@evident-ai/cli 3.1.1-dev.4c17a8b → 3.1.1-dev.4d3cdb1
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 +60 -11
- package/dist/index.js +2587 -611
- 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
|
|
@@ -285,14 +324,14 @@ function blank() {
|
|
|
285
324
|
console.log();
|
|
286
325
|
}
|
|
287
326
|
function waitForEnter(prompt = "Press Enter to continue...") {
|
|
288
|
-
return new Promise((
|
|
327
|
+
return new Promise((resolve3) => {
|
|
289
328
|
process.stdout.write(chalk.dim(prompt));
|
|
290
329
|
const handler = () => {
|
|
291
330
|
process.stdin.removeListener("data", handler);
|
|
292
331
|
process.stdin.setRawMode?.(false);
|
|
293
332
|
process.stdin.pause();
|
|
294
333
|
console.log();
|
|
295
|
-
|
|
334
|
+
resolve3();
|
|
296
335
|
};
|
|
297
336
|
if (process.stdin.isTTY) {
|
|
298
337
|
process.stdin.setRawMode?.(true);
|
|
@@ -302,7 +341,7 @@ function waitForEnter(prompt = "Press Enter to continue...") {
|
|
|
302
341
|
});
|
|
303
342
|
}
|
|
304
343
|
function sleep(ms) {
|
|
305
|
-
return new Promise((
|
|
344
|
+
return new Promise((resolve3) => setTimeout(resolve3, ms));
|
|
306
345
|
}
|
|
307
346
|
|
|
308
347
|
// src/commands/login.ts
|
|
@@ -373,22 +412,25 @@ async function deviceFlowLogin(options) {
|
|
|
373
412
|
}
|
|
374
413
|
async function tokenLogin() {
|
|
375
414
|
console.log("Token login mode.");
|
|
376
|
-
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
|
+
);
|
|
377
419
|
blank();
|
|
378
420
|
process.stdout.write("Paste token: ");
|
|
379
|
-
const token = await new Promise((
|
|
421
|
+
const token = await new Promise((resolve3) => {
|
|
380
422
|
let data = "";
|
|
381
423
|
process.stdin.setEncoding("utf8");
|
|
382
424
|
process.stdin.on("data", (chunk) => {
|
|
383
425
|
data += chunk;
|
|
384
426
|
});
|
|
385
427
|
process.stdin.on("end", () => {
|
|
386
|
-
|
|
428
|
+
resolve3(data.trim());
|
|
387
429
|
});
|
|
388
430
|
if (process.stdin.isTTY) {
|
|
389
431
|
process.stdin.once("data", (chunk) => {
|
|
390
432
|
process.stdin.pause();
|
|
391
|
-
|
|
433
|
+
resolve3(chunk.toString().trim());
|
|
392
434
|
});
|
|
393
435
|
process.stdin.resume();
|
|
394
436
|
}
|
|
@@ -397,13 +439,22 @@ async function tokenLogin() {
|
|
|
397
439
|
printError("No token provided.");
|
|
398
440
|
process.exit(1);
|
|
399
441
|
}
|
|
442
|
+
await validateAndStoreToken(token);
|
|
443
|
+
}
|
|
444
|
+
async function validateAndStoreToken(token) {
|
|
400
445
|
const spinner = ora("Validating token...").start();
|
|
401
446
|
try {
|
|
402
|
-
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
|
+
}
|
|
403
455
|
await storeToken({
|
|
404
456
|
token,
|
|
405
|
-
user: result.user
|
|
406
|
-
expiresAt: result.expires_at
|
|
457
|
+
user: { email: result.user.email }
|
|
407
458
|
});
|
|
408
459
|
spinner.stop();
|
|
409
460
|
printSuccess(`Logged in as ${chalk2.bold(result.user.email)}`);
|
|
@@ -423,9 +474,22 @@ async function login(options) {
|
|
|
423
474
|
}
|
|
424
475
|
|
|
425
476
|
// src/commands/logout.ts
|
|
477
|
+
function describeFailure(failure) {
|
|
478
|
+
if (failure.type === "enumerate") {
|
|
479
|
+
return `could not list stored keychain entries (${failure.error.message})`;
|
|
480
|
+
}
|
|
481
|
+
return `${failure.account} (${failure.error.message})`;
|
|
482
|
+
}
|
|
426
483
|
async function logout(options = {}) {
|
|
427
484
|
if (options.all) {
|
|
428
|
-
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
|
+
}
|
|
429
493
|
printSuccess("Logged out of all endpoints.");
|
|
430
494
|
return;
|
|
431
495
|
}
|
|
@@ -450,7 +514,9 @@ async function whoami() {
|
|
|
450
514
|
blank();
|
|
451
515
|
console.log(keyValue("Endpoint", apiUrl));
|
|
452
516
|
console.log(keyValue("User", chalk3.bold(credentials2.user.email)));
|
|
453
|
-
|
|
517
|
+
if (credentials2.user.id) {
|
|
518
|
+
console.log(keyValue("User ID", credentials2.user.id));
|
|
519
|
+
}
|
|
454
520
|
if (credentials2.expiresAt) {
|
|
455
521
|
const expiresAt = new Date(credentials2.expiresAt);
|
|
456
522
|
const now = /* @__PURE__ */ new Date();
|
|
@@ -466,198 +532,6 @@ async function whoami() {
|
|
|
466
532
|
blank();
|
|
467
533
|
}
|
|
468
534
|
|
|
469
|
-
// src/commands/run.ts
|
|
470
|
-
import chalk6 from "chalk";
|
|
471
|
-
import ora3 from "ora";
|
|
472
|
-
import { select as select3 } from "@inquirer/prompts";
|
|
473
|
-
|
|
474
|
-
// ../../packages/types/src/telemetry/index.ts
|
|
475
|
-
var TelemetryEventTypes = {
|
|
476
|
-
// Agent activity events (shown in web UI activity log)
|
|
477
|
-
AGENT_CONNECTED: "agent.connected",
|
|
478
|
-
AGENT_DISCONNECTED: "agent.disconnected",
|
|
479
|
-
AGENT_MESSAGE_PROCESSING: "agent.message_processing",
|
|
480
|
-
AGENT_MESSAGE_DONE: "agent.message_done",
|
|
481
|
-
AGENT_MESSAGE_FAILED: "agent.message_failed"
|
|
482
|
-
};
|
|
483
|
-
|
|
484
|
-
// ../../packages/types/src/tunnel/index.ts
|
|
485
|
-
var MAX_FRAME_BYTES = 256 * 1024;
|
|
486
|
-
var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
|
|
487
|
-
|
|
488
|
-
// ../../packages/types/src/logging/index.ts
|
|
489
|
-
var CORRELATION_ID_HEADER = "x-evident-correlation-id";
|
|
490
|
-
function log(level, event, fields) {
|
|
491
|
-
const method = level === "debug" ? "log" : level;
|
|
492
|
-
try {
|
|
493
|
-
console[method]("[evident]", JSON.stringify({ level, event, ...fields }));
|
|
494
|
-
} catch (err) {
|
|
495
|
-
console.error(
|
|
496
|
-
"[evident] log_serialize_failed",
|
|
497
|
-
event,
|
|
498
|
-
err instanceof Error ? err.message : String(err)
|
|
499
|
-
);
|
|
500
|
-
}
|
|
501
|
-
}
|
|
502
|
-
function errorFields(err) {
|
|
503
|
-
if (err instanceof Error) {
|
|
504
|
-
return { error: err.message, error_name: err.name };
|
|
505
|
-
}
|
|
506
|
-
return { error: String(err) };
|
|
507
|
-
}
|
|
508
|
-
function stripQuery(url) {
|
|
509
|
-
try {
|
|
510
|
-
return new URL(url).pathname;
|
|
511
|
-
} catch {
|
|
512
|
-
const q = url.indexOf("?");
|
|
513
|
-
return q === -1 ? url : url.slice(0, q);
|
|
514
|
-
}
|
|
515
|
-
}
|
|
516
|
-
|
|
517
|
-
// src/lib/telemetry.ts
|
|
518
|
-
var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
|
|
519
|
-
function getCliVersion() {
|
|
520
|
-
return CLI_VERSION;
|
|
521
|
-
}
|
|
522
|
-
var eventBuffer = [];
|
|
523
|
-
var flushTimeout = null;
|
|
524
|
-
var isShuttingDown = false;
|
|
525
|
-
var FLUSH_INTERVAL_MS = 5e3;
|
|
526
|
-
var MAX_BUFFER_SIZE = 50;
|
|
527
|
-
var FLUSH_TIMEOUT_MS = 3e3;
|
|
528
|
-
function logEvent(eventType, options = {}) {
|
|
529
|
-
const event = {
|
|
530
|
-
event_type: eventType,
|
|
531
|
-
severity: options.severity || "info",
|
|
532
|
-
message: options.message,
|
|
533
|
-
metadata: options.metadata,
|
|
534
|
-
agent_id: options.agentId,
|
|
535
|
-
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
536
|
-
};
|
|
537
|
-
eventBuffer.push(event);
|
|
538
|
-
if (options.severity === "error" || eventBuffer.length >= MAX_BUFFER_SIZE) {
|
|
539
|
-
void flushEvents();
|
|
540
|
-
} else if (!flushTimeout && !isShuttingDown) {
|
|
541
|
-
flushTimeout = setTimeout(() => {
|
|
542
|
-
flushTimeout = null;
|
|
543
|
-
void flushEvents();
|
|
544
|
-
}, FLUSH_INTERVAL_MS);
|
|
545
|
-
}
|
|
546
|
-
}
|
|
547
|
-
var telemetry = {
|
|
548
|
-
debug: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "debug", message, metadata, agentId }),
|
|
549
|
-
info: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "info", message, metadata, agentId }),
|
|
550
|
-
warn: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "warning", message, metadata, agentId }),
|
|
551
|
-
error: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "error", message, metadata, agentId })
|
|
552
|
-
};
|
|
553
|
-
async function flushEvents() {
|
|
554
|
-
if (eventBuffer.length === 0) return;
|
|
555
|
-
const events = eventBuffer;
|
|
556
|
-
eventBuffer = [];
|
|
557
|
-
if (flushTimeout) {
|
|
558
|
-
clearTimeout(flushTimeout);
|
|
559
|
-
flushTimeout = null;
|
|
560
|
-
}
|
|
561
|
-
try {
|
|
562
|
-
const credentials2 = await getToken();
|
|
563
|
-
if (!credentials2) {
|
|
564
|
-
return;
|
|
565
|
-
}
|
|
566
|
-
const apiUrl = getApiUrlConfig();
|
|
567
|
-
const controller = new AbortController();
|
|
568
|
-
const timeout = setTimeout(() => controller.abort(), FLUSH_TIMEOUT_MS);
|
|
569
|
-
try {
|
|
570
|
-
const request = {
|
|
571
|
-
events,
|
|
572
|
-
client_type: "cli",
|
|
573
|
-
client_version: CLI_VERSION
|
|
574
|
-
};
|
|
575
|
-
const response = await fetch(`${apiUrl}/telemetry/events`, {
|
|
576
|
-
method: "POST",
|
|
577
|
-
headers: {
|
|
578
|
-
"Content-Type": "application/json",
|
|
579
|
-
Authorization: `Bearer ${credentials2.token}`
|
|
580
|
-
},
|
|
581
|
-
body: JSON.stringify(request),
|
|
582
|
-
signal: controller.signal
|
|
583
|
-
});
|
|
584
|
-
if (!response.ok) {
|
|
585
|
-
console.error(`Telemetry flush failed: ${response.status}`);
|
|
586
|
-
}
|
|
587
|
-
} finally {
|
|
588
|
-
clearTimeout(timeout);
|
|
589
|
-
}
|
|
590
|
-
} catch (error2) {
|
|
591
|
-
if (process.env.DEBUG) {
|
|
592
|
-
console.error("Telemetry error:", error2);
|
|
593
|
-
}
|
|
594
|
-
}
|
|
595
|
-
}
|
|
596
|
-
async function shutdownTelemetry() {
|
|
597
|
-
isShuttingDown = true;
|
|
598
|
-
if (flushTimeout) {
|
|
599
|
-
clearTimeout(flushTimeout);
|
|
600
|
-
flushTimeout = null;
|
|
601
|
-
}
|
|
602
|
-
await flushEvents();
|
|
603
|
-
}
|
|
604
|
-
function emitEvent(event) {
|
|
605
|
-
logEvent(event.event_type, {
|
|
606
|
-
severity: event.severity,
|
|
607
|
-
message: event.message,
|
|
608
|
-
metadata: event.metadata,
|
|
609
|
-
agentId: event.agent_id
|
|
610
|
-
});
|
|
611
|
-
}
|
|
612
|
-
function emitAgentConnected(agentId, metadata) {
|
|
613
|
-
emitEvent({
|
|
614
|
-
event_type: TelemetryEventTypes.AGENT_CONNECTED,
|
|
615
|
-
severity: "info",
|
|
616
|
-
message: "Agent CLI connected",
|
|
617
|
-
metadata,
|
|
618
|
-
agent_id: agentId
|
|
619
|
-
});
|
|
620
|
-
}
|
|
621
|
-
function emitAgentDisconnected(agentId, metadata) {
|
|
622
|
-
emitEvent({
|
|
623
|
-
event_type: TelemetryEventTypes.AGENT_DISCONNECTED,
|
|
624
|
-
severity: "info",
|
|
625
|
-
message: `Agent CLI disconnected (code: ${metadata.code})`,
|
|
626
|
-
metadata,
|
|
627
|
-
agent_id: agentId
|
|
628
|
-
});
|
|
629
|
-
}
|
|
630
|
-
var EventTypes = {
|
|
631
|
-
// Tunnel lifecycle
|
|
632
|
-
TUNNEL_STARTING: "tunnel.starting",
|
|
633
|
-
TUNNEL_CONNECTED: "tunnel.connected",
|
|
634
|
-
TUNNEL_DISCONNECTED: "tunnel.disconnected",
|
|
635
|
-
TUNNEL_RECONNECTING: "tunnel.reconnecting",
|
|
636
|
-
TUNNEL_ERROR: "tunnel.error",
|
|
637
|
-
// OpenCode communication
|
|
638
|
-
OPENCODE_HEALTH_CHECK: "opencode.health_check",
|
|
639
|
-
OPENCODE_HEALTH_OK: "opencode.health_ok",
|
|
640
|
-
OPENCODE_HEALTH_FAILED: "opencode.health_failed",
|
|
641
|
-
OPENCODE_REQUEST_RECEIVED: "opencode.request_received",
|
|
642
|
-
OPENCODE_REQUEST_FORWARDED: "opencode.request_forwarded",
|
|
643
|
-
OPENCODE_RESPONSE_SENT: "opencode.response_sent",
|
|
644
|
-
OPENCODE_UNREACHABLE: "opencode.unreachable",
|
|
645
|
-
OPENCODE_ERROR: "opencode.error",
|
|
646
|
-
// Authentication
|
|
647
|
-
AUTH_LOGIN_STARTED: "auth.login_started",
|
|
648
|
-
AUTH_LOGIN_SUCCESS: "auth.login_success",
|
|
649
|
-
AUTH_LOGIN_FAILED: "auth.login_failed",
|
|
650
|
-
AUTH_LOGOUT: "auth.logout",
|
|
651
|
-
// CLI lifecycle
|
|
652
|
-
CLI_STARTED: "cli.started",
|
|
653
|
-
CLI_COMMAND: "cli.command",
|
|
654
|
-
CLI_ERROR: "cli.error",
|
|
655
|
-
// Deprecation telemetry (#412) — usage of the old `--agent`/`EVIDENT_AGENT_KEY`
|
|
656
|
-
// names instead of the preferred `--runner`/`EVIDENT_RUNNER_KEY` (#409).
|
|
657
|
-
DEPRECATED_AGENT_FLAG_USED: "cli.deprecated_agent_flag_used",
|
|
658
|
-
DEPRECATED_AGENT_KEY_ENV_USED: "cli.deprecated_agent_key_env_used"
|
|
659
|
-
};
|
|
660
|
-
|
|
661
535
|
// src/lib/auth.ts
|
|
662
536
|
async function getAuthCredentials() {
|
|
663
537
|
const runnerKey = process.env.EVIDENT_RUNNER_KEY;
|
|
@@ -701,33 +575,740 @@ function isInteractive(jsonOutput) {
|
|
|
701
575
|
return true;
|
|
702
576
|
}
|
|
703
577
|
|
|
704
|
-
// src/
|
|
705
|
-
async function
|
|
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;
|
|
706
582
|
try {
|
|
707
|
-
const
|
|
708
|
-
|
|
709
|
-
|
|
710
|
-
|
|
711
|
-
if (!response.ok) {
|
|
712
|
-
return { healthy: false, error: `HTTP ${response.status}` };
|
|
713
|
-
}
|
|
714
|
-
const data = await response.json().catch(() => ({}));
|
|
715
|
-
return { healthy: true, version: data.version };
|
|
716
|
-
} catch (error2) {
|
|
717
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
718
|
-
return { healthy: false, error: message };
|
|
719
|
-
}
|
|
720
|
-
}
|
|
721
|
-
async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
|
|
722
|
-
const startTime = Date.now();
|
|
723
|
-
while (Date.now() - startTime < timeoutMs) {
|
|
724
|
-
const health = await checkOpenCodeHealth(port);
|
|
725
|
-
if (health.healthy) {
|
|
726
|
-
return health;
|
|
583
|
+
const data = JSON.parse(text);
|
|
584
|
+
const message = data.message ?? data.error;
|
|
585
|
+
if (typeof message === "string" && message.trim()) {
|
|
586
|
+
return message;
|
|
727
587
|
}
|
|
728
|
-
|
|
588
|
+
} catch {
|
|
729
589
|
}
|
|
730
|
-
return
|
|
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
|
+
}
|
|
993
|
+
}
|
|
994
|
+
|
|
995
|
+
// src/commands/run.ts
|
|
996
|
+
import { homedir as homedir3 } from "os";
|
|
997
|
+
import { isAbsolute as isAbsolute2, join as join3, parse, resolve as resolvePath } from "path";
|
|
998
|
+
import chalk6 from "chalk";
|
|
999
|
+
|
|
1000
|
+
// ../../packages/types/src/agents/index.ts
|
|
1001
|
+
var MICROVM_MAX_LIFETIME_MS = 8 * 60 * 6e4;
|
|
1002
|
+
|
|
1003
|
+
// ../../packages/types/src/telemetry/index.ts
|
|
1004
|
+
var TelemetryEventTypes = {
|
|
1005
|
+
// Agent activity events (shown in web UI activity log)
|
|
1006
|
+
AGENT_CONNECTED: "agent.connected",
|
|
1007
|
+
AGENT_DISCONNECTED: "agent.disconnected",
|
|
1008
|
+
AGENT_MESSAGE_PROCESSING: "agent.message_processing",
|
|
1009
|
+
AGENT_MESSAGE_DONE: "agent.message_done",
|
|
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"
|
|
1014
|
+
};
|
|
1015
|
+
|
|
1016
|
+
// ../../packages/types/src/tunnel/index.ts
|
|
1017
|
+
var MAX_FRAME_BYTES = 256 * 1024;
|
|
1018
|
+
var TUNNEL_DRAIN_PING_PATH = "/__evident/drain";
|
|
1019
|
+
|
|
1020
|
+
// ../../packages/types/src/runner-files.ts
|
|
1021
|
+
var MAX_FILE_PUSH_BYTES = 64 * 1024;
|
|
1022
|
+
var MAX_FILE_SYNC_DIRECTORIES = 16;
|
|
1023
|
+
|
|
1024
|
+
// ../../packages/types/src/logging/index.ts
|
|
1025
|
+
var CORRELATION_ID_HEADER = "x-evident-correlation-id";
|
|
1026
|
+
function log(level, event, fields) {
|
|
1027
|
+
const method = level === "debug" ? "log" : level;
|
|
1028
|
+
try {
|
|
1029
|
+
console[method]("[evident]", JSON.stringify({ level, event, ...fields }));
|
|
1030
|
+
} catch (err) {
|
|
1031
|
+
console.error(
|
|
1032
|
+
"[evident] log_serialize_failed",
|
|
1033
|
+
event,
|
|
1034
|
+
err instanceof Error ? err.message : String(err)
|
|
1035
|
+
);
|
|
1036
|
+
}
|
|
1037
|
+
}
|
|
1038
|
+
function errorFields(err) {
|
|
1039
|
+
if (err instanceof Error) {
|
|
1040
|
+
return { error: err.message, error_name: err.name };
|
|
1041
|
+
}
|
|
1042
|
+
return { error: String(err) };
|
|
1043
|
+
}
|
|
1044
|
+
function stripQuery(url) {
|
|
1045
|
+
try {
|
|
1046
|
+
return new URL(url).pathname;
|
|
1047
|
+
} catch {
|
|
1048
|
+
const q = url.indexOf("?");
|
|
1049
|
+
return q === -1 ? url : url.slice(0, q);
|
|
1050
|
+
}
|
|
1051
|
+
}
|
|
1052
|
+
|
|
1053
|
+
// src/commands/run.ts
|
|
1054
|
+
import ora3 from "ora";
|
|
1055
|
+
import { select as select3 } from "@inquirer/prompts";
|
|
1056
|
+
|
|
1057
|
+
// src/lib/telemetry.ts
|
|
1058
|
+
var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
|
|
1059
|
+
function getCliVersion() {
|
|
1060
|
+
return CLI_VERSION;
|
|
1061
|
+
}
|
|
1062
|
+
var eventBuffer = [];
|
|
1063
|
+
var flushTimeout = null;
|
|
1064
|
+
var isShuttingDown = false;
|
|
1065
|
+
var FLUSH_INTERVAL_MS = 5e3;
|
|
1066
|
+
var MAX_BUFFER_SIZE = 50;
|
|
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;
|
|
1075
|
+
function logEvent(eventType, options = {}) {
|
|
1076
|
+
const event = {
|
|
1077
|
+
event_type: eventType,
|
|
1078
|
+
severity: options.severity || "info",
|
|
1079
|
+
message: options.message,
|
|
1080
|
+
metadata: options.metadata,
|
|
1081
|
+
agent_id: options.agentId,
|
|
1082
|
+
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
1083
|
+
};
|
|
1084
|
+
eventBuffer.push(event);
|
|
1085
|
+
if (options.severity === "error" || eventBuffer.length >= MAX_BUFFER_SIZE) {
|
|
1086
|
+
void flushEvents();
|
|
1087
|
+
} else if (!flushTimeout && !isShuttingDown) {
|
|
1088
|
+
flushTimeout = setTimeout(() => {
|
|
1089
|
+
flushTimeout = null;
|
|
1090
|
+
void flushEvents();
|
|
1091
|
+
}, FLUSH_INTERVAL_MS);
|
|
1092
|
+
}
|
|
1093
|
+
}
|
|
1094
|
+
var telemetry = {
|
|
1095
|
+
debug: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "debug", message, metadata, agentId }),
|
|
1096
|
+
info: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "info", message, metadata, agentId }),
|
|
1097
|
+
warn: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "warning", message, metadata, agentId }),
|
|
1098
|
+
error: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "error", message, metadata, agentId })
|
|
1099
|
+
};
|
|
1100
|
+
async function flushEvents() {
|
|
1101
|
+
if (eventBuffer.length === 0) return;
|
|
1102
|
+
const events = eventBuffer;
|
|
1103
|
+
eventBuffer = [];
|
|
1104
|
+
if (flushTimeout) {
|
|
1105
|
+
clearTimeout(flushTimeout);
|
|
1106
|
+
flushTimeout = null;
|
|
1107
|
+
}
|
|
1108
|
+
try {
|
|
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 ${credentials2.token}`;
|
|
1119
|
+
}
|
|
1120
|
+
const apiUrl = getApiUrlConfig();
|
|
1121
|
+
const controller = new AbortController();
|
|
1122
|
+
const timeout = setTimeout(() => controller.abort(), FLUSH_TIMEOUT_MS);
|
|
1123
|
+
try {
|
|
1124
|
+
const request = {
|
|
1125
|
+
events,
|
|
1126
|
+
client_type: "cli",
|
|
1127
|
+
client_version: CLI_VERSION
|
|
1128
|
+
};
|
|
1129
|
+
const response = await fetch(`${apiUrl}/telemetry/events`, {
|
|
1130
|
+
method: "POST",
|
|
1131
|
+
headers: {
|
|
1132
|
+
"Content-Type": "application/json",
|
|
1133
|
+
Authorization: authHeader
|
|
1134
|
+
},
|
|
1135
|
+
body: JSON.stringify(request),
|
|
1136
|
+
signal: controller.signal
|
|
1137
|
+
});
|
|
1138
|
+
if (!response.ok) {
|
|
1139
|
+
console.error(`Telemetry flush failed: ${response.status}`);
|
|
1140
|
+
}
|
|
1141
|
+
} finally {
|
|
1142
|
+
clearTimeout(timeout);
|
|
1143
|
+
}
|
|
1144
|
+
} catch (error2) {
|
|
1145
|
+
const now = Date.now();
|
|
1146
|
+
if (now - lastFlushFailureLoggedAt >= FLUSH_FAILURE_LOG_INTERVAL_MS) {
|
|
1147
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
1148
|
+
const suffix = suppressedFlushFailureCount > 0 ? ` (${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++;
|
|
1154
|
+
}
|
|
1155
|
+
}
|
|
1156
|
+
}
|
|
1157
|
+
async function shutdownTelemetry() {
|
|
1158
|
+
isShuttingDown = true;
|
|
1159
|
+
if (flushTimeout) {
|
|
1160
|
+
clearTimeout(flushTimeout);
|
|
1161
|
+
flushTimeout = null;
|
|
1162
|
+
}
|
|
1163
|
+
await flushEvents();
|
|
1164
|
+
}
|
|
1165
|
+
function emitEvent(event) {
|
|
1166
|
+
logEvent(event.event_type, {
|
|
1167
|
+
severity: event.severity,
|
|
1168
|
+
message: event.message,
|
|
1169
|
+
metadata: event.metadata,
|
|
1170
|
+
agentId: event.agent_id
|
|
1171
|
+
});
|
|
1172
|
+
}
|
|
1173
|
+
function emitAgentConnected(agentId, metadata) {
|
|
1174
|
+
emitEvent({
|
|
1175
|
+
event_type: TelemetryEventTypes.AGENT_CONNECTED,
|
|
1176
|
+
severity: "info",
|
|
1177
|
+
message: "Agent CLI connected",
|
|
1178
|
+
metadata,
|
|
1179
|
+
agent_id: agentId
|
|
1180
|
+
});
|
|
1181
|
+
}
|
|
1182
|
+
function emitAgentDisconnected(agentId, metadata) {
|
|
1183
|
+
emitEvent({
|
|
1184
|
+
event_type: TelemetryEventTypes.AGENT_DISCONNECTED,
|
|
1185
|
+
severity: "info",
|
|
1186
|
+
message: `Agent CLI disconnected (code: ${metadata.code})`,
|
|
1187
|
+
metadata,
|
|
1188
|
+
agent_id: agentId
|
|
1189
|
+
});
|
|
1190
|
+
}
|
|
1191
|
+
var EventTypes = {
|
|
1192
|
+
// Tunnel lifecycle
|
|
1193
|
+
TUNNEL_STARTING: "tunnel.starting",
|
|
1194
|
+
TUNNEL_CONNECTED: "tunnel.connected",
|
|
1195
|
+
TUNNEL_DISCONNECTED: "tunnel.disconnected",
|
|
1196
|
+
TUNNEL_RECONNECTING: "tunnel.reconnecting",
|
|
1197
|
+
TUNNEL_ERROR: "tunnel.error",
|
|
1198
|
+
// OpenCode communication
|
|
1199
|
+
OPENCODE_HEALTH_CHECK: "opencode.health_check",
|
|
1200
|
+
OPENCODE_HEALTH_OK: "opencode.health_ok",
|
|
1201
|
+
OPENCODE_HEALTH_FAILED: "opencode.health_failed",
|
|
1202
|
+
OPENCODE_REQUEST_RECEIVED: "opencode.request_received",
|
|
1203
|
+
OPENCODE_REQUEST_FORWARDED: "opencode.request_forwarded",
|
|
1204
|
+
OPENCODE_RESPONSE_SENT: "opencode.response_sent",
|
|
1205
|
+
OPENCODE_UNREACHABLE: "opencode.unreachable",
|
|
1206
|
+
OPENCODE_ERROR: "opencode.error",
|
|
1207
|
+
// Authentication
|
|
1208
|
+
AUTH_LOGIN_STARTED: "auth.login_started",
|
|
1209
|
+
AUTH_LOGIN_SUCCESS: "auth.login_success",
|
|
1210
|
+
AUTH_LOGIN_FAILED: "auth.login_failed",
|
|
1211
|
+
AUTH_LOGOUT: "auth.logout",
|
|
1212
|
+
// CLI lifecycle
|
|
1213
|
+
CLI_STARTED: "cli.started",
|
|
1214
|
+
CLI_COMMAND: "cli.command",
|
|
1215
|
+
CLI_ERROR: "cli.error",
|
|
1216
|
+
// Deprecation telemetry (#412) — usage of the old `--agent`/`EVIDENT_AGENT_KEY`
|
|
1217
|
+
// names instead of the preferred `--runner`/`EVIDENT_RUNNER_KEY` (#409).
|
|
1218
|
+
DEPRECATED_AGENT_FLAG_USED: "cli.deprecated_agent_flag_used",
|
|
1219
|
+
DEPRECATED_AGENT_KEY_ENV_USED: "cli.deprecated_agent_key_env_used"
|
|
1220
|
+
};
|
|
1221
|
+
|
|
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 ${RATE_LIMIT_MAX_EVENTS}/min)`
|
|
1247
|
+
);
|
|
1248
|
+
}
|
|
1249
|
+
windowStartedAt = now;
|
|
1250
|
+
windowCount = 0;
|
|
1251
|
+
windowDroppedCount = 0;
|
|
1252
|
+
}
|
|
1253
|
+
if (windowCount >= RATE_LIMIT_MAX_EVENTS) {
|
|
1254
|
+
windowDroppedCount++;
|
|
1255
|
+
if (windowDroppedCount === 1) {
|
|
1256
|
+
console.error(
|
|
1257
|
+
`[runner-activity-telemetry] rate cap reached (${RATE_LIMIT_MAX_EVENTS} per ${RATE_LIMIT_WINDOW_MS / 1e3}s) \u2014 dropping further entries this window`
|
|
1258
|
+
);
|
|
1259
|
+
}
|
|
1260
|
+
return false;
|
|
1261
|
+
}
|
|
1262
|
+
windowCount++;
|
|
1263
|
+
return true;
|
|
1264
|
+
}
|
|
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: ${err instanceof Error ? err.message : String(err)}`
|
|
1281
|
+
);
|
|
1282
|
+
}
|
|
1283
|
+
}
|
|
1284
|
+
|
|
1285
|
+
// src/lib/opencode/health.ts
|
|
1286
|
+
async function checkOpenCodeHealth(port) {
|
|
1287
|
+
try {
|
|
1288
|
+
const response = await fetch(`http://127.0.0.1:${port}/global/health`, {
|
|
1289
|
+
signal: AbortSignal.timeout(2e3)
|
|
1290
|
+
// 2 second timeout
|
|
1291
|
+
});
|
|
1292
|
+
if (!response.ok) {
|
|
1293
|
+
return { healthy: false, error: `HTTP ${response.status}` };
|
|
1294
|
+
}
|
|
1295
|
+
const data = await response.json().catch(() => ({}));
|
|
1296
|
+
return { healthy: true, version: data.version };
|
|
1297
|
+
} catch (error2) {
|
|
1298
|
+
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
1299
|
+
return { healthy: false, error: message };
|
|
1300
|
+
}
|
|
1301
|
+
}
|
|
1302
|
+
async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
|
|
1303
|
+
const startTime = Date.now();
|
|
1304
|
+
while (Date.now() - startTime < timeoutMs) {
|
|
1305
|
+
const health = await checkOpenCodeHealth(port);
|
|
1306
|
+
if (health.healthy) {
|
|
1307
|
+
return health;
|
|
1308
|
+
}
|
|
1309
|
+
await new Promise((resolve3) => setTimeout(resolve3, 1e3));
|
|
1310
|
+
}
|
|
1311
|
+
return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
|
|
731
1312
|
}
|
|
732
1313
|
|
|
733
1314
|
// src/lib/opencode/opencode-version-gate.ts
|
|
@@ -1364,7 +1945,7 @@ async function sendPromptAsync(port, sessionId, content, options, attachments) {
|
|
|
1364
1945
|
}
|
|
1365
1946
|
}
|
|
1366
1947
|
if (attempt < READ_BACK_ATTEMPTS - 1) {
|
|
1367
|
-
await new Promise((
|
|
1948
|
+
await new Promise((resolve3) => setTimeout(resolve3, READ_BACK_DELAY_MS));
|
|
1368
1949
|
}
|
|
1369
1950
|
}
|
|
1370
1951
|
return null;
|
|
@@ -1492,6 +2073,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
|
|
|
1492
2073
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1493
2074
|
return completedOf(reply) != null && finishOf(reply) === "tool-calls";
|
|
1494
2075
|
}
|
|
2076
|
+
function isB2AbandonmentConfirmed(params) {
|
|
2077
|
+
return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
|
|
2078
|
+
}
|
|
1495
2079
|
function messageError(messages, userMessageId) {
|
|
1496
2080
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1497
2081
|
const error2 = errorOf(reply);
|
|
@@ -1505,6 +2089,57 @@ function messageError(messages, userMessageId) {
|
|
|
1505
2089
|
}
|
|
1506
2090
|
return "The agent run failed.";
|
|
1507
2091
|
}
|
|
2092
|
+
function isAbortedTerminalReply(messages, userMessageId) {
|
|
2093
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
2094
|
+
const error2 = errorOf(reply);
|
|
2095
|
+
if (error2 == null) return false;
|
|
2096
|
+
if (typeof error2 === "string") return error2.trim() === "Aborted";
|
|
2097
|
+
if (typeof error2 === "object") {
|
|
2098
|
+
const e = error2;
|
|
2099
|
+
if (e.name === "MessageAbortedError") return true;
|
|
2100
|
+
if (e.name === "AbortError") return true;
|
|
2101
|
+
const dataMessage = e.data?.message;
|
|
2102
|
+
const rendered = typeof dataMessage === "string" ? dataMessage : typeof e.message === "string" ? e.message : null;
|
|
2103
|
+
return rendered != null && rendered.trim() === "Aborted";
|
|
2104
|
+
}
|
|
2105
|
+
return false;
|
|
2106
|
+
}
|
|
2107
|
+
function messageFailure(messages, userMessageId) {
|
|
2108
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
2109
|
+
const error2 = errorOf(reply);
|
|
2110
|
+
if (error2 == null || typeof error2 !== "object") return null;
|
|
2111
|
+
const e = error2;
|
|
2112
|
+
const replyProviderId = reply?.info?.providerID ?? null;
|
|
2113
|
+
const replyModelId = reply?.info?.modelID ?? null;
|
|
2114
|
+
if (e.name === "ProviderAuthError") {
|
|
2115
|
+
const data = e.data;
|
|
2116
|
+
const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
|
|
2117
|
+
return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
|
|
2118
|
+
}
|
|
2119
|
+
if (e.name === "APIError") {
|
|
2120
|
+
const data = e.data;
|
|
2121
|
+
const statusCode = data?.statusCode;
|
|
2122
|
+
if (statusCode === 401 || statusCode === 403) {
|
|
2123
|
+
return {
|
|
2124
|
+
kind: "model_auth",
|
|
2125
|
+
providerId: replyProviderId,
|
|
2126
|
+
modelId: replyModelId,
|
|
2127
|
+
reason: "rejected"
|
|
2128
|
+
};
|
|
2129
|
+
}
|
|
2130
|
+
}
|
|
2131
|
+
return null;
|
|
2132
|
+
}
|
|
2133
|
+
function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
|
|
2134
|
+
if (classified != null) return classified;
|
|
2135
|
+
if (hasConfiguredProvider !== false) return null;
|
|
2136
|
+
return {
|
|
2137
|
+
kind: "model_auth",
|
|
2138
|
+
providerId: replyProviderId,
|
|
2139
|
+
modelId: replyModelId,
|
|
2140
|
+
reason: "missing"
|
|
2141
|
+
};
|
|
2142
|
+
}
|
|
1508
2143
|
function hasRunningAssistantExcept(messages, exceptUserMessageId) {
|
|
1509
2144
|
if (!messages || messages.length === 0) return false;
|
|
1510
2145
|
return messages.some(
|
|
@@ -1738,12 +2373,12 @@ var StreamForwarder = class {
|
|
|
1738
2373
|
let endBody;
|
|
1739
2374
|
if (has_body) {
|
|
1740
2375
|
const chunks = [];
|
|
1741
|
-
bodyPromise = new Promise((
|
|
2376
|
+
bodyPromise = new Promise((resolve3) => {
|
|
1742
2377
|
pushBody = (buf) => {
|
|
1743
2378
|
chunks.push(buf);
|
|
1744
2379
|
};
|
|
1745
2380
|
endBody = () => {
|
|
1746
|
-
|
|
2381
|
+
resolve3(Buffer.concat(chunks));
|
|
1747
2382
|
};
|
|
1748
2383
|
});
|
|
1749
2384
|
}
|
|
@@ -1860,7 +2495,7 @@ function connectTunnel(options) {
|
|
|
1860
2495
|
} = options;
|
|
1861
2496
|
const tunnelUrl = getTunnelUrlConfig();
|
|
1862
2497
|
const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
|
|
1863
|
-
return new Promise((
|
|
2498
|
+
return new Promise((resolve3, reject) => {
|
|
1864
2499
|
const ws = new WebSocket2(url, {
|
|
1865
2500
|
headers: {
|
|
1866
2501
|
Authorization: authHeader
|
|
@@ -1915,7 +2550,7 @@ function connectTunnel(options) {
|
|
|
1915
2550
|
clearTimeout(connectionTimeout);
|
|
1916
2551
|
const connectedAgentId = message.agent_id ?? agentId;
|
|
1917
2552
|
onConnected?.(connectedAgentId);
|
|
1918
|
-
|
|
2553
|
+
resolve3({
|
|
1919
2554
|
ws,
|
|
1920
2555
|
close: () => ws.close(1e3, "CLI shutdown")
|
|
1921
2556
|
});
|
|
@@ -1934,104 +2569,545 @@ function connectTunnel(options) {
|
|
|
1934
2569
|
break;
|
|
1935
2570
|
}
|
|
1936
2571
|
});
|
|
1937
|
-
ws.on("error", (error2) => {
|
|
1938
|
-
clearTimeout(connectionTimeout);
|
|
1939
|
-
const detail = upgradeRejection ?? describeSocketError(error2, url);
|
|
1940
|
-
onError?.(`Connection error: ${detail}`);
|
|
1941
|
-
reject(upgradeRejection ? new Error(upgradeRejection) : new Error(detail));
|
|
2572
|
+
ws.on("error", (error2) => {
|
|
2573
|
+
clearTimeout(connectionTimeout);
|
|
2574
|
+
const detail = upgradeRejection ?? describeSocketError(error2, url);
|
|
2575
|
+
onError?.(`Connection error: ${detail}`);
|
|
2576
|
+
reject(upgradeRejection ? new Error(upgradeRejection) : new Error(detail));
|
|
2577
|
+
});
|
|
2578
|
+
ws.on("close", (code, reason) => {
|
|
2579
|
+
const reasonStr = reason.toString() || upgradeRejection || (code === 1006 ? "abnormal closure" : "No reason provided");
|
|
2580
|
+
forwarder.abortAll();
|
|
2581
|
+
onDisconnected?.(code, reasonStr);
|
|
2582
|
+
});
|
|
2583
|
+
});
|
|
2584
|
+
}
|
|
2585
|
+
|
|
2586
|
+
// src/lib/tunnel/runner-connection.ts
|
|
2587
|
+
var RunnerConnection = class {
|
|
2588
|
+
opts;
|
|
2589
|
+
sleep;
|
|
2590
|
+
connection = null;
|
|
2591
|
+
resolvedAgentId;
|
|
2592
|
+
/** True while a (re)connect loop is in flight. */
|
|
2593
|
+
reconnecting = false;
|
|
2594
|
+
/** The in-flight reconnect promise, awaitable by the caller. */
|
|
2595
|
+
reconnectPromise = null;
|
|
2596
|
+
/** 1-based count of the current reconnect attempt streak. */
|
|
2597
|
+
reconnectAttempt = 0;
|
|
2598
|
+
constructor(opts) {
|
|
2599
|
+
this.opts = opts;
|
|
2600
|
+
this.resolvedAgentId = opts.agentId;
|
|
2601
|
+
this.sleep = opts.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
|
|
2602
|
+
}
|
|
2603
|
+
get agentId() {
|
|
2604
|
+
return this.resolvedAgentId;
|
|
2605
|
+
}
|
|
2606
|
+
/** Establish the initial tunnel connection (with retry/backoff). */
|
|
2607
|
+
async connect() {
|
|
2608
|
+
await this.connectWithRetry(false);
|
|
2609
|
+
}
|
|
2610
|
+
/** Close the active connection (idempotent). */
|
|
2611
|
+
close() {
|
|
2612
|
+
if (this.connection) {
|
|
2613
|
+
try {
|
|
2614
|
+
this.connection.close();
|
|
2615
|
+
} catch (err) {
|
|
2616
|
+
log("error", "runner_connection_close_failed", {
|
|
2617
|
+
agent_id: this.resolvedAgentId,
|
|
2618
|
+
...errorFields(err)
|
|
2619
|
+
});
|
|
2620
|
+
}
|
|
2621
|
+
this.connection = null;
|
|
2622
|
+
}
|
|
2623
|
+
}
|
|
2624
|
+
async connectWithRetry(isReconnect) {
|
|
2625
|
+
if (isReconnect && this.reconnecting) return;
|
|
2626
|
+
this.reconnecting = true;
|
|
2627
|
+
this.close();
|
|
2628
|
+
const { events } = this.opts;
|
|
2629
|
+
while (this.opts.isRunning()) {
|
|
2630
|
+
try {
|
|
2631
|
+
this.connection = await connectTunnel({
|
|
2632
|
+
agentId: this.resolvedAgentId,
|
|
2633
|
+
authHeader: this.opts.getAuthHeader(),
|
|
2634
|
+
port: this.opts.port,
|
|
2635
|
+
onConnected: (agentId) => {
|
|
2636
|
+
this.reconnectAttempt = 0;
|
|
2637
|
+
this.reconnecting = false;
|
|
2638
|
+
this.resolvedAgentId = agentId;
|
|
2639
|
+
events.onConnected(agentId, isReconnect);
|
|
2640
|
+
},
|
|
2641
|
+
onDisconnected: (code, reason) => {
|
|
2642
|
+
events.onDisconnected(code, reason);
|
|
2643
|
+
if (this.opts.isRunning() && code !== 1e3 && !this.reconnecting) {
|
|
2644
|
+
this.reconnectPromise = this.connectWithRetry(true).catch((err) => {
|
|
2645
|
+
events.onError?.(`Reconnection failed: ${err.message}`);
|
|
2646
|
+
});
|
|
2647
|
+
}
|
|
2648
|
+
},
|
|
2649
|
+
onError: (error2) => events.onError?.(error2),
|
|
2650
|
+
onResponse: () => events.onResponse?.(),
|
|
2651
|
+
onDrainPing: () => events.onDrainPing?.(),
|
|
2652
|
+
onInfo: (message) => events.onInfo?.(message)
|
|
2653
|
+
});
|
|
2654
|
+
return;
|
|
2655
|
+
} catch (error2) {
|
|
2656
|
+
this.reconnectAttempt++;
|
|
2657
|
+
if (error2.message === "Unauthorized") {
|
|
2658
|
+
this.reconnecting = false;
|
|
2659
|
+
throw error2;
|
|
2660
|
+
}
|
|
2661
|
+
const delay = getReconnectDelay(this.reconnectAttempt);
|
|
2662
|
+
events.onReconnecting?.(this.reconnectAttempt);
|
|
2663
|
+
events.onError?.(`Connection failed, retrying in ${Math.round(delay / 1e3)}s...`);
|
|
2664
|
+
await this.sleep(delay);
|
|
2665
|
+
}
|
|
2666
|
+
}
|
|
2667
|
+
this.reconnecting = false;
|
|
2668
|
+
}
|
|
2669
|
+
};
|
|
2670
|
+
|
|
2671
|
+
// src/lib/tunnel/ready-marker.ts
|
|
2672
|
+
import { writeFileSync } from "fs";
|
|
2673
|
+
function writeTunnelReadyMarker(path, agentId) {
|
|
2674
|
+
try {
|
|
2675
|
+
writeFileSync(path, `${agentId}
|
|
2676
|
+
`);
|
|
2677
|
+
return { ok: true };
|
|
2678
|
+
} catch (error2) {
|
|
2679
|
+
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
2680
|
+
}
|
|
2681
|
+
}
|
|
2682
|
+
|
|
2683
|
+
// src/lib/claude-usage-reporting.ts
|
|
2684
|
+
var VALID_MODES = ["auto", "on", "off"];
|
|
2685
|
+
function resolveClaudeUsageReportingMode(flagValue, env) {
|
|
2686
|
+
const raw = flagValue ?? env.EVIDENT_CLAUDE_USAGE_REPORTING;
|
|
2687
|
+
if (raw === void 0 || raw === "") {
|
|
2688
|
+
return { mode: "auto", warnings: [] };
|
|
2689
|
+
}
|
|
2690
|
+
const normalized = raw.trim().toLowerCase();
|
|
2691
|
+
if (VALID_MODES.includes(normalized)) {
|
|
2692
|
+
return { mode: normalized, warnings: [] };
|
|
2693
|
+
}
|
|
2694
|
+
const source = flagValue !== void 0 ? "--claude-usage-reporting" : "EVIDENT_CLAUDE_USAGE_REPORTING";
|
|
2695
|
+
return {
|
|
2696
|
+
mode: "auto",
|
|
2697
|
+
warnings: [
|
|
2698
|
+
`Ignoring invalid ${source} "${raw}": expected one of ${VALID_MODES.join(", ")}; using auto`
|
|
2699
|
+
]
|
|
2700
|
+
};
|
|
2701
|
+
}
|
|
2702
|
+
var BASE_REPORT_DELAY_MS = 10 * 6e4;
|
|
2703
|
+
var REPORT_DELAY_JITTER_FRACTION = 0.2;
|
|
2704
|
+
function nextReportDelayMs(random = Math.random) {
|
|
2705
|
+
const jitterRangeMs = BASE_REPORT_DELAY_MS * REPORT_DELAY_JITTER_FRACTION;
|
|
2706
|
+
return BASE_REPORT_DELAY_MS - jitterRangeMs + random() * (2 * jitterRangeMs);
|
|
2707
|
+
}
|
|
2708
|
+
var FIRST_REPORT_DELAY_MS = 5e3 + Math.random() * 1e4;
|
|
2709
|
+
var CLAUDE_USAGE_FAILURE_REESCALATION_TICKS = 6;
|
|
2710
|
+
function claudeUsageFailureLogLevel(consecutiveFailures) {
|
|
2711
|
+
return consecutiveFailures === 1 || consecutiveFailures % CLAUDE_USAGE_FAILURE_REESCALATION_TICKS === 0 ? "warn" : "debug";
|
|
2712
|
+
}
|
|
2713
|
+
|
|
2714
|
+
// src/lib/channels/driver.ts
|
|
2715
|
+
import { homedir as homedir2 } from "os";
|
|
2716
|
+
|
|
2717
|
+
// src/lib/file-push.ts
|
|
2718
|
+
import { randomUUID } from "crypto";
|
|
2719
|
+
import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
|
|
2720
|
+
import { basename, dirname as dirname2, isAbsolute, join as join2, relative, resolve as resolve2, sep } from "path";
|
|
2721
|
+
var FILE_MODE = 384;
|
|
2722
|
+
var DIRECTORY_MODE = 448;
|
|
2723
|
+
async function writePushedFile(request) {
|
|
2724
|
+
const { requestedPath, content, allowedDirectories, homeDir } = request;
|
|
2725
|
+
const bytes = content.byteLength;
|
|
2726
|
+
if (allowedDirectories.length === 0) {
|
|
2727
|
+
return refuse("file_sync_disabled", "File sync is not enabled on this runner.", {
|
|
2728
|
+
path: requestedPath,
|
|
2729
|
+
bytes
|
|
2730
|
+
});
|
|
2731
|
+
}
|
|
2732
|
+
if (bytes > MAX_FILE_PUSH_BYTES) {
|
|
2733
|
+
return refuse(
|
|
2734
|
+
"file_too_large",
|
|
2735
|
+
`File is ${bytes} bytes; the limit is ${MAX_FILE_PUSH_BYTES}.`,
|
|
2736
|
+
{
|
|
2737
|
+
path: requestedPath,
|
|
2738
|
+
bytes
|
|
2739
|
+
}
|
|
2740
|
+
);
|
|
2741
|
+
}
|
|
2742
|
+
const candidate = expandAndValidate(requestedPath, homeDir);
|
|
2743
|
+
if (candidate === null) {
|
|
2744
|
+
return refuse("invalid_path", "The requested path is not a valid absolute file path.", {
|
|
2745
|
+
path: requestedPath,
|
|
2746
|
+
bytes
|
|
2747
|
+
});
|
|
2748
|
+
}
|
|
2749
|
+
try {
|
|
2750
|
+
const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
|
|
2751
|
+
dirname2(candidate)
|
|
2752
|
+
);
|
|
2753
|
+
const realTarget = join2(existingAncestor, ...missingSegments, basename(candidate));
|
|
2754
|
+
const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
|
|
2755
|
+
if (allowedDirectory === null) {
|
|
2756
|
+
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
2757
|
+
path: realTarget,
|
|
2758
|
+
bytes
|
|
2759
|
+
});
|
|
2760
|
+
}
|
|
2761
|
+
if (missingSegments.length > 0) {
|
|
2762
|
+
await createMissingDirectories(existingAncestor, missingSegments);
|
|
2763
|
+
const realParent = await realpath(dirname2(realTarget));
|
|
2764
|
+
if (realParent !== dirname2(realTarget) || !contains(allowedDirectory, realTarget)) {
|
|
2765
|
+
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
2766
|
+
path: realTarget,
|
|
2767
|
+
bytes,
|
|
2768
|
+
reason: "parent_changed_after_create"
|
|
2769
|
+
});
|
|
2770
|
+
}
|
|
2771
|
+
}
|
|
2772
|
+
await writeAtomically(realTarget, content);
|
|
2773
|
+
log("info", "file_push_written", { path: realTarget, bytes });
|
|
2774
|
+
return { ok: true, path: realTarget };
|
|
2775
|
+
} catch (err) {
|
|
2776
|
+
const errno = err.code ?? "UNKNOWN";
|
|
2777
|
+
return refuse("write_failed", `The runner could not write the file (${errno}).`, {
|
|
2778
|
+
path: candidate,
|
|
2779
|
+
bytes,
|
|
2780
|
+
errno,
|
|
2781
|
+
...errorFields(err)
|
|
2782
|
+
});
|
|
2783
|
+
}
|
|
2784
|
+
}
|
|
2785
|
+
function expandAndValidate(requestedPath, homeDir) {
|
|
2786
|
+
if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
|
|
2787
|
+
return null;
|
|
2788
|
+
}
|
|
2789
|
+
const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join2(homeDir, requestedPath.slice(2)) : requestedPath;
|
|
2790
|
+
if (expanded.split(/[/\\]/).includes("..")) {
|
|
2791
|
+
return null;
|
|
2792
|
+
}
|
|
2793
|
+
if (!isAbsolute(expanded)) {
|
|
2794
|
+
return null;
|
|
2795
|
+
}
|
|
2796
|
+
const candidate = resolve2(expanded);
|
|
2797
|
+
const name = basename(candidate);
|
|
2798
|
+
return name === "" || name === "." || name === ".." ? null : candidate;
|
|
2799
|
+
}
|
|
2800
|
+
async function resolveNearestExistingAncestor(directory) {
|
|
2801
|
+
const missingSegments = [];
|
|
2802
|
+
let current = directory;
|
|
2803
|
+
for (; ; ) {
|
|
2804
|
+
try {
|
|
2805
|
+
return { existingAncestor: await realpath(current), missingSegments };
|
|
2806
|
+
} catch (err) {
|
|
2807
|
+
const parent = dirname2(current);
|
|
2808
|
+
if (err.code !== "ENOENT" || parent === current) {
|
|
2809
|
+
throw err;
|
|
2810
|
+
}
|
|
2811
|
+
missingSegments.unshift(basename(current));
|
|
2812
|
+
current = parent;
|
|
2813
|
+
}
|
|
2814
|
+
}
|
|
2815
|
+
}
|
|
2816
|
+
async function findContainingAllowedDirectory(allowedDirectories, realTarget) {
|
|
2817
|
+
for (const directory of allowedDirectories) {
|
|
2818
|
+
if (!isAbsolute(directory)) {
|
|
2819
|
+
log("warn", "file_push_allowed_directory_skipped", { directory, reason: "not_absolute" });
|
|
2820
|
+
continue;
|
|
2821
|
+
}
|
|
2822
|
+
const realDirectory = await realpathCreatingIfMissing(directory);
|
|
2823
|
+
if (realDirectory !== null && contains(realDirectory, realTarget)) {
|
|
2824
|
+
return realDirectory;
|
|
2825
|
+
}
|
|
2826
|
+
}
|
|
2827
|
+
return null;
|
|
2828
|
+
}
|
|
2829
|
+
async function realpathCreatingIfMissing(directory) {
|
|
2830
|
+
try {
|
|
2831
|
+
return await realpath(directory);
|
|
2832
|
+
} catch (err) {
|
|
2833
|
+
if (err.code !== "ENOENT") {
|
|
2834
|
+
log("warn", "file_push_allowed_directory_skipped", {
|
|
2835
|
+
directory,
|
|
2836
|
+
reason: "unresolvable",
|
|
2837
|
+
...errorFields(err)
|
|
2838
|
+
});
|
|
2839
|
+
return null;
|
|
2840
|
+
}
|
|
2841
|
+
}
|
|
2842
|
+
try {
|
|
2843
|
+
await mkdir(directory, { recursive: true, mode: DIRECTORY_MODE });
|
|
2844
|
+
await chmod(directory, DIRECTORY_MODE);
|
|
2845
|
+
return await realpath(directory);
|
|
2846
|
+
} catch (err) {
|
|
2847
|
+
log("warn", "file_push_allowed_directory_skipped", {
|
|
2848
|
+
directory,
|
|
2849
|
+
reason: "create_failed",
|
|
2850
|
+
...errorFields(err)
|
|
2851
|
+
});
|
|
2852
|
+
return null;
|
|
2853
|
+
}
|
|
2854
|
+
}
|
|
2855
|
+
function contains(realDirectory, realTarget) {
|
|
2856
|
+
const rel = relative(realDirectory, realTarget);
|
|
2857
|
+
return rel !== "" && rel !== ".." && !rel.startsWith(`..${sep}`) && !isAbsolute(rel);
|
|
2858
|
+
}
|
|
2859
|
+
async function createMissingDirectories(existingAncestor, missingSegments) {
|
|
2860
|
+
let current = existingAncestor;
|
|
2861
|
+
for (const segment of missingSegments) {
|
|
2862
|
+
current = join2(current, segment);
|
|
2863
|
+
await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
|
|
2864
|
+
await chmod(current, DIRECTORY_MODE);
|
|
2865
|
+
}
|
|
2866
|
+
}
|
|
2867
|
+
async function writeAtomically(realTarget, content) {
|
|
2868
|
+
const temporaryPath = join2(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
|
|
2869
|
+
let handle;
|
|
2870
|
+
try {
|
|
2871
|
+
handle = await open2(temporaryPath, "wx", FILE_MODE);
|
|
2872
|
+
await handle.writeFile(content);
|
|
2873
|
+
await handle.chmod(FILE_MODE);
|
|
2874
|
+
await handle.close();
|
|
2875
|
+
handle = void 0;
|
|
2876
|
+
await rename(temporaryPath, realTarget);
|
|
2877
|
+
} catch (err) {
|
|
2878
|
+
await discardTemporaryFile(temporaryPath, handle);
|
|
2879
|
+
throw err;
|
|
2880
|
+
}
|
|
2881
|
+
}
|
|
2882
|
+
async function discardTemporaryFile(temporaryPath, handle) {
|
|
2883
|
+
try {
|
|
2884
|
+
await handle?.close();
|
|
2885
|
+
} catch (err) {
|
|
2886
|
+
log("warn", "file_push_temp_close_failed", { path: temporaryPath, ...errorFields(err) });
|
|
2887
|
+
}
|
|
2888
|
+
try {
|
|
2889
|
+
await unlink(temporaryPath);
|
|
2890
|
+
} catch (err) {
|
|
2891
|
+
const errno = err.code;
|
|
2892
|
+
if (errno !== "ENOENT" && errno !== "ENOTDIR") {
|
|
2893
|
+
log("warn", "file_push_temp_cleanup_failed", { path: temporaryPath, ...errorFields(err) });
|
|
2894
|
+
}
|
|
2895
|
+
}
|
|
2896
|
+
}
|
|
2897
|
+
function refuse(code, message, fields) {
|
|
2898
|
+
log(code === "write_failed" ? "error" : "warn", "file_push_refused", { code, ...fields });
|
|
2899
|
+
return { ok: false, code, message };
|
|
2900
|
+
}
|
|
2901
|
+
|
|
2902
|
+
// src/lib/runner-file-sync.ts
|
|
2903
|
+
var MAX_ACK_ATTEMPTS = 5;
|
|
2904
|
+
async function syncPendingRunnerFiles(options) {
|
|
2905
|
+
const pending = await listPendingFiles(options);
|
|
2906
|
+
const pendingIds = new Set(pending.map((file) => file.id));
|
|
2907
|
+
for (const id of options.ackFailures.keys()) {
|
|
2908
|
+
if (!pendingIds.has(id)) options.ackFailures.delete(id);
|
|
2909
|
+
}
|
|
2910
|
+
if (pending.length === 0) return 0;
|
|
2911
|
+
options.log({
|
|
2912
|
+
level: "info",
|
|
2913
|
+
message: `Runner file sync: ${pending.length} file(s) queued for this runner`
|
|
2914
|
+
});
|
|
2915
|
+
let applied = 0;
|
|
2916
|
+
for (const file of pending) {
|
|
2917
|
+
if ((options.ackFailures.get(file.id) ?? 0) >= MAX_ACK_ATTEMPTS) continue;
|
|
2918
|
+
if (await applyOne(options, file)) applied += 1;
|
|
2919
|
+
}
|
|
2920
|
+
return applied;
|
|
2921
|
+
}
|
|
2922
|
+
async function listPendingFiles(options) {
|
|
2923
|
+
let res;
|
|
2924
|
+
try {
|
|
2925
|
+
res = await options.fetchImpl(`${options.apiUrl}/runners/${options.agentId}/files/pending`, {
|
|
2926
|
+
headers: { Authorization: options.getAuthHeader() }
|
|
2927
|
+
});
|
|
2928
|
+
} catch (err) {
|
|
2929
|
+
options.log({
|
|
2930
|
+
level: "warn",
|
|
2931
|
+
message: `Could not list pending runner files \u2014 retrying on the next drain: ${describe(err)}`
|
|
1942
2932
|
});
|
|
1943
|
-
|
|
1944
|
-
|
|
1945
|
-
|
|
1946
|
-
|
|
2933
|
+
return [];
|
|
2934
|
+
}
|
|
2935
|
+
if (!res.ok) {
|
|
2936
|
+
options.log({
|
|
2937
|
+
level: res.status === 404 ? "debug" : "warn",
|
|
2938
|
+
message: `Listing pending runner files returned HTTP ${res.status} \u2014 retrying on the next drain`
|
|
1947
2939
|
});
|
|
1948
|
-
|
|
2940
|
+
return [];
|
|
2941
|
+
}
|
|
2942
|
+
let body;
|
|
2943
|
+
try {
|
|
2944
|
+
body = await res.json();
|
|
2945
|
+
} catch (err) {
|
|
2946
|
+
options.log({
|
|
2947
|
+
level: "warn",
|
|
2948
|
+
message: `Pending runner file list was not readable JSON \u2014 retrying on the next drain: ${describe(err)}`
|
|
2949
|
+
});
|
|
2950
|
+
return [];
|
|
2951
|
+
}
|
|
2952
|
+
if (!Array.isArray(body)) {
|
|
2953
|
+
options.log({
|
|
2954
|
+
level: "warn",
|
|
2955
|
+
message: "Pending runner file list was not an array \u2014 ignoring it for this drain"
|
|
2956
|
+
});
|
|
2957
|
+
return [];
|
|
2958
|
+
}
|
|
2959
|
+
const files = [];
|
|
2960
|
+
for (const entry of body) {
|
|
2961
|
+
const file = asPendingFile(entry);
|
|
2962
|
+
if (file === null) {
|
|
2963
|
+
options.log({
|
|
2964
|
+
level: "warn",
|
|
2965
|
+
message: "Ignoring a malformed pending runner file entry (expected id, path and size)"
|
|
2966
|
+
});
|
|
2967
|
+
continue;
|
|
2968
|
+
}
|
|
2969
|
+
files.push(file);
|
|
2970
|
+
}
|
|
2971
|
+
return files;
|
|
1949
2972
|
}
|
|
1950
|
-
|
|
1951
|
-
|
|
1952
|
-
|
|
1953
|
-
|
|
1954
|
-
|
|
1955
|
-
|
|
1956
|
-
|
|
1957
|
-
|
|
1958
|
-
|
|
1959
|
-
|
|
1960
|
-
|
|
1961
|
-
|
|
1962
|
-
|
|
1963
|
-
|
|
1964
|
-
|
|
1965
|
-
|
|
1966
|
-
|
|
2973
|
+
function asPendingFile(entry) {
|
|
2974
|
+
if (entry === null || typeof entry !== "object") return null;
|
|
2975
|
+
const { id, path, size } = entry;
|
|
2976
|
+
if (typeof id !== "string" || id === "") return null;
|
|
2977
|
+
if (typeof path !== "string" || path === "") return null;
|
|
2978
|
+
if (typeof size !== "number" || !Number.isFinite(size) || size < 0) return null;
|
|
2979
|
+
return { id, path, size };
|
|
2980
|
+
}
|
|
2981
|
+
async function applyOne(options, file) {
|
|
2982
|
+
const label = `${file.id.slice(0, 8)} (${file.path})`;
|
|
2983
|
+
if (options.allowedDirectories.length === 0) {
|
|
2984
|
+
options.log({
|
|
2985
|
+
level: "warn",
|
|
2986
|
+
message: `Runner file ${label} rejected: file sync is not enabled on this runner (start it with --enable-file-sync-to)`
|
|
2987
|
+
});
|
|
2988
|
+
await ack(options, file, "rejected", "file_sync_disabled");
|
|
2989
|
+
return false;
|
|
1967
2990
|
}
|
|
1968
|
-
|
|
1969
|
-
|
|
2991
|
+
if (file.size > MAX_FILE_PUSH_BYTES) {
|
|
2992
|
+
options.log({
|
|
2993
|
+
level: "warn",
|
|
2994
|
+
message: `Runner file ${label} rejected: declared ${file.size} bytes, the limit is ${MAX_FILE_PUSH_BYTES}`
|
|
2995
|
+
});
|
|
2996
|
+
await ack(options, file, "rejected", "file_too_large");
|
|
2997
|
+
return false;
|
|
1970
2998
|
}
|
|
1971
|
-
|
|
1972
|
-
|
|
1973
|
-
await
|
|
2999
|
+
const download = await downloadContent(options, file, label);
|
|
3000
|
+
if (!download.ok) {
|
|
3001
|
+
if (download.terminal) await ack(options, file, "rejected", download.code);
|
|
3002
|
+
return false;
|
|
1974
3003
|
}
|
|
1975
|
-
|
|
1976
|
-
|
|
1977
|
-
|
|
1978
|
-
|
|
1979
|
-
|
|
1980
|
-
|
|
1981
|
-
|
|
1982
|
-
|
|
1983
|
-
|
|
3004
|
+
let outcome;
|
|
3005
|
+
try {
|
|
3006
|
+
outcome = await writePushedFile({
|
|
3007
|
+
requestedPath: file.path,
|
|
3008
|
+
content: download.content,
|
|
3009
|
+
allowedDirectories: options.allowedDirectories,
|
|
3010
|
+
homeDir: options.homeDir
|
|
3011
|
+
});
|
|
3012
|
+
} catch (err) {
|
|
3013
|
+
options.log({
|
|
3014
|
+
level: "error",
|
|
3015
|
+
message: `Runner file ${label} could not be written: ${describe(err)}`
|
|
3016
|
+
});
|
|
3017
|
+
await ack(options, file, "rejected", "write_failed");
|
|
3018
|
+
return false;
|
|
3019
|
+
}
|
|
3020
|
+
if (!outcome.ok) {
|
|
3021
|
+
options.log({
|
|
3022
|
+
level: "warn",
|
|
3023
|
+
message: `Runner file ${label} rejected (${outcome.code}): ${outcome.message}`
|
|
3024
|
+
});
|
|
3025
|
+
await ack(options, file, "rejected", outcome.code);
|
|
3026
|
+
return false;
|
|
3027
|
+
}
|
|
3028
|
+
options.log({
|
|
3029
|
+
level: "info",
|
|
3030
|
+
message: `Runner file ${label} applied (${download.content.byteLength} bytes)`
|
|
3031
|
+
});
|
|
3032
|
+
await ack(options, file, "applied");
|
|
3033
|
+
return true;
|
|
3034
|
+
}
|
|
3035
|
+
function durableDownloadCode(status2) {
|
|
3036
|
+
return status2 === 413 ? "file_too_large" : "write_failed";
|
|
3037
|
+
}
|
|
3038
|
+
async function downloadContent(options, file, label) {
|
|
3039
|
+
try {
|
|
3040
|
+
const res = await options.fetchImpl(
|
|
3041
|
+
`${options.apiUrl}/runners/${options.agentId}/files/${file.id}/content`,
|
|
3042
|
+
{ headers: { Authorization: options.getAuthHeader() } }
|
|
3043
|
+
);
|
|
3044
|
+
if (!res.ok) {
|
|
3045
|
+
const terminal = res.status >= 400 && res.status < 500 && res.status !== 401 && res.status !== 403 && res.status !== 408 && res.status !== 429;
|
|
3046
|
+
if (!terminal) {
|
|
3047
|
+
options.log({
|
|
3048
|
+
level: "warn",
|
|
3049
|
+
message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 retrying on the next drain`
|
|
1984
3050
|
});
|
|
3051
|
+
return { ok: false, terminal: false };
|
|
1985
3052
|
}
|
|
1986
|
-
|
|
3053
|
+
const code = durableDownloadCode(res.status);
|
|
3054
|
+
options.log({
|
|
3055
|
+
level: "error",
|
|
3056
|
+
message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 rejecting it as ${code} (the bytes never reached the writer)`
|
|
3057
|
+
});
|
|
3058
|
+
return { ok: false, terminal: true, code };
|
|
1987
3059
|
}
|
|
3060
|
+
return { ok: true, content: Buffer.from(await res.arrayBuffer()) };
|
|
3061
|
+
} catch (err) {
|
|
3062
|
+
options.log({
|
|
3063
|
+
level: "warn",
|
|
3064
|
+
message: `Downloading runner file ${label} failed \u2014 retrying on the next drain: ${describe(err)}`
|
|
3065
|
+
});
|
|
3066
|
+
return { ok: false, terminal: false };
|
|
1988
3067
|
}
|
|
1989
|
-
|
|
1990
|
-
|
|
1991
|
-
|
|
1992
|
-
|
|
1993
|
-
const
|
|
1994
|
-
|
|
1995
|
-
|
|
1996
|
-
|
|
1997
|
-
|
|
1998
|
-
|
|
1999
|
-
|
|
2000
|
-
|
|
2001
|
-
|
|
2002
|
-
this.reconnecting = false;
|
|
2003
|
-
this.resolvedAgentId = agentId;
|
|
2004
|
-
events.onConnected(agentId, isReconnect);
|
|
2005
|
-
},
|
|
2006
|
-
onDisconnected: (code, reason) => {
|
|
2007
|
-
events.onDisconnected(code, reason);
|
|
2008
|
-
if (this.opts.isRunning() && code !== 1e3 && !this.reconnecting) {
|
|
2009
|
-
this.reconnectPromise = this.connectWithRetry(true).catch((err) => {
|
|
2010
|
-
events.onError?.(`Reconnection failed: ${err.message}`);
|
|
2011
|
-
});
|
|
2012
|
-
}
|
|
2013
|
-
},
|
|
2014
|
-
onError: (error2) => events.onError?.(error2),
|
|
2015
|
-
onResponse: () => events.onResponse?.(),
|
|
2016
|
-
onDrainPing: () => events.onDrainPing?.(),
|
|
2017
|
-
onInfo: (message) => events.onInfo?.(message)
|
|
2018
|
-
});
|
|
2019
|
-
return;
|
|
2020
|
-
} catch (error2) {
|
|
2021
|
-
this.reconnectAttempt++;
|
|
2022
|
-
if (error2.message === "Unauthorized") {
|
|
2023
|
-
this.reconnecting = false;
|
|
2024
|
-
throw error2;
|
|
2025
|
-
}
|
|
2026
|
-
const delay = getReconnectDelay(this.reconnectAttempt);
|
|
2027
|
-
events.onReconnecting?.(this.reconnectAttempt);
|
|
2028
|
-
events.onError?.(`Connection failed, retrying in ${Math.round(delay / 1e3)}s...`);
|
|
2029
|
-
await this.sleep(delay);
|
|
3068
|
+
}
|
|
3069
|
+
async function ack(options, file, status2, reason) {
|
|
3070
|
+
const outcome = `${status2}${reason ? ` (${reason})` : ""}`;
|
|
3071
|
+
try {
|
|
3072
|
+
const res = await options.fetchImpl(
|
|
3073
|
+
`${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
|
|
3074
|
+
{
|
|
3075
|
+
method: "POST",
|
|
3076
|
+
headers: {
|
|
3077
|
+
Authorization: options.getAuthHeader(),
|
|
3078
|
+
"Content-Type": "application/json"
|
|
3079
|
+
},
|
|
3080
|
+
body: JSON.stringify(reason ? { status: status2, reason } : { status: status2 })
|
|
2030
3081
|
}
|
|
3082
|
+
);
|
|
3083
|
+
if (!res.ok) {
|
|
3084
|
+
recordAckFailure(
|
|
3085
|
+
options,
|
|
3086
|
+
file,
|
|
3087
|
+
`Acking runner file ${file.id.slice(0, 8)} as ${outcome} returned HTTP ${res.status}`
|
|
3088
|
+
);
|
|
3089
|
+
return;
|
|
2031
3090
|
}
|
|
2032
|
-
|
|
3091
|
+
options.ackFailures.delete(file.id);
|
|
3092
|
+
} catch (err) {
|
|
3093
|
+
recordAckFailure(
|
|
3094
|
+
options,
|
|
3095
|
+
file,
|
|
3096
|
+
`Acking runner file ${file.id.slice(0, 8)} as ${outcome} failed: ${describe(err)}`
|
|
3097
|
+
);
|
|
2033
3098
|
}
|
|
2034
|
-
}
|
|
3099
|
+
}
|
|
3100
|
+
function recordAckFailure(options, file, what) {
|
|
3101
|
+
const attempts = (options.ackFailures.get(file.id) ?? 0) + 1;
|
|
3102
|
+
options.ackFailures.set(file.id, attempts);
|
|
3103
|
+
options.log({
|
|
3104
|
+
level: "error",
|
|
3105
|
+
message: attempts >= MAX_ACK_ATTEMPTS ? `${what} \u2014 giving up after ${attempts} attempts. It stays pending until the server expires it; restart the runner to retry.` : `${what} \u2014 it stays pending until a later drain re-acks it (attempt ${attempts} of ${MAX_ACK_ATTEMPTS})`
|
|
3106
|
+
});
|
|
3107
|
+
}
|
|
3108
|
+
function describe(err) {
|
|
3109
|
+
return err instanceof Error ? err.message : String(err);
|
|
3110
|
+
}
|
|
2035
3111
|
|
|
2036
3112
|
// src/lib/channels/driver.ts
|
|
2037
3113
|
function messageIdOf(m) {
|
|
@@ -2061,6 +3137,8 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
|
|
|
2061
3137
|
var DEFAULT_STUCK_QUEUED_MS = 6e4;
|
|
2062
3138
|
var HEARTBEAT_MS = 6e4;
|
|
2063
3139
|
var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
|
|
3140
|
+
var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
|
|
3141
|
+
var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
|
|
2064
3142
|
var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
|
|
2065
3143
|
var MAX_SUPERSEDED_CONVERSATIONS = 256;
|
|
2066
3144
|
var ChannelAuthError = class extends Error {
|
|
@@ -2071,10 +3149,10 @@ var ChannelAuthError = class extends Error {
|
|
|
2071
3149
|
};
|
|
2072
3150
|
var ChannelTerminalError = class extends Error {
|
|
2073
3151
|
status;
|
|
2074
|
-
constructor(message,
|
|
3152
|
+
constructor(message, status2) {
|
|
2075
3153
|
super(message);
|
|
2076
3154
|
this.name = "ChannelTerminalError";
|
|
2077
|
-
this.status =
|
|
3155
|
+
this.status = status2;
|
|
2078
3156
|
}
|
|
2079
3157
|
};
|
|
2080
3158
|
function backoffDelay(attempt, policy) {
|
|
@@ -2082,8 +3160,8 @@ function backoffDelay(attempt, policy) {
|
|
|
2082
3160
|
const capped = Math.min(policy.maxDelayMs, exp);
|
|
2083
3161
|
return Math.floor(Math.random() * capped);
|
|
2084
3162
|
}
|
|
2085
|
-
function isRetryableStatus(
|
|
2086
|
-
return
|
|
3163
|
+
function isRetryableStatus(status2) {
|
|
3164
|
+
return status2 === 429 || status2 >= 500 && status2 <= 599;
|
|
2087
3165
|
}
|
|
2088
3166
|
var ChannelDriver = class _ChannelDriver {
|
|
2089
3167
|
agentId;
|
|
@@ -2099,6 +3177,8 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2099
3177
|
pausedMaxWaitMs;
|
|
2100
3178
|
stuckQueuedMs;
|
|
2101
3179
|
now;
|
|
3180
|
+
fileSyncDirectories;
|
|
3181
|
+
homeDir;
|
|
2102
3182
|
/** Cache of conversationId → opencode sessionId. */
|
|
2103
3183
|
sessions = /* @__PURE__ */ new Map();
|
|
2104
3184
|
/**
|
|
@@ -2196,6 +3276,24 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2196
3276
|
* processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
|
|
2197
3277
|
*/
|
|
2198
3278
|
readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
3279
|
+
/**
|
|
3280
|
+
* "Already emitted `redrive_unresolved` for this row" (#965). Mirrors
|
|
3281
|
+
* `readoptPollUnresolvedSignalled`: `resolveRedrive`'s `unresolved` leaf recurs
|
|
3282
|
+
* every ~2s drain until opencode's status becomes readable, but the
|
|
3283
|
+
* server-visible signal is an OUTCOME, so it fires at most once per row. Cleared
|
|
3284
|
+
* on any non-`unresolved` outcome so the set cannot grow beyond the currently
|
|
3285
|
+
* unresolvable rows.
|
|
3286
|
+
*/
|
|
3287
|
+
redriveUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
3288
|
+
/**
|
|
3289
|
+
* First `now()` a `pending` row's re-drive was observed `unresolved` (#965). A
|
|
3290
|
+
* `pending` row is invisible to every cron arm (all require `status =
|
|
3291
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
3292
|
+
* nothing driving it. Once `now - since >= pausedMaxWaitMs`, `resolveRedrive`
|
|
3293
|
+
* takes `dispatch` instead of `unresolved` (reusing the existing knob — see
|
|
3294
|
+
* ADR-0047's own "unreachable ⇒ bounded" rule). Cleared on any other outcome.
|
|
3295
|
+
*/
|
|
3296
|
+
redriveUnresolvedSince = /* @__PURE__ */ new Map();
|
|
2199
3297
|
/**
|
|
2200
3298
|
* "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
|
|
2201
3299
|
* Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
|
|
@@ -2251,6 +3349,24 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2251
3349
|
sessionTitles = /* @__PURE__ */ new Map();
|
|
2252
3350
|
/** Serialises drains so a reconnect during a drain doesn't double-process. */
|
|
2253
3351
|
draining = false;
|
|
3352
|
+
/**
|
|
3353
|
+
* Serialises runner-file syncs (#559) so the ~2s poll tick and a concurrent
|
|
3354
|
+
* drain ping don't download, write and ack the same file twice.
|
|
3355
|
+
*/
|
|
3356
|
+
syncingFiles = false;
|
|
3357
|
+
/**
|
|
3358
|
+
* Consecutive failed acks per pending file (#559). Lives on the driver so it
|
|
3359
|
+
* survives across drains — without it, a file whose ack keeps failing is
|
|
3360
|
+
* re-downloaded and re-written every ~2s until the server expires it.
|
|
3361
|
+
*/
|
|
3362
|
+
fileAckFailures = /* @__PURE__ */ new Map();
|
|
3363
|
+
/**
|
|
3364
|
+
* Monotonic count of files this runner has pulled and written (#559). Only
|
|
3365
|
+
* ever increases, so `run.ts` detects work by comparing it against the value
|
|
3366
|
+
* it saw on the previous cycle — including work that landed mid-sleep, the
|
|
3367
|
+
* same trick `lastProxiedActivityAt` uses.
|
|
3368
|
+
*/
|
|
3369
|
+
appliedFileCount = 0;
|
|
2254
3370
|
/**
|
|
2255
3371
|
* The currently-executing `drainPending()` promise, or null when idle. Lets a
|
|
2256
3372
|
* graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
|
|
@@ -2266,21 +3382,23 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2266
3382
|
* and stops opencode.
|
|
2267
3383
|
*/
|
|
2268
3384
|
stopped = false;
|
|
2269
|
-
constructor(
|
|
2270
|
-
this.agentId =
|
|
2271
|
-
this.port =
|
|
2272
|
-
this.apiUrl =
|
|
2273
|
-
this.getAuthHeader =
|
|
2274
|
-
this.conversationFilter =
|
|
2275
|
-
this.retry = { ...DEFAULT_RETRY_POLICY, ...
|
|
2276
|
-
this.log =
|
|
3385
|
+
constructor(config) {
|
|
3386
|
+
this.agentId = config.agentId;
|
|
3387
|
+
this.port = config.port;
|
|
3388
|
+
this.apiUrl = config.apiUrl.replace(/\/$/, "");
|
|
3389
|
+
this.getAuthHeader = config.getAuthHeader;
|
|
3390
|
+
this.conversationFilter = config.conversationFilter ?? null;
|
|
3391
|
+
this.retry = { ...DEFAULT_RETRY_POLICY, ...config.retry };
|
|
3392
|
+
this.log = config.log ?? (() => {
|
|
2277
3393
|
});
|
|
2278
|
-
this.fetchImpl =
|
|
2279
|
-
this.sleep =
|
|
2280
|
-
this.pausedPollIntervalMs =
|
|
2281
|
-
this.pausedMaxWaitMs =
|
|
2282
|
-
this.stuckQueuedMs =
|
|
2283
|
-
this.now =
|
|
3394
|
+
this.fetchImpl = config.fetchImpl ?? fetch;
|
|
3395
|
+
this.sleep = config.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
|
|
3396
|
+
this.pausedPollIntervalMs = config.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
|
|
3397
|
+
this.pausedMaxWaitMs = config.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
|
|
3398
|
+
this.stuckQueuedMs = config.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
|
|
3399
|
+
this.now = config.now ?? (() => Date.now());
|
|
3400
|
+
this.fileSyncDirectories = config.fileSyncDirectories ?? [];
|
|
3401
|
+
this.homeDir = config.homeDir ?? homedir2();
|
|
2284
3402
|
}
|
|
2285
3403
|
/** The IPv4-loopback base URL for the local `opencode serve`. */
|
|
2286
3404
|
get opencodeBase() {
|
|
@@ -2308,6 +3426,47 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2308
3426
|
);
|
|
2309
3427
|
return run2;
|
|
2310
3428
|
}
|
|
3429
|
+
/**
|
|
3430
|
+
* Pull-and-apply any files Evident has queued for this runner (#559), riding
|
|
3431
|
+
* the EXISTING drain cycle — `run.ts` calls it from the same ~2s channel poll
|
|
3432
|
+
* and drain ping that call `drainPending()`. There is deliberately no channel,
|
|
3433
|
+
* control frame or poll loop of its own: worst-case latency is one poll tick.
|
|
3434
|
+
*
|
|
3435
|
+
* NEVER throws and never surfaces a `ChannelAuthError`: a file failure must not
|
|
3436
|
+
* cost a conversation turn. Failures are logged and either acked as a terminal
|
|
3437
|
+
* outcome or left pending for the next drain (see `runner-file-sync.ts`).
|
|
3438
|
+
*
|
|
3439
|
+
* Re-entrant calls are skipped (the poll tick and a drain ping can overlap).
|
|
3440
|
+
*
|
|
3441
|
+
* @returns the number of files written to disk.
|
|
3442
|
+
*/
|
|
3443
|
+
async syncPendingFiles() {
|
|
3444
|
+
if (this.stopped) return 0;
|
|
3445
|
+
if (this.syncingFiles) return 0;
|
|
3446
|
+
this.syncingFiles = true;
|
|
3447
|
+
try {
|
|
3448
|
+
const applied = await syncPendingRunnerFiles({
|
|
3449
|
+
agentId: this.agentId,
|
|
3450
|
+
apiUrl: this.apiUrl,
|
|
3451
|
+
getAuthHeader: this.getAuthHeader,
|
|
3452
|
+
fetchImpl: this.fetchImpl,
|
|
3453
|
+
allowedDirectories: this.fileSyncDirectories,
|
|
3454
|
+
homeDir: this.homeDir,
|
|
3455
|
+
ackFailures: this.fileAckFailures,
|
|
3456
|
+
log: this.log
|
|
3457
|
+
});
|
|
3458
|
+
this.appliedFileCount += applied;
|
|
3459
|
+
return applied;
|
|
3460
|
+
} catch (err) {
|
|
3461
|
+
this.log({
|
|
3462
|
+
level: "error",
|
|
3463
|
+
message: `Runner file sync failed unexpectedly (message processing is unaffected): ${err instanceof Error ? err.message : String(err)}`
|
|
3464
|
+
});
|
|
3465
|
+
return 0;
|
|
3466
|
+
} finally {
|
|
3467
|
+
this.syncingFiles = false;
|
|
3468
|
+
}
|
|
3469
|
+
}
|
|
2311
3470
|
async runDrain() {
|
|
2312
3471
|
let dispatched = 0;
|
|
2313
3472
|
try {
|
|
@@ -2341,6 +3500,28 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2341
3500
|
}
|
|
2342
3501
|
return false;
|
|
2343
3502
|
}
|
|
3503
|
+
/**
|
|
3504
|
+
* File-pull work, for `run.ts`'s idle accounting (#559).
|
|
3505
|
+
*
|
|
3506
|
+
* Pulling a file is real work that `drainPending()` knows nothing about, so
|
|
3507
|
+
* without this a near-idle runner counts a credential pull as an empty tick
|
|
3508
|
+
* and `--idle-timeout` can `process.exit` mid-pull — leaving a
|
|
3509
|
+
* `.evident-push-*.tmp` behind — or immediately after the write, before the
|
|
3510
|
+
* browser has run the authorize/callback that activates it (the user then sees
|
|
3511
|
+
* `saved_not_activated` for a runner that was fine).
|
|
3512
|
+
*
|
|
3513
|
+
* Two signals because one cannot cover both cases: `inFlight` is the pull
|
|
3514
|
+
* happening RIGHT NOW (it may outlive the tick that started it), and
|
|
3515
|
+
* `appliedFiles` is monotonic so a pull that started AND finished between two
|
|
3516
|
+
* idle checks still shows up as an advance.
|
|
3517
|
+
*
|
|
3518
|
+
* CALLER CONTRACT: sample `inFlight` BEFORE calling `syncPendingFiles()` for
|
|
3519
|
+
* the cycle. `syncPendingFiles` sets the flag synchronously, so a caller that
|
|
3520
|
+
* samples afterwards reads `true` every single cycle and can never idle out.
|
|
3521
|
+
*/
|
|
3522
|
+
fileSyncActivity() {
|
|
3523
|
+
return { appliedFiles: this.appliedFileCount, inFlight: this.syncingFiles };
|
|
3524
|
+
}
|
|
2344
3525
|
/**
|
|
2345
3526
|
* OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
|
|
2346
3527
|
* exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
|
|
@@ -2402,7 +3583,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2402
3583
|
await this.sleep(step);
|
|
2403
3584
|
}
|
|
2404
3585
|
}
|
|
2405
|
-
while (this.hasInFlightWatchers()) {
|
|
3586
|
+
while (this.hasInFlightWatchers() || this.syncingFiles) {
|
|
2406
3587
|
if (this.now() >= deadline) return false;
|
|
2407
3588
|
await this.sleep(step);
|
|
2408
3589
|
}
|
|
@@ -2452,6 +3633,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2452
3633
|
skippedAlreadyDispatched += 1;
|
|
2453
3634
|
continue;
|
|
2454
3635
|
}
|
|
3636
|
+
if (message.opencode_message_id) {
|
|
3637
|
+
const outcome = await this.resolveRedrive(conv, sessionId, message, refusedSessionId);
|
|
3638
|
+
if (outcome !== "dispatch") {
|
|
3639
|
+
break;
|
|
3640
|
+
}
|
|
3641
|
+
}
|
|
2455
3642
|
const options = {
|
|
2456
3643
|
agent: message.opencode_agent ?? void 0,
|
|
2457
3644
|
model: message.opencode_model ?? void 0
|
|
@@ -2496,50 +3683,278 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2496
3683
|
this.sessions.delete(conv.id);
|
|
2497
3684
|
this.supersede(conv.id, sessionId);
|
|
2498
3685
|
this.log({
|
|
2499
|
-
level: "warn",
|
|
2500
|
-
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.`,
|
|
3686
|
+
level: "warn",
|
|
3687
|
+
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.`,
|
|
3688
|
+
conversation_id: conv.id,
|
|
3689
|
+
message_id: message.id
|
|
3690
|
+
});
|
|
3691
|
+
await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
|
|
3692
|
+
this.log({
|
|
3693
|
+
level: "warn",
|
|
3694
|
+
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)}`,
|
|
3695
|
+
conversation_id: conv.id,
|
|
3696
|
+
message_id: message.id
|
|
3697
|
+
});
|
|
3698
|
+
});
|
|
3699
|
+
this.log({
|
|
3700
|
+
level: "error",
|
|
3701
|
+
message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
|
|
3702
|
+
conversation_id: conv.id,
|
|
3703
|
+
message_id: message.id
|
|
3704
|
+
});
|
|
3705
|
+
break;
|
|
3706
|
+
}
|
|
3707
|
+
if (opencodeMessageId === null) {
|
|
3708
|
+
this.log({
|
|
3709
|
+
level: "warn",
|
|
3710
|
+
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`,
|
|
3711
|
+
conversation_id: conv.id,
|
|
3712
|
+
message_id: message.id
|
|
3713
|
+
});
|
|
3714
|
+
continue;
|
|
3715
|
+
}
|
|
3716
|
+
this.dispatched.add(message.id);
|
|
3717
|
+
this.registerInFlight(conv, sessionId, message, opencodeMessageId);
|
|
3718
|
+
dispatched += 1;
|
|
3719
|
+
void this.postSignal(conv.id, message.id, "dispatched");
|
|
3720
|
+
}
|
|
3721
|
+
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3722
|
+
this.log({
|
|
3723
|
+
level: "warn",
|
|
3724
|
+
message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
|
|
3725
|
+
conversation_id: conv.id
|
|
3726
|
+
});
|
|
3727
|
+
}
|
|
3728
|
+
this.ensureWatcherRunning(sessionId);
|
|
3729
|
+
return dispatched;
|
|
3730
|
+
}
|
|
3731
|
+
/**
|
|
3732
|
+
* Poll a session's message list for the re-drive fence (#965), via the
|
|
3733
|
+
* INJECTED `fetchImpl` — NOT the imported `getSessionMessages` helper, which
|
|
3734
|
+
* hits the global `fetch` and would bypass the same override every other
|
|
3735
|
+
* opencode poll in this file respects. Mirrors `readoptProcessing`'s own
|
|
3736
|
+
* snapshot fetch (`:3081-3111`). `null` = unreadable (non-OK response,
|
|
3737
|
+
* non-array body, or a network exception) — treated as "can't observe",
|
|
3738
|
+
* never as "confirmed gone".
|
|
3739
|
+
*/
|
|
3740
|
+
async pollSessionMessagesForRedrive(conv, message, sessionId) {
|
|
3741
|
+
try {
|
|
3742
|
+
const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
|
|
3743
|
+
if (!res.ok) {
|
|
3744
|
+
this.log({
|
|
3745
|
+
level: "warn",
|
|
3746
|
+
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`,
|
|
3747
|
+
conversation_id: conv.id,
|
|
3748
|
+
message_id: message.id
|
|
3749
|
+
});
|
|
3750
|
+
return null;
|
|
3751
|
+
}
|
|
3752
|
+
const body = await res.json();
|
|
3753
|
+
if (!Array.isArray(body)) {
|
|
3754
|
+
this.log({
|
|
3755
|
+
level: "warn",
|
|
3756
|
+
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`,
|
|
3757
|
+
conversation_id: conv.id,
|
|
3758
|
+
message_id: message.id
|
|
3759
|
+
});
|
|
3760
|
+
return null;
|
|
3761
|
+
}
|
|
3762
|
+
return body;
|
|
3763
|
+
} catch (err) {
|
|
3764
|
+
this.log({
|
|
3765
|
+
level: "warn",
|
|
3766
|
+
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)}`,
|
|
3767
|
+
conversation_id: conv.id,
|
|
3768
|
+
message_id: message.id
|
|
3769
|
+
});
|
|
3770
|
+
return null;
|
|
3771
|
+
}
|
|
3772
|
+
}
|
|
3773
|
+
/**
|
|
3774
|
+
* The re-drive fence for a `pending` row that already carries a stored
|
|
3775
|
+
* `opencode_message_id` (#965) — i.e. it has already been handed to opencode at
|
|
3776
|
+
* least once (see the invariant at `QueuedMessage.opencode_message_id`'s doc).
|
|
3777
|
+
* The lifecycle cron can falsely reclaim a `processing` row back to `pending`
|
|
3778
|
+
* mid-turn (a 5-minute liveness-staleness check racing a still-running turn);
|
|
3779
|
+
* without this fence the drain loop would re-`prompt_async` the SAME turn a
|
|
3780
|
+
* second time against live GitHub state. Mirrors `readoptOne`'s job for the
|
|
3781
|
+
* `processing` re-adopt path, but simpler: no b1/b2 preamble cross-check is
|
|
3782
|
+
* needed here because `refusedSessionId` already handles the one case
|
|
3783
|
+
* (#553 abandoned session) that path exists for.
|
|
3784
|
+
*
|
|
3785
|
+
* Only `ChannelAuthError` propagates; every other failure resolves to
|
|
3786
|
+
* `unresolved` and is retried whole on the next ~2s drain tick.
|
|
3787
|
+
*/
|
|
3788
|
+
async resolveRedrive(conv, sessionId, message, refusedSessionId) {
|
|
3789
|
+
const ocId = message.opencode_message_id ?? null;
|
|
3790
|
+
if (refusedSessionId) {
|
|
3791
|
+
this.clearRedriveUnresolved(message.id);
|
|
3792
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3793
|
+
return "dispatch";
|
|
3794
|
+
}
|
|
3795
|
+
const messages = await this.pollSessionMessagesForRedrive(conv, message, sessionId);
|
|
3796
|
+
if (messages == null || messages.length === 0) {
|
|
3797
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3798
|
+
}
|
|
3799
|
+
const state = messageRunState(messages, ocId ?? "");
|
|
3800
|
+
if (state === "failed" && isAbortedTerminalReply(messages, ocId ?? "")) {
|
|
3801
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3802
|
+
if (ongoing === false) {
|
|
3803
|
+
this.log({
|
|
3804
|
+
level: "info",
|
|
3805
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} carries an abort-shaped terminal error and session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status \u2014 restart orphan, re-dispatching instead of marking it permanently failed`,
|
|
2501
3806
|
conversation_id: conv.id,
|
|
2502
3807
|
message_id: message.id
|
|
2503
3808
|
});
|
|
2504
|
-
|
|
2505
|
-
|
|
2506
|
-
|
|
2507
|
-
|
|
2508
|
-
|
|
2509
|
-
|
|
2510
|
-
|
|
2511
|
-
|
|
3809
|
+
this.clearRedriveUnresolved(message.id);
|
|
3810
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3811
|
+
return "dispatch";
|
|
3812
|
+
}
|
|
3813
|
+
}
|
|
3814
|
+
if (state === "done" || state === "failed") {
|
|
3815
|
+
return this.settleRedrive(conv, sessionId, message, ocId, messages, state);
|
|
3816
|
+
}
|
|
3817
|
+
if (state === "running" || state === "queued") {
|
|
3818
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3819
|
+
if (ongoing === true) {
|
|
3820
|
+
return this.reattachRedrive(conv, sessionId, message, ocId);
|
|
3821
|
+
}
|
|
3822
|
+
if (ongoing === false) {
|
|
3823
|
+
this.clearRedriveUnresolved(message.id);
|
|
3824
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3825
|
+
return "dispatch";
|
|
3826
|
+
}
|
|
3827
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3828
|
+
}
|
|
3829
|
+
this.clearRedriveUnresolved(message.id);
|
|
3830
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3831
|
+
return "dispatch";
|
|
3832
|
+
}
|
|
3833
|
+
/**
|
|
3834
|
+
* The `reattached` outcome (Task 3.3): the prior turn is STILL ONGOING per
|
|
3835
|
+
* opencode's own status map — undo the false reclaim instead of starting a
|
|
3836
|
+
* second turn.
|
|
3837
|
+
*/
|
|
3838
|
+
async reattachRedrive(conv, sessionId, message, ocId) {
|
|
3839
|
+
let anchorMs;
|
|
3840
|
+
const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
|
|
3841
|
+
if (!Number.isNaN(parsed)) {
|
|
3842
|
+
anchorMs = parsed;
|
|
3843
|
+
} else {
|
|
3844
|
+
anchorMs = this.now();
|
|
3845
|
+
this.log({
|
|
3846
|
+
level: "error",
|
|
3847
|
+
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)`,
|
|
3848
|
+
conversation_id: conv.id,
|
|
3849
|
+
message_id: message.id
|
|
3850
|
+
});
|
|
3851
|
+
}
|
|
3852
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3853
|
+
try {
|
|
3854
|
+
await this.markProcessing(conv.id, message.id, sessionId, ocId, title);
|
|
3855
|
+
} catch (err) {
|
|
3856
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3857
|
+
this.log({
|
|
3858
|
+
level: "warn",
|
|
3859
|
+
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)}`,
|
|
3860
|
+
conversation_id: conv.id,
|
|
3861
|
+
message_id: message.id
|
|
3862
|
+
});
|
|
3863
|
+
return "unresolved";
|
|
3864
|
+
}
|
|
3865
|
+
this.clearRedriveUnresolved(message.id);
|
|
3866
|
+
this.registerReadopted(conv, sessionId, message, ocId ?? "", anchorMs);
|
|
3867
|
+
this.dispatched.add(message.id);
|
|
3868
|
+
this.readopted.add(message.id);
|
|
3869
|
+
this.ensureWatcherRunning(sessionId);
|
|
3870
|
+
const watchedForMs = this.now() - anchorMs;
|
|
3871
|
+
void this.postSignal(conv.id, message.id, "redrive_reattached", {
|
|
3872
|
+
watched_for_ms: watchedForMs
|
|
3873
|
+
});
|
|
3874
|
+
this.log({
|
|
3875
|
+
level: "warn",
|
|
3876
|
+
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`,
|
|
3877
|
+
conversation_id: conv.id,
|
|
3878
|
+
message_id: message.id
|
|
3879
|
+
});
|
|
3880
|
+
return "reattached";
|
|
3881
|
+
}
|
|
3882
|
+
/**
|
|
3883
|
+
* The `settled` outcome (Task 3.2): the prior turn already finished (or
|
|
3884
|
+
* errored) while nobody was watching — deliver/report it instead of re-running.
|
|
3885
|
+
* Mirrors `readoptOne`'s `done`/`failed` branches' error discipline, simplified
|
|
3886
|
+
* (no `doneUndeliverable` park: a terminal PATCH failure here just retries next
|
|
3887
|
+
* drain, same as any other non-auth failure). The restart-abort carve-out that
|
|
3888
|
+
* keeps the two in step for `failed` lives in the caller (`resolveRedrive`, #1310),
|
|
3889
|
+
* so a row reaching this `failed` branch is a GENUINE failure.
|
|
3890
|
+
*/
|
|
3891
|
+
async settleRedrive(conv, sessionId, message, ocId, messages, state) {
|
|
3892
|
+
try {
|
|
3893
|
+
if (state === "done") {
|
|
3894
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3895
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
2512
3896
|
this.log({
|
|
2513
|
-
level: "
|
|
2514
|
-
message: `
|
|
3897
|
+
level: "info",
|
|
3898
|
+
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`,
|
|
2515
3899
|
conversation_id: conv.id,
|
|
2516
3900
|
message_id: message.id
|
|
2517
3901
|
});
|
|
2518
|
-
|
|
2519
|
-
}
|
|
2520
|
-
|
|
3902
|
+
await this.markDone(conv.id, message.id, sessionId, ocId, title, usage);
|
|
3903
|
+
} else {
|
|
3904
|
+
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3905
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3906
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
2521
3907
|
this.log({
|
|
2522
|
-
level: "
|
|
2523
|
-
message: `
|
|
3908
|
+
level: "error",
|
|
3909
|
+
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)"}`,
|
|
2524
3910
|
conversation_id: conv.id,
|
|
2525
3911
|
message_id: message.id
|
|
2526
3912
|
});
|
|
2527
|
-
|
|
3913
|
+
await this.markFailed(conv.id, message.id, sessionId, error2, usage, failure);
|
|
2528
3914
|
}
|
|
2529
|
-
|
|
2530
|
-
|
|
2531
|
-
dispatched += 1;
|
|
2532
|
-
void this.postSignal(conv.id, message.id, "dispatched");
|
|
2533
|
-
}
|
|
2534
|
-
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3915
|
+
} catch (err) {
|
|
3916
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
2535
3917
|
this.log({
|
|
2536
3918
|
level: "warn",
|
|
2537
|
-
message: `
|
|
2538
|
-
conversation_id: conv.id
|
|
3919
|
+
message: `Re-drive: failed to report message ${message.id.slice(0, 8)} ${state} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3920
|
+
conversation_id: conv.id,
|
|
3921
|
+
message_id: message.id
|
|
2539
3922
|
});
|
|
3923
|
+
return "unresolved";
|
|
2540
3924
|
}
|
|
2541
|
-
this.
|
|
2542
|
-
|
|
3925
|
+
this.clearRedriveUnresolved(message.id);
|
|
3926
|
+
void this.postSignal(conv.id, message.id, "redrive_settled");
|
|
3927
|
+
return "settled";
|
|
3928
|
+
}
|
|
3929
|
+
/**
|
|
3930
|
+
* The bounded `unresolved` outcome (Task 3.4): opencode's state could not be
|
|
3931
|
+
* observed (snapshot unreadable/empty, or `isSessionOngoing` returned `null`).
|
|
3932
|
+
* A `pending` row is invisible to every cron arm (all require `status =
|
|
3933
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
3934
|
+
* nothing driving it — bound it to the existing `pausedMaxWaitMs` window
|
|
3935
|
+
* (reusing the knob, not a new constant) and take `dispatch` once elapsed.
|
|
3936
|
+
*/
|
|
3937
|
+
resolveRedriveUnresolved(conv, message) {
|
|
3938
|
+
const now = this.now();
|
|
3939
|
+
const since = this.redriveUnresolvedSince.get(message.id);
|
|
3940
|
+
if (since !== void 0 && now - since >= this.pausedMaxWaitMs) {
|
|
3941
|
+
this.clearRedriveUnresolved(message.id);
|
|
3942
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3943
|
+
return "dispatch";
|
|
3944
|
+
}
|
|
3945
|
+
if (since === void 0) {
|
|
3946
|
+
this.redriveUnresolvedSince.set(message.id, now);
|
|
3947
|
+
}
|
|
3948
|
+
if (!this.redriveUnresolvedSignalled.has(message.id)) {
|
|
3949
|
+
this.redriveUnresolvedSignalled.add(message.id);
|
|
3950
|
+
void this.postSignal(conv.id, message.id, "redrive_unresolved");
|
|
3951
|
+
}
|
|
3952
|
+
return "unresolved";
|
|
3953
|
+
}
|
|
3954
|
+
/** Clear both `unresolved`-bound trackers for a row (any non-`unresolved` outcome). */
|
|
3955
|
+
clearRedriveUnresolved(messageId) {
|
|
3956
|
+
this.redriveUnresolvedSince.delete(messageId);
|
|
3957
|
+
this.redriveUnresolvedSignalled.delete(messageId);
|
|
2543
3958
|
}
|
|
2544
3959
|
/**
|
|
2545
3960
|
* Record that `sessionId` is no longer a valid binding for `conversationId`
|
|
@@ -2602,7 +4017,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2602
4017
|
const directory = await this.resolveOpenCodeDirectory();
|
|
2603
4018
|
const sessionId = await createOpenCodeSession(this.port, directory);
|
|
2604
4019
|
this.sessions.set(conversationId, sessionId);
|
|
2605
|
-
await this.persistSession(conversationId, sessionId).catch(() => {
|
|
4020
|
+
await this.persistSession(conversationId, sessionId).catch((err) => {
|
|
4021
|
+
this.log({
|
|
4022
|
+
level: "warn",
|
|
4023
|
+
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)}`,
|
|
4024
|
+
conversation_id: conversationId
|
|
4025
|
+
});
|
|
2606
4026
|
});
|
|
2607
4027
|
return sessionId;
|
|
2608
4028
|
}
|
|
@@ -2794,12 +4214,17 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2794
4214
|
stuckReported: false,
|
|
2795
4215
|
lastAliveAt: 0,
|
|
2796
4216
|
aliveInFlight: false,
|
|
4217
|
+
titleSynced: false,
|
|
4218
|
+
titleSyncInFlight: false,
|
|
2797
4219
|
awaitingHumanLatched: false,
|
|
2798
4220
|
pausedOnQuestion: false,
|
|
2799
4221
|
pausedOnPermission: false,
|
|
2800
4222
|
pausedClearConfirmed: false,
|
|
2801
4223
|
pausedInFlight: false,
|
|
2802
|
-
deliveryDeadlineAnchored: false
|
|
4224
|
+
deliveryDeadlineAnchored: false,
|
|
4225
|
+
b2PinnedSinceMs: 0,
|
|
4226
|
+
b2LastDescendantCheckMs: 0,
|
|
4227
|
+
b2AbandonedSignalled: false
|
|
2803
4228
|
});
|
|
2804
4229
|
}
|
|
2805
4230
|
/**
|
|
@@ -2814,9 +4239,10 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2814
4239
|
* opencode reports ACTIVELY `running` is watched to completion (its liveness
|
|
2815
4240
|
* heartbeat keeps the cron off its row), while a re-adopted turn that is paused
|
|
2816
4241
|
* awaiting a human — or queued/unreachable — is still bounded by `deadline` and
|
|
2817
|
-
* handed to the cron.
|
|
2818
|
-
*
|
|
2819
|
-
*
|
|
4242
|
+
* handed to the cron. Real invariant (#965): the cron MAY reclaim a row this
|
|
4243
|
+
* runner still holds; a reclaimed row that already ran is never re-dispatched
|
|
4244
|
+
* while opencode reports its turn ongoing (readopt's own gate here, and the
|
|
4245
|
+
* `pending`-row re-drive fence, `resolveRedrive`). `dispatchedAt` stays `now`
|
|
2820
4246
|
* (only the appear-guard uses it).
|
|
2821
4247
|
*
|
|
2822
4248
|
* `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
|
|
@@ -2867,12 +4293,17 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
2867
4293
|
// with no extra `re_adopted` signal needed (folds old WI-6).
|
|
2868
4294
|
lastAliveAt: 0,
|
|
2869
4295
|
aliveInFlight: false,
|
|
4296
|
+
titleSynced: false,
|
|
4297
|
+
titleSyncInFlight: false,
|
|
2870
4298
|
awaitingHumanLatched: false,
|
|
2871
4299
|
pausedOnQuestion: false,
|
|
2872
4300
|
pausedOnPermission: false,
|
|
2873
4301
|
pausedClearConfirmed: false,
|
|
2874
4302
|
pausedInFlight: false,
|
|
2875
|
-
deliveryDeadlineAnchored: false
|
|
4303
|
+
deliveryDeadlineAnchored: false,
|
|
4304
|
+
b2PinnedSinceMs: 0,
|
|
4305
|
+
b2LastDescendantCheckMs: 0,
|
|
4306
|
+
b2AbandonedSignalled: false
|
|
2876
4307
|
});
|
|
2877
4308
|
}
|
|
2878
4309
|
/**
|
|
@@ -3034,58 +4465,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3034
4465
|
}
|
|
3035
4466
|
}
|
|
3036
4467
|
if (state === "done") {
|
|
3037
|
-
this.
|
|
3038
|
-
if (!inFlight.done) {
|
|
3039
|
-
this.log({
|
|
3040
|
-
level: "info",
|
|
3041
|
-
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
3042
|
-
conversation_id: conv.id,
|
|
3043
|
-
message_id: inFlight.evidentMessageId
|
|
3044
|
-
});
|
|
3045
|
-
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
3046
|
-
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
3047
|
-
try {
|
|
3048
|
-
await this.markDone(
|
|
3049
|
-
conv.id,
|
|
3050
|
-
inFlight.evidentMessageId,
|
|
3051
|
-
sessionId,
|
|
3052
|
-
inFlight.opencodeMessageId,
|
|
3053
|
-
title,
|
|
3054
|
-
usage
|
|
3055
|
-
);
|
|
3056
|
-
} catch (err) {
|
|
3057
|
-
if (err instanceof ChannelAuthError) throw err;
|
|
3058
|
-
if (err instanceof ChannelTerminalError) {
|
|
3059
|
-
this.log({
|
|
3060
|
-
level: "warn",
|
|
3061
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
3062
|
-
conversation_id: conv.id,
|
|
3063
|
-
message_id: inFlight.evidentMessageId
|
|
3064
|
-
});
|
|
3065
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3066
|
-
return;
|
|
3067
|
-
}
|
|
3068
|
-
if (this.now() >= inFlight.deadline) {
|
|
3069
|
-
this.log({
|
|
3070
|
-
level: "warn",
|
|
3071
|
-
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)}`,
|
|
3072
|
-
conversation_id: conv.id,
|
|
3073
|
-
message_id: inFlight.evidentMessageId
|
|
3074
|
-
});
|
|
3075
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3076
|
-
return;
|
|
3077
|
-
}
|
|
3078
|
-
this.log({
|
|
3079
|
-
level: "warn",
|
|
3080
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
3081
|
-
conversation_id: conv.id,
|
|
3082
|
-
message_id: inFlight.evidentMessageId
|
|
3083
|
-
});
|
|
3084
|
-
return;
|
|
3085
|
-
}
|
|
3086
|
-
inFlight.done = true;
|
|
3087
|
-
}
|
|
3088
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4468
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
3089
4469
|
return;
|
|
3090
4470
|
}
|
|
3091
4471
|
if (state === "failed") {
|
|
@@ -3099,8 +4479,16 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3099
4479
|
message_id: inFlight.evidentMessageId
|
|
3100
4480
|
});
|
|
3101
4481
|
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4482
|
+
const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
|
|
3102
4483
|
try {
|
|
3103
|
-
await this.markFailed(
|
|
4484
|
+
await this.markFailed(
|
|
4485
|
+
conv.id,
|
|
4486
|
+
inFlight.evidentMessageId,
|
|
4487
|
+
sessionId,
|
|
4488
|
+
error2,
|
|
4489
|
+
usage,
|
|
4490
|
+
failure
|
|
4491
|
+
);
|
|
3104
4492
|
} catch (err) {
|
|
3105
4493
|
if (err instanceof ChannelAuthError) throw err;
|
|
3106
4494
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -3145,6 +4533,44 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3145
4533
|
});
|
|
3146
4534
|
}
|
|
3147
4535
|
const activelyRunning = state === "running" && !awaitingHuman;
|
|
4536
|
+
const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
|
|
4537
|
+
const snapshotReadable = messages != null && messages.length > 0;
|
|
4538
|
+
if (!pinnedNow) {
|
|
4539
|
+
if (snapshotReadable) {
|
|
4540
|
+
inFlight.b2PinnedSinceMs = 0;
|
|
4541
|
+
inFlight.b2LastDescendantCheckMs = 0;
|
|
4542
|
+
inFlight.b2AbandonedSignalled = false;
|
|
4543
|
+
}
|
|
4544
|
+
} else {
|
|
4545
|
+
if (inFlight.b2AbandonedSignalled) {
|
|
4546
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4547
|
+
return;
|
|
4548
|
+
}
|
|
4549
|
+
if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
|
|
4550
|
+
const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
|
|
4551
|
+
if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
|
|
4552
|
+
inFlight.b2LastDescendantCheckMs = this.now();
|
|
4553
|
+
const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
|
|
4554
|
+
if (isB2AbandonmentConfirmed({
|
|
4555
|
+
pinnedForMs,
|
|
4556
|
+
minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
|
|
4557
|
+
descendantOngoing
|
|
4558
|
+
})) {
|
|
4559
|
+
inFlight.b2AbandonedSignalled = true;
|
|
4560
|
+
this.log({
|
|
4561
|
+
level: "warn",
|
|
4562
|
+
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`,
|
|
4563
|
+
conversation_id: conv.id,
|
|
4564
|
+
message_id: id
|
|
4565
|
+
});
|
|
4566
|
+
void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
|
|
4567
|
+
watched_for_ms: pinnedForMs
|
|
4568
|
+
});
|
|
4569
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4570
|
+
return;
|
|
4571
|
+
}
|
|
4572
|
+
}
|
|
4573
|
+
}
|
|
3148
4574
|
if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
|
|
3149
4575
|
this.log({
|
|
3150
4576
|
level: "warn",
|
|
@@ -3164,6 +4590,18 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3164
4590
|
inFlight.aliveInFlight = false;
|
|
3165
4591
|
if (ok) inFlight.lastAliveAt = this.now();
|
|
3166
4592
|
});
|
|
4593
|
+
if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
|
|
4594
|
+
inFlight.titleSyncInFlight = true;
|
|
4595
|
+
void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
|
|
4596
|
+
if (!title) {
|
|
4597
|
+
inFlight.titleSyncInFlight = false;
|
|
4598
|
+
return;
|
|
4599
|
+
}
|
|
4600
|
+
const ok = await this.patchConversationTitle(conv.id, title);
|
|
4601
|
+
inFlight.titleSyncInFlight = false;
|
|
4602
|
+
if (ok) inFlight.titleSynced = true;
|
|
4603
|
+
});
|
|
4604
|
+
}
|
|
3167
4605
|
}
|
|
3168
4606
|
if (awaitingHuman) {
|
|
3169
4607
|
if (!inFlight.awaitingHumanLatched) {
|
|
@@ -3201,6 +4639,70 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3201
4639
|
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3202
4640
|
}
|
|
3203
4641
|
}
|
|
4642
|
+
/**
|
|
4643
|
+
* Settle a message whose run-state has resolved `'done'` — extracted verbatim
|
|
4644
|
+
* (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
|
|
4645
|
+
* inline `state === 'done'` branch body, so a SECOND caller (the #721
|
|
4646
|
+
* b2-abandonment resolution) can reach the exact same completion behavior
|
|
4647
|
+
* (delivery-deadline anchoring, title resolution, usage extraction, and
|
|
4648
|
+
* `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
|
|
4649
|
+
* and risking the two copies silently drifting apart.
|
|
4650
|
+
*/
|
|
4651
|
+
async settleMessageDone(sessionId, watcher, inFlight, messages) {
|
|
4652
|
+
const conv = watcher.conv;
|
|
4653
|
+
this.anchorDeliveryDeadline(inFlight);
|
|
4654
|
+
if (!inFlight.done) {
|
|
4655
|
+
this.log({
|
|
4656
|
+
level: "info",
|
|
4657
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
4658
|
+
conversation_id: conv.id,
|
|
4659
|
+
message_id: inFlight.evidentMessageId
|
|
4660
|
+
});
|
|
4661
|
+
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
4662
|
+
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4663
|
+
try {
|
|
4664
|
+
await this.markDone(
|
|
4665
|
+
conv.id,
|
|
4666
|
+
inFlight.evidentMessageId,
|
|
4667
|
+
sessionId,
|
|
4668
|
+
inFlight.opencodeMessageId,
|
|
4669
|
+
title,
|
|
4670
|
+
usage
|
|
4671
|
+
);
|
|
4672
|
+
} catch (err) {
|
|
4673
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
4674
|
+
if (err instanceof ChannelTerminalError) {
|
|
4675
|
+
this.log({
|
|
4676
|
+
level: "warn",
|
|
4677
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
4678
|
+
conversation_id: conv.id,
|
|
4679
|
+
message_id: inFlight.evidentMessageId
|
|
4680
|
+
});
|
|
4681
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4682
|
+
return;
|
|
4683
|
+
}
|
|
4684
|
+
if (this.now() >= inFlight.deadline) {
|
|
4685
|
+
this.log({
|
|
4686
|
+
level: "warn",
|
|
4687
|
+
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)}`,
|
|
4688
|
+
conversation_id: conv.id,
|
|
4689
|
+
message_id: inFlight.evidentMessageId
|
|
4690
|
+
});
|
|
4691
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4692
|
+
return;
|
|
4693
|
+
}
|
|
4694
|
+
this.log({
|
|
4695
|
+
level: "warn",
|
|
4696
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
4697
|
+
conversation_id: conv.id,
|
|
4698
|
+
message_id: inFlight.evidentMessageId
|
|
4699
|
+
});
|
|
4700
|
+
return;
|
|
4701
|
+
}
|
|
4702
|
+
inFlight.done = true;
|
|
4703
|
+
}
|
|
4704
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4705
|
+
}
|
|
3204
4706
|
// Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
|
|
3205
4707
|
/**
|
|
3206
4708
|
* Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
|
|
@@ -3300,7 +4802,10 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3300
4802
|
* re-dispatched (at most once, see `forceReadoptRun`):
|
|
3301
4803
|
* - `done` → `markDone` now (guarded like the watcher's done branch);
|
|
3302
4804
|
* - `failed` → `markFailed` with the surfaced error (issue #182), so an
|
|
3303
|
-
* errored turn is reported failed on restart, NOT re-dispatched
|
|
4805
|
+
* errored turn is reported failed on restart, NOT re-dispatched —
|
|
4806
|
+
* EXCEPT a restart-ABORTED turn under a not-ongoing session,
|
|
4807
|
+
* which is a restart orphan wearing a terminal error and is
|
|
4808
|
+
* re-dispatched instead (issue #1310, see the branch below);
|
|
3304
4809
|
* - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
|
|
3305
4810
|
* tracking the stored id so the reply correlates by it;
|
|
3306
4811
|
* - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
|
|
@@ -3364,9 +4869,19 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3364
4869
|
void this.postSignal(row.conversation_id, row.id, "readopt_done");
|
|
3365
4870
|
return;
|
|
3366
4871
|
}
|
|
3367
|
-
|
|
4872
|
+
const restartAborted = state === "failed" && sessionOngoing === false && isAbortedTerminalReply(messages, ocId ?? "");
|
|
4873
|
+
if (restartAborted) {
|
|
4874
|
+
this.log({
|
|
4875
|
+
level: "info",
|
|
4876
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} carries an abort-shaped terminal error and session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status \u2014 restart orphan, re-dispatching instead of marking it permanently failed`,
|
|
4877
|
+
conversation_id: row.conversation_id,
|
|
4878
|
+
message_id: row.id
|
|
4879
|
+
});
|
|
4880
|
+
}
|
|
4881
|
+
if (state === "failed" && !restartAborted) {
|
|
3368
4882
|
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3369
4883
|
const usage = messageUsage(messages, ocId ?? "");
|
|
4884
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3370
4885
|
this.log({
|
|
3371
4886
|
level: "error",
|
|
3372
4887
|
message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
|
|
@@ -3374,7 +4889,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3374
4889
|
message_id: row.id
|
|
3375
4890
|
});
|
|
3376
4891
|
try {
|
|
3377
|
-
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
|
|
4892
|
+
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
|
|
3378
4893
|
} catch (err) {
|
|
3379
4894
|
if (err instanceof ChannelAuthError) throw err;
|
|
3380
4895
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -3641,7 +5156,8 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3641
5156
|
opencode_model: row.opencode_model,
|
|
3642
5157
|
source_message_id: row.source_message_id,
|
|
3643
5158
|
slack_user_id: row.slack_user_id,
|
|
3644
|
-
attachments: row.attachments ?? null
|
|
5159
|
+
attachments: row.attachments ?? null,
|
|
5160
|
+
opencode_message_id: row.opencode_message_id
|
|
3645
5161
|
};
|
|
3646
5162
|
}
|
|
3647
5163
|
/**
|
|
@@ -3780,6 +5296,47 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3780
5296
|
}
|
|
3781
5297
|
return false;
|
|
3782
5298
|
}
|
|
5299
|
+
/**
|
|
5300
|
+
* Tri-state variant of the upward parentID membership walk (#721), used ONLY
|
|
5301
|
+
* by `isAnyDescendantSessionOngoing`. Walks the SAME cached
|
|
5302
|
+
* `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
|
|
5303
|
+
* `sessionBelongsTo`, which deliberately collapses "confirmed not a
|
|
5304
|
+
* descendant" and "the walk's fetch failed" into the same `false` (safe for
|
|
5305
|
+
* its OTHER callers: interaction attribution and the recovery-path
|
|
5306
|
+
* `isAnyDescendantSessionAlive`, both of which just retry next tick with no
|
|
5307
|
+
* safety consequence either way) — this variant keeps those two outcomes
|
|
5308
|
+
* SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
|
|
5309
|
+
* (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
|
|
5310
|
+
* not ongoing".
|
|
5311
|
+
*
|
|
5312
|
+
* Return contract:
|
|
5313
|
+
* - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
|
|
5314
|
+
* - `false` → the walk reached a definitive, parent-less root session
|
|
5315
|
+
* WITHOUT ever matching `rootSessionId` — `sessionId` is
|
|
5316
|
+
* CONFIRMED NOT a descendant of it.
|
|
5317
|
+
* - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
|
|
5318
|
+
* through the walk (`resolveSessionParent` returned `undefined`),
|
|
5319
|
+
* or the depth cap (32) was hit without a definitive answer (a
|
|
5320
|
+
* pathological/cyclic chain proves nothing either way). NEVER
|
|
5321
|
+
* treat this the same as `false` — see `sessionBelongsTo`'s own
|
|
5322
|
+
* doc comment above for why that collapse is safe THERE but not
|
|
5323
|
+
* here.
|
|
5324
|
+
*
|
|
5325
|
+
* `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
|
|
5326
|
+
* to the live-path descendant check, not a modification of shared code used
|
|
5327
|
+
* by interaction attribution or the recovery path.
|
|
5328
|
+
*/
|
|
5329
|
+
async resolveSessionMembership(sessionId, rootSessionId) {
|
|
5330
|
+
let current = sessionId;
|
|
5331
|
+
for (let depth = 0; current && depth < 32; depth++) {
|
|
5332
|
+
if (current === rootSessionId) return true;
|
|
5333
|
+
const parent = await this.resolveSessionParent(current);
|
|
5334
|
+
if (parent === void 0) return null;
|
|
5335
|
+
if (parent === null) return false;
|
|
5336
|
+
current = parent;
|
|
5337
|
+
}
|
|
5338
|
+
return null;
|
|
5339
|
+
}
|
|
3783
5340
|
/**
|
|
3784
5341
|
* Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
|
|
3785
5342
|
* `null` for a root session (no parent) and `undefined` when opencode is
|
|
@@ -3864,6 +5421,54 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3864
5421
|
}
|
|
3865
5422
|
return null;
|
|
3866
5423
|
}
|
|
5424
|
+
/**
|
|
5425
|
+
* Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
|
|
5426
|
+
* session title onto the conversation via the PLAIN conversation-update
|
|
5427
|
+
* endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
|
|
5428
|
+
* message-status endpoint `markProcessing`/`markDone` use. Deliberately a
|
|
5429
|
+
* separate, lighter call: it carries no `status`, so it cannot re-trigger the
|
|
5430
|
+
* `processing`/`done` transition side effects (Slack notices, activity-log
|
|
5431
|
+
* rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
|
|
5432
|
+
* ever touches `conversations.title`. That route (`routes/conversations.ts`)
|
|
5433
|
+
* skips a title write matching the stored value, so a redundant call with the
|
|
5434
|
+
* same title is a real no-op — it does not bump `updated_at`, which the
|
|
5435
|
+
* conversation list sorts and paginates on. (Note this is a DIFFERENT guard
|
|
5436
|
+
* from `threads.ts`'s "non-empty AND changed" one, which only covers the
|
|
5437
|
+
* message-status PATCH; the non-empty half is enforced here instead, by
|
|
5438
|
+
* `resolveSessionTitle` never returning an empty/placeholder title.)
|
|
5439
|
+
*
|
|
5440
|
+
* Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
|
|
5441
|
+
* is logged and the title is simply retried on the next heartbeat tick (the
|
|
5442
|
+
* caller only latches `titleSynced` on `true`).
|
|
5443
|
+
*/
|
|
5444
|
+
async patchConversationTitle(conversationId, title) {
|
|
5445
|
+
try {
|
|
5446
|
+
const res = await this.fetchImpl(
|
|
5447
|
+
`${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
|
|
5448
|
+
{
|
|
5449
|
+
method: "PATCH",
|
|
5450
|
+
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
5451
|
+
body: JSON.stringify({ title })
|
|
5452
|
+
}
|
|
5453
|
+
);
|
|
5454
|
+
if (!res.ok) {
|
|
5455
|
+
this.log({
|
|
5456
|
+
level: "debug",
|
|
5457
|
+
message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
|
|
5458
|
+
conversation_id: conversationId
|
|
5459
|
+
});
|
|
5460
|
+
return false;
|
|
5461
|
+
}
|
|
5462
|
+
return true;
|
|
5463
|
+
} catch (err) {
|
|
5464
|
+
this.log({
|
|
5465
|
+
level: "debug",
|
|
5466
|
+
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)}`,
|
|
5467
|
+
conversation_id: conversationId
|
|
5468
|
+
});
|
|
5469
|
+
return false;
|
|
5470
|
+
}
|
|
5471
|
+
}
|
|
3867
5472
|
/**
|
|
3868
5473
|
* DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
|
|
3869
5474
|
* (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
|
|
@@ -3920,7 +5525,85 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
3920
5525
|
return true;
|
|
3921
5526
|
}
|
|
3922
5527
|
}
|
|
3923
|
-
return false;
|
|
5528
|
+
return false;
|
|
5529
|
+
}
|
|
5530
|
+
/**
|
|
5531
|
+
* LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
|
|
5532
|
+
* sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
|
|
5533
|
+
* in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
|
|
5534
|
+
*
|
|
5535
|
+
* Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
|
|
5536
|
+
* cross-check above): that method judges liveness from the child's OWN
|
|
5537
|
+
* TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
|
|
5538
|
+
* on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
|
|
5539
|
+
* path the local opencode server IS running, so its in-memory status map is
|
|
5540
|
+
* live and authoritative — and per ADR-0047 §4a ("the child has its own entry
|
|
5541
|
+
* [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
|
|
5542
|
+
* ENTIRE turn (including any tool call it is itself executing), not a
|
|
5543
|
+
* per-message transcript snapshot. This sidesteps the "child's own tool is
|
|
5544
|
+
* executing, between its step's completion and the next generation step"
|
|
5545
|
+
* transcript gap that a transcript-based check would need a second,
|
|
5546
|
+
* sustained-window bound to guard against — it is simply not derived from
|
|
5547
|
+
* message timestamps at all.
|
|
5548
|
+
*
|
|
5549
|
+
* Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
|
|
5550
|
+
* status, as the recovery path does per §4a)? Because on the LIVE path the
|
|
5551
|
+
* root session can be shared: a SECOND, unrelated user message can land on the
|
|
5552
|
+
* SAME session (issue #721's own root cause) and keep the root `busy` for a
|
|
5553
|
+
* reason that has nothing to do with THIS message's delegation. A `task`
|
|
5554
|
+
* descendant session is spawned for exactly one delegated turn and never
|
|
5555
|
+
* reused, so its OWN status-map entry is unambiguous evidence about that one
|
|
5556
|
+
* delegation — which the root's status is not.
|
|
5557
|
+
*
|
|
5558
|
+
* Why membership is checked via `resolveSessionMembership`, NOT
|
|
5559
|
+
* `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
|
|
5560
|
+
* `GET /session/:id` fetch failure into "not a descendant", which would
|
|
5561
|
+
* silently drop a genuinely-live candidate from consideration on the one
|
|
5562
|
+
* unlucky tick its membership-walk fetch hiccups (#721).
|
|
5563
|
+
* `resolveSessionMembership` keeps that failure mode as a distinct `null`
|
|
5564
|
+
* (indeterminate) so it is folded into THIS method's own `indeterminate` flag
|
|
5565
|
+
* instead.
|
|
5566
|
+
*
|
|
5567
|
+
* Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
|
|
5568
|
+
* - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
|
|
5569
|
+
* - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
|
|
5570
|
+
* confirmed either way (`resolveSessionMembership` never
|
|
5571
|
+
* returned `null`), and every CONFIRMED descendant's status read
|
|
5572
|
+
* succeeded and is not ongoing (includes "no descendant session
|
|
5573
|
+
* exists at all" — e.g. a plain, non-`task` tool call).
|
|
5574
|
+
* - `null` → INDETERMINATE: `listSessions` failed, OR at least one
|
|
5575
|
+
* candidate's MEMBERSHIP could not be confirmed
|
|
5576
|
+
* (`resolveSessionMembership` returned `null` — a fetch failure
|
|
5577
|
+
* or pathological chain partway through the parent walk), OR at
|
|
5578
|
+
* least one CONFIRMED descendant's `isSessionOngoing` read
|
|
5579
|
+
* failed — and no OTHER candidate was already confirmed `true`.
|
|
5580
|
+
* The caller MUST NOT treat `null` the same as `false` here
|
|
5581
|
+
* (unlike the recovery cross-check's contract) — see
|
|
5582
|
+
* `isB2AbandonmentConfirmed`.
|
|
5583
|
+
*/
|
|
5584
|
+
async isAnyDescendantSessionOngoing(rootSessionId) {
|
|
5585
|
+
const sessions = await listSessions(this.port);
|
|
5586
|
+
if (!sessions) {
|
|
5587
|
+
this.log({
|
|
5588
|
+
level: "warn",
|
|
5589
|
+
message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
|
|
5590
|
+
});
|
|
5591
|
+
return null;
|
|
5592
|
+
}
|
|
5593
|
+
let indeterminate = false;
|
|
5594
|
+
for (const candidate of sessions) {
|
|
5595
|
+
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
5596
|
+
const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
|
|
5597
|
+
if (membership === null) {
|
|
5598
|
+
indeterminate = true;
|
|
5599
|
+
continue;
|
|
5600
|
+
}
|
|
5601
|
+
if (membership === false) continue;
|
|
5602
|
+
const ongoing = await isSessionOngoing(this.port, candidate.id);
|
|
5603
|
+
if (ongoing === true) return true;
|
|
5604
|
+
if (ongoing === null) indeterminate = true;
|
|
5605
|
+
}
|
|
5606
|
+
return indeterminate ? null : false;
|
|
3924
5607
|
}
|
|
3925
5608
|
/**
|
|
3926
5609
|
* Cheap decision-telemetry label for a running row's LAST correlated reply
|
|
@@ -4066,11 +5749,11 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4066
5749
|
* `ensureSession` guard, not this suppression, is what makes the self-heal
|
|
4067
5750
|
* stick.
|
|
4068
5751
|
*/
|
|
4069
|
-
sessionIdBody(sessionId, conversationId, messageId,
|
|
5752
|
+
sessionIdBody(sessionId, conversationId, messageId, status2) {
|
|
4070
5753
|
if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
|
|
4071
5754
|
this.log({
|
|
4072
5755
|
level: "debug",
|
|
4073
|
-
message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${
|
|
5756
|
+
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)}`,
|
|
4074
5757
|
conversation_id: conversationId,
|
|
4075
5758
|
message_id: messageId
|
|
4076
5759
|
});
|
|
@@ -4185,7 +5868,7 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4185
5868
|
* exists but is wedged, so the next attempt must get a fresh one
|
|
4186
5869
|
* instead of reusing it — see WI-1's server-side null-clearing PATCH).
|
|
4187
5870
|
*/
|
|
4188
|
-
async markFailed(conversationId, messageId, sessionId, error2, usage) {
|
|
5871
|
+
async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
|
|
4189
5872
|
const body = { status: "failed" };
|
|
4190
5873
|
if (sessionId === null) {
|
|
4191
5874
|
body.opencode_session_id = null;
|
|
@@ -4194,6 +5877,12 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4194
5877
|
}
|
|
4195
5878
|
if (error2 !== void 0) body.error = error2;
|
|
4196
5879
|
if (usage) Object.assign(body, usage);
|
|
5880
|
+
if (failure) {
|
|
5881
|
+
body.failure_kind = failure.kind;
|
|
5882
|
+
body.failure_provider_id = failure.providerId;
|
|
5883
|
+
body.failure_model_id = failure.modelId;
|
|
5884
|
+
body.failure_reason = failure.reason;
|
|
5885
|
+
}
|
|
4197
5886
|
await this.callWithRetry(
|
|
4198
5887
|
"marking message as failed",
|
|
4199
5888
|
() => this.fetchImpl(
|
|
@@ -4206,6 +5895,29 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4206
5895
|
)
|
|
4207
5896
|
);
|
|
4208
5897
|
}
|
|
5898
|
+
/**
|
|
5899
|
+
* Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
|
|
5900
|
+
*
|
|
5901
|
+
* `messageFailure` alone (structured OpenCode error → `model_auth`) covers
|
|
5902
|
+
* most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
|
|
5903
|
+
* to the P1-2b zero-provider check — one extra loopback call to
|
|
5904
|
+
* `hasAnyConfiguredProvider`, only reached when the structured classifier
|
|
5905
|
+
* couldn't place it. Fails open (never throws): a fallback probe failure
|
|
5906
|
+
* (`null`/indeterminate) leaves the classification `null`, which produces
|
|
5907
|
+
* today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
|
|
5908
|
+
*/
|
|
5909
|
+
async classifyModelAuthFailure(messages, userMessageId) {
|
|
5910
|
+
const classified = messageFailure(messages, userMessageId);
|
|
5911
|
+
if (classified != null) return classified;
|
|
5912
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
5913
|
+
const hasProvider = await hasAnyConfiguredProvider(this.port);
|
|
5914
|
+
return applyZeroProviderFallback(
|
|
5915
|
+
classified,
|
|
5916
|
+
hasProvider,
|
|
5917
|
+
reply?.info?.providerID ?? null,
|
|
5918
|
+
reply?.info?.modelID ?? null
|
|
5919
|
+
);
|
|
5920
|
+
}
|
|
4209
5921
|
/**
|
|
4210
5922
|
* Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
|
|
4211
5923
|
* (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
|
|
@@ -4358,10 +6070,16 @@ var ChannelDriver = class _ChannelDriver {
|
|
|
4358
6070
|
import chalk5 from "chalk";
|
|
4359
6071
|
import ora2 from "ora";
|
|
4360
6072
|
import { select as select2 } from "@inquirer/prompts";
|
|
6073
|
+
var INTERACTIVE_START_TIMEOUT_MS = 3e4;
|
|
4361
6074
|
async function ensureOpenCodeRunning(ctx) {
|
|
4362
6075
|
const healthCheck = await checkOpenCodeHealth(ctx.port);
|
|
4363
6076
|
if (healthCheck.healthy) {
|
|
4364
|
-
return {
|
|
6077
|
+
return {
|
|
6078
|
+
port: ctx.port,
|
|
6079
|
+
process: null,
|
|
6080
|
+
version: healthCheck.version ?? null,
|
|
6081
|
+
notReadyReason: null
|
|
6082
|
+
};
|
|
4365
6083
|
}
|
|
4366
6084
|
const runningInstances = await findHealthyOpenCodeInstances();
|
|
4367
6085
|
if (runningInstances.length > 0) {
|
|
@@ -4382,7 +6100,7 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4382
6100
|
console.log(chalk5.yellow("Tip: Run with the correct port:"));
|
|
4383
6101
|
console.log(
|
|
4384
6102
|
chalk5.dim(
|
|
4385
|
-
` ${getCliName()} run --
|
|
6103
|
+
` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
|
|
4386
6104
|
)
|
|
4387
6105
|
);
|
|
4388
6106
|
}
|
|
@@ -4402,14 +6120,22 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4402
6120
|
if (!ctx.interactive) {
|
|
4403
6121
|
ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
|
|
4404
6122
|
const proc = await startOpenCode(ctx.port);
|
|
4405
|
-
const health = await waitForOpenCodeHealth(ctx.port,
|
|
6123
|
+
const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
|
|
4406
6124
|
if (!health.healthy) {
|
|
4407
|
-
|
|
4408
|
-
|
|
4409
|
-
|
|
6125
|
+
return {
|
|
6126
|
+
port: ctx.port,
|
|
6127
|
+
process: proc,
|
|
6128
|
+
version: null,
|
|
6129
|
+
notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
|
|
6130
|
+
};
|
|
4410
6131
|
}
|
|
4411
6132
|
ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
|
|
4412
|
-
return {
|
|
6133
|
+
return {
|
|
6134
|
+
port: ctx.port,
|
|
6135
|
+
process: proc,
|
|
6136
|
+
version: health.version ?? null,
|
|
6137
|
+
notReadyReason: null
|
|
6138
|
+
};
|
|
4413
6139
|
}
|
|
4414
6140
|
let port = ctx.port;
|
|
4415
6141
|
if (isPortInUse(port)) {
|
|
@@ -4462,122 +6188,15 @@ Port ${port} is already in use.`));
|
|
|
4462
6188
|
if (action === "start") {
|
|
4463
6189
|
const spinner = ora2("Starting OpenCode...").start();
|
|
4464
6190
|
const proc = await startOpenCode(port);
|
|
4465
|
-
const health = await waitForOpenCodeHealth(port,
|
|
6191
|
+
const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
|
|
4466
6192
|
if (!health.healthy) {
|
|
4467
6193
|
spinner.fail("Failed to start OpenCode");
|
|
4468
6194
|
throw new Error("OpenCode failed to start");
|
|
4469
6195
|
}
|
|
4470
6196
|
spinner.stop();
|
|
4471
|
-
return { port, process: proc, version: health.version ?? null };
|
|
4472
|
-
}
|
|
4473
|
-
return { port, process: null, version: null };
|
|
4474
|
-
}
|
|
4475
|
-
|
|
4476
|
-
// src/commands/agent-lookup.ts
|
|
4477
|
-
async function readErrorMessage(response) {
|
|
4478
|
-
const text = await response.text().catch(() => "");
|
|
4479
|
-
if (!text) return response.statusText || void 0;
|
|
4480
|
-
try {
|
|
4481
|
-
const data = JSON.parse(text);
|
|
4482
|
-
const message = data.message ?? data.error;
|
|
4483
|
-
if (typeof message === "string" && message.trim()) {
|
|
4484
|
-
return message;
|
|
4485
|
-
}
|
|
4486
|
-
} catch {
|
|
4487
|
-
}
|
|
4488
|
-
return text.trim() || response.statusText || void 0;
|
|
4489
|
-
}
|
|
4490
|
-
function authFailureHint(apiUrl, serverMessage) {
|
|
4491
|
-
const reason = serverMessage ? `: ${serverMessage}` : "";
|
|
4492
|
-
return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
|
|
4493
|
-
}
|
|
4494
|
-
async function resolveAgentIdFromKey(authHeader) {
|
|
4495
|
-
const apiUrl = getApiUrlConfig();
|
|
4496
|
-
try {
|
|
4497
|
-
const response = await fetch(`${apiUrl}/me`, {
|
|
4498
|
-
headers: { Authorization: authHeader }
|
|
4499
|
-
});
|
|
4500
|
-
if (response.status === 401) {
|
|
4501
|
-
const serverMessage = await readErrorMessage(response);
|
|
4502
|
-
return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
4503
|
-
}
|
|
4504
|
-
if (!response.ok) {
|
|
4505
|
-
const serverMessage = await readErrorMessage(response);
|
|
4506
|
-
return {
|
|
4507
|
-
error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
4508
|
-
};
|
|
4509
|
-
}
|
|
4510
|
-
const data = await response.json();
|
|
4511
|
-
if (data.auth_type === "agent_key" && data.agent_id) {
|
|
4512
|
-
return { agent_id: data.agent_id };
|
|
4513
|
-
}
|
|
4514
|
-
return {
|
|
4515
|
-
error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
|
|
4516
|
-
};
|
|
4517
|
-
} catch (error2) {
|
|
4518
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
4519
|
-
return { error: `Failed to resolve runner from key: ${message}` };
|
|
4520
|
-
}
|
|
4521
|
-
}
|
|
4522
|
-
async function notifyAgentDisconnected(agentId, authHeader) {
|
|
4523
|
-
const apiUrl = getApiUrlConfig();
|
|
4524
|
-
try {
|
|
4525
|
-
const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
|
|
4526
|
-
method: "POST",
|
|
4527
|
-
headers: { Authorization: authHeader }
|
|
4528
|
-
});
|
|
4529
|
-
if (!response.ok) {
|
|
4530
|
-
const serverMessage = await readErrorMessage(response);
|
|
4531
|
-
return {
|
|
4532
|
-
ok: false,
|
|
4533
|
-
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
4534
|
-
};
|
|
4535
|
-
}
|
|
4536
|
-
return { ok: true };
|
|
4537
|
-
} catch (error2) {
|
|
4538
|
-
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
4539
|
-
}
|
|
4540
|
-
}
|
|
4541
|
-
async function getAgentInfo(agentId, authHeader) {
|
|
4542
|
-
const apiUrl = getApiUrlConfig();
|
|
4543
|
-
try {
|
|
4544
|
-
const response = await fetch(`${apiUrl}/runners/${agentId}`, {
|
|
4545
|
-
headers: { Authorization: authHeader }
|
|
4546
|
-
});
|
|
4547
|
-
if (response.status === 401) {
|
|
4548
|
-
const serverMessage = await readErrorMessage(response);
|
|
4549
|
-
return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
4550
|
-
}
|
|
4551
|
-
if (response.status === 403) {
|
|
4552
|
-
const serverMessage = await readErrorMessage(response);
|
|
4553
|
-
return {
|
|
4554
|
-
valid: false,
|
|
4555
|
-
error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
|
|
4556
|
-
};
|
|
4557
|
-
}
|
|
4558
|
-
if (response.status === 404) {
|
|
4559
|
-
const serverMessage = await readErrorMessage(response);
|
|
4560
|
-
return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
|
|
4561
|
-
}
|
|
4562
|
-
if (!response.ok) {
|
|
4563
|
-
const serverMessage = await readErrorMessage(response);
|
|
4564
|
-
return {
|
|
4565
|
-
valid: false,
|
|
4566
|
-
error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
4567
|
-
};
|
|
4568
|
-
}
|
|
4569
|
-
const agent = await response.json();
|
|
4570
|
-
if (agent.agent_type !== "local") {
|
|
4571
|
-
return {
|
|
4572
|
-
valid: false,
|
|
4573
|
-
error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
|
|
4574
|
-
};
|
|
4575
|
-
}
|
|
4576
|
-
return { valid: true, agent };
|
|
4577
|
-
} catch (error2) {
|
|
4578
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
4579
|
-
return { valid: false, error: `Failed to validate runner: ${message}` };
|
|
6197
|
+
return { port, process: proc, version: health.version ?? null, notReadyReason: null };
|
|
4580
6198
|
}
|
|
6199
|
+
return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
|
|
4581
6200
|
}
|
|
4582
6201
|
|
|
4583
6202
|
// src/commands/run.ts
|
|
@@ -4585,6 +6204,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
|
|
|
4585
6204
|
var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
|
|
4586
6205
|
var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
|
|
4587
6206
|
var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
|
|
6207
|
+
var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
|
|
4588
6208
|
function resolveLogLevel(options) {
|
|
4589
6209
|
const accepted = Object.keys(LOG_LEVELS);
|
|
4590
6210
|
const validate = (value, source) => {
|
|
@@ -4608,6 +6228,63 @@ function resolveLogLevel(options) {
|
|
|
4608
6228
|
}
|
|
4609
6229
|
return "info";
|
|
4610
6230
|
}
|
|
6231
|
+
function resolveFileSyncDirectories(raw, homeDir) {
|
|
6232
|
+
const directories = [];
|
|
6233
|
+
for (const entry of raw ?? []) {
|
|
6234
|
+
const trimmed = entry.trim();
|
|
6235
|
+
if (trimmed === "") {
|
|
6236
|
+
throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
|
|
6237
|
+
}
|
|
6238
|
+
const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join3(homeDir, trimmed.slice(2)) : trimmed;
|
|
6239
|
+
if (!isAbsolute2(expanded)) {
|
|
6240
|
+
throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
|
|
6241
|
+
}
|
|
6242
|
+
const normalized = resolvePath(expanded);
|
|
6243
|
+
if (parse(normalized).root === normalized) {
|
|
6244
|
+
throw new Error(
|
|
6245
|
+
`--enable-file-sync-to will not allow-list the filesystem root ("${entry}"); name the specific directory the credentials belong in (for example ~/.claude)`
|
|
6246
|
+
);
|
|
6247
|
+
}
|
|
6248
|
+
if (!directories.includes(normalized)) {
|
|
6249
|
+
directories.push(normalized);
|
|
6250
|
+
}
|
|
6251
|
+
}
|
|
6252
|
+
if (directories.length > MAX_FILE_SYNC_DIRECTORIES) {
|
|
6253
|
+
throw new Error(
|
|
6254
|
+
`--enable-file-sync-to accepts at most ${MAX_FILE_SYNC_DIRECTORIES} directories; got ${directories.length}`
|
|
6255
|
+
);
|
|
6256
|
+
}
|
|
6257
|
+
return directories;
|
|
6258
|
+
}
|
|
6259
|
+
var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
|
|
6260
|
+
var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
|
|
6261
|
+
var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
|
|
6262
|
+
function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
|
|
6263
|
+
const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
|
|
6264
|
+
let raw;
|
|
6265
|
+
let source;
|
|
6266
|
+
if (options.opencodeStartTimeout !== void 0) {
|
|
6267
|
+
raw = options.opencodeStartTimeout;
|
|
6268
|
+
source = "--opencode-start-timeout";
|
|
6269
|
+
} else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
|
|
6270
|
+
raw = env[OPENCODE_START_TIMEOUT_ENV];
|
|
6271
|
+
source = OPENCODE_START_TIMEOUT_ENV;
|
|
6272
|
+
} else {
|
|
6273
|
+
return { timeoutMs: defaultMs, warnings: [] };
|
|
6274
|
+
}
|
|
6275
|
+
const trimmed = raw.trim();
|
|
6276
|
+
const seconds = Number(trimmed);
|
|
6277
|
+
const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
|
|
6278
|
+
if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
|
|
6279
|
+
return {
|
|
6280
|
+
timeoutMs: defaultMs,
|
|
6281
|
+
warnings: [
|
|
6282
|
+
`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`
|
|
6283
|
+
]
|
|
6284
|
+
};
|
|
6285
|
+
}
|
|
6286
|
+
return { timeoutMs: seconds * 1e3, warnings: [] };
|
|
6287
|
+
}
|
|
4611
6288
|
function meetsThreshold(state, level) {
|
|
4612
6289
|
return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
|
|
4613
6290
|
}
|
|
@@ -4629,6 +6306,10 @@ function log2(state, message, level = "info") {
|
|
|
4629
6306
|
function logActivity(state, entry) {
|
|
4630
6307
|
const level = entry.level ?? (entry.type === "error" ? "error" : "info");
|
|
4631
6308
|
if (!meetsThreshold(state, level)) return;
|
|
6309
|
+
forwardRunnerActivity(
|
|
6310
|
+
{ level, message: entry.message, error: entry.error },
|
|
6311
|
+
{ agentId: state.agentId, authHeader: state.authHeader }
|
|
6312
|
+
);
|
|
4632
6313
|
const fullEntry = {
|
|
4633
6314
|
...entry,
|
|
4634
6315
|
level,
|
|
@@ -4728,23 +6409,46 @@ async function handleAuthError(state, error2) {
|
|
|
4728
6409
|
}
|
|
4729
6410
|
async function driveChannels(state, driver) {
|
|
4730
6411
|
let idlePolls = 0;
|
|
6412
|
+
let idleMs = 0;
|
|
6413
|
+
let consecutiveDrainFailures = 0;
|
|
6414
|
+
let unreachableMs = 0;
|
|
4731
6415
|
let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
6416
|
+
let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
4732
6417
|
while (state.running) {
|
|
6418
|
+
const cycleStartedAtMs = performance.now();
|
|
6419
|
+
let idleThisCycle = false;
|
|
6420
|
+
let unreachableThisCycle = false;
|
|
4733
6421
|
if (state.connection?.reconnecting && state.connection.reconnectPromise) {
|
|
4734
6422
|
logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
|
|
4735
6423
|
if (state.interactive) displayStatus(state);
|
|
4736
6424
|
await state.connection.reconnectPromise;
|
|
4737
6425
|
}
|
|
6426
|
+
const carriedOverFileSync = driver.fileSyncActivity().inFlight;
|
|
6427
|
+
void driver.syncPendingFiles().catch(
|
|
6428
|
+
(error2) => logActivity(state, {
|
|
6429
|
+
type: "error",
|
|
6430
|
+
error: `Runner file sync failed: ${error2 instanceof Error ? error2.message : String(error2)}`
|
|
6431
|
+
})
|
|
6432
|
+
);
|
|
4738
6433
|
try {
|
|
4739
6434
|
const processed = await driver.drainPending();
|
|
6435
|
+
consecutiveDrainFailures = 0;
|
|
6436
|
+
unreachableMs = 0;
|
|
4740
6437
|
state.messageCount += processed;
|
|
4741
6438
|
const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
|
|
4742
6439
|
lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
4743
|
-
|
|
6440
|
+
const appliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
6441
|
+
const filesApplied = appliedFiles !== lastSeenAppliedFiles;
|
|
6442
|
+
const fileActivity = carriedOverFileSync || filesApplied;
|
|
6443
|
+
lastSeenAppliedFiles = appliedFiles;
|
|
6444
|
+
if (filesApplied) state.claudeUsageRearm?.();
|
|
6445
|
+
if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
|
|
4744
6446
|
idlePolls = 0;
|
|
6447
|
+
idleMs = 0;
|
|
4745
6448
|
if (processed > 0 && state.interactive) displayStatus(state);
|
|
4746
6449
|
} else if (state.idleTimeout !== null) {
|
|
4747
6450
|
idlePolls++;
|
|
6451
|
+
idleThisCycle = true;
|
|
4748
6452
|
if (idlePolls === 1) {
|
|
4749
6453
|
logActivity(state, {
|
|
4750
6454
|
type: "info",
|
|
@@ -4768,21 +6472,44 @@ async function driveChannels(state, driver) {
|
|
|
4768
6472
|
const errorMessage = error2 instanceof Error ? error2.message : String(error2);
|
|
4769
6473
|
logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
|
|
4770
6474
|
if (state.interactive) displayStatus(state);
|
|
4771
|
-
|
|
4772
|
-
|
|
4773
|
-
|
|
4774
|
-
|
|
4775
|
-
|
|
4776
|
-
|
|
4777
|
-
if (
|
|
4778
|
-
|
|
6475
|
+
if (driver.hasInFlightWatchers()) {
|
|
6476
|
+
consecutiveDrainFailures = 0;
|
|
6477
|
+
unreachableMs = 0;
|
|
6478
|
+
} else if (state.idleTimeout !== null) {
|
|
6479
|
+
consecutiveDrainFailures++;
|
|
6480
|
+
unreachableThisCycle = true;
|
|
6481
|
+
if (consecutiveDrainFailures === 1) {
|
|
6482
|
+
logActivity(state, {
|
|
6483
|
+
type: "info",
|
|
6484
|
+
message: `Cannot reach Evident, will exit if this persists past the idle timeout (timeout: ${state.idleTimeout}s)...`
|
|
6485
|
+
});
|
|
6486
|
+
if (state.interactive) displayStatus(state);
|
|
6487
|
+
}
|
|
4779
6488
|
}
|
|
4780
6489
|
}
|
|
6490
|
+
await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
|
|
6491
|
+
const cycleMs = performance.now() - cycleStartedAtMs;
|
|
6492
|
+
if (idleThisCycle) idleMs += cycleMs;
|
|
6493
|
+
if (unreachableThisCycle) unreachableMs += cycleMs;
|
|
6494
|
+
if (state.idleTimeout !== null && consecutiveDrainFailures >= 2 && unreachableMs > state.idleTimeout * 1e3) {
|
|
6495
|
+
logActivity(state, {
|
|
6496
|
+
type: "info",
|
|
6497
|
+
level: "warn",
|
|
6498
|
+
message: `Exiting: could not reach Evident for ${consecutiveDrainFailures} consecutive polls (${Math.round(unreachableMs / 1e3)}s)`
|
|
6499
|
+
});
|
|
6500
|
+
if (state.interactive) displayStatus(state);
|
|
6501
|
+
break;
|
|
6502
|
+
}
|
|
6503
|
+
if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
|
|
6504
|
+
logActivity(state, { type: "info", message: "Idle timeout reached" });
|
|
6505
|
+
if (state.interactive) displayStatus(state);
|
|
6506
|
+
break;
|
|
6507
|
+
}
|
|
4781
6508
|
}
|
|
4782
6509
|
}
|
|
4783
6510
|
var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
|
|
4784
|
-
async function runSweep(state, driver,
|
|
4785
|
-
const mode = `age=${
|
|
6511
|
+
async function runSweep(state, driver, config) {
|
|
6512
|
+
const mode = `age=${config.maxAgeMs ?? "\u2014"} count=${config.maxCount ?? "\u2014"}`;
|
|
4786
6513
|
try {
|
|
4787
6514
|
const sessions = await listSessions(state.port);
|
|
4788
6515
|
if (sessions === null) {
|
|
@@ -4795,8 +6522,8 @@ async function runSweep(state, driver, config2) {
|
|
|
4795
6522
|
const toDelete = selectSessionsToDelete(
|
|
4796
6523
|
sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
|
|
4797
6524
|
{
|
|
4798
|
-
maxAgeMs:
|
|
4799
|
-
maxCount:
|
|
6525
|
+
maxAgeMs: config.maxAgeMs,
|
|
6526
|
+
maxCount: config.maxCount,
|
|
4800
6527
|
nowMs: Date.now(),
|
|
4801
6528
|
protectedIds: driver.protectedSessionIds()
|
|
4802
6529
|
}
|
|
@@ -4832,7 +6559,7 @@ async function runSweep(state, driver, config2) {
|
|
|
4832
6559
|
}
|
|
4833
6560
|
}
|
|
4834
6561
|
function scheduleSessionCleanup(state, driver, options) {
|
|
4835
|
-
const
|
|
6562
|
+
const config = resolveSessionCleanupConfig(
|
|
4836
6563
|
{
|
|
4837
6564
|
maxAge: options.sessionCleanupMaxAge,
|
|
4838
6565
|
maxCount: options.sessionCleanupMaxCount,
|
|
@@ -4840,21 +6567,129 @@ function scheduleSessionCleanup(state, driver, options) {
|
|
|
4840
6567
|
},
|
|
4841
6568
|
process.env
|
|
4842
6569
|
);
|
|
4843
|
-
for (const warning2 of
|
|
6570
|
+
for (const warning2 of config.warnings) {
|
|
4844
6571
|
logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
|
|
4845
6572
|
}
|
|
4846
|
-
if (!
|
|
6573
|
+
if (!config.enabled) return;
|
|
4847
6574
|
logActivity(state, {
|
|
4848
6575
|
type: "info",
|
|
4849
|
-
message: `Session cleanup enabled (age=${
|
|
6576
|
+
message: `Session cleanup enabled (age=${config.maxAgeMs ?? "\u2014"}, count=${config.maxCount ?? "\u2014"}, interval=${config.intervalMs}ms)`
|
|
4850
6577
|
});
|
|
4851
|
-
const interval = setInterval(() => void runSweep(state, driver,
|
|
6578
|
+
const interval = setInterval(() => void runSweep(state, driver, config), config.intervalMs);
|
|
4852
6579
|
const firstSweep = setTimeout(
|
|
4853
|
-
() => void runSweep(state, driver,
|
|
6580
|
+
() => void runSweep(state, driver, config),
|
|
4854
6581
|
SESSION_CLEANUP_FIRST_SWEEP_MS
|
|
4855
6582
|
);
|
|
4856
6583
|
state.sessionCleanupTimers.push(interval, firstSweep);
|
|
4857
6584
|
}
|
|
6585
|
+
function claudeUsageFailureStreakSuffix(consecutiveFailures) {
|
|
6586
|
+
return consecutiveFailures > 1 ? ` (${consecutiveFailures} consecutive failures)` : "";
|
|
6587
|
+
}
|
|
6588
|
+
function scheduleClaudeUsageReporting(state, options) {
|
|
6589
|
+
const { mode, warnings } = resolveClaudeUsageReportingMode(
|
|
6590
|
+
options.claudeUsageReporting,
|
|
6591
|
+
process.env
|
|
6592
|
+
);
|
|
6593
|
+
for (const warning2 of warnings) {
|
|
6594
|
+
logActivity(state, {
|
|
6595
|
+
type: "info",
|
|
6596
|
+
level: "warn",
|
|
6597
|
+
message: `Claude usage reporting: ${warning2}`
|
|
6598
|
+
});
|
|
6599
|
+
}
|
|
6600
|
+
if (mode === "off") {
|
|
6601
|
+
logActivity(state, {
|
|
6602
|
+
type: "info",
|
|
6603
|
+
level: "debug",
|
|
6604
|
+
message: "Claude usage reporting is off (--claude-usage-reporting off)"
|
|
6605
|
+
});
|
|
6606
|
+
return null;
|
|
6607
|
+
}
|
|
6608
|
+
let consecutiveFailures = 0;
|
|
6609
|
+
let armed = false;
|
|
6610
|
+
let rearmRequested = false;
|
|
6611
|
+
const scheduleNextTick = () => {
|
|
6612
|
+
armed = true;
|
|
6613
|
+
rearmRequested = false;
|
|
6614
|
+
state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
|
|
6615
|
+
};
|
|
6616
|
+
const rearm = () => {
|
|
6617
|
+
if (armed) {
|
|
6618
|
+
rearmRequested = true;
|
|
6619
|
+
return;
|
|
6620
|
+
}
|
|
6621
|
+
rearmRequested = false;
|
|
6622
|
+
armed = true;
|
|
6623
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6624
|
+
};
|
|
6625
|
+
const tick = async (isProbe) => {
|
|
6626
|
+
try {
|
|
6627
|
+
const usage = await getClaudeUsage();
|
|
6628
|
+
const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
|
|
6629
|
+
if (result.ok) {
|
|
6630
|
+
if (consecutiveFailures > 0) {
|
|
6631
|
+
logActivity(state, {
|
|
6632
|
+
type: "info",
|
|
6633
|
+
level: "info",
|
|
6634
|
+
message: "Claude usage reporting recovered"
|
|
6635
|
+
});
|
|
6636
|
+
}
|
|
6637
|
+
consecutiveFailures = 0;
|
|
6638
|
+
logActivity(state, {
|
|
6639
|
+
type: "info",
|
|
6640
|
+
level: "debug",
|
|
6641
|
+
message: "Reported Claude usage to Evident"
|
|
6642
|
+
});
|
|
6643
|
+
} else {
|
|
6644
|
+
consecutiveFailures++;
|
|
6645
|
+
logActivity(state, {
|
|
6646
|
+
type: "info",
|
|
6647
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6648
|
+
message: `Failed to report Claude usage: ${result.error}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6649
|
+
});
|
|
6650
|
+
}
|
|
6651
|
+
scheduleNextTick();
|
|
6652
|
+
} catch (error2) {
|
|
6653
|
+
if (error2 instanceof ClaudeUsageError && isLocalCredentialProblem(error2)) {
|
|
6654
|
+
if (mode === "on") {
|
|
6655
|
+
logActivity(state, {
|
|
6656
|
+
type: "info",
|
|
6657
|
+
level: "warn",
|
|
6658
|
+
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"
|
|
6659
|
+
});
|
|
6660
|
+
scheduleNextTick();
|
|
6661
|
+
} else if (isProbe) {
|
|
6662
|
+
logActivity(state, {
|
|
6663
|
+
type: "info",
|
|
6664
|
+
level: "debug",
|
|
6665
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6666
|
+
});
|
|
6667
|
+
armed = false;
|
|
6668
|
+
if (rearmRequested) rearm();
|
|
6669
|
+
} else {
|
|
6670
|
+
logActivity(state, {
|
|
6671
|
+
type: "info",
|
|
6672
|
+
level: "debug",
|
|
6673
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6674
|
+
});
|
|
6675
|
+
scheduleNextTick();
|
|
6676
|
+
}
|
|
6677
|
+
} else {
|
|
6678
|
+
consecutiveFailures++;
|
|
6679
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
6680
|
+
logActivity(state, {
|
|
6681
|
+
type: "info",
|
|
6682
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6683
|
+
message: `Claude usage reporting failed: ${message}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6684
|
+
});
|
|
6685
|
+
scheduleNextTick();
|
|
6686
|
+
}
|
|
6687
|
+
}
|
|
6688
|
+
};
|
|
6689
|
+
armed = true;
|
|
6690
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6691
|
+
return rearm;
|
|
6692
|
+
}
|
|
4858
6693
|
async function notifyOffline(state) {
|
|
4859
6694
|
if (!state.agentId || !state.authHeader) return;
|
|
4860
6695
|
if (!state.connected) {
|
|
@@ -4872,13 +6707,29 @@ async function notifyOffline(state) {
|
|
|
4872
6707
|
if (state.interactive) displayStatus(state);
|
|
4873
6708
|
}
|
|
4874
6709
|
}
|
|
6710
|
+
async function timeShutdownPhase(state, durations, name, run2) {
|
|
6711
|
+
const startedAt = Date.now();
|
|
6712
|
+
try {
|
|
6713
|
+
return await run2();
|
|
6714
|
+
} finally {
|
|
6715
|
+
const elapsedMs = Date.now() - startedAt;
|
|
6716
|
+
durations[name] = elapsedMs;
|
|
6717
|
+
log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
|
|
6718
|
+
}
|
|
6719
|
+
}
|
|
4875
6720
|
async function cleanup(state, opts = {}) {
|
|
6721
|
+
const durations = {};
|
|
4876
6722
|
state.running = false;
|
|
4877
6723
|
for (const timer of state.sessionCleanupTimers) {
|
|
4878
6724
|
clearInterval(timer);
|
|
4879
6725
|
clearTimeout(timer);
|
|
4880
6726
|
}
|
|
4881
6727
|
state.sessionCleanupTimers = [];
|
|
6728
|
+
if (state.claudeUsageTimer) {
|
|
6729
|
+
clearTimeout(state.claudeUsageTimer);
|
|
6730
|
+
state.claudeUsageTimer = null;
|
|
6731
|
+
}
|
|
6732
|
+
state.claudeUsageRearm = null;
|
|
4882
6733
|
if (opts.graceful && state.channelDriver) {
|
|
4883
6734
|
state.channelDriver.stop();
|
|
4884
6735
|
log2(state, "Draining in-flight channel work before shutdown...");
|
|
@@ -4886,7 +6737,13 @@ async function cleanup(state, opts = {}) {
|
|
|
4886
6737
|
logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
|
|
4887
6738
|
displayStatus(state);
|
|
4888
6739
|
}
|
|
4889
|
-
const
|
|
6740
|
+
const driver = state.channelDriver;
|
|
6741
|
+
const settled = await timeShutdownPhase(
|
|
6742
|
+
state,
|
|
6743
|
+
durations,
|
|
6744
|
+
"drain",
|
|
6745
|
+
() => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
|
|
6746
|
+
);
|
|
4890
6747
|
if (!settled) {
|
|
4891
6748
|
logActivity(state, {
|
|
4892
6749
|
type: "info",
|
|
@@ -4895,13 +6752,15 @@ async function cleanup(state, opts = {}) {
|
|
|
4895
6752
|
if (state.interactive) displayStatus(state);
|
|
4896
6753
|
}
|
|
4897
6754
|
}
|
|
4898
|
-
await notifyOffline(state);
|
|
6755
|
+
await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
|
|
4899
6756
|
if (state.connection) {
|
|
4900
|
-
state.connection
|
|
6757
|
+
const connection = state.connection;
|
|
6758
|
+
await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
|
|
4901
6759
|
state.connection = null;
|
|
4902
6760
|
}
|
|
4903
6761
|
if (state.opencodeProcess) {
|
|
4904
|
-
|
|
6762
|
+
const opencodeProcess = state.opencodeProcess;
|
|
6763
|
+
await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
|
|
4905
6764
|
if (state.interactive) {
|
|
4906
6765
|
logActivity(state, { type: "info", message: "Stopped OpenCode process" });
|
|
4907
6766
|
displayStatus(state);
|
|
@@ -4910,12 +6769,15 @@ async function cleanup(state, opts = {}) {
|
|
|
4910
6769
|
}
|
|
4911
6770
|
state.opencodeProcess = null;
|
|
4912
6771
|
}
|
|
6772
|
+
return durations;
|
|
4913
6773
|
}
|
|
4914
6774
|
async function run(options) {
|
|
4915
6775
|
const interactive = isInteractive(options.json);
|
|
4916
6776
|
let logLevel;
|
|
6777
|
+
let fileSyncDirectories;
|
|
4917
6778
|
try {
|
|
4918
6779
|
logLevel = resolveLogLevel(options);
|
|
6780
|
+
fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir3());
|
|
4919
6781
|
} catch (error2) {
|
|
4920
6782
|
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
4921
6783
|
if (options.json) {
|
|
@@ -4948,8 +6810,16 @@ async function run(options) {
|
|
|
4948
6810
|
messageCount: 0,
|
|
4949
6811
|
lastProxiedActivityAt: null,
|
|
4950
6812
|
sessionCleanupTimers: [],
|
|
6813
|
+
claudeUsageTimer: null,
|
|
6814
|
+
claudeUsageRearm: null,
|
|
4951
6815
|
authHeader: ""
|
|
4952
6816
|
};
|
|
6817
|
+
setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
|
|
6818
|
+
if (fileSyncDirectories.length > 0) {
|
|
6819
|
+
log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
|
|
6820
|
+
} else {
|
|
6821
|
+
log2(state, "File sync is disabled (no --enable-file-sync-to given)", "debug");
|
|
6822
|
+
}
|
|
4953
6823
|
if (!options.runner && options.agent) {
|
|
4954
6824
|
telemetry.info(
|
|
4955
6825
|
EventTypes.DEPRECATED_AGENT_FLAG_USED,
|
|
@@ -4973,14 +6843,38 @@ async function run(options) {
|
|
|
4973
6843
|
const handleSignal = async () => {
|
|
4974
6844
|
if (state.shuttingDown) return;
|
|
4975
6845
|
state.shuttingDown = true;
|
|
6846
|
+
const shutdownStartedAt = Date.now();
|
|
4976
6847
|
if (state.interactive) {
|
|
4977
6848
|
logActivity(state, { type: "info", message: "Shutting down..." });
|
|
4978
6849
|
displayStatus(state);
|
|
4979
6850
|
} else {
|
|
4980
6851
|
log2(state, "Shutting down...");
|
|
4981
6852
|
}
|
|
4982
|
-
await cleanup(state, { graceful: true });
|
|
4983
|
-
|
|
6853
|
+
const durations = await cleanup(state, { graceful: true });
|
|
6854
|
+
const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
|
|
6855
|
+
await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
|
|
6856
|
+
let timer;
|
|
6857
|
+
const flushed = shutdownTelemetry().then(
|
|
6858
|
+
() => true,
|
|
6859
|
+
(error2) => {
|
|
6860
|
+
log2(
|
|
6861
|
+
state,
|
|
6862
|
+
`Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
|
|
6863
|
+
"warn"
|
|
6864
|
+
);
|
|
6865
|
+
return true;
|
|
6866
|
+
}
|
|
6867
|
+
);
|
|
6868
|
+
const timedOut = new Promise((resolve3) => {
|
|
6869
|
+
timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
|
|
6870
|
+
});
|
|
6871
|
+
if (!await Promise.race([flushed, timedOut])) {
|
|
6872
|
+
log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
|
|
6873
|
+
}
|
|
6874
|
+
clearTimeout(timer);
|
|
6875
|
+
});
|
|
6876
|
+
const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
|
|
6877
|
+
log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
|
|
4984
6878
|
process.exit(0);
|
|
4985
6879
|
};
|
|
4986
6880
|
process.on("SIGINT", handleSignal);
|
|
@@ -4997,6 +6891,7 @@ async function run(options) {
|
|
|
4997
6891
|
console.log(chalk6.dim("Or run `evident login` for interactive authentication"));
|
|
4998
6892
|
blank();
|
|
4999
6893
|
process.exit(1);
|
|
6894
|
+
return;
|
|
5000
6895
|
}
|
|
5001
6896
|
blank();
|
|
5002
6897
|
console.log(chalk6.yellow("You are not logged in to Evident."));
|
|
@@ -5041,6 +6936,7 @@ async function run(options) {
|
|
|
5041
6936
|
} else {
|
|
5042
6937
|
printError(resolved.error || "Failed to resolve runner ID from key");
|
|
5043
6938
|
process.exit(1);
|
|
6939
|
+
return;
|
|
5044
6940
|
}
|
|
5045
6941
|
} else {
|
|
5046
6942
|
printError(
|
|
@@ -5054,6 +6950,7 @@ async function run(options) {
|
|
|
5054
6950
|
);
|
|
5055
6951
|
blank();
|
|
5056
6952
|
process.exit(1);
|
|
6953
|
+
return;
|
|
5057
6954
|
}
|
|
5058
6955
|
}
|
|
5059
6956
|
telemetry.info(
|
|
@@ -5094,40 +6991,67 @@ async function run(options) {
|
|
|
5094
6991
|
}
|
|
5095
6992
|
spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
|
|
5096
6993
|
state.agentName = validation.agent.name;
|
|
6994
|
+
const microvmId = process.env.MICROVM_ID?.trim();
|
|
6995
|
+
if (microvmId) {
|
|
6996
|
+
const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
|
|
6997
|
+
if (reported.ok) {
|
|
6998
|
+
log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
|
|
6999
|
+
} else {
|
|
7000
|
+
const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
|
|
7001
|
+
log2(state, message, "warn");
|
|
7002
|
+
if (state.interactive && !state.json) {
|
|
7003
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
7004
|
+
}
|
|
7005
|
+
}
|
|
7006
|
+
} else {
|
|
7007
|
+
log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
|
|
7008
|
+
}
|
|
7009
|
+
const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
|
|
7010
|
+
for (const warning2 of opencodeStartTimeoutWarnings) {
|
|
7011
|
+
logActivity(state, { type: "info", level: "warn", message: warning2 });
|
|
7012
|
+
}
|
|
5097
7013
|
const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
|
|
5098
7014
|
try {
|
|
5099
7015
|
const oc = await ensureOpenCodeRunning({
|
|
5100
7016
|
port: state.port,
|
|
5101
7017
|
interactive: state.interactive,
|
|
5102
7018
|
agentId: state.agentId,
|
|
5103
|
-
log: (message) => log2(state, message)
|
|
7019
|
+
log: (message) => log2(state, message),
|
|
7020
|
+
startTimeoutMs: opencodeStartTimeoutMs
|
|
5104
7021
|
});
|
|
5105
7022
|
state.port = oc.port;
|
|
5106
7023
|
state.opencodeProcess = oc.process;
|
|
5107
7024
|
state.opencodeVersion = oc.version;
|
|
5108
|
-
state.opencodeConnected = oc.
|
|
7025
|
+
state.opencodeConnected = oc.notReadyReason === null;
|
|
5109
7026
|
const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
|
|
5110
7027
|
ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
|
|
5111
|
-
|
|
5112
|
-
|
|
5113
|
-
|
|
5114
|
-
|
|
5115
|
-
|
|
7028
|
+
if (!state.interactive && oc.notReadyReason !== null) {
|
|
7029
|
+
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}).`;
|
|
7030
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
7031
|
+
} else {
|
|
7032
|
+
const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
|
|
7033
|
+
if (versionWarning) {
|
|
7034
|
+
log2(state, versionWarning, "warn");
|
|
7035
|
+
if (state.interactive && !state.json) {
|
|
7036
|
+
logActivity(state, { type: "info", level: "warn", message: versionWarning });
|
|
7037
|
+
}
|
|
5116
7038
|
}
|
|
5117
|
-
|
|
5118
|
-
|
|
5119
|
-
|
|
5120
|
-
|
|
5121
|
-
|
|
5122
|
-
|
|
5123
|
-
|
|
5124
|
-
|
|
5125
|
-
|
|
5126
|
-
|
|
5127
|
-
|
|
5128
|
-
|
|
5129
|
-
|
|
5130
|
-
|
|
7039
|
+
const noProviderWarning = buildNoProviderWarning(
|
|
7040
|
+
await hasAnyConfiguredProvider(state.port)
|
|
7041
|
+
);
|
|
7042
|
+
if (noProviderWarning) {
|
|
7043
|
+
log2(state, noProviderWarning, "warn");
|
|
7044
|
+
if (state.interactive && !state.json) {
|
|
7045
|
+
logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
|
|
7046
|
+
blank();
|
|
7047
|
+
console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
|
|
7048
|
+
console.log(
|
|
7049
|
+
chalk6.dim(
|
|
7050
|
+
`Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
|
|
7051
|
+
)
|
|
7052
|
+
);
|
|
7053
|
+
blank();
|
|
7054
|
+
}
|
|
5131
7055
|
}
|
|
5132
7056
|
}
|
|
5133
7057
|
} catch (error2) {
|
|
@@ -5142,6 +7066,10 @@ async function run(options) {
|
|
|
5142
7066
|
getAuthHeader: () => state.authHeader,
|
|
5143
7067
|
conversationFilter: state.conversationFilter,
|
|
5144
7068
|
stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
|
|
7069
|
+
// #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
|
|
7070
|
+
// REJECTED with `file_sync_disabled` on the ack, not silently ignored.
|
|
7071
|
+
fileSyncDirectories,
|
|
7072
|
+
homeDir: homedir3(),
|
|
5145
7073
|
log: (entry) => (
|
|
5146
7074
|
// Thread the driver's real level straight through so `debug`/`warn`
|
|
5147
7075
|
// survive the sink filter (they no longer collapse to info). `type`
|
|
@@ -5168,6 +7096,18 @@ async function run(options) {
|
|
|
5168
7096
|
type: "info",
|
|
5169
7097
|
message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
|
|
5170
7098
|
});
|
|
7099
|
+
if (options.tunnelReadyFile) {
|
|
7100
|
+
const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
|
|
7101
|
+
if (marker.ok) {
|
|
7102
|
+
log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
|
|
7103
|
+
} else {
|
|
7104
|
+
log2(
|
|
7105
|
+
state,
|
|
7106
|
+
`Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
|
|
7107
|
+
"error"
|
|
7108
|
+
);
|
|
7109
|
+
}
|
|
7110
|
+
}
|
|
5171
7111
|
emitAgentConnected(state.agentId, {
|
|
5172
7112
|
port: state.port,
|
|
5173
7113
|
cli_version: getCliVersion(),
|
|
@@ -5223,6 +7163,12 @@ async function run(options) {
|
|
|
5223
7163
|
onDrainPing: () => {
|
|
5224
7164
|
if (!state.running) return;
|
|
5225
7165
|
logActivity(state, { type: "info", message: "Drain ping received \u2014 draining" });
|
|
7166
|
+
void channelDriver.syncPendingFiles().catch(
|
|
7167
|
+
(error2) => logActivity(state, {
|
|
7168
|
+
type: "error",
|
|
7169
|
+
error: `Runner file sync failed on ping: ${error2 instanceof Error ? error2.message : String(error2)}`
|
|
7170
|
+
})
|
|
7171
|
+
);
|
|
5226
7172
|
channelDriver.drainPending().then((processed) => {
|
|
5227
7173
|
if (processed > 0) {
|
|
5228
7174
|
state.messageCount += processed;
|
|
@@ -5252,6 +7198,7 @@ async function run(options) {
|
|
|
5252
7198
|
throw error2;
|
|
5253
7199
|
}
|
|
5254
7200
|
scheduleSessionCleanup(state, channelDriver, options);
|
|
7201
|
+
state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
|
|
5255
7202
|
if (!interactive || state.json) {
|
|
5256
7203
|
log2(state, "Driving channel messages...");
|
|
5257
7204
|
}
|
|
@@ -5306,10 +7253,18 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
|
|
|
5306
7253
|
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);
|
|
5307
7254
|
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 }));
|
|
5308
7255
|
program.command("whoami").description("Show the currently logged in user").action(whoami);
|
|
5309
|
-
program.command("
|
|
7256
|
+
program.command("status").description("Check whether the configured credentials can reach Evident").option("--json", "Output in JSON format").action((options) => status({ json: options.json }));
|
|
7257
|
+
program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
|
|
7258
|
+
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(
|
|
7259
|
+
"-a, --agent [id]",
|
|
7260
|
+
"Deprecated alias for --runner (still supported; --runner wins if both are given)"
|
|
7261
|
+
).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
|
|
5310
7262
|
"--log-level <level>",
|
|
5311
7263
|
"Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
|
|
5312
|
-
).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(
|
|
7264
|
+
).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(
|
|
7265
|
+
"--opencode-start-timeout <seconds>",
|
|
7266
|
+
"Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
|
|
7267
|
+
).option("--json", "Output in JSON format").option(
|
|
5313
7268
|
"--session-cleanup-max-age <duration>",
|
|
5314
7269
|
"Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
|
|
5315
7270
|
).option(
|
|
@@ -5318,6 +7273,17 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5318
7273
|
).option(
|
|
5319
7274
|
"--session-cleanup-interval <duration>",
|
|
5320
7275
|
"How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
|
|
7276
|
+
).option(
|
|
7277
|
+
"--claude-usage-reporting <mode>",
|
|
7278
|
+
"Report Claude subscription usage to Evident: auto | on | off (default: auto). Env: EVIDENT_CLAUDE_USAGE_REPORTING"
|
|
7279
|
+
).option(
|
|
7280
|
+
"--enable-file-sync-to <dir>",
|
|
7281
|
+
"Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
|
|
7282
|
+
(value, previous) => previous.concat([value]),
|
|
7283
|
+
[]
|
|
7284
|
+
).option(
|
|
7285
|
+
"--tunnel-ready-file <path>",
|
|
7286
|
+
"Path to write once the tunnel is connected (opt-in; unused unless an operator/image sets it)"
|
|
5321
7287
|
).action(
|
|
5322
7288
|
(options) => {
|
|
5323
7289
|
run({
|
|
@@ -5330,11 +7296,21 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5330
7296
|
verbose: options.verbose,
|
|
5331
7297
|
conversation: options.conversation,
|
|
5332
7298
|
idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
|
|
7299
|
+
// Raw string — validation/precedence is single-sourced in run.ts's
|
|
7300
|
+
// resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
|
|
7301
|
+
opencodeStartTimeout: options.opencodeStartTimeout,
|
|
5333
7302
|
json: options.json,
|
|
5334
7303
|
// Raw strings — the resolver in run.ts single-sources parsing (M1).
|
|
5335
7304
|
sessionCleanupMaxAge: options.sessionCleanupMaxAge,
|
|
5336
7305
|
sessionCleanupMaxCount: options.sessionCleanupMaxCount,
|
|
5337
|
-
sessionCleanupInterval: options.sessionCleanupInterval
|
|
7306
|
+
sessionCleanupInterval: options.sessionCleanupInterval,
|
|
7307
|
+
// Raw string — the resolver in run.ts single-sources parsing
|
|
7308
|
+
// (resolveClaudeUsageReportingMode).
|
|
7309
|
+
claudeUsageReporting: options.claudeUsageReporting,
|
|
7310
|
+
// Raw values — expansion/validation is single-sourced in run.ts's
|
|
7311
|
+
// resolveFileSyncDirectories.
|
|
7312
|
+
enableFileSyncTo: options.enableFileSyncTo,
|
|
7313
|
+
tunnelReadyFile: options.tunnelReadyFile
|
|
5338
7314
|
});
|
|
5339
7315
|
}
|
|
5340
7316
|
);
|