
|
4668
|
+
});
|
|
4669
|
+
return;
|
|
4670
|
+
}
|
|
4671
|
+
inFlight.done = true;
|
|
4672
|
+
}
|
|
4673
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4674
|
+
}
|
|
3696
4675
|
// Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
|
|
3697
4676
|
/**
|
|
3698
4677
|
* Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
|
|
@@ -3859,6 +4838,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3859
4838
|
if (state === "failed") {
|
|
3860
4839
|
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3861
4840
|
const usage = messageUsage(messages, ocId ?? "");
|
|
4841
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3862
4842
|
this.log({
|
|
3863
4843
|
level: "error",
|
|
3864
4844
|
message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
|
|
@@ -3866,7 +4846,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3866
4846
|
message_id: row.id
|
|
3867
4847
|
});
|
|
3868
4848
|
try {
|
|
3869
|
-
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
|
|
4849
|
+
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
|
|
3870
4850
|
} catch (err) {
|
|
3871
4851
|
if (err instanceof ChannelAuthError) throw err;
|
|
3872
4852
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -4133,7 +5113,8 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4133
5113
|
opencode_model: row.opencode_model,
|
|
4134
5114
|
source_message_id: row.source_message_id,
|
|
4135
5115
|
slack_user_id: row.slack_user_id,
|
|
4136
|
-
attachments: row.attachments ?? null
|
|
5116
|
+
attachments: row.attachments ?? null,
|
|
5117
|
+
opencode_message_id: row.opencode_message_id
|
|
4137
5118
|
};
|
|
4138
5119
|
}
|
|
4139
5120
|
/**
|
|
@@ -4272,6 +5253,47 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4272
5253
|
}
|
|
4273
5254
|
return false;
|
|
4274
5255
|
}
|
|
5256
|
+
/**
|
|
5257
|
+
* Tri-state variant of the upward parentID membership walk (#721), used ONLY
|
|
5258
|
+
* by `isAnyDescendantSessionOngoing`. Walks the SAME cached
|
|
5259
|
+
* `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
|
|
5260
|
+
* `sessionBelongsTo`, which deliberately collapses "confirmed not a
|
|
5261
|
+
* descendant" and "the walk's fetch failed" into the same `false` (safe for
|
|
5262
|
+
* its OTHER callers: interaction attribution and the recovery-path
|
|
5263
|
+
* `isAnyDescendantSessionAlive`, both of which just retry next tick with no
|
|
5264
|
+
* safety consequence either way) — this variant keeps those two outcomes
|
|
5265
|
+
* SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
|
|
5266
|
+
* (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
|
|
5267
|
+
* not ongoing".
|
|
5268
|
+
*
|
|
5269
|
+
* Return contract:
|
|
5270
|
+
* - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
|
|
5271
|
+
* - `false` → the walk reached a definitive, parent-less root session
|
|
5272
|
+
* WITHOUT ever matching `rootSessionId` — `sessionId` is
|
|
5273
|
+
* CONFIRMED NOT a descendant of it.
|
|
5274
|
+
* - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
|
|
5275
|
+
* through the walk (`resolveSessionParent` returned `undefined`),
|
|
5276
|
+
* or the depth cap (32) was hit without a definitive answer (a
|
|
5277
|
+
* pathological/cyclic chain proves nothing either way). NEVER
|
|
5278
|
+
* treat this the same as `false` — see `sessionBelongsTo`'s own
|
|
5279
|
+
* doc comment above for why that collapse is safe THERE but not
|
|
5280
|
+
* here.
|
|
5281
|
+
*
|
|
5282
|
+
* `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
|
|
5283
|
+
* to the live-path descendant check, not a modification of shared code used
|
|
5284
|
+
* by interaction attribution or the recovery path.
|
|
5285
|
+
*/
|
|
5286
|
+
async resolveSessionMembership(sessionId, rootSessionId) {
|
|
5287
|
+
let current = sessionId;
|
|
5288
|
+
for (let depth = 0; current && depth < 32; depth++) {
|
|
5289
|
+
if (current === rootSessionId) return true;
|
|
5290
|
+
const parent = await this.resolveSessionParent(current);
|
|
5291
|
+
if (parent === void 0) return null;
|
|
5292
|
+
if (parent === null) return false;
|
|
5293
|
+
current = parent;
|
|
5294
|
+
}
|
|
5295
|
+
return null;
|
|
5296
|
+
}
|
|
4275
5297
|
/**
|
|
4276
5298
|
* Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
|
|
4277
5299
|
* `null` for a root session (no parent) and `undefined` when opencode is
|
|
@@ -4356,6 +5378,54 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4356
5378
|
}
|
|
4357
5379
|
return null;
|
|
4358
5380
|
}
|
|
5381
|
+
/**
|
|
5382
|
+
* Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
|
|
5383
|
+
* session title onto the conversation via the PLAIN conversation-update
|
|
5384
|
+
* endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
|
|
5385
|
+
* message-status endpoint `markProcessing`/`markDone` use. Deliberately a
|
|
5386
|
+
* separate, lighter call: it carries no `status`, so it cannot re-trigger the
|
|
5387
|
+
* `processing`/`done` transition side effects (Slack notices, activity-log
|
|
5388
|
+
* rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
|
|
5389
|
+
* ever touches `conversations.title`. That route (`routes/conversations.ts`)
|
|
5390
|
+
* skips a title write matching the stored value, so a redundant call with the
|
|
5391
|
+
* same title is a real no-op — it does not bump `updated_at`, which the
|
|
5392
|
+
* conversation list sorts and paginates on. (Note this is a DIFFERENT guard
|
|
5393
|
+
* from `threads.ts`'s "non-empty AND changed" one, which only covers the
|
|
5394
|
+
* message-status PATCH; the non-empty half is enforced here instead, by
|
|
5395
|
+
* `resolveSessionTitle` never returning an empty/placeholder title.)
|
|
5396
|
+
*
|
|
5397
|
+
* Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
|
|
5398
|
+
* is logged and the title is simply retried on the next heartbeat tick (the
|
|
5399
|
+
* caller only latches `titleSynced` on `true`).
|
|
5400
|
+
*/
|
|
5401
|
+
async patchConversationTitle(conversationId, title) {
|
|
5402
|
+
try {
|
|
5403
|
+
const res = await this.fetchImpl(
|
|
5404
|
+
`${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
|
|
5405
|
+
{
|
|
5406
|
+
method: "PATCH",
|
|
5407
|
+
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
5408
|
+
body: JSON.stringify({ title })
|
|
5409
|
+
}
|
|
5410
|
+
);
|
|
5411
|
+
if (!res.ok) {
|
|
5412
|
+
this.log({
|
|
5413
|
+
level: "debug",
|
|
5414
|
+
message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
|
|
5415
|
+
conversation_id: conversationId
|
|
5416
|
+
});
|
|
5417
|
+
return false;
|
|
5418
|
+
}
|
|
5419
|
+
return true;
|
|
5420
|
+
} catch (err) {
|
|
5421
|
+
this.log({
|
|
5422
|
+
level: "debug",
|
|
5423
|
+
message: `Best-effort mid-turn title sync PATCH failed for conversation ${conversationId.slice(0, 8)} (will retry next heartbeat): ${err instanceof Error ? err.message : String(err)}`,
|
|
5424
|
+
conversation_id: conversationId
|
|
5425
|
+
});
|
|
5426
|
+
return false;
|
|
5427
|
+
}
|
|
5428
|
+
}
|
|
4359
5429
|
/**
|
|
4360
5430
|
* DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
|
|
4361
5431
|
* (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
|
|
@@ -4388,31 +5458,109 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4388
5458
|
* longer exists". The inversion lives in the caller; this method just reports
|
|
4389
5459
|
* true/false/null faithfully.
|
|
4390
5460
|
*
|
|
4391
|
-
* VERIFY-BEFORE-DEPEND: we depend ONLY on (a) `parentID` from `GET /session/:id`
|
|
4392
|
-
* (already proven by the existing child-session interaction tests, via
|
|
4393
|
-
* `resolveSessionParent`/`sessionBelongsTo`) and (b) the child's own message-list
|
|
4394
|
-
* terminal state. We do NOT depend on any session-level `busy`/`idle` field —
|
|
4395
|
-
* there is none on `GET /session/:id`; OpenCode's busy state is in-memory
|
|
4396
|
-
* `SessionStatus` only.
|
|
5461
|
+
* VERIFY-BEFORE-DEPEND: we depend ONLY on (a) `parentID` from `GET /session/:id`
|
|
5462
|
+
* (already proven by the existing child-session interaction tests, via
|
|
5463
|
+
* `resolveSessionParent`/`sessionBelongsTo`) and (b) the child's own message-list
|
|
5464
|
+
* terminal state. We do NOT depend on any session-level `busy`/`idle` field —
|
|
5465
|
+
* there is none on `GET /session/:id`; OpenCode's busy state is in-memory
|
|
5466
|
+
* `SessionStatus` only.
|
|
5467
|
+
*/
|
|
5468
|
+
async isAnyDescendantSessionAlive(rootSessionId) {
|
|
5469
|
+
const sessions = await listSessions(this.port);
|
|
5470
|
+
if (!sessions) {
|
|
5471
|
+
this.log({
|
|
5472
|
+
level: "warn",
|
|
5473
|
+
message: `Re-adopt: could not enumerate sessions to cross-check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating child liveness as indeterminate`
|
|
5474
|
+
});
|
|
5475
|
+
return null;
|
|
5476
|
+
}
|
|
5477
|
+
for (const candidate of sessions) {
|
|
5478
|
+
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
5479
|
+
if (!await this.sessionBelongsTo(candidate.id, rootSessionId)) continue;
|
|
5480
|
+
const childMsgs = await getSessionMessages(this.port, candidate.id);
|
|
5481
|
+
if (isSessionActivelyGenerating(childMsgs)) {
|
|
5482
|
+
return true;
|
|
5483
|
+
}
|
|
5484
|
+
}
|
|
5485
|
+
return false;
|
|
5486
|
+
}
|
|
5487
|
+
/**
|
|
5488
|
+
* LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
|
|
5489
|
+
* sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
|
|
5490
|
+
* in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
|
|
5491
|
+
*
|
|
5492
|
+
* Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
|
|
5493
|
+
* cross-check above): that method judges liveness from the child's OWN
|
|
5494
|
+
* TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
|
|
5495
|
+
* on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
|
|
5496
|
+
* path the local opencode server IS running, so its in-memory status map is
|
|
5497
|
+
* live and authoritative — and per ADR-0047 §4a ("the child has its own entry
|
|
5498
|
+
* [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
|
|
5499
|
+
* ENTIRE turn (including any tool call it is itself executing), not a
|
|
5500
|
+
* per-message transcript snapshot. This sidesteps the "child's own tool is
|
|
5501
|
+
* executing, between its step's completion and the next generation step"
|
|
5502
|
+
* transcript gap that a transcript-based check would need a second,
|
|
5503
|
+
* sustained-window bound to guard against — it is simply not derived from
|
|
5504
|
+
* message timestamps at all.
|
|
5505
|
+
*
|
|
5506
|
+
* Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
|
|
5507
|
+
* status, as the recovery path does per §4a)? Because on the LIVE path the
|
|
5508
|
+
* root session can be shared: a SECOND, unrelated user message can land on the
|
|
5509
|
+
* SAME session (issue #721's own root cause) and keep the root `busy` for a
|
|
5510
|
+
* reason that has nothing to do with THIS message's delegation. A `task`
|
|
5511
|
+
* descendant session is spawned for exactly one delegated turn and never
|
|
5512
|
+
* reused, so its OWN status-map entry is unambiguous evidence about that one
|
|
5513
|
+
* delegation — which the root's status is not.
|
|
5514
|
+
*
|
|
5515
|
+
* Why membership is checked via `resolveSessionMembership`, NOT
|
|
5516
|
+
* `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
|
|
5517
|
+
* `GET /session/:id` fetch failure into "not a descendant", which would
|
|
5518
|
+
* silently drop a genuinely-live candidate from consideration on the one
|
|
5519
|
+
* unlucky tick its membership-walk fetch hiccups (#721).
|
|
5520
|
+
* `resolveSessionMembership` keeps that failure mode as a distinct `null`
|
|
5521
|
+
* (indeterminate) so it is folded into THIS method's own `indeterminate` flag
|
|
5522
|
+
* instead.
|
|
5523
|
+
*
|
|
5524
|
+
* Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
|
|
5525
|
+
* - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
|
|
5526
|
+
* - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
|
|
5527
|
+
* confirmed either way (`resolveSessionMembership` never
|
|
5528
|
+
* returned `null`), and every CONFIRMED descendant's status read
|
|
5529
|
+
* succeeded and is not ongoing (includes "no descendant session
|
|
5530
|
+
* exists at all" — e.g. a plain, non-`task` tool call).
|
|
5531
|
+
* - `null` → INDETERMINATE: `listSessions` failed, OR at least one
|
|
5532
|
+
* candidate's MEMBERSHIP could not be confirmed
|
|
5533
|
+
* (`resolveSessionMembership` returned `null` — a fetch failure
|
|
5534
|
+
* or pathological chain partway through the parent walk), OR at
|
|
5535
|
+
* least one CONFIRMED descendant's `isSessionOngoing` read
|
|
5536
|
+
* failed — and no OTHER candidate was already confirmed `true`.
|
|
5537
|
+
* The caller MUST NOT treat `null` the same as `false` here
|
|
5538
|
+
* (unlike the recovery cross-check's contract) — see
|
|
5539
|
+
* `isB2AbandonmentConfirmed`.
|
|
4397
5540
|
*/
|
|
4398
|
-
async
|
|
5541
|
+
async isAnyDescendantSessionOngoing(rootSessionId) {
|
|
4399
5542
|
const sessions = await listSessions(this.port);
|
|
4400
5543
|
if (!sessions) {
|
|
4401
5544
|
this.log({
|
|
4402
5545
|
level: "warn",
|
|
4403
|
-
message: `
|
|
5546
|
+
message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
|
|
4404
5547
|
});
|
|
4405
5548
|
return null;
|
|
4406
5549
|
}
|
|
5550
|
+
let indeterminate = false;
|
|
4407
5551
|
for (const candidate of sessions) {
|
|
4408
5552
|
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
4409
|
-
|
|
4410
|
-
|
|
4411
|
-
|
|
4412
|
-
|
|
5553
|
+
const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
|
|
5554
|
+
if (membership === null) {
|
|
5555
|
+
indeterminate = true;
|
|
5556
|
+
continue;
|
|
4413
5557
|
}
|
|
5558
|
+
if (membership === false) continue;
|
|
5559
|
+
const ongoing = await isSessionOngoing(this.port, candidate.id);
|
|
5560
|
+
if (ongoing === true) return true;
|
|
5561
|
+
if (ongoing === null) indeterminate = true;
|
|
4414
5562
|
}
|
|
4415
|
-
return false;
|
|
5563
|
+
return indeterminate ? null : false;
|
|
4416
5564
|
}
|
|
4417
5565
|
/**
|
|
4418
5566
|
* Cheap decision-telemetry label for a running row's LAST correlated reply
|
|
@@ -4558,11 +5706,11 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4558
5706
|
* `ensureSession` guard, not this suppression, is what makes the self-heal
|
|
4559
5707
|
* stick.
|
|
4560
5708
|
*/
|
|
4561
|
-
sessionIdBody(sessionId, conversationId, messageId,
|
|
5709
|
+
sessionIdBody(sessionId, conversationId, messageId, status2) {
|
|
4562
5710
|
if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
|
|
4563
5711
|
this.log({
|
|
4564
5712
|
level: "debug",
|
|
4565
|
-
message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${
|
|
5713
|
+
message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${status2}' update for message ${messageId.slice(0, 8)} so it is not re-bound to conversation ${conversationId.slice(0, 8)}`,
|
|
4566
5714
|
conversation_id: conversationId,
|
|
4567
5715
|
message_id: messageId
|
|
4568
5716
|
});
|
|
@@ -4677,7 +5825,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4677
5825
|
* exists but is wedged, so the next attempt must get a fresh one
|
|
4678
5826
|
* instead of reusing it — see WI-1's server-side null-clearing PATCH).
|
|
4679
5827
|
*/
|
|
4680
|
-
async markFailed(conversationId, messageId, sessionId, error2, usage) {
|
|
5828
|
+
async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
|
|
4681
5829
|
const body = { status: "failed" };
|
|
4682
5830
|
if (sessionId === null) {
|
|
4683
5831
|
body.opencode_session_id = null;
|
|
@@ -4686,6 +5834,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4686
5834
|
}
|
|
4687
5835
|
if (error2 !== void 0) body.error = error2;
|
|
4688
5836
|
if (usage) Object.assign(body, usage);
|
|
5837
|
+
if (failure) {
|
|
5838
|
+
body.failure_kind = failure.kind;
|
|
5839
|
+
body.failure_provider_id = failure.providerId;
|
|
5840
|
+
body.failure_model_id = failure.modelId;
|
|
5841
|
+
body.failure_reason = failure.reason;
|
|
5842
|
+
}
|
|
4689
5843
|
await this.callWithRetry(
|
|
4690
5844
|
"marking message as failed",
|
|
4691
5845
|
() => this.fetchImpl(
|
|
@@ -4698,6 +5852,29 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4698
5852
|
)
|
|
4699
5853
|
);
|
|
4700
5854
|
}
|
|
5855
|
+
/**
|
|
5856
|
+
* Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
|
|
5857
|
+
*
|
|
5858
|
+
* `messageFailure` alone (structured OpenCode error → `model_auth`) covers
|
|
5859
|
+
* most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
|
|
5860
|
+
* to the P1-2b zero-provider check — one extra loopback call to
|
|
5861
|
+
* `hasAnyConfiguredProvider`, only reached when the structured classifier
|
|
5862
|
+
* couldn't place it. Fails open (never throws): a fallback probe failure
|
|
5863
|
+
* (`null`/indeterminate) leaves the classification `null`, which produces
|
|
5864
|
+
* today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
|
|
5865
|
+
*/
|
|
5866
|
+
async classifyModelAuthFailure(messages, userMessageId) {
|
|
5867
|
+
const classified = messageFailure(messages, userMessageId);
|
|
5868
|
+
if (classified != null) return classified;
|
|
5869
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
5870
|
+
const hasProvider = await hasAnyConfiguredProvider(this.port);
|
|
5871
|
+
return applyZeroProviderFallback(
|
|
5872
|
+
classified,
|
|
5873
|
+
hasProvider,
|
|
5874
|
+
reply?.info?.providerID ?? null,
|
|
5875
|
+
reply?.info?.modelID ?? null
|
|
5876
|
+
);
|
|
5877
|
+
}
|
|
4701
5878
|
/**
|
|
4702
5879
|
* Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
|
|
4703
5880
|
* (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
|
|
@@ -4850,10 +6027,16 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4850
6027
|
import chalk5 from "chalk";
|
|
4851
6028
|
import ora2 from "ora";
|
|
4852
6029
|
import { select as select2 } from "@inquirer/prompts";
|
|
6030
|
+
var INTERACTIVE_START_TIMEOUT_MS = 3e4;
|
|
4853
6031
|
async function ensureOpenCodeRunning(ctx) {
|
|
4854
6032
|
const healthCheck = await checkOpenCodeHealth(ctx.port);
|
|
4855
6033
|
if (healthCheck.healthy) {
|
|
4856
|
-
return {
|
|
6034
|
+
return {
|
|
6035
|
+
port: ctx.port,
|
|
6036
|
+
process: null,
|
|
6037
|
+
version: healthCheck.version ?? null,
|
|
6038
|
+
notReadyReason: null
|
|
6039
|
+
};
|
|
4857
6040
|
}
|
|
4858
6041
|
const runningInstances = await findHealthyOpenCodeInstances();
|
|
4859
6042
|
if (runningInstances.length > 0) {
|
|
@@ -4874,7 +6057,7 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4874
6057
|
console.log(chalk5.yellow("Tip: Run with the correct port:"));
|
|
4875
6058
|
console.log(
|
|
4876
6059
|
chalk5.dim(
|
|
4877
|
-
` ${getCliName()} run --
|
|
6060
|
+
` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
|
|
4878
6061
|
)
|
|
4879
6062
|
);
|
|
4880
6063
|
}
|
|
@@ -4894,14 +6077,22 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4894
6077
|
if (!ctx.interactive) {
|
|
4895
6078
|
ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
|
|
4896
6079
|
const proc = await startOpenCode(ctx.port);
|
|
4897
|
-
const health = await waitForOpenCodeHealth(ctx.port,
|
|
6080
|
+
const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
|
|
4898
6081
|
if (!health.healthy) {
|
|
4899
|
-
|
|
4900
|
-
|
|
4901
|
-
|
|
6082
|
+
return {
|
|
6083
|
+
port: ctx.port,
|
|
6084
|
+
process: proc,
|
|
6085
|
+
version: null,
|
|
6086
|
+
notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
|
|
6087
|
+
};
|
|
4902
6088
|
}
|
|
4903
6089
|
ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
|
|
4904
|
-
return {
|
|
6090
|
+
return {
|
|
6091
|
+
port: ctx.port,
|
|
6092
|
+
process: proc,
|
|
6093
|
+
version: health.version ?? null,
|
|
6094
|
+
notReadyReason: null
|
|
6095
|
+
};
|
|
4905
6096
|
}
|
|
4906
6097
|
let port = ctx.port;
|
|
4907
6098
|
if (isPortInUse(port)) {
|
|
@@ -4954,122 +6145,15 @@ Port ${port} is already in use.`));
|
|
|
4954
6145
|
if (action === "start") {
|
|
4955
6146
|
const spinner = ora2("Starting OpenCode...").start();
|
|
4956
6147
|
const proc = await startOpenCode(port);
|
|
4957
|
-
const health = await waitForOpenCodeHealth(port,
|
|
6148
|
+
const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
|
|
4958
6149
|
if (!health.healthy) {
|
|
4959
6150
|
spinner.fail("Failed to start OpenCode");
|
|
4960
6151
|
throw new Error("OpenCode failed to start");
|
|
4961
6152
|
}
|
|
4962
6153
|
spinner.stop();
|
|
4963
|
-
return { port, process: proc, version: health.version ?? null };
|
|
4964
|
-
}
|
|
4965
|
-
return { port, process: null, version: null };
|
|
4966
|
-
}
|
|
4967
|
-
|
|
4968
|
-
// src/commands/agent-lookup.ts
|
|
4969
|
-
async function readErrorMessage(response) {
|
|
4970
|
-
const text = await response.text().catch(() => "");
|
|
4971
|
-
if (!text) return response.statusText || void 0;
|
|
4972
|
-
try {
|
|
4973
|
-
const data = JSON.parse(text);
|
|
4974
|
-
const message = data.message ?? data.error;
|
|
4975
|
-
if (typeof message === "string" && message.trim()) {
|
|
4976
|
-
return message;
|
|
4977
|
-
}
|
|
4978
|
-
} catch {
|
|
4979
|
-
}
|
|
4980
|
-
return text.trim() || response.statusText || void 0;
|
|
4981
|
-
}
|
|
4982
|
-
function authFailureHint(apiUrl, serverMessage) {
|
|
4983
|
-
const reason = serverMessage ? `: ${serverMessage}` : "";
|
|
4984
|
-
return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
|
|
4985
|
-
}
|
|
4986
|
-
async function resolveAgentIdFromKey(authHeader) {
|
|
4987
|
-
const apiUrl = getApiUrlConfig();
|
|
4988
|
-
try {
|
|
4989
|
-
const response = await fetch(`${apiUrl}/me`, {
|
|
4990
|
-
headers: { Authorization: authHeader }
|
|
4991
|
-
});
|
|
4992
|
-
if (response.status === 401) {
|
|
4993
|
-
const serverMessage = await readErrorMessage(response);
|
|
4994
|
-
return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
4995
|
-
}
|
|
4996
|
-
if (!response.ok) {
|
|
4997
|
-
const serverMessage = await readErrorMessage(response);
|
|
4998
|
-
return {
|
|
4999
|
-
error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
5000
|
-
};
|
|
5001
|
-
}
|
|
5002
|
-
const data = await response.json();
|
|
5003
|
-
if (data.auth_type === "agent_key" && data.agent_id) {
|
|
5004
|
-
return { agent_id: data.agent_id };
|
|
5005
|
-
}
|
|
5006
|
-
return {
|
|
5007
|
-
error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
|
|
5008
|
-
};
|
|
5009
|
-
} catch (error2) {
|
|
5010
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
5011
|
-
return { error: `Failed to resolve runner from key: ${message}` };
|
|
5012
|
-
}
|
|
5013
|
-
}
|
|
5014
|
-
async function notifyAgentDisconnected(agentId, authHeader) {
|
|
5015
|
-
const apiUrl = getApiUrlConfig();
|
|
5016
|
-
try {
|
|
5017
|
-
const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
|
|
5018
|
-
method: "POST",
|
|
5019
|
-
headers: { Authorization: authHeader }
|
|
5020
|
-
});
|
|
5021
|
-
if (!response.ok) {
|
|
5022
|
-
const serverMessage = await readErrorMessage(response);
|
|
5023
|
-
return {
|
|
5024
|
-
ok: false,
|
|
5025
|
-
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
5026
|
-
};
|
|
5027
|
-
}
|
|
5028
|
-
return { ok: true };
|
|
5029
|
-
} catch (error2) {
|
|
5030
|
-
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
5031
|
-
}
|
|
5032
|
-
}
|
|
5033
|
-
async function getAgentInfo(agentId, authHeader) {
|
|
5034
|
-
const apiUrl = getApiUrlConfig();
|
|
5035
|
-
try {
|
|
5036
|
-
const response = await fetch(`${apiUrl}/runners/${agentId}`, {
|
|
5037
|
-
headers: { Authorization: authHeader }
|
|
5038
|
-
});
|
|
5039
|
-
if (response.status === 401) {
|
|
5040
|
-
const serverMessage = await readErrorMessage(response);
|
|
5041
|
-
return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
5042
|
-
}
|
|
5043
|
-
if (response.status === 403) {
|
|
5044
|
-
const serverMessage = await readErrorMessage(response);
|
|
5045
|
-
return {
|
|
5046
|
-
valid: false,
|
|
5047
|
-
error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
|
|
5048
|
-
};
|
|
5049
|
-
}
|
|
5050
|
-
if (response.status === 404) {
|
|
5051
|
-
const serverMessage = await readErrorMessage(response);
|
|
5052
|
-
return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
|
|
5053
|
-
}
|
|
5054
|
-
if (!response.ok) {
|
|
5055
|
-
const serverMessage = await readErrorMessage(response);
|
|
5056
|
-
return {
|
|
5057
|
-
valid: false,
|
|
5058
|
-
error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
5059
|
-
};
|
|
5060
|
-
}
|
|
5061
|
-
const agent = await response.json();
|
|
5062
|
-
if (agent.agent_type !== "local") {
|
|
5063
|
-
return {
|
|
5064
|
-
valid: false,
|
|
5065
|
-
error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
|
|
5066
|
-
};
|
|
5067
|
-
}
|
|
5068
|
-
return { valid: true, agent };
|
|
5069
|
-
} catch (error2) {
|
|
5070
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
5071
|
-
return { valid: false, error: `Failed to validate runner: ${message}` };
|
|
6154
|
+
return { port, process: proc, version: health.version ?? null, notReadyReason: null };
|
|
5072
6155
|
}
|
|
6156
|
+
return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
|
|
5073
6157
|
}
|
|
5074
6158
|
|
|
5075
6159
|
// src/commands/run.ts
|
|
@@ -5077,6 +6161,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
|
|
|
5077
6161
|
var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
|
|
5078
6162
|
var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
|
|
5079
6163
|
var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
|
|
6164
|
+
var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
|
|
5080
6165
|
function resolveLogLevel(options) {
|
|
5081
6166
|
const accepted = Object.keys(LOG_LEVELS);
|
|
5082
6167
|
const validate = (value, source) => {
|
|
@@ -5128,6 +6213,35 @@ function resolveFileSyncDirectories(raw, homeDir) {
|
|
|
5128
6213
|
}
|
|
5129
6214
|
return directories;
|
|
5130
6215
|
}
|
|
6216
|
+
var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
|
|
6217
|
+
var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
|
|
6218
|
+
var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
|
|
6219
|
+
function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
|
|
6220
|
+
const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
|
|
6221
|
+
let raw;
|
|
6222
|
+
let source;
|
|
6223
|
+
if (options.opencodeStartTimeout !== void 0) {
|
|
6224
|
+
raw = options.opencodeStartTimeout;
|
|
6225
|
+
source = "--opencode-start-timeout";
|
|
6226
|
+
} else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
|
|
6227
|
+
raw = env[OPENCODE_START_TIMEOUT_ENV];
|
|
6228
|
+
source = OPENCODE_START_TIMEOUT_ENV;
|
|
6229
|
+
} else {
|
|
6230
|
+
return { timeoutMs: defaultMs, warnings: [] };
|
|
6231
|
+
}
|
|
6232
|
+
const trimmed = raw.trim();
|
|
6233
|
+
const seconds = Number(trimmed);
|
|
6234
|
+
const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
|
|
6235
|
+
if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
|
|
6236
|
+
return {
|
|
6237
|
+
timeoutMs: defaultMs,
|
|
6238
|
+
warnings: [
|
|
6239
|
+
`Ignoring invalid ${source} "${raw}": expected a positive integer number of seconds (at most ${MAX_OPENCODE_START_TIMEOUT_SECONDS}); using the default ${DEFAULT_OPENCODE_START_TIMEOUT_SECONDS}s`
|
|
6240
|
+
]
|
|
6241
|
+
};
|
|
6242
|
+
}
|
|
6243
|
+
return { timeoutMs: seconds * 1e3, warnings: [] };
|
|
6244
|
+
}
|
|
5131
6245
|
function meetsThreshold(state, level) {
|
|
5132
6246
|
return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
|
|
5133
6247
|
}
|
|
@@ -5149,6 +6263,10 @@ function log2(state, message, level = "info") {
|
|
|
5149
6263
|
function logActivity(state, entry) {
|
|
5150
6264
|
const level = entry.level ?? (entry.type === "error" ? "error" : "info");
|
|
5151
6265
|
if (!meetsThreshold(state, level)) return;
|
|
6266
|
+
forwardRunnerActivity(
|
|
6267
|
+
{ level, message: entry.message, error: entry.error },
|
|
6268
|
+
{ agentId: state.agentId, authHeader: state.authHeader }
|
|
6269
|
+
);
|
|
5152
6270
|
const fullEntry = {
|
|
5153
6271
|
...entry,
|
|
5154
6272
|
level,
|
|
@@ -5248,9 +6366,15 @@ async function handleAuthError(state, error2) {
|
|
|
5248
6366
|
}
|
|
5249
6367
|
async function driveChannels(state, driver) {
|
|
5250
6368
|
let idlePolls = 0;
|
|
6369
|
+
let idleMs = 0;
|
|
6370
|
+
let consecutiveDrainFailures = 0;
|
|
6371
|
+
let unreachableMs = 0;
|
|
5251
6372
|
let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
5252
6373
|
let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
5253
6374
|
while (state.running) {
|
|
6375
|
+
const cycleStartedAtMs = performance.now();
|
|
6376
|
+
let idleThisCycle = false;
|
|
6377
|
+
let unreachableThisCycle = false;
|
|
5254
6378
|
if (state.connection?.reconnecting && state.connection.reconnectPromise) {
|
|
5255
6379
|
logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
|
|
5256
6380
|
if (state.interactive) displayStatus(state);
|
|
@@ -5265,17 +6389,23 @@ async function driveChannels(state, driver) {
|
|
|
5265
6389
|
);
|
|
5266
6390
|
try {
|
|
5267
6391
|
const processed = await driver.drainPending();
|
|
6392
|
+
consecutiveDrainFailures = 0;
|
|
6393
|
+
unreachableMs = 0;
|
|
5268
6394
|
state.messageCount += processed;
|
|
5269
6395
|
const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
|
|
5270
6396
|
lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
5271
6397
|
const appliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
5272
|
-
const
|
|
6398
|
+
const filesApplied = appliedFiles !== lastSeenAppliedFiles;
|
|
6399
|
+
const fileActivity = carriedOverFileSync || filesApplied;
|
|
5273
6400
|
lastSeenAppliedFiles = appliedFiles;
|
|
6401
|
+
if (filesApplied) state.claudeUsageRearm?.();
|
|
5274
6402
|
if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
|
|
5275
6403
|
idlePolls = 0;
|
|
6404
|
+
idleMs = 0;
|
|
5276
6405
|
if (processed > 0 && state.interactive) displayStatus(state);
|
|
5277
6406
|
} else if (state.idleTimeout !== null) {
|
|
5278
6407
|
idlePolls++;
|
|
6408
|
+
idleThisCycle = true;
|
|
5279
6409
|
if (idlePolls === 1) {
|
|
5280
6410
|
logActivity(state, {
|
|
5281
6411
|
type: "info",
|
|
@@ -5299,21 +6429,44 @@ async function driveChannels(state, driver) {
|
|
|
5299
6429
|
const errorMessage = error2 instanceof Error ? error2.message : String(error2);
|
|
5300
6430
|
logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
|
|
5301
6431
|
if (state.interactive) displayStatus(state);
|
|
6432
|
+
if (driver.hasInFlightWatchers()) {
|
|
6433
|
+
consecutiveDrainFailures = 0;
|
|
6434
|
+
unreachableMs = 0;
|
|
6435
|
+
} else if (state.idleTimeout !== null) {
|
|
6436
|
+
consecutiveDrainFailures++;
|
|
6437
|
+
unreachableThisCycle = true;
|
|
6438
|
+
if (consecutiveDrainFailures === 1) {
|
|
6439
|
+
logActivity(state, {
|
|
6440
|
+
type: "info",
|
|
6441
|
+
message: `Cannot reach Evident, will exit if this persists past the idle timeout (timeout: ${state.idleTimeout}s)...`
|
|
6442
|
+
});
|
|
6443
|
+
if (state.interactive) displayStatus(state);
|
|
6444
|
+
}
|
|
6445
|
+
}
|
|
5302
6446
|
}
|
|
5303
6447
|
await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
|
|
5304
|
-
|
|
5305
|
-
|
|
5306
|
-
|
|
5307
|
-
|
|
5308
|
-
|
|
5309
|
-
|
|
5310
|
-
|
|
6448
|
+
const cycleMs = performance.now() - cycleStartedAtMs;
|
|
6449
|
+
if (idleThisCycle) idleMs += cycleMs;
|
|
6450
|
+
if (unreachableThisCycle) unreachableMs += cycleMs;
|
|
6451
|
+
if (state.idleTimeout !== null && consecutiveDrainFailures >= 2 && unreachableMs > state.idleTimeout * 1e3) {
|
|
6452
|
+
logActivity(state, {
|
|
6453
|
+
type: "info",
|
|
6454
|
+
level: "warn",
|
|
6455
|
+
message: `Exiting: could not reach Evident for ${consecutiveDrainFailures} consecutive polls (${Math.round(unreachableMs / 1e3)}s)`
|
|
6456
|
+
});
|
|
6457
|
+
if (state.interactive) displayStatus(state);
|
|
6458
|
+
break;
|
|
6459
|
+
}
|
|
6460
|
+
if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
|
|
6461
|
+
logActivity(state, { type: "info", message: "Idle timeout reached" });
|
|
6462
|
+
if (state.interactive) displayStatus(state);
|
|
6463
|
+
break;
|
|
5311
6464
|
}
|
|
5312
6465
|
}
|
|
5313
6466
|
}
|
|
5314
6467
|
var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
|
|
5315
|
-
async function runSweep(state, driver,
|
|
5316
|
-
const mode = `age=${
|
|
6468
|
+
async function runSweep(state, driver, config) {
|
|
6469
|
+
const mode = `age=${config.maxAgeMs ?? "\u2014"} count=${config.maxCount ?? "\u2014"}`;
|
|
5317
6470
|
try {
|
|
5318
6471
|
const sessions = await listSessions(state.port);
|
|
5319
6472
|
if (sessions === null) {
|
|
@@ -5326,8 +6479,8 @@ async function runSweep(state, driver, config2) {
|
|
|
5326
6479
|
const toDelete = selectSessionsToDelete(
|
|
5327
6480
|
sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
|
|
5328
6481
|
{
|
|
5329
|
-
maxAgeMs:
|
|
5330
|
-
maxCount:
|
|
6482
|
+
maxAgeMs: config.maxAgeMs,
|
|
6483
|
+
maxCount: config.maxCount,
|
|
5331
6484
|
nowMs: Date.now(),
|
|
5332
6485
|
protectedIds: driver.protectedSessionIds()
|
|
5333
6486
|
}
|
|
@@ -5363,7 +6516,7 @@ async function runSweep(state, driver, config2) {
|
|
|
5363
6516
|
}
|
|
5364
6517
|
}
|
|
5365
6518
|
function scheduleSessionCleanup(state, driver, options) {
|
|
5366
|
-
const
|
|
6519
|
+
const config = resolveSessionCleanupConfig(
|
|
5367
6520
|
{
|
|
5368
6521
|
maxAge: options.sessionCleanupMaxAge,
|
|
5369
6522
|
maxCount: options.sessionCleanupMaxCount,
|
|
@@ -5371,21 +6524,129 @@ function scheduleSessionCleanup(state, driver, options) {
|
|
|
5371
6524
|
},
|
|
5372
6525
|
process.env
|
|
5373
6526
|
);
|
|
5374
|
-
for (const warning2 of
|
|
6527
|
+
for (const warning2 of config.warnings) {
|
|
5375
6528
|
logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
|
|
5376
6529
|
}
|
|
5377
|
-
if (!
|
|
6530
|
+
if (!config.enabled) return;
|
|
5378
6531
|
logActivity(state, {
|
|
5379
6532
|
type: "info",
|
|
5380
|
-
message: `Session cleanup enabled (age=${
|
|
6533
|
+
message: `Session cleanup enabled (age=${config.maxAgeMs ?? "\u2014"}, count=${config.maxCount ?? "\u2014"}, interval=${config.intervalMs}ms)`
|
|
5381
6534
|
});
|
|
5382
|
-
const interval = setInterval(() => void runSweep(state, driver,
|
|
6535
|
+
const interval = setInterval(() => void runSweep(state, driver, config), config.intervalMs);
|
|
5383
6536
|
const firstSweep = setTimeout(
|
|
5384
|
-
() => void runSweep(state, driver,
|
|
6537
|
+
() => void runSweep(state, driver, config),
|
|
5385
6538
|
SESSION_CLEANUP_FIRST_SWEEP_MS
|
|
5386
6539
|
);
|
|
5387
6540
|
state.sessionCleanupTimers.push(interval, firstSweep);
|
|
5388
6541
|
}
|
|
6542
|
+
function claudeUsageFailureStreakSuffix(consecutiveFailures) {
|
|
6543
|
+
return consecutiveFailures > 1 ? ` (${consecutiveFailures} consecutive failures)` : "";
|
|
6544
|
+
}
|
|
6545
|
+
function scheduleClaudeUsageReporting(state, options) {
|
|
6546
|
+
const { mode, warnings } = resolveClaudeUsageReportingMode(
|
|
6547
|
+
options.claudeUsageReporting,
|
|
6548
|
+
process.env
|
|
6549
|
+
);
|
|
6550
|
+
for (const warning2 of warnings) {
|
|
6551
|
+
logActivity(state, {
|
|
6552
|
+
type: "info",
|
|
6553
|
+
level: "warn",
|
|
6554
|
+
message: `Claude usage reporting: ${warning2}`
|
|
6555
|
+
});
|
|
6556
|
+
}
|
|
6557
|
+
if (mode === "off") {
|
|
6558
|
+
logActivity(state, {
|
|
6559
|
+
type: "info",
|
|
6560
|
+
level: "debug",
|
|
6561
|
+
message: "Claude usage reporting is off (--claude-usage-reporting off)"
|
|
6562
|
+
});
|
|
6563
|
+
return null;
|
|
6564
|
+
}
|
|
6565
|
+
let consecutiveFailures = 0;
|
|
6566
|
+
let armed = false;
|
|
6567
|
+
let rearmRequested = false;
|
|
6568
|
+
const scheduleNextTick = () => {
|
|
6569
|
+
armed = true;
|
|
6570
|
+
rearmRequested = false;
|
|
6571
|
+
state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
|
|
6572
|
+
};
|
|
6573
|
+
const rearm = () => {
|
|
6574
|
+
if (armed) {
|
|
6575
|
+
rearmRequested = true;
|
|
6576
|
+
return;
|
|
6577
|
+
}
|
|
6578
|
+
rearmRequested = false;
|
|
6579
|
+
armed = true;
|
|
6580
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6581
|
+
};
|
|
6582
|
+
const tick = async (isProbe) => {
|
|
6583
|
+
try {
|
|
6584
|
+
const usage = await getClaudeUsage();
|
|
6585
|
+
const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
|
|
6586
|
+
if (result.ok) {
|
|
6587
|
+
if (consecutiveFailures > 0) {
|
|
6588
|
+
logActivity(state, {
|
|
6589
|
+
type: "info",
|
|
6590
|
+
level: "info",
|
|
6591
|
+
message: "Claude usage reporting recovered"
|
|
6592
|
+
});
|
|
6593
|
+
}
|
|
6594
|
+
consecutiveFailures = 0;
|
|
6595
|
+
logActivity(state, {
|
|
6596
|
+
type: "info",
|
|
6597
|
+
level: "debug",
|
|
6598
|
+
message: "Reported Claude usage to Evident"
|
|
6599
|
+
});
|
|
6600
|
+
} else {
|
|
6601
|
+
consecutiveFailures++;
|
|
6602
|
+
logActivity(state, {
|
|
6603
|
+
type: "info",
|
|
6604
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6605
|
+
message: `Failed to report Claude usage: ${result.error}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6606
|
+
});
|
|
6607
|
+
}
|
|
6608
|
+
scheduleNextTick();
|
|
6609
|
+
} catch (error2) {
|
|
6610
|
+
if (error2 instanceof ClaudeUsageError && isLocalCredentialProblem(error2)) {
|
|
6611
|
+
if (mode === "on") {
|
|
6612
|
+
logActivity(state, {
|
|
6613
|
+
type: "info",
|
|
6614
|
+
level: "warn",
|
|
6615
|
+
message: "Claude usage reporting is forced on but no usable Claude Code login was found \u2014 run `claude` to sign in; reporting will keep retrying"
|
|
6616
|
+
});
|
|
6617
|
+
scheduleNextTick();
|
|
6618
|
+
} else if (isProbe) {
|
|
6619
|
+
logActivity(state, {
|
|
6620
|
+
type: "info",
|
|
6621
|
+
level: "debug",
|
|
6622
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6623
|
+
});
|
|
6624
|
+
armed = false;
|
|
6625
|
+
if (rearmRequested) rearm();
|
|
6626
|
+
} else {
|
|
6627
|
+
logActivity(state, {
|
|
6628
|
+
type: "info",
|
|
6629
|
+
level: "debug",
|
|
6630
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6631
|
+
});
|
|
6632
|
+
scheduleNextTick();
|
|
6633
|
+
}
|
|
6634
|
+
} else {
|
|
6635
|
+
consecutiveFailures++;
|
|
6636
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
6637
|
+
logActivity(state, {
|
|
6638
|
+
type: "info",
|
|
6639
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6640
|
+
message: `Claude usage reporting failed: ${message}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6641
|
+
});
|
|
6642
|
+
scheduleNextTick();
|
|
6643
|
+
}
|
|
6644
|
+
}
|
|
6645
|
+
};
|
|
6646
|
+
armed = true;
|
|
6647
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6648
|
+
return rearm;
|
|
6649
|
+
}
|
|
5389
6650
|
async function notifyOffline(state) {
|
|
5390
6651
|
if (!state.agentId || !state.authHeader) return;
|
|
5391
6652
|
if (!state.connected) {
|
|
@@ -5403,13 +6664,29 @@ async function notifyOffline(state) {
|
|
|
5403
6664
|
if (state.interactive) displayStatus(state);
|
|
5404
6665
|
}
|
|
5405
6666
|
}
|
|
6667
|
+
async function timeShutdownPhase(state, durations, name, run2) {
|
|
6668
|
+
const startedAt = Date.now();
|
|
6669
|
+
try {
|
|
6670
|
+
return await run2();
|
|
6671
|
+
} finally {
|
|
6672
|
+
const elapsedMs = Date.now() - startedAt;
|
|
6673
|
+
durations[name] = elapsedMs;
|
|
6674
|
+
log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
|
|
6675
|
+
}
|
|
6676
|
+
}
|
|
5406
6677
|
async function cleanup(state, opts = {}) {
|
|
6678
|
+
const durations = {};
|
|
5407
6679
|
state.running = false;
|
|
5408
6680
|
for (const timer of state.sessionCleanupTimers) {
|
|
5409
6681
|
clearInterval(timer);
|
|
5410
6682
|
clearTimeout(timer);
|
|
5411
6683
|
}
|
|
5412
6684
|
state.sessionCleanupTimers = [];
|
|
6685
|
+
if (state.claudeUsageTimer) {
|
|
6686
|
+
clearTimeout(state.claudeUsageTimer);
|
|
6687
|
+
state.claudeUsageTimer = null;
|
|
6688
|
+
}
|
|
6689
|
+
state.claudeUsageRearm = null;
|
|
5413
6690
|
if (opts.graceful && state.channelDriver) {
|
|
5414
6691
|
state.channelDriver.stop();
|
|
5415
6692
|
log2(state, "Draining in-flight channel work before shutdown...");
|
|
@@ -5417,7 +6694,13 @@ async function cleanup(state, opts = {}) {
|
|
|
5417
6694
|
logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
|
|
5418
6695
|
displayStatus(state);
|
|
5419
6696
|
}
|
|
5420
|
-
const
|
|
6697
|
+
const driver = state.channelDriver;
|
|
6698
|
+
const settled = await timeShutdownPhase(
|
|
6699
|
+
state,
|
|
6700
|
+
durations,
|
|
6701
|
+
"drain",
|
|
6702
|
+
() => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
|
|
6703
|
+
);
|
|
5421
6704
|
if (!settled) {
|
|
5422
6705
|
logActivity(state, {
|
|
5423
6706
|
type: "info",
|
|
@@ -5426,13 +6709,15 @@ async function cleanup(state, opts = {}) {
|
|
|
5426
6709
|
if (state.interactive) displayStatus(state);
|
|
5427
6710
|
}
|
|
5428
6711
|
}
|
|
5429
|
-
await notifyOffline(state);
|
|
6712
|
+
await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
|
|
5430
6713
|
if (state.connection) {
|
|
5431
|
-
state.connection
|
|
6714
|
+
const connection = state.connection;
|
|
6715
|
+
await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
|
|
5432
6716
|
state.connection = null;
|
|
5433
6717
|
}
|
|
5434
6718
|
if (state.opencodeProcess) {
|
|
5435
|
-
|
|
6719
|
+
const opencodeProcess = state.opencodeProcess;
|
|
6720
|
+
await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
|
|
5436
6721
|
if (state.interactive) {
|
|
5437
6722
|
logActivity(state, { type: "info", message: "Stopped OpenCode process" });
|
|
5438
6723
|
displayStatus(state);
|
|
@@ -5441,6 +6726,7 @@ async function cleanup(state, opts = {}) {
|
|
|
5441
6726
|
}
|
|
5442
6727
|
state.opencodeProcess = null;
|
|
5443
6728
|
}
|
|
6729
|
+
return durations;
|
|
5444
6730
|
}
|
|
5445
6731
|
async function run(options) {
|
|
5446
6732
|
const interactive = isInteractive(options.json);
|
|
@@ -5481,8 +6767,11 @@ async function run(options) {
|
|
|
5481
6767
|
messageCount: 0,
|
|
5482
6768
|
lastProxiedActivityAt: null,
|
|
5483
6769
|
sessionCleanupTimers: [],
|
|
6770
|
+
claudeUsageTimer: null,
|
|
6771
|
+
claudeUsageRearm: null,
|
|
5484
6772
|
authHeader: ""
|
|
5485
6773
|
};
|
|
6774
|
+
setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
|
|
5486
6775
|
if (fileSyncDirectories.length > 0) {
|
|
5487
6776
|
log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
|
|
5488
6777
|
} else {
|
|
@@ -5511,14 +6800,38 @@ async function run(options) {
|
|
|
5511
6800
|
const handleSignal = async () => {
|
|
5512
6801
|
if (state.shuttingDown) return;
|
|
5513
6802
|
state.shuttingDown = true;
|
|
6803
|
+
const shutdownStartedAt = Date.now();
|
|
5514
6804
|
if (state.interactive) {
|
|
5515
6805
|
logActivity(state, { type: "info", message: "Shutting down..." });
|
|
5516
6806
|
displayStatus(state);
|
|
5517
6807
|
} else {
|
|
5518
6808
|
log2(state, "Shutting down...");
|
|
5519
6809
|
}
|
|
5520
|
-
await cleanup(state, { graceful: true });
|
|
5521
|
-
|
|
6810
|
+
const durations = await cleanup(state, { graceful: true });
|
|
6811
|
+
const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
|
|
6812
|
+
await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
|
|
6813
|
+
let timer;
|
|
6814
|
+
const flushed = shutdownTelemetry().then(
|
|
6815
|
+
() => true,
|
|
6816
|
+
(error2) => {
|
|
6817
|
+
log2(
|
|
6818
|
+
state,
|
|
6819
|
+
`Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
|
|
6820
|
+
"warn"
|
|
6821
|
+
);
|
|
6822
|
+
return true;
|
|
6823
|
+
}
|
|
6824
|
+
);
|
|
6825
|
+
const timedOut = new Promise((resolve3) => {
|
|
6826
|
+
timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
|
|
6827
|
+
});
|
|
6828
|
+
if (!await Promise.race([flushed, timedOut])) {
|
|
6829
|
+
log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
|
|
6830
|
+
}
|
|
6831
|
+
clearTimeout(timer);
|
|
6832
|
+
});
|
|
6833
|
+
const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
|
|
6834
|
+
log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
|
|
5522
6835
|
process.exit(0);
|
|
5523
6836
|
};
|
|
5524
6837
|
process.on("SIGINT", handleSignal);
|
|
@@ -5535,6 +6848,7 @@ async function run(options) {
|
|
|
5535
6848
|
console.log(chalk6.dim("Or run `evident login` for interactive authentication"));
|
|
5536
6849
|
blank();
|
|
5537
6850
|
process.exit(1);
|
|
6851
|
+
return;
|
|
5538
6852
|
}
|
|
5539
6853
|
blank();
|
|
5540
6854
|
console.log(chalk6.yellow("You are not logged in to Evident."));
|
|
@@ -5579,6 +6893,7 @@ async function run(options) {
|
|
|
5579
6893
|
} else {
|
|
5580
6894
|
printError(resolved.error || "Failed to resolve runner ID from key");
|
|
5581
6895
|
process.exit(1);
|
|
6896
|
+
return;
|
|
5582
6897
|
}
|
|
5583
6898
|
} else {
|
|
5584
6899
|
printError(
|
|
@@ -5592,6 +6907,7 @@ async function run(options) {
|
|
|
5592
6907
|
);
|
|
5593
6908
|
blank();
|
|
5594
6909
|
process.exit(1);
|
|
6910
|
+
return;
|
|
5595
6911
|
}
|
|
5596
6912
|
}
|
|
5597
6913
|
telemetry.info(
|
|
@@ -5632,40 +6948,67 @@ async function run(options) {
|
|
|
5632
6948
|
}
|
|
5633
6949
|
spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
|
|
5634
6950
|
state.agentName = validation.agent.name;
|
|
6951
|
+
const microvmId = process.env.MICROVM_ID?.trim();
|
|
6952
|
+
if (microvmId) {
|
|
6953
|
+
const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
|
|
6954
|
+
if (reported.ok) {
|
|
6955
|
+
log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
|
|
6956
|
+
} else {
|
|
6957
|
+
const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
|
|
6958
|
+
log2(state, message, "warn");
|
|
6959
|
+
if (state.interactive && !state.json) {
|
|
6960
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
6961
|
+
}
|
|
6962
|
+
}
|
|
6963
|
+
} else {
|
|
6964
|
+
log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
|
|
6965
|
+
}
|
|
6966
|
+
const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
|
|
6967
|
+
for (const warning2 of opencodeStartTimeoutWarnings) {
|
|
6968
|
+
logActivity(state, { type: "info", level: "warn", message: warning2 });
|
|
6969
|
+
}
|
|
5635
6970
|
const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
|
|
5636
6971
|
try {
|
|
5637
6972
|
const oc = await ensureOpenCodeRunning({
|
|
5638
6973
|
port: state.port,
|
|
5639
6974
|
interactive: state.interactive,
|
|
5640
6975
|
agentId: state.agentId,
|
|
5641
|
-
log: (message) => log2(state, message)
|
|
6976
|
+
log: (message) => log2(state, message),
|
|
6977
|
+
startTimeoutMs: opencodeStartTimeoutMs
|
|
5642
6978
|
});
|
|
5643
6979
|
state.port = oc.port;
|
|
5644
6980
|
state.opencodeProcess = oc.process;
|
|
5645
6981
|
state.opencodeVersion = oc.version;
|
|
5646
|
-
state.opencodeConnected = oc.
|
|
6982
|
+
state.opencodeConnected = oc.notReadyReason === null;
|
|
5647
6983
|
const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
|
|
5648
6984
|
ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
|
|
5649
|
-
|
|
5650
|
-
|
|
5651
|
-
|
|
5652
|
-
|
|
5653
|
-
|
|
6985
|
+
if (!state.interactive && oc.notReadyReason !== null) {
|
|
6986
|
+
const message = `OpenCode is not ready on port ${state.port}: ${oc.notReadyReason}. The runner will still come online, but messages will fail until opencode answers \u2014 raise the wait with --opencode-start-timeout <seconds> (env ${OPENCODE_START_TIMEOUT_ENV}).`;
|
|
6987
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
6988
|
+
} else {
|
|
6989
|
+
const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
|
|
6990
|
+
if (versionWarning) {
|
|
6991
|
+
log2(state, versionWarning, "warn");
|
|
6992
|
+
if (state.interactive && !state.json) {
|
|
6993
|
+
logActivity(state, { type: "info", level: "warn", message: versionWarning });
|
|
6994
|
+
}
|
|
5654
6995
|
}
|
|
5655
|
-
|
|
5656
|
-
|
|
5657
|
-
|
|
5658
|
-
|
|
5659
|
-
|
|
5660
|
-
|
|
5661
|
-
|
|
5662
|
-
|
|
5663
|
-
|
|
5664
|
-
|
|
5665
|
-
|
|
5666
|
-
|
|
5667
|
-
|
|
5668
|
-
|
|
6996
|
+
const noProviderWarning = buildNoProviderWarning(
|
|
6997
|
+
await hasAnyConfiguredProvider(state.port)
|
|
6998
|
+
);
|
|
6999
|
+
if (noProviderWarning) {
|
|
7000
|
+
log2(state, noProviderWarning, "warn");
|
|
7001
|
+
if (state.interactive && !state.json) {
|
|
7002
|
+
logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
|
|
7003
|
+
blank();
|
|
7004
|
+
console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
|
|
7005
|
+
console.log(
|
|
7006
|
+
chalk6.dim(
|
|
7007
|
+
`Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
|
|
7008
|
+
)
|
|
7009
|
+
);
|
|
7010
|
+
blank();
|
|
7011
|
+
}
|
|
5669
7012
|
}
|
|
5670
7013
|
}
|
|
5671
7014
|
} catch (error2) {
|
|
@@ -5710,6 +7053,18 @@ async function run(options) {
|
|
|
5710
7053
|
type: "info",
|
|
5711
7054
|
message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
|
|
5712
7055
|
});
|
|
7056
|
+
if (options.tunnelReadyFile) {
|
|
7057
|
+
const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
|
|
7058
|
+
if (marker.ok) {
|
|
7059
|
+
log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
|
|
7060
|
+
} else {
|
|
7061
|
+
log2(
|
|
7062
|
+
state,
|
|
7063
|
+
`Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
|
|
7064
|
+
"error"
|
|
7065
|
+
);
|
|
7066
|
+
}
|
|
7067
|
+
}
|
|
5713
7068
|
emitAgentConnected(state.agentId, {
|
|
5714
7069
|
port: state.port,
|
|
5715
7070
|
cli_version: getCliVersion(),
|
|
@@ -5800,6 +7155,7 @@ async function run(options) {
|
|
|
5800
7155
|
throw error2;
|
|
5801
7156
|
}
|
|
5802
7157
|
scheduleSessionCleanup(state, channelDriver, options);
|
|
7158
|
+
state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
|
|
5803
7159
|
if (!interactive || state.json) {
|
|
5804
7160
|
log2(state, "Driving channel messages...");
|
|
5805
7161
|
}
|
|
@@ -5854,13 +7210,18 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
|
|
|
5854
7210
|
program.command("login").description("Authenticate with Evident").option("--token", "Use token-based authentication (for CI/CD)").option("--no-browser", "Do not open the browser automatically").action(login);
|
|
5855
7211
|
program.command("logout").description("Remove stored credentials for the current endpoint").option("--all", "Remove stored credentials for all endpoints").action((options) => logout({ all: options.all }));
|
|
5856
7212
|
program.command("whoami").description("Show the currently logged in user").action(whoami);
|
|
7213
|
+
program.command("status").description("Check whether the configured credentials can reach Evident").option("--json", "Output in JSON format").action((options) => status({ json: options.json }));
|
|
7214
|
+
program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
|
|
5857
7215
|
program.command("run").description("Connect to Evident and process messages").option("--runner [id]", "Runner ID to connect to (optional when EVIDENT_RUNNER_KEY is set)").option(
|
|
5858
7216
|
"-a, --agent [id]",
|
|
5859
7217
|
"Deprecated alias for --runner (still supported; --runner wins if both are given)"
|
|
5860
7218
|
).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
|
|
5861
7219
|
"--log-level <level>",
|
|
5862
7220
|
"Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
|
|
5863
|
-
).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option(
|
|
7221
|
+
).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option(
|
|
7222
|
+
"--opencode-start-timeout <seconds>",
|
|
7223
|
+
"Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
|
|
7224
|
+
).option("--json", "Output in JSON format").option(
|
|
5864
7225
|
"--session-cleanup-max-age <duration>",
|
|
5865
7226
|
"Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
|
|
5866
7227
|
).option(
|
|
@@ -5869,11 +7230,17 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5869
7230
|
).option(
|
|
5870
7231
|
"--session-cleanup-interval <duration>",
|
|
5871
7232
|
"How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
|
|
7233
|
+
).option(
|
|
7234
|
+
"--claude-usage-reporting <mode>",
|
|
7235
|
+
"Report Claude subscription usage to Evident: auto | on | off (default: auto). Env: EVIDENT_CLAUDE_USAGE_REPORTING"
|
|
5872
7236
|
).option(
|
|
5873
7237
|
"--enable-file-sync-to <dir>",
|
|
5874
7238
|
"Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
|
|
5875
7239
|
(value, previous) => previous.concat([value]),
|
|
5876
7240
|
[]
|
|
7241
|
+
).option(
|
|
7242
|
+
"--tunnel-ready-file <path>",
|
|
7243
|
+
"Path to write once the tunnel is connected (opt-in; unused unless an operator/image sets it)"
|
|
5877
7244
|
).action(
|
|
5878
7245
|
(options) => {
|
|
5879
7246
|
run({
|
|
@@ -5886,14 +7253,21 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5886
7253
|
verbose: options.verbose,
|
|
5887
7254
|
conversation: options.conversation,
|
|
5888
7255
|
idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
|
|
7256
|
+
// Raw string — validation/precedence is single-sourced in run.ts's
|
|
7257
|
+
// resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
|
|
7258
|
+
opencodeStartTimeout: options.opencodeStartTimeout,
|
|
5889
7259
|
json: options.json,
|
|
5890
7260
|
// Raw strings — the resolver in run.ts single-sources parsing (M1).
|
|
5891
7261
|
sessionCleanupMaxAge: options.sessionCleanupMaxAge,
|
|
5892
7262
|
sessionCleanupMaxCount: options.sessionCleanupMaxCount,
|
|
5893
7263
|
sessionCleanupInterval: options.sessionCleanupInterval,
|
|
7264
|
+
// Raw string — the resolver in run.ts single-sources parsing
|
|
7265
|
+
// (resolveClaudeUsageReportingMode).
|
|
7266
|
+
claudeUsageReporting: options.claudeUsageReporting,
|
|
5894
7267
|
// Raw values — expansion/validation is single-sourced in run.ts's
|
|
5895
7268
|
// resolveFileSyncDirectories.
|
|
5896
|
-
enableFileSyncTo: options.enableFileSyncTo
|
|
7269
|
+
enableFileSyncTo: options.enableFileSyncTo,
|
|
7270
|
+
tunnelReadyFile: options.tunnelReadyFile
|
|
5897
7271
|
});
|
|
5898
7272
|
}
|
|
5899
7273
|
);
|