@evident-ai/cli 3.1.1-dev.4159d3a → 3.1.1-dev.43b2e9a
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +60 -11
- package/dist/index.js +2744 -610
- 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();
|
|
@@ -466,192 +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 stripQuery(url) {
|
|
503
|
-
try {
|
|
504
|
-
return new URL(url).pathname;
|
|
505
|
-
} catch {
|
|
506
|
-
const q = url.indexOf("?");
|
|
507
|
-
return q === -1 ? url : url.slice(0, q);
|
|
508
|
-
}
|
|
509
|
-
}
|
|
510
|
-
|
|
511
|
-
// src/lib/telemetry.ts
|
|
512
|
-
var CLI_VERSION = (true ? "3.0.0" : void 0) ?? process.env.npm_package_version ?? "unknown";
|
|
513
|
-
function getCliVersion() {
|
|
514
|
-
return CLI_VERSION;
|
|
515
|
-
}
|
|
516
|
-
var eventBuffer = [];
|
|
517
|
-
var flushTimeout = null;
|
|
518
|
-
var isShuttingDown = false;
|
|
519
|
-
var FLUSH_INTERVAL_MS = 5e3;
|
|
520
|
-
var MAX_BUFFER_SIZE = 50;
|
|
521
|
-
var FLUSH_TIMEOUT_MS = 3e3;
|
|
522
|
-
function logEvent(eventType, options = {}) {
|
|
523
|
-
const event = {
|
|
524
|
-
event_type: eventType,
|
|
525
|
-
severity: options.severity || "info",
|
|
526
|
-
message: options.message,
|
|
527
|
-
metadata: options.metadata,
|
|
528
|
-
agent_id: options.agentId,
|
|
529
|
-
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
530
|
-
};
|
|
531
|
-
eventBuffer.push(event);
|
|
532
|
-
if (options.severity === "error" || eventBuffer.length >= MAX_BUFFER_SIZE) {
|
|
533
|
-
void flushEvents();
|
|
534
|
-
} else if (!flushTimeout && !isShuttingDown) {
|
|
535
|
-
flushTimeout = setTimeout(() => {
|
|
536
|
-
flushTimeout = null;
|
|
537
|
-
void flushEvents();
|
|
538
|
-
}, FLUSH_INTERVAL_MS);
|
|
539
|
-
}
|
|
540
|
-
}
|
|
541
|
-
var telemetry = {
|
|
542
|
-
debug: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "debug", message, metadata, agentId }),
|
|
543
|
-
info: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "info", message, metadata, agentId }),
|
|
544
|
-
warn: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "warning", message, metadata, agentId }),
|
|
545
|
-
error: (eventType, message, metadata, agentId) => logEvent(eventType, { severity: "error", message, metadata, agentId })
|
|
546
|
-
};
|
|
547
|
-
async function flushEvents() {
|
|
548
|
-
if (eventBuffer.length === 0) return;
|
|
549
|
-
const events = eventBuffer;
|
|
550
|
-
eventBuffer = [];
|
|
551
|
-
if (flushTimeout) {
|
|
552
|
-
clearTimeout(flushTimeout);
|
|
553
|
-
flushTimeout = null;
|
|
554
|
-
}
|
|
555
|
-
try {
|
|
556
|
-
const credentials2 = await getToken();
|
|
557
|
-
if (!credentials2) {
|
|
558
|
-
return;
|
|
559
|
-
}
|
|
560
|
-
const apiUrl = getApiUrlConfig();
|
|
561
|
-
const controller = new AbortController();
|
|
562
|
-
const timeout = setTimeout(() => controller.abort(), FLUSH_TIMEOUT_MS);
|
|
563
|
-
try {
|
|
564
|
-
const request = {
|
|
565
|
-
events,
|
|
566
|
-
client_type: "cli",
|
|
567
|
-
client_version: CLI_VERSION
|
|
568
|
-
};
|
|
569
|
-
const response = await fetch(`${apiUrl}/telemetry/events`, {
|
|
570
|
-
method: "POST",
|
|
571
|
-
headers: {
|
|
572
|
-
"Content-Type": "application/json",
|
|
573
|
-
Authorization: `Bearer ${credentials2.token}`
|
|
574
|
-
},
|
|
575
|
-
body: JSON.stringify(request),
|
|
576
|
-
signal: controller.signal
|
|
577
|
-
});
|
|
578
|
-
if (!response.ok) {
|
|
579
|
-
console.error(`Telemetry flush failed: ${response.status}`);
|
|
580
|
-
}
|
|
581
|
-
} finally {
|
|
582
|
-
clearTimeout(timeout);
|
|
583
|
-
}
|
|
584
|
-
} catch (error2) {
|
|
585
|
-
if (process.env.DEBUG) {
|
|
586
|
-
console.error("Telemetry error:", error2);
|
|
587
|
-
}
|
|
588
|
-
}
|
|
589
|
-
}
|
|
590
|
-
async function shutdownTelemetry() {
|
|
591
|
-
isShuttingDown = true;
|
|
592
|
-
if (flushTimeout) {
|
|
593
|
-
clearTimeout(flushTimeout);
|
|
594
|
-
flushTimeout = null;
|
|
595
|
-
}
|
|
596
|
-
await flushEvents();
|
|
597
|
-
}
|
|
598
|
-
function emitEvent(event) {
|
|
599
|
-
logEvent(event.event_type, {
|
|
600
|
-
severity: event.severity,
|
|
601
|
-
message: event.message,
|
|
602
|
-
metadata: event.metadata,
|
|
603
|
-
agentId: event.agent_id
|
|
604
|
-
});
|
|
605
|
-
}
|
|
606
|
-
function emitAgentConnected(agentId, metadata) {
|
|
607
|
-
emitEvent({
|
|
608
|
-
event_type: TelemetryEventTypes.AGENT_CONNECTED,
|
|
609
|
-
severity: "info",
|
|
610
|
-
message: "Agent CLI connected",
|
|
611
|
-
metadata,
|
|
612
|
-
agent_id: agentId
|
|
613
|
-
});
|
|
614
|
-
}
|
|
615
|
-
function emitAgentDisconnected(agentId, metadata) {
|
|
616
|
-
emitEvent({
|
|
617
|
-
event_type: TelemetryEventTypes.AGENT_DISCONNECTED,
|
|
618
|
-
severity: "info",
|
|
619
|
-
message: `Agent CLI disconnected (code: ${metadata.code})`,
|
|
620
|
-
metadata,
|
|
621
|
-
agent_id: agentId
|
|
622
|
-
});
|
|
623
|
-
}
|
|
624
|
-
var EventTypes = {
|
|
625
|
-
// Tunnel lifecycle
|
|
626
|
-
TUNNEL_STARTING: "tunnel.starting",
|
|
627
|
-
TUNNEL_CONNECTED: "tunnel.connected",
|
|
628
|
-
TUNNEL_DISCONNECTED: "tunnel.disconnected",
|
|
629
|
-
TUNNEL_RECONNECTING: "tunnel.reconnecting",
|
|
630
|
-
TUNNEL_ERROR: "tunnel.error",
|
|
631
|
-
// OpenCode communication
|
|
632
|
-
OPENCODE_HEALTH_CHECK: "opencode.health_check",
|
|
633
|
-
OPENCODE_HEALTH_OK: "opencode.health_ok",
|
|
634
|
-
OPENCODE_HEALTH_FAILED: "opencode.health_failed",
|
|
635
|
-
OPENCODE_REQUEST_RECEIVED: "opencode.request_received",
|
|
636
|
-
OPENCODE_REQUEST_FORWARDED: "opencode.request_forwarded",
|
|
637
|
-
OPENCODE_RESPONSE_SENT: "opencode.response_sent",
|
|
638
|
-
OPENCODE_UNREACHABLE: "opencode.unreachable",
|
|
639
|
-
OPENCODE_ERROR: "opencode.error",
|
|
640
|
-
// Authentication
|
|
641
|
-
AUTH_LOGIN_STARTED: "auth.login_started",
|
|
642
|
-
AUTH_LOGIN_SUCCESS: "auth.login_success",
|
|
643
|
-
AUTH_LOGIN_FAILED: "auth.login_failed",
|
|
644
|
-
AUTH_LOGOUT: "auth.logout",
|
|
645
|
-
// CLI lifecycle
|
|
646
|
-
CLI_STARTED: "cli.started",
|
|
647
|
-
CLI_COMMAND: "cli.command",
|
|
648
|
-
CLI_ERROR: "cli.error",
|
|
649
|
-
// Deprecation telemetry (#412) — usage of the old `--agent`/`EVIDENT_AGENT_KEY`
|
|
650
|
-
// names instead of the preferred `--runner`/`EVIDENT_RUNNER_KEY` (#409).
|
|
651
|
-
DEPRECATED_AGENT_FLAG_USED: "cli.deprecated_agent_flag_used",
|
|
652
|
-
DEPRECATED_AGENT_KEY_ENV_USED: "cli.deprecated_agent_key_env_used"
|
|
653
|
-
};
|
|
654
|
-
|
|
655
535
|
// src/lib/auth.ts
|
|
656
536
|
async function getAuthCredentials() {
|
|
657
537
|
const runnerKey = process.env.EVIDENT_RUNNER_KEY;
|
|
@@ -695,38 +575,745 @@ function isInteractive(jsonOutput) {
|
|
|
695
575
|
return true;
|
|
696
576
|
}
|
|
697
577
|
|
|
698
|
-
// src/
|
|
699
|
-
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;
|
|
700
582
|
try {
|
|
701
|
-
const
|
|
702
|
-
|
|
703
|
-
|
|
704
|
-
|
|
705
|
-
if (!response.ok) {
|
|
706
|
-
return { healthy: false, error: `HTTP ${response.status}` };
|
|
583
|
+
const data = JSON.parse(text);
|
|
584
|
+
const message = data.message ?? data.error;
|
|
585
|
+
if (typeof message === "string" && message.trim()) {
|
|
586
|
+
return message;
|
|
707
587
|
}
|
|
708
|
-
|
|
709
|
-
return { healthy: true, version: data.version };
|
|
710
|
-
} catch (error2) {
|
|
711
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
712
|
-
return { healthy: false, error: message };
|
|
588
|
+
} catch {
|
|
713
589
|
}
|
|
590
|
+
return text.trim() || response.statusText || void 0;
|
|
714
591
|
}
|
|
715
|
-
|
|
716
|
-
const
|
|
717
|
-
|
|
718
|
-
const health = await checkOpenCodeHealth(port);
|
|
719
|
-
if (health.healthy) {
|
|
720
|
-
return health;
|
|
721
|
-
}
|
|
722
|
-
await new Promise((resolve2) => setTimeout(resolve2, 1e3));
|
|
723
|
-
}
|
|
724
|
-
return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
|
|
592
|
+
function authFailureHint(apiUrl, serverMessage) {
|
|
593
|
+
const reason = serverMessage ? `: ${serverMessage}` : "";
|
|
594
|
+
return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
|
|
725
595
|
}
|
|
726
|
-
|
|
727
|
-
|
|
728
|
-
|
|
729
|
-
|
|
596
|
+
async function resolveAgentIdFromKey(authHeader) {
|
|
597
|
+
const apiUrl = getApiUrlConfig();
|
|
598
|
+
try {
|
|
599
|
+
const response = await fetch(`${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" };
|
|
1312
|
+
}
|
|
1313
|
+
|
|
1314
|
+
// src/lib/opencode/opencode-version-gate.ts
|
|
1315
|
+
var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11", "1.18.3"];
|
|
1316
|
+
function isQueueValidatedVersion(version2) {
|
|
730
1317
|
if (!version2) return false;
|
|
731
1318
|
return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version2);
|
|
732
1319
|
}
|
|
@@ -1267,6 +1854,16 @@ async function buildFileParts(attachments, capable) {
|
|
|
1267
1854
|
);
|
|
1268
1855
|
dataUrl = null;
|
|
1269
1856
|
}
|
|
1857
|
+
if (dataUrl !== null && typeof dataUrl === "object") {
|
|
1858
|
+
outcomes.push({
|
|
1859
|
+
index: a.index,
|
|
1860
|
+
mime: a.mime,
|
|
1861
|
+
filename: a.filename,
|
|
1862
|
+
status: "failed",
|
|
1863
|
+
reason: "needs_reauth"
|
|
1864
|
+
});
|
|
1865
|
+
continue;
|
|
1866
|
+
}
|
|
1270
1867
|
if (dataUrl == null) {
|
|
1271
1868
|
outcomes.push({ index: a.index, mime: a.mime, filename: a.filename, status: "failed" });
|
|
1272
1869
|
continue;
|
|
@@ -1348,7 +1945,7 @@ async function sendPromptAsync(port, sessionId, content, options, attachments) {
|
|
|
1348
1945
|
}
|
|
1349
1946
|
}
|
|
1350
1947
|
if (attempt < READ_BACK_ATTEMPTS - 1) {
|
|
1351
|
-
await new Promise((
|
|
1948
|
+
await new Promise((resolve3) => setTimeout(resolve3, READ_BACK_DELAY_MS));
|
|
1352
1949
|
}
|
|
1353
1950
|
}
|
|
1354
1951
|
return null;
|
|
@@ -1476,6 +2073,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
|
|
|
1476
2073
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1477
2074
|
return completedOf(reply) != null && finishOf(reply) === "tool-calls";
|
|
1478
2075
|
}
|
|
2076
|
+
function isB2AbandonmentConfirmed(params) {
|
|
2077
|
+
return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
|
|
2078
|
+
}
|
|
1479
2079
|
function messageError(messages, userMessageId) {
|
|
1480
2080
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1481
2081
|
const error2 = errorOf(reply);
|
|
@@ -1489,6 +2089,42 @@ function messageError(messages, userMessageId) {
|
|
|
1489
2089
|
}
|
|
1490
2090
|
return "The agent run failed.";
|
|
1491
2091
|
}
|
|
2092
|
+
function messageFailure(messages, userMessageId) {
|
|
2093
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
2094
|
+
const error2 = errorOf(reply);
|
|
2095
|
+
if (error2 == null || typeof error2 !== "object") return null;
|
|
2096
|
+
const e = error2;
|
|
2097
|
+
const replyProviderId = reply?.info?.providerID ?? null;
|
|
2098
|
+
const replyModelId = reply?.info?.modelID ?? null;
|
|
2099
|
+
if (e.name === "ProviderAuthError") {
|
|
2100
|
+
const data = e.data;
|
|
2101
|
+
const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
|
|
2102
|
+
return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
|
|
2103
|
+
}
|
|
2104
|
+
if (e.name === "APIError") {
|
|
2105
|
+
const data = e.data;
|
|
2106
|
+
const statusCode = data?.statusCode;
|
|
2107
|
+
if (statusCode === 401 || statusCode === 403) {
|
|
2108
|
+
return {
|
|
2109
|
+
kind: "model_auth",
|
|
2110
|
+
providerId: replyProviderId,
|
|
2111
|
+
modelId: replyModelId,
|
|
2112
|
+
reason: "rejected"
|
|
2113
|
+
};
|
|
2114
|
+
}
|
|
2115
|
+
}
|
|
2116
|
+
return null;
|
|
2117
|
+
}
|
|
2118
|
+
function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
|
|
2119
|
+
if (classified != null) return classified;
|
|
2120
|
+
if (hasConfiguredProvider !== false) return null;
|
|
2121
|
+
return {
|
|
2122
|
+
kind: "model_auth",
|
|
2123
|
+
providerId: replyProviderId,
|
|
2124
|
+
modelId: replyModelId,
|
|
2125
|
+
reason: "missing"
|
|
2126
|
+
};
|
|
2127
|
+
}
|
|
1492
2128
|
function hasRunningAssistantExcept(messages, exceptUserMessageId) {
|
|
1493
2129
|
if (!messages || messages.length === 0) return false;
|
|
1494
2130
|
return messages.some(
|
|
@@ -1684,10 +2320,11 @@ var StreamForwarder = class {
|
|
|
1684
2320
|
* Abort every in-flight stream (e.g. on WebSocket close).
|
|
1685
2321
|
*/
|
|
1686
2322
|
abortAll() {
|
|
1687
|
-
for (const stream of this.inflight.
|
|
2323
|
+
for (const [sid, stream] of this.inflight.entries()) {
|
|
1688
2324
|
try {
|
|
1689
2325
|
stream.abort();
|
|
1690
|
-
} catch {
|
|
2326
|
+
} catch (err) {
|
|
2327
|
+
log("error", "forwarder_abort_failed", { sid, ...errorFields(err) });
|
|
1691
2328
|
}
|
|
1692
2329
|
}
|
|
1693
2330
|
this.inflight.clear();
|
|
@@ -1721,12 +2358,12 @@ var StreamForwarder = class {
|
|
|
1721
2358
|
let endBody;
|
|
1722
2359
|
if (has_body) {
|
|
1723
2360
|
const chunks = [];
|
|
1724
|
-
bodyPromise = new Promise((
|
|
2361
|
+
bodyPromise = new Promise((resolve3) => {
|
|
1725
2362
|
pushBody = (buf) => {
|
|
1726
2363
|
chunks.push(buf);
|
|
1727
2364
|
};
|
|
1728
2365
|
endBody = () => {
|
|
1729
|
-
|
|
2366
|
+
resolve3(Buffer.concat(chunks));
|
|
1730
2367
|
};
|
|
1731
2368
|
});
|
|
1732
2369
|
}
|
|
@@ -1843,7 +2480,7 @@ function connectTunnel(options) {
|
|
|
1843
2480
|
} = options;
|
|
1844
2481
|
const tunnelUrl = getTunnelUrlConfig();
|
|
1845
2482
|
const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
|
|
1846
|
-
return new Promise((
|
|
2483
|
+
return new Promise((resolve3, reject) => {
|
|
1847
2484
|
const ws = new WebSocket2(url, {
|
|
1848
2485
|
headers: {
|
|
1849
2486
|
Authorization: authHeader
|
|
@@ -1898,7 +2535,7 @@ function connectTunnel(options) {
|
|
|
1898
2535
|
clearTimeout(connectionTimeout);
|
|
1899
2536
|
const connectedAgentId = message.agent_id ?? agentId;
|
|
1900
2537
|
onConnected?.(connectedAgentId);
|
|
1901
|
-
|
|
2538
|
+
resolve3({
|
|
1902
2539
|
ws,
|
|
1903
2540
|
close: () => ws.close(1e3, "CLI shutdown")
|
|
1904
2541
|
});
|
|
@@ -1951,66 +2588,511 @@ var RunnerConnection = class {
|
|
|
1951
2588
|
get agentId() {
|
|
1952
2589
|
return this.resolvedAgentId;
|
|
1953
2590
|
}
|
|
1954
|
-
/** Establish the initial tunnel connection (with retry/backoff). */
|
|
1955
|
-
async connect() {
|
|
1956
|
-
await this.connectWithRetry(false);
|
|
2591
|
+
/** Establish the initial tunnel connection (with retry/backoff). */
|
|
2592
|
+
async connect() {
|
|
2593
|
+
await this.connectWithRetry(false);
|
|
2594
|
+
}
|
|
2595
|
+
/** Close the active connection (idempotent). */
|
|
2596
|
+
close() {
|
|
2597
|
+
if (this.connection) {
|
|
2598
|
+
try {
|
|
2599
|
+
this.connection.close();
|
|
2600
|
+
} catch (err) {
|
|
2601
|
+
log("error", "runner_connection_close_failed", {
|
|
2602
|
+
agent_id: this.resolvedAgentId,
|
|
2603
|
+
...errorFields(err)
|
|
2604
|
+
});
|
|
2605
|
+
}
|
|
2606
|
+
this.connection = null;
|
|
2607
|
+
}
|
|
2608
|
+
}
|
|
2609
|
+
async connectWithRetry(isReconnect) {
|
|
2610
|
+
if (isReconnect && this.reconnecting) return;
|
|
2611
|
+
this.reconnecting = true;
|
|
2612
|
+
this.close();
|
|
2613
|
+
const { events } = this.opts;
|
|
2614
|
+
while (this.opts.isRunning()) {
|
|
2615
|
+
try {
|
|
2616
|
+
this.connection = await connectTunnel({
|
|
2617
|
+
agentId: this.resolvedAgentId,
|
|
2618
|
+
authHeader: this.opts.getAuthHeader(),
|
|
2619
|
+
port: this.opts.port,
|
|
2620
|
+
onConnected: (agentId) => {
|
|
2621
|
+
this.reconnectAttempt = 0;
|
|
2622
|
+
this.reconnecting = false;
|
|
2623
|
+
this.resolvedAgentId = agentId;
|
|
2624
|
+
events.onConnected(agentId, isReconnect);
|
|
2625
|
+
},
|
|
2626
|
+
onDisconnected: (code, reason) => {
|
|
2627
|
+
events.onDisconnected(code, reason);
|
|
2628
|
+
if (this.opts.isRunning() && code !== 1e3 && !this.reconnecting) {
|
|
2629
|
+
this.reconnectPromise = this.connectWithRetry(true).catch((err) => {
|
|
2630
|
+
events.onError?.(`Reconnection failed: ${err.message}`);
|
|
2631
|
+
});
|
|
2632
|
+
}
|
|
2633
|
+
},
|
|
2634
|
+
onError: (error2) => events.onError?.(error2),
|
|
2635
|
+
onResponse: () => events.onResponse?.(),
|
|
2636
|
+
onDrainPing: () => events.onDrainPing?.(),
|
|
2637
|
+
onInfo: (message) => events.onInfo?.(message)
|
|
2638
|
+
});
|
|
2639
|
+
return;
|
|
2640
|
+
} catch (error2) {
|
|
2641
|
+
this.reconnectAttempt++;
|
|
2642
|
+
if (error2.message === "Unauthorized") {
|
|
2643
|
+
this.reconnecting = false;
|
|
2644
|
+
throw error2;
|
|
2645
|
+
}
|
|
2646
|
+
const delay = getReconnectDelay(this.reconnectAttempt);
|
|
2647
|
+
events.onReconnecting?.(this.reconnectAttempt);
|
|
2648
|
+
events.onError?.(`Connection failed, retrying in ${Math.round(delay / 1e3)}s...`);
|
|
2649
|
+
await this.sleep(delay);
|
|
2650
|
+
}
|
|
2651
|
+
}
|
|
2652
|
+
this.reconnecting = false;
|
|
2653
|
+
}
|
|
2654
|
+
};
|
|
2655
|
+
|
|
2656
|
+
// src/lib/tunnel/ready-marker.ts
|
|
2657
|
+
import { writeFileSync } from "fs";
|
|
2658
|
+
function writeTunnelReadyMarker(path, agentId) {
|
|
2659
|
+
try {
|
|
2660
|
+
writeFileSync(path, `${agentId}
|
|
2661
|
+
`);
|
|
2662
|
+
return { ok: true };
|
|
2663
|
+
} catch (error2) {
|
|
2664
|
+
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
2665
|
+
}
|
|
2666
|
+
}
|
|
2667
|
+
|
|
2668
|
+
// src/lib/claude-usage-reporting.ts
|
|
2669
|
+
var VALID_MODES = ["auto", "on", "off"];
|
|
2670
|
+
function resolveClaudeUsageReportingMode(flagValue, env) {
|
|
2671
|
+
const raw = flagValue ?? env.EVIDENT_CLAUDE_USAGE_REPORTING;
|
|
2672
|
+
if (raw === void 0 || raw === "") {
|
|
2673
|
+
return { mode: "auto", warnings: [] };
|
|
2674
|
+
}
|
|
2675
|
+
const normalized = raw.trim().toLowerCase();
|
|
2676
|
+
if (VALID_MODES.includes(normalized)) {
|
|
2677
|
+
return { mode: normalized, warnings: [] };
|
|
2678
|
+
}
|
|
2679
|
+
const source = flagValue !== void 0 ? "--claude-usage-reporting" : "EVIDENT_CLAUDE_USAGE_REPORTING";
|
|
2680
|
+
return {
|
|
2681
|
+
mode: "auto",
|
|
2682
|
+
warnings: [
|
|
2683
|
+
`Ignoring invalid ${source} "${raw}": expected one of ${VALID_MODES.join(", ")}; using auto`
|
|
2684
|
+
]
|
|
2685
|
+
};
|
|
2686
|
+
}
|
|
2687
|
+
var BASE_REPORT_DELAY_MS = 10 * 6e4;
|
|
2688
|
+
var REPORT_DELAY_JITTER_FRACTION = 0.2;
|
|
2689
|
+
function nextReportDelayMs(random = Math.random) {
|
|
2690
|
+
const jitterRangeMs = BASE_REPORT_DELAY_MS * REPORT_DELAY_JITTER_FRACTION;
|
|
2691
|
+
return BASE_REPORT_DELAY_MS - jitterRangeMs + random() * (2 * jitterRangeMs);
|
|
2692
|
+
}
|
|
2693
|
+
var FIRST_REPORT_DELAY_MS = 5e3 + Math.random() * 1e4;
|
|
2694
|
+
var CLAUDE_USAGE_FAILURE_REESCALATION_TICKS = 6;
|
|
2695
|
+
function claudeUsageFailureLogLevel(consecutiveFailures) {
|
|
2696
|
+
return consecutiveFailures === 1 || consecutiveFailures % CLAUDE_USAGE_FAILURE_REESCALATION_TICKS === 0 ? "warn" : "debug";
|
|
2697
|
+
}
|
|
2698
|
+
|
|
2699
|
+
// src/lib/channels/driver.ts
|
|
2700
|
+
import { homedir as homedir2 } from "os";
|
|
2701
|
+
|
|
2702
|
+
// src/lib/file-push.ts
|
|
2703
|
+
import { randomUUID } from "crypto";
|
|
2704
|
+
import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
|
|
2705
|
+
import { basename, dirname as dirname2, isAbsolute, join as join2, relative, resolve as resolve2, sep } from "path";
|
|
2706
|
+
var FILE_MODE = 384;
|
|
2707
|
+
var DIRECTORY_MODE = 448;
|
|
2708
|
+
async function writePushedFile(request) {
|
|
2709
|
+
const { requestedPath, content, allowedDirectories, homeDir } = request;
|
|
2710
|
+
const bytes = content.byteLength;
|
|
2711
|
+
if (allowedDirectories.length === 0) {
|
|
2712
|
+
return refuse("file_sync_disabled", "File sync is not enabled on this runner.", {
|
|
2713
|
+
path: requestedPath,
|
|
2714
|
+
bytes
|
|
2715
|
+
});
|
|
2716
|
+
}
|
|
2717
|
+
if (bytes > MAX_FILE_PUSH_BYTES) {
|
|
2718
|
+
return refuse(
|
|
2719
|
+
"file_too_large",
|
|
2720
|
+
`File is ${bytes} bytes; the limit is ${MAX_FILE_PUSH_BYTES}.`,
|
|
2721
|
+
{
|
|
2722
|
+
path: requestedPath,
|
|
2723
|
+
bytes
|
|
2724
|
+
}
|
|
2725
|
+
);
|
|
2726
|
+
}
|
|
2727
|
+
const candidate = expandAndValidate(requestedPath, homeDir);
|
|
2728
|
+
if (candidate === null) {
|
|
2729
|
+
return refuse("invalid_path", "The requested path is not a valid absolute file path.", {
|
|
2730
|
+
path: requestedPath,
|
|
2731
|
+
bytes
|
|
2732
|
+
});
|
|
2733
|
+
}
|
|
2734
|
+
try {
|
|
2735
|
+
const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
|
|
2736
|
+
dirname2(candidate)
|
|
2737
|
+
);
|
|
2738
|
+
const realTarget = join2(existingAncestor, ...missingSegments, basename(candidate));
|
|
2739
|
+
const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
|
|
2740
|
+
if (allowedDirectory === null) {
|
|
2741
|
+
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
2742
|
+
path: realTarget,
|
|
2743
|
+
bytes
|
|
2744
|
+
});
|
|
2745
|
+
}
|
|
2746
|
+
if (missingSegments.length > 0) {
|
|
2747
|
+
await createMissingDirectories(existingAncestor, missingSegments);
|
|
2748
|
+
const realParent = await realpath(dirname2(realTarget));
|
|
2749
|
+
if (realParent !== dirname2(realTarget) || !contains(allowedDirectory, realTarget)) {
|
|
2750
|
+
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
2751
|
+
path: realTarget,
|
|
2752
|
+
bytes,
|
|
2753
|
+
reason: "parent_changed_after_create"
|
|
2754
|
+
});
|
|
2755
|
+
}
|
|
2756
|
+
}
|
|
2757
|
+
await writeAtomically(realTarget, content);
|
|
2758
|
+
log("info", "file_push_written", { path: realTarget, bytes });
|
|
2759
|
+
return { ok: true, path: realTarget };
|
|
2760
|
+
} catch (err) {
|
|
2761
|
+
const errno = err.code ?? "UNKNOWN";
|
|
2762
|
+
return refuse("write_failed", `The runner could not write the file (${errno}).`, {
|
|
2763
|
+
path: candidate,
|
|
2764
|
+
bytes,
|
|
2765
|
+
errno,
|
|
2766
|
+
...errorFields(err)
|
|
2767
|
+
});
|
|
2768
|
+
}
|
|
2769
|
+
}
|
|
2770
|
+
function expandAndValidate(requestedPath, homeDir) {
|
|
2771
|
+
if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
|
|
2772
|
+
return null;
|
|
2773
|
+
}
|
|
2774
|
+
const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join2(homeDir, requestedPath.slice(2)) : requestedPath;
|
|
2775
|
+
if (expanded.split(/[/\\]/).includes("..")) {
|
|
2776
|
+
return null;
|
|
2777
|
+
}
|
|
2778
|
+
if (!isAbsolute(expanded)) {
|
|
2779
|
+
return null;
|
|
2780
|
+
}
|
|
2781
|
+
const candidate = resolve2(expanded);
|
|
2782
|
+
const name = basename(candidate);
|
|
2783
|
+
return name === "" || name === "." || name === ".." ? null : candidate;
|
|
2784
|
+
}
|
|
2785
|
+
async function resolveNearestExistingAncestor(directory) {
|
|
2786
|
+
const missingSegments = [];
|
|
2787
|
+
let current = directory;
|
|
2788
|
+
for (; ; ) {
|
|
2789
|
+
try {
|
|
2790
|
+
return { existingAncestor: await realpath(current), missingSegments };
|
|
2791
|
+
} catch (err) {
|
|
2792
|
+
const parent = dirname2(current);
|
|
2793
|
+
if (err.code !== "ENOENT" || parent === current) {
|
|
2794
|
+
throw err;
|
|
2795
|
+
}
|
|
2796
|
+
missingSegments.unshift(basename(current));
|
|
2797
|
+
current = parent;
|
|
2798
|
+
}
|
|
2799
|
+
}
|
|
2800
|
+
}
|
|
2801
|
+
async function findContainingAllowedDirectory(allowedDirectories, realTarget) {
|
|
2802
|
+
for (const directory of allowedDirectories) {
|
|
2803
|
+
if (!isAbsolute(directory)) {
|
|
2804
|
+
log("warn", "file_push_allowed_directory_skipped", { directory, reason: "not_absolute" });
|
|
2805
|
+
continue;
|
|
2806
|
+
}
|
|
2807
|
+
const realDirectory = await realpathCreatingIfMissing(directory);
|
|
2808
|
+
if (realDirectory !== null && contains(realDirectory, realTarget)) {
|
|
2809
|
+
return realDirectory;
|
|
2810
|
+
}
|
|
2811
|
+
}
|
|
2812
|
+
return null;
|
|
2813
|
+
}
|
|
2814
|
+
async function realpathCreatingIfMissing(directory) {
|
|
2815
|
+
try {
|
|
2816
|
+
return await realpath(directory);
|
|
2817
|
+
} catch (err) {
|
|
2818
|
+
if (err.code !== "ENOENT") {
|
|
2819
|
+
log("warn", "file_push_allowed_directory_skipped", {
|
|
2820
|
+
directory,
|
|
2821
|
+
reason: "unresolvable",
|
|
2822
|
+
...errorFields(err)
|
|
2823
|
+
});
|
|
2824
|
+
return null;
|
|
2825
|
+
}
|
|
2826
|
+
}
|
|
2827
|
+
try {
|
|
2828
|
+
await mkdir(directory, { recursive: true, mode: DIRECTORY_MODE });
|
|
2829
|
+
await chmod(directory, DIRECTORY_MODE);
|
|
2830
|
+
return await realpath(directory);
|
|
2831
|
+
} catch (err) {
|
|
2832
|
+
log("warn", "file_push_allowed_directory_skipped", {
|
|
2833
|
+
directory,
|
|
2834
|
+
reason: "create_failed",
|
|
2835
|
+
...errorFields(err)
|
|
2836
|
+
});
|
|
2837
|
+
return null;
|
|
2838
|
+
}
|
|
2839
|
+
}
|
|
2840
|
+
function contains(realDirectory, realTarget) {
|
|
2841
|
+
const rel = relative(realDirectory, realTarget);
|
|
2842
|
+
return rel !== "" && rel !== ".." && !rel.startsWith(`..${sep}`) && !isAbsolute(rel);
|
|
2843
|
+
}
|
|
2844
|
+
async function createMissingDirectories(existingAncestor, missingSegments) {
|
|
2845
|
+
let current = existingAncestor;
|
|
2846
|
+
for (const segment of missingSegments) {
|
|
2847
|
+
current = join2(current, segment);
|
|
2848
|
+
await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
|
|
2849
|
+
await chmod(current, DIRECTORY_MODE);
|
|
2850
|
+
}
|
|
2851
|
+
}
|
|
2852
|
+
async function writeAtomically(realTarget, content) {
|
|
2853
|
+
const temporaryPath = join2(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
|
|
2854
|
+
let handle;
|
|
2855
|
+
try {
|
|
2856
|
+
handle = await open2(temporaryPath, "wx", FILE_MODE);
|
|
2857
|
+
await handle.writeFile(content);
|
|
2858
|
+
await handle.chmod(FILE_MODE);
|
|
2859
|
+
await handle.close();
|
|
2860
|
+
handle = void 0;
|
|
2861
|
+
await rename(temporaryPath, realTarget);
|
|
2862
|
+
} catch (err) {
|
|
2863
|
+
await discardTemporaryFile(temporaryPath, handle);
|
|
2864
|
+
throw err;
|
|
2865
|
+
}
|
|
2866
|
+
}
|
|
2867
|
+
async function discardTemporaryFile(temporaryPath, handle) {
|
|
2868
|
+
try {
|
|
2869
|
+
await handle?.close();
|
|
2870
|
+
} catch (err) {
|
|
2871
|
+
log("warn", "file_push_temp_close_failed", { path: temporaryPath, ...errorFields(err) });
|
|
2872
|
+
}
|
|
2873
|
+
try {
|
|
2874
|
+
await unlink(temporaryPath);
|
|
2875
|
+
} catch (err) {
|
|
2876
|
+
const errno = err.code;
|
|
2877
|
+
if (errno !== "ENOENT" && errno !== "ENOTDIR") {
|
|
2878
|
+
log("warn", "file_push_temp_cleanup_failed", { path: temporaryPath, ...errorFields(err) });
|
|
2879
|
+
}
|
|
2880
|
+
}
|
|
2881
|
+
}
|
|
2882
|
+
function refuse(code, message, fields) {
|
|
2883
|
+
log(code === "write_failed" ? "error" : "warn", "file_push_refused", { code, ...fields });
|
|
2884
|
+
return { ok: false, code, message };
|
|
2885
|
+
}
|
|
2886
|
+
|
|
2887
|
+
// src/lib/runner-file-sync.ts
|
|
2888
|
+
var MAX_ACK_ATTEMPTS = 5;
|
|
2889
|
+
async function syncPendingRunnerFiles(options) {
|
|
2890
|
+
const pending = await listPendingFiles(options);
|
|
2891
|
+
const pendingIds = new Set(pending.map((file) => file.id));
|
|
2892
|
+
for (const id of options.ackFailures.keys()) {
|
|
2893
|
+
if (!pendingIds.has(id)) options.ackFailures.delete(id);
|
|
2894
|
+
}
|
|
2895
|
+
if (pending.length === 0) return 0;
|
|
2896
|
+
options.log({
|
|
2897
|
+
level: "info",
|
|
2898
|
+
message: `Runner file sync: ${pending.length} file(s) queued for this runner`
|
|
2899
|
+
});
|
|
2900
|
+
let applied = 0;
|
|
2901
|
+
for (const file of pending) {
|
|
2902
|
+
if ((options.ackFailures.get(file.id) ?? 0) >= MAX_ACK_ATTEMPTS) continue;
|
|
2903
|
+
if (await applyOne(options, file)) applied += 1;
|
|
2904
|
+
}
|
|
2905
|
+
return applied;
|
|
2906
|
+
}
|
|
2907
|
+
async function listPendingFiles(options) {
|
|
2908
|
+
let res;
|
|
2909
|
+
try {
|
|
2910
|
+
res = await options.fetchImpl(`${options.apiUrl}/runners/${options.agentId}/files/pending`, {
|
|
2911
|
+
headers: { Authorization: options.getAuthHeader() }
|
|
2912
|
+
});
|
|
2913
|
+
} catch (err) {
|
|
2914
|
+
options.log({
|
|
2915
|
+
level: "warn",
|
|
2916
|
+
message: `Could not list pending runner files \u2014 retrying on the next drain: ${describe(err)}`
|
|
2917
|
+
});
|
|
2918
|
+
return [];
|
|
2919
|
+
}
|
|
2920
|
+
if (!res.ok) {
|
|
2921
|
+
options.log({
|
|
2922
|
+
level: res.status === 404 ? "debug" : "warn",
|
|
2923
|
+
message: `Listing pending runner files returned HTTP ${res.status} \u2014 retrying on the next drain`
|
|
2924
|
+
});
|
|
2925
|
+
return [];
|
|
2926
|
+
}
|
|
2927
|
+
let body;
|
|
2928
|
+
try {
|
|
2929
|
+
body = await res.json();
|
|
2930
|
+
} catch (err) {
|
|
2931
|
+
options.log({
|
|
2932
|
+
level: "warn",
|
|
2933
|
+
message: `Pending runner file list was not readable JSON \u2014 retrying on the next drain: ${describe(err)}`
|
|
2934
|
+
});
|
|
2935
|
+
return [];
|
|
2936
|
+
}
|
|
2937
|
+
if (!Array.isArray(body)) {
|
|
2938
|
+
options.log({
|
|
2939
|
+
level: "warn",
|
|
2940
|
+
message: "Pending runner file list was not an array \u2014 ignoring it for this drain"
|
|
2941
|
+
});
|
|
2942
|
+
return [];
|
|
2943
|
+
}
|
|
2944
|
+
const files = [];
|
|
2945
|
+
for (const entry of body) {
|
|
2946
|
+
const file = asPendingFile(entry);
|
|
2947
|
+
if (file === null) {
|
|
2948
|
+
options.log({
|
|
2949
|
+
level: "warn",
|
|
2950
|
+
message: "Ignoring a malformed pending runner file entry (expected id, path and size)"
|
|
2951
|
+
});
|
|
2952
|
+
continue;
|
|
2953
|
+
}
|
|
2954
|
+
files.push(file);
|
|
2955
|
+
}
|
|
2956
|
+
return files;
|
|
2957
|
+
}
|
|
2958
|
+
function asPendingFile(entry) {
|
|
2959
|
+
if (entry === null || typeof entry !== "object") return null;
|
|
2960
|
+
const { id, path, size } = entry;
|
|
2961
|
+
if (typeof id !== "string" || id === "") return null;
|
|
2962
|
+
if (typeof path !== "string" || path === "") return null;
|
|
2963
|
+
if (typeof size !== "number" || !Number.isFinite(size) || size < 0) return null;
|
|
2964
|
+
return { id, path, size };
|
|
2965
|
+
}
|
|
2966
|
+
async function applyOne(options, file) {
|
|
2967
|
+
const label = `${file.id.slice(0, 8)} (${file.path})`;
|
|
2968
|
+
if (options.allowedDirectories.length === 0) {
|
|
2969
|
+
options.log({
|
|
2970
|
+
level: "warn",
|
|
2971
|
+
message: `Runner file ${label} rejected: file sync is not enabled on this runner (start it with --enable-file-sync-to)`
|
|
2972
|
+
});
|
|
2973
|
+
await ack(options, file, "rejected", "file_sync_disabled");
|
|
2974
|
+
return false;
|
|
2975
|
+
}
|
|
2976
|
+
if (file.size > MAX_FILE_PUSH_BYTES) {
|
|
2977
|
+
options.log({
|
|
2978
|
+
level: "warn",
|
|
2979
|
+
message: `Runner file ${label} rejected: declared ${file.size} bytes, the limit is ${MAX_FILE_PUSH_BYTES}`
|
|
2980
|
+
});
|
|
2981
|
+
await ack(options, file, "rejected", "file_too_large");
|
|
2982
|
+
return false;
|
|
2983
|
+
}
|
|
2984
|
+
const download = await downloadContent(options, file, label);
|
|
2985
|
+
if (!download.ok) {
|
|
2986
|
+
if (download.terminal) await ack(options, file, "rejected", download.code);
|
|
2987
|
+
return false;
|
|
2988
|
+
}
|
|
2989
|
+
let outcome;
|
|
2990
|
+
try {
|
|
2991
|
+
outcome = await writePushedFile({
|
|
2992
|
+
requestedPath: file.path,
|
|
2993
|
+
content: download.content,
|
|
2994
|
+
allowedDirectories: options.allowedDirectories,
|
|
2995
|
+
homeDir: options.homeDir
|
|
2996
|
+
});
|
|
2997
|
+
} catch (err) {
|
|
2998
|
+
options.log({
|
|
2999
|
+
level: "error",
|
|
3000
|
+
message: `Runner file ${label} could not be written: ${describe(err)}`
|
|
3001
|
+
});
|
|
3002
|
+
await ack(options, file, "rejected", "write_failed");
|
|
3003
|
+
return false;
|
|
3004
|
+
}
|
|
3005
|
+
if (!outcome.ok) {
|
|
3006
|
+
options.log({
|
|
3007
|
+
level: "warn",
|
|
3008
|
+
message: `Runner file ${label} rejected (${outcome.code}): ${outcome.message}`
|
|
3009
|
+
});
|
|
3010
|
+
await ack(options, file, "rejected", outcome.code);
|
|
3011
|
+
return false;
|
|
1957
3012
|
}
|
|
1958
|
-
|
|
1959
|
-
|
|
1960
|
-
|
|
1961
|
-
|
|
1962
|
-
|
|
1963
|
-
|
|
3013
|
+
options.log({
|
|
3014
|
+
level: "info",
|
|
3015
|
+
message: `Runner file ${label} applied (${download.content.byteLength} bytes)`
|
|
3016
|
+
});
|
|
3017
|
+
await ack(options, file, "applied");
|
|
3018
|
+
return true;
|
|
3019
|
+
}
|
|
3020
|
+
function durableDownloadCode(status2) {
|
|
3021
|
+
return status2 === 413 ? "file_too_large" : "write_failed";
|
|
3022
|
+
}
|
|
3023
|
+
async function downloadContent(options, file, label) {
|
|
3024
|
+
try {
|
|
3025
|
+
const res = await options.fetchImpl(
|
|
3026
|
+
`${options.apiUrl}/runners/${options.agentId}/files/${file.id}/content`,
|
|
3027
|
+
{ headers: { Authorization: options.getAuthHeader() } }
|
|
3028
|
+
);
|
|
3029
|
+
if (!res.ok) {
|
|
3030
|
+
const terminal = res.status >= 400 && res.status < 500 && res.status !== 401 && res.status !== 403 && res.status !== 408 && res.status !== 429;
|
|
3031
|
+
if (!terminal) {
|
|
3032
|
+
options.log({
|
|
3033
|
+
level: "warn",
|
|
3034
|
+
message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 retrying on the next drain`
|
|
3035
|
+
});
|
|
3036
|
+
return { ok: false, terminal: false };
|
|
1964
3037
|
}
|
|
1965
|
-
|
|
3038
|
+
const code = durableDownloadCode(res.status);
|
|
3039
|
+
options.log({
|
|
3040
|
+
level: "error",
|
|
3041
|
+
message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 rejecting it as ${code} (the bytes never reached the writer)`
|
|
3042
|
+
});
|
|
3043
|
+
return { ok: false, terminal: true, code };
|
|
1966
3044
|
}
|
|
3045
|
+
return { ok: true, content: Buffer.from(await res.arrayBuffer()) };
|
|
3046
|
+
} catch (err) {
|
|
3047
|
+
options.log({
|
|
3048
|
+
level: "warn",
|
|
3049
|
+
message: `Downloading runner file ${label} failed \u2014 retrying on the next drain: ${describe(err)}`
|
|
3050
|
+
});
|
|
3051
|
+
return { ok: false, terminal: false };
|
|
1967
3052
|
}
|
|
1968
|
-
|
|
1969
|
-
|
|
1970
|
-
|
|
1971
|
-
|
|
1972
|
-
const
|
|
1973
|
-
|
|
1974
|
-
|
|
1975
|
-
|
|
1976
|
-
|
|
1977
|
-
|
|
1978
|
-
|
|
1979
|
-
|
|
1980
|
-
|
|
1981
|
-
this.reconnecting = false;
|
|
1982
|
-
this.resolvedAgentId = agentId;
|
|
1983
|
-
events.onConnected(agentId, isReconnect);
|
|
1984
|
-
},
|
|
1985
|
-
onDisconnected: (code, reason) => {
|
|
1986
|
-
events.onDisconnected(code, reason);
|
|
1987
|
-
if (this.opts.isRunning() && code !== 1e3 && !this.reconnecting) {
|
|
1988
|
-
this.reconnectPromise = this.connectWithRetry(true).catch((err) => {
|
|
1989
|
-
events.onError?.(`Reconnection failed: ${err.message}`);
|
|
1990
|
-
});
|
|
1991
|
-
}
|
|
1992
|
-
},
|
|
1993
|
-
onError: (error2) => events.onError?.(error2),
|
|
1994
|
-
onResponse: () => events.onResponse?.(),
|
|
1995
|
-
onDrainPing: () => events.onDrainPing?.(),
|
|
1996
|
-
onInfo: (message) => events.onInfo?.(message)
|
|
1997
|
-
});
|
|
1998
|
-
return;
|
|
1999
|
-
} catch (error2) {
|
|
2000
|
-
this.reconnectAttempt++;
|
|
2001
|
-
if (error2.message === "Unauthorized") {
|
|
2002
|
-
this.reconnecting = false;
|
|
2003
|
-
throw error2;
|
|
2004
|
-
}
|
|
2005
|
-
const delay = getReconnectDelay(this.reconnectAttempt);
|
|
2006
|
-
events.onReconnecting?.(this.reconnectAttempt);
|
|
2007
|
-
events.onError?.(`Connection failed, retrying in ${Math.round(delay / 1e3)}s...`);
|
|
2008
|
-
await this.sleep(delay);
|
|
3053
|
+
}
|
|
3054
|
+
async function ack(options, file, status2, reason) {
|
|
3055
|
+
const outcome = `${status2}${reason ? ` (${reason})` : ""}`;
|
|
3056
|
+
try {
|
|
3057
|
+
const res = await options.fetchImpl(
|
|
3058
|
+
`${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
|
|
3059
|
+
{
|
|
3060
|
+
method: "POST",
|
|
3061
|
+
headers: {
|
|
3062
|
+
Authorization: options.getAuthHeader(),
|
|
3063
|
+
"Content-Type": "application/json"
|
|
3064
|
+
},
|
|
3065
|
+
body: JSON.stringify(reason ? { status: status2, reason } : { status: status2 })
|
|
2009
3066
|
}
|
|
3067
|
+
);
|
|
3068
|
+
if (!res.ok) {
|
|
3069
|
+
recordAckFailure(
|
|
3070
|
+
options,
|
|
3071
|
+
file,
|
|
3072
|
+
`Acking runner file ${file.id.slice(0, 8)} as ${outcome} returned HTTP ${res.status}`
|
|
3073
|
+
);
|
|
3074
|
+
return;
|
|
2010
3075
|
}
|
|
2011
|
-
|
|
3076
|
+
options.ackFailures.delete(file.id);
|
|
3077
|
+
} catch (err) {
|
|
3078
|
+
recordAckFailure(
|
|
3079
|
+
options,
|
|
3080
|
+
file,
|
|
3081
|
+
`Acking runner file ${file.id.slice(0, 8)} as ${outcome} failed: ${describe(err)}`
|
|
3082
|
+
);
|
|
2012
3083
|
}
|
|
2013
|
-
}
|
|
3084
|
+
}
|
|
3085
|
+
function recordAckFailure(options, file, what) {
|
|
3086
|
+
const attempts = (options.ackFailures.get(file.id) ?? 0) + 1;
|
|
3087
|
+
options.ackFailures.set(file.id, attempts);
|
|
3088
|
+
options.log({
|
|
3089
|
+
level: "error",
|
|
3090
|
+
message: attempts >= MAX_ACK_ATTEMPTS ? `${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})`
|
|
3091
|
+
});
|
|
3092
|
+
}
|
|
3093
|
+
function describe(err) {
|
|
3094
|
+
return err instanceof Error ? err.message : String(err);
|
|
3095
|
+
}
|
|
2014
3096
|
|
|
2015
3097
|
// src/lib/channels/driver.ts
|
|
2016
3098
|
function messageIdOf(m) {
|
|
@@ -2040,7 +3122,10 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
|
|
|
2040
3122
|
var DEFAULT_STUCK_QUEUED_MS = 6e4;
|
|
2041
3123
|
var HEARTBEAT_MS = 6e4;
|
|
2042
3124
|
var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
|
|
3125
|
+
var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
|
|
3126
|
+
var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
|
|
2043
3127
|
var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
|
|
3128
|
+
var MAX_SUPERSEDED_CONVERSATIONS = 256;
|
|
2044
3129
|
var ChannelAuthError = class extends Error {
|
|
2045
3130
|
constructor(message) {
|
|
2046
3131
|
super(message);
|
|
@@ -2049,10 +3134,10 @@ var ChannelAuthError = class extends Error {
|
|
|
2049
3134
|
};
|
|
2050
3135
|
var ChannelTerminalError = class extends Error {
|
|
2051
3136
|
status;
|
|
2052
|
-
constructor(message,
|
|
3137
|
+
constructor(message, status2) {
|
|
2053
3138
|
super(message);
|
|
2054
3139
|
this.name = "ChannelTerminalError";
|
|
2055
|
-
this.status =
|
|
3140
|
+
this.status = status2;
|
|
2056
3141
|
}
|
|
2057
3142
|
};
|
|
2058
3143
|
function backoffDelay(attempt, policy) {
|
|
@@ -2060,10 +3145,10 @@ function backoffDelay(attempt, policy) {
|
|
|
2060
3145
|
const capped = Math.min(policy.maxDelayMs, exp);
|
|
2061
3146
|
return Math.floor(Math.random() * capped);
|
|
2062
3147
|
}
|
|
2063
|
-
function isRetryableStatus(
|
|
2064
|
-
return
|
|
3148
|
+
function isRetryableStatus(status2) {
|
|
3149
|
+
return status2 === 429 || status2 >= 500 && status2 <= 599;
|
|
2065
3150
|
}
|
|
2066
|
-
var ChannelDriver = class {
|
|
3151
|
+
var ChannelDriver = class _ChannelDriver {
|
|
2067
3152
|
agentId;
|
|
2068
3153
|
port;
|
|
2069
3154
|
apiUrl;
|
|
@@ -2077,8 +3162,38 @@ var ChannelDriver = class {
|
|
|
2077
3162
|
pausedMaxWaitMs;
|
|
2078
3163
|
stuckQueuedMs;
|
|
2079
3164
|
now;
|
|
3165
|
+
fileSyncDirectories;
|
|
3166
|
+
homeDir;
|
|
2080
3167
|
/** Cache of conversationId → opencode sessionId. */
|
|
2081
3168
|
sessions = /* @__PURE__ */ new Map();
|
|
3169
|
+
/**
|
|
3170
|
+
* conversationId → the opencode session this runner has ABANDONED as that
|
|
3171
|
+
* conversation's binding (#553), after a genuine (`sessionExists === true`)
|
|
3172
|
+
* dispatch failure: the session still exists but is wedged, so #485's self-heal
|
|
3173
|
+
* must bind a fresh one.
|
|
3174
|
+
*
|
|
3175
|
+
* Dropping the local binding + clearing the server row is not enough on its own:
|
|
3176
|
+
* a SIBLING message dispatched earlier in the same drain is still in-flight under
|
|
3177
|
+
* the same session, and its watcher's routine status writes carry
|
|
3178
|
+
* `opencode_session_id`, RESURRECTING the wedged id server-side after the clear —
|
|
3179
|
+
* and `ensureSession`'s persisted-id fallback then reuses it, defeating the
|
|
3180
|
+
* self-heal. This map makes the runner authoritative instead of racing those
|
|
3181
|
+
* writes: *`ensureSession` never reuses an abandoned id for that conversation,
|
|
3182
|
+
* whatever the server row says* — which holds even when the resurrecting write
|
|
3183
|
+
* is one we deliberately keep (see `markDone`).
|
|
3184
|
+
*
|
|
3185
|
+
* Bounded by construction, on both axes: keyed by CONVERSATION, so N failures on
|
|
3186
|
+
* one conversation hold ONE entry (the newest abandonment replaces the older), and
|
|
3187
|
+
* hard-capped at `MAX_SUPERSEDED_CONVERSATIONS` with FIFO eviction. Only the
|
|
3188
|
+
* NEWEST abandoned id per conversation is guarded: after a second abandonment a
|
|
3189
|
+
* late sibling of the FIRST session can write that id back and `ensureSession`
|
|
3190
|
+
* will reuse it — costing ONE repeat failure, which re-supersedes it. Deliberately
|
|
3191
|
+
* NOT dropped when the session's watcher tears down: `markDone` still writes the
|
|
3192
|
+
* abandoned id back (it must, or the reply is lost), so the guard has to outlive
|
|
3193
|
+
* the turn that resurrects it. In-memory only — a restart forgets it, at the same
|
|
3194
|
+
* bounded cost.
|
|
3195
|
+
*/
|
|
3196
|
+
supersededSessions = /* @__PURE__ */ new Map();
|
|
2082
3197
|
/**
|
|
2083
3198
|
* Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
|
|
2084
3199
|
* longer idempotent (no caller-supplied `messageID`), and its read-back picks
|
|
@@ -2146,6 +3261,24 @@ var ChannelDriver = class {
|
|
|
2146
3261
|
* processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
|
|
2147
3262
|
*/
|
|
2148
3263
|
readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
3264
|
+
/**
|
|
3265
|
+
* "Already emitted `redrive_unresolved` for this row" (#965). Mirrors
|
|
3266
|
+
* `readoptPollUnresolvedSignalled`: `resolveRedrive`'s `unresolved` leaf recurs
|
|
3267
|
+
* every ~2s drain until opencode's status becomes readable, but the
|
|
3268
|
+
* server-visible signal is an OUTCOME, so it fires at most once per row. Cleared
|
|
3269
|
+
* on any non-`unresolved` outcome so the set cannot grow beyond the currently
|
|
3270
|
+
* unresolvable rows.
|
|
3271
|
+
*/
|
|
3272
|
+
redriveUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
3273
|
+
/**
|
|
3274
|
+
* First `now()` a `pending` row's re-drive was observed `unresolved` (#965). A
|
|
3275
|
+
* `pending` row is invisible to every cron arm (all require `status =
|
|
3276
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
3277
|
+
* nothing driving it. Once `now - since >= pausedMaxWaitMs`, `resolveRedrive`
|
|
3278
|
+
* takes `dispatch` instead of `unresolved` (reusing the existing knob — see
|
|
3279
|
+
* ADR-0047's own "unreachable ⇒ bounded" rule). Cleared on any other outcome.
|
|
3280
|
+
*/
|
|
3281
|
+
redriveUnresolvedSince = /* @__PURE__ */ new Map();
|
|
2149
3282
|
/**
|
|
2150
3283
|
* "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
|
|
2151
3284
|
* Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
|
|
@@ -2188,9 +3321,12 @@ var ChannelDriver = class {
|
|
|
2188
3321
|
sessionParents = /* @__PURE__ */ new Map();
|
|
2189
3322
|
/**
|
|
2190
3323
|
* Per-session OpenCode title cache (#310), keyed by sessionId. Only a resolved
|
|
2191
|
-
* NON-EMPTY name is stored (terminal — a real session name
|
|
2192
|
-
* so we do NOT re-GET `/session/:id` every tick.
|
|
2193
|
-
*
|
|
3324
|
+
* NON-EMPTY, non-placeholder name is stored (terminal — a real session name
|
|
3325
|
+
* won't later un-name), so we do NOT re-GET `/session/:id` every tick. "Non-empty"
|
|
3326
|
+
* excludes OpenCode's synchronous default title (see
|
|
3327
|
+
* `OPENCODE_DEFAULT_TITLE_PREFIX`, #549) — that placeholder is treated the same
|
|
3328
|
+
* as an empty title so it never latches. A missing entry = not yet resolved OR
|
|
3329
|
+
* resolved-but-still-empty/placeholder → re-fetch on next need, since OpenCode
|
|
2194
3330
|
* names sessions asynchronously mid-turn. Driver-level (not per-watcher) so both
|
|
2195
3331
|
* the watcher completion path AND the restart-recovery re-adopt path (which has
|
|
2196
3332
|
* no watcher) can resolve the title.
|
|
@@ -2198,6 +3334,24 @@ var ChannelDriver = class {
|
|
|
2198
3334
|
sessionTitles = /* @__PURE__ */ new Map();
|
|
2199
3335
|
/** Serialises drains so a reconnect during a drain doesn't double-process. */
|
|
2200
3336
|
draining = false;
|
|
3337
|
+
/**
|
|
3338
|
+
* Serialises runner-file syncs (#559) so the ~2s poll tick and a concurrent
|
|
3339
|
+
* drain ping don't download, write and ack the same file twice.
|
|
3340
|
+
*/
|
|
3341
|
+
syncingFiles = false;
|
|
3342
|
+
/**
|
|
3343
|
+
* Consecutive failed acks per pending file (#559). Lives on the driver so it
|
|
3344
|
+
* survives across drains — without it, a file whose ack keeps failing is
|
|
3345
|
+
* re-downloaded and re-written every ~2s until the server expires it.
|
|
3346
|
+
*/
|
|
3347
|
+
fileAckFailures = /* @__PURE__ */ new Map();
|
|
3348
|
+
/**
|
|
3349
|
+
* Monotonic count of files this runner has pulled and written (#559). Only
|
|
3350
|
+
* ever increases, so `run.ts` detects work by comparing it against the value
|
|
3351
|
+
* it saw on the previous cycle — including work that landed mid-sleep, the
|
|
3352
|
+
* same trick `lastProxiedActivityAt` uses.
|
|
3353
|
+
*/
|
|
3354
|
+
appliedFileCount = 0;
|
|
2201
3355
|
/**
|
|
2202
3356
|
* The currently-executing `drainPending()` promise, or null when idle. Lets a
|
|
2203
3357
|
* graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
|
|
@@ -2213,21 +3367,23 @@ var ChannelDriver = class {
|
|
|
2213
3367
|
* and stops opencode.
|
|
2214
3368
|
*/
|
|
2215
3369
|
stopped = false;
|
|
2216
|
-
constructor(
|
|
2217
|
-
this.agentId =
|
|
2218
|
-
this.port =
|
|
2219
|
-
this.apiUrl =
|
|
2220
|
-
this.getAuthHeader =
|
|
2221
|
-
this.conversationFilter =
|
|
2222
|
-
this.retry = { ...DEFAULT_RETRY_POLICY, ...
|
|
2223
|
-
this.log =
|
|
3370
|
+
constructor(config) {
|
|
3371
|
+
this.agentId = config.agentId;
|
|
3372
|
+
this.port = config.port;
|
|
3373
|
+
this.apiUrl = config.apiUrl.replace(/\/$/, "");
|
|
3374
|
+
this.getAuthHeader = config.getAuthHeader;
|
|
3375
|
+
this.conversationFilter = config.conversationFilter ?? null;
|
|
3376
|
+
this.retry = { ...DEFAULT_RETRY_POLICY, ...config.retry };
|
|
3377
|
+
this.log = config.log ?? (() => {
|
|
2224
3378
|
});
|
|
2225
|
-
this.fetchImpl =
|
|
2226
|
-
this.sleep =
|
|
2227
|
-
this.pausedPollIntervalMs =
|
|
2228
|
-
this.pausedMaxWaitMs =
|
|
2229
|
-
this.stuckQueuedMs =
|
|
2230
|
-
this.now =
|
|
3379
|
+
this.fetchImpl = config.fetchImpl ?? fetch;
|
|
3380
|
+
this.sleep = config.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
|
|
3381
|
+
this.pausedPollIntervalMs = config.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
|
|
3382
|
+
this.pausedMaxWaitMs = config.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
|
|
3383
|
+
this.stuckQueuedMs = config.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
|
|
3384
|
+
this.now = config.now ?? (() => Date.now());
|
|
3385
|
+
this.fileSyncDirectories = config.fileSyncDirectories ?? [];
|
|
3386
|
+
this.homeDir = config.homeDir ?? homedir2();
|
|
2231
3387
|
}
|
|
2232
3388
|
/** The IPv4-loopback base URL for the local `opencode serve`. */
|
|
2233
3389
|
get opencodeBase() {
|
|
@@ -2255,6 +3411,47 @@ var ChannelDriver = class {
|
|
|
2255
3411
|
);
|
|
2256
3412
|
return run2;
|
|
2257
3413
|
}
|
|
3414
|
+
/**
|
|
3415
|
+
* Pull-and-apply any files Evident has queued for this runner (#559), riding
|
|
3416
|
+
* the EXISTING drain cycle — `run.ts` calls it from the same ~2s channel poll
|
|
3417
|
+
* and drain ping that call `drainPending()`. There is deliberately no channel,
|
|
3418
|
+
* control frame or poll loop of its own: worst-case latency is one poll tick.
|
|
3419
|
+
*
|
|
3420
|
+
* NEVER throws and never surfaces a `ChannelAuthError`: a file failure must not
|
|
3421
|
+
* cost a conversation turn. Failures are logged and either acked as a terminal
|
|
3422
|
+
* outcome or left pending for the next drain (see `runner-file-sync.ts`).
|
|
3423
|
+
*
|
|
3424
|
+
* Re-entrant calls are skipped (the poll tick and a drain ping can overlap).
|
|
3425
|
+
*
|
|
3426
|
+
* @returns the number of files written to disk.
|
|
3427
|
+
*/
|
|
3428
|
+
async syncPendingFiles() {
|
|
3429
|
+
if (this.stopped) return 0;
|
|
3430
|
+
if (this.syncingFiles) return 0;
|
|
3431
|
+
this.syncingFiles = true;
|
|
3432
|
+
try {
|
|
3433
|
+
const applied = await syncPendingRunnerFiles({
|
|
3434
|
+
agentId: this.agentId,
|
|
3435
|
+
apiUrl: this.apiUrl,
|
|
3436
|
+
getAuthHeader: this.getAuthHeader,
|
|
3437
|
+
fetchImpl: this.fetchImpl,
|
|
3438
|
+
allowedDirectories: this.fileSyncDirectories,
|
|
3439
|
+
homeDir: this.homeDir,
|
|
3440
|
+
ackFailures: this.fileAckFailures,
|
|
3441
|
+
log: this.log
|
|
3442
|
+
});
|
|
3443
|
+
this.appliedFileCount += applied;
|
|
3444
|
+
return applied;
|
|
3445
|
+
} catch (err) {
|
|
3446
|
+
this.log({
|
|
3447
|
+
level: "error",
|
|
3448
|
+
message: `Runner file sync failed unexpectedly (message processing is unaffected): ${err instanceof Error ? err.message : String(err)}`
|
|
3449
|
+
});
|
|
3450
|
+
return 0;
|
|
3451
|
+
} finally {
|
|
3452
|
+
this.syncingFiles = false;
|
|
3453
|
+
}
|
|
3454
|
+
}
|
|
2258
3455
|
async runDrain() {
|
|
2259
3456
|
let dispatched = 0;
|
|
2260
3457
|
try {
|
|
@@ -2288,6 +3485,28 @@ var ChannelDriver = class {
|
|
|
2288
3485
|
}
|
|
2289
3486
|
return false;
|
|
2290
3487
|
}
|
|
3488
|
+
/**
|
|
3489
|
+
* File-pull work, for `run.ts`'s idle accounting (#559).
|
|
3490
|
+
*
|
|
3491
|
+
* Pulling a file is real work that `drainPending()` knows nothing about, so
|
|
3492
|
+
* without this a near-idle runner counts a credential pull as an empty tick
|
|
3493
|
+
* and `--idle-timeout` can `process.exit` mid-pull — leaving a
|
|
3494
|
+
* `.evident-push-*.tmp` behind — or immediately after the write, before the
|
|
3495
|
+
* browser has run the authorize/callback that activates it (the user then sees
|
|
3496
|
+
* `saved_not_activated` for a runner that was fine).
|
|
3497
|
+
*
|
|
3498
|
+
* Two signals because one cannot cover both cases: `inFlight` is the pull
|
|
3499
|
+
* happening RIGHT NOW (it may outlive the tick that started it), and
|
|
3500
|
+
* `appliedFiles` is monotonic so a pull that started AND finished between two
|
|
3501
|
+
* idle checks still shows up as an advance.
|
|
3502
|
+
*
|
|
3503
|
+
* CALLER CONTRACT: sample `inFlight` BEFORE calling `syncPendingFiles()` for
|
|
3504
|
+
* the cycle. `syncPendingFiles` sets the flag synchronously, so a caller that
|
|
3505
|
+
* samples afterwards reads `true` every single cycle and can never idle out.
|
|
3506
|
+
*/
|
|
3507
|
+
fileSyncActivity() {
|
|
3508
|
+
return { appliedFiles: this.appliedFileCount, inFlight: this.syncingFiles };
|
|
3509
|
+
}
|
|
2291
3510
|
/**
|
|
2292
3511
|
* OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
|
|
2293
3512
|
* exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
|
|
@@ -2349,7 +3568,7 @@ var ChannelDriver = class {
|
|
|
2349
3568
|
await this.sleep(step);
|
|
2350
3569
|
}
|
|
2351
3570
|
}
|
|
2352
|
-
while (this.hasInFlightWatchers()) {
|
|
3571
|
+
while (this.hasInFlightWatchers() || this.syncingFiles) {
|
|
2353
3572
|
if (this.now() >= deadline) return false;
|
|
2354
3573
|
await this.sleep(step);
|
|
2355
3574
|
}
|
|
@@ -2384,16 +3603,27 @@ var ChannelDriver = class {
|
|
|
2384
3603
|
* @returns the count of messages NEWLY dispatched (not already in-flight).
|
|
2385
3604
|
*/
|
|
2386
3605
|
async processConversation(conv) {
|
|
2387
|
-
const sessionId = await this.ensureSession(conv);
|
|
3606
|
+
const { sessionId, refusedSessionId } = await this.ensureSession(conv);
|
|
2388
3607
|
const messages = await this.getPendingMessages(conv.id);
|
|
2389
3608
|
let dispatched = 0;
|
|
2390
3609
|
let skippedAlreadyDispatched = 0;
|
|
3610
|
+
if (refusedSessionId && messages.length > 0) {
|
|
3611
|
+
void this.postSignal(conv.id, messages[0].id, "session_superseded", {
|
|
3612
|
+
superseded_session_id: refusedSessionId
|
|
3613
|
+
});
|
|
3614
|
+
}
|
|
2391
3615
|
for (const message of messages) {
|
|
2392
3616
|
if (this.stopped) break;
|
|
2393
3617
|
if (this.dispatched.has(message.id)) {
|
|
2394
3618
|
skippedAlreadyDispatched += 1;
|
|
2395
3619
|
continue;
|
|
2396
3620
|
}
|
|
3621
|
+
if (message.opencode_message_id) {
|
|
3622
|
+
const outcome = await this.resolveRedrive(conv, sessionId, message, refusedSessionId);
|
|
3623
|
+
if (outcome !== "dispatch") {
|
|
3624
|
+
break;
|
|
3625
|
+
}
|
|
3626
|
+
}
|
|
2397
3627
|
const options = {
|
|
2398
3628
|
agent: message.opencode_agent ?? void 0,
|
|
2399
3629
|
model: message.opencode_model ?? void 0
|
|
@@ -2402,64 +3632,335 @@ var ChannelDriver = class {
|
|
|
2402
3632
|
try {
|
|
2403
3633
|
this.log({
|
|
2404
3634
|
level: "info",
|
|
2405
|
-
message: `Dispatching message ${message.id.slice(0, 8)} to OpenCode native queue (session ${sessionId.slice(0, 8)})`,
|
|
3635
|
+
message: `Dispatching message ${message.id.slice(0, 8)} to OpenCode native queue (session ${sessionId.slice(0, 8)})`,
|
|
3636
|
+
conversation_id: conv.id,
|
|
3637
|
+
message_id: message.id
|
|
3638
|
+
});
|
|
3639
|
+
const sendAttachments = this.buildSendAttachments(conv, message);
|
|
3640
|
+
opencodeMessageId = await this.dispatchLocked(
|
|
3641
|
+
sessionId,
|
|
3642
|
+
() => sendPromptAsync(this.port, sessionId, message.content, options, sendAttachments)
|
|
3643
|
+
);
|
|
3644
|
+
} catch (err) {
|
|
3645
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3646
|
+
this.dispatched.delete(message.id);
|
|
3647
|
+
const exists = await sessionExists(this.port, sessionId);
|
|
3648
|
+
if (exists === false) {
|
|
3649
|
+
this.sessions.delete(conv.id);
|
|
3650
|
+
this.log({
|
|
3651
|
+
level: "warn",
|
|
3652
|
+
message: `Message ${message.id.slice(0, 8)} dispatch hit a session (${sessionId.slice(0, 8)}) that was deleted mid-dispatch (cleanup race) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick (recreated then, in order). Already-dispatched turns keep their watcher.`,
|
|
3653
|
+
conversation_id: conv.id,
|
|
3654
|
+
message_id: message.id
|
|
3655
|
+
});
|
|
3656
|
+
break;
|
|
3657
|
+
}
|
|
3658
|
+
if (exists === null) {
|
|
3659
|
+
this.log({
|
|
3660
|
+
level: "warn",
|
|
3661
|
+
message: `Message ${message.id.slice(0, 8)} dispatch failed and session (${sessionId.slice(0, 8)}) existence could not be confirmed (opencode momentarily unreachable) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick rather than treating it as a genuine failure.`,
|
|
3662
|
+
conversation_id: conv.id,
|
|
3663
|
+
message_id: message.id
|
|
3664
|
+
});
|
|
3665
|
+
break;
|
|
3666
|
+
}
|
|
3667
|
+
const errorMessage = err instanceof Error ? err.message : String(err);
|
|
3668
|
+
this.sessions.delete(conv.id);
|
|
3669
|
+
this.supersede(conv.id, sessionId);
|
|
3670
|
+
this.log({
|
|
3671
|
+
level: "warn",
|
|
3672
|
+
message: `Abandoning OpenCode session ${sessionId.slice(0, 8)} as the binding for conversation ${conv.id.slice(0, 8)} (it exists but failed to run a turn) \u2014 a fresh session is created on the next tick, whatever the persisted binding says by then.`,
|
|
3673
|
+
conversation_id: conv.id,
|
|
3674
|
+
message_id: message.id
|
|
3675
|
+
});
|
|
3676
|
+
await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
|
|
3677
|
+
this.log({
|
|
3678
|
+
level: "warn",
|
|
3679
|
+
message: `markFailed PATCH for message ${message.id.slice(0, 8)} (conversation ${conv.id.slice(0, 8)}) failed (best-effort, not retried): ${markErr instanceof Error ? markErr.message : String(markErr)}`,
|
|
3680
|
+
conversation_id: conv.id,
|
|
3681
|
+
message_id: message.id
|
|
3682
|
+
});
|
|
3683
|
+
});
|
|
3684
|
+
this.log({
|
|
3685
|
+
level: "error",
|
|
3686
|
+
message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
|
|
3687
|
+
conversation_id: conv.id,
|
|
3688
|
+
message_id: message.id
|
|
3689
|
+
});
|
|
3690
|
+
break;
|
|
3691
|
+
}
|
|
3692
|
+
if (opencodeMessageId === null) {
|
|
3693
|
+
this.log({
|
|
3694
|
+
level: "warn",
|
|
3695
|
+
message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
|
|
3696
|
+
conversation_id: conv.id,
|
|
3697
|
+
message_id: message.id
|
|
3698
|
+
});
|
|
3699
|
+
continue;
|
|
3700
|
+
}
|
|
3701
|
+
this.dispatched.add(message.id);
|
|
3702
|
+
this.registerInFlight(conv, sessionId, message, opencodeMessageId);
|
|
3703
|
+
dispatched += 1;
|
|
3704
|
+
void this.postSignal(conv.id, message.id, "dispatched");
|
|
3705
|
+
}
|
|
3706
|
+
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3707
|
+
this.log({
|
|
3708
|
+
level: "warn",
|
|
3709
|
+
message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
|
|
3710
|
+
conversation_id: conv.id
|
|
3711
|
+
});
|
|
3712
|
+
}
|
|
3713
|
+
this.ensureWatcherRunning(sessionId);
|
|
3714
|
+
return dispatched;
|
|
3715
|
+
}
|
|
3716
|
+
/**
|
|
3717
|
+
* Poll a session's message list for the re-drive fence (#965), via the
|
|
3718
|
+
* INJECTED `fetchImpl` — NOT the imported `getSessionMessages` helper, which
|
|
3719
|
+
* hits the global `fetch` and would bypass the same override every other
|
|
3720
|
+
* opencode poll in this file respects. Mirrors `readoptProcessing`'s own
|
|
3721
|
+
* snapshot fetch (`:3081-3111`). `null` = unreadable (non-OK response,
|
|
3722
|
+
* non-array body, or a network exception) — treated as "can't observe",
|
|
3723
|
+
* never as "confirmed gone".
|
|
3724
|
+
*/
|
|
3725
|
+
async pollSessionMessagesForRedrive(conv, message, sessionId) {
|
|
3726
|
+
try {
|
|
3727
|
+
const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
|
|
3728
|
+
if (!res.ok) {
|
|
3729
|
+
this.log({
|
|
3730
|
+
level: "warn",
|
|
3731
|
+
message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned HTTP ${res.status} \u2014 treating as unreadable this tick`,
|
|
3732
|
+
conversation_id: conv.id,
|
|
3733
|
+
message_id: message.id
|
|
3734
|
+
});
|
|
3735
|
+
return null;
|
|
3736
|
+
}
|
|
3737
|
+
const body = await res.json();
|
|
3738
|
+
if (!Array.isArray(body)) {
|
|
3739
|
+
this.log({
|
|
3740
|
+
level: "warn",
|
|
3741
|
+
message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned a non-array message body \u2014 treating as unreadable this tick`,
|
|
3742
|
+
conversation_id: conv.id,
|
|
3743
|
+
message_id: message.id
|
|
3744
|
+
});
|
|
3745
|
+
return null;
|
|
3746
|
+
}
|
|
3747
|
+
return body;
|
|
3748
|
+
} catch (err) {
|
|
3749
|
+
this.log({
|
|
3750
|
+
level: "warn",
|
|
3751
|
+
message: `Re-drive: failed to poll session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3752
|
+
conversation_id: conv.id,
|
|
3753
|
+
message_id: message.id
|
|
3754
|
+
});
|
|
3755
|
+
return null;
|
|
3756
|
+
}
|
|
3757
|
+
}
|
|
3758
|
+
/**
|
|
3759
|
+
* The re-drive fence for a `pending` row that already carries a stored
|
|
3760
|
+
* `opencode_message_id` (#965) — i.e. it has already been handed to opencode at
|
|
3761
|
+
* least once (see the invariant at `QueuedMessage.opencode_message_id`'s doc).
|
|
3762
|
+
* The lifecycle cron can falsely reclaim a `processing` row back to `pending`
|
|
3763
|
+
* mid-turn (a 5-minute liveness-staleness check racing a still-running turn);
|
|
3764
|
+
* without this fence the drain loop would re-`prompt_async` the SAME turn a
|
|
3765
|
+
* second time against live GitHub state. Mirrors `readoptOne`'s job for the
|
|
3766
|
+
* `processing` re-adopt path, but simpler: no b1/b2 preamble cross-check is
|
|
3767
|
+
* needed here because `refusedSessionId` already handles the one case
|
|
3768
|
+
* (#553 abandoned session) that path exists for.
|
|
3769
|
+
*
|
|
3770
|
+
* Only `ChannelAuthError` propagates; every other failure resolves to
|
|
3771
|
+
* `unresolved` and is retried whole on the next ~2s drain tick.
|
|
3772
|
+
*/
|
|
3773
|
+
async resolveRedrive(conv, sessionId, message, refusedSessionId) {
|
|
3774
|
+
const ocId = message.opencode_message_id ?? null;
|
|
3775
|
+
if (refusedSessionId) {
|
|
3776
|
+
this.clearRedriveUnresolved(message.id);
|
|
3777
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3778
|
+
return "dispatch";
|
|
3779
|
+
}
|
|
3780
|
+
const messages = await this.pollSessionMessagesForRedrive(conv, message, sessionId);
|
|
3781
|
+
if (messages == null || messages.length === 0) {
|
|
3782
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3783
|
+
}
|
|
3784
|
+
const state = messageRunState(messages, ocId ?? "");
|
|
3785
|
+
if (state === "done" || state === "failed") {
|
|
3786
|
+
return this.settleRedrive(conv, sessionId, message, ocId, messages, state);
|
|
3787
|
+
}
|
|
3788
|
+
if (state === "running" || state === "queued") {
|
|
3789
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3790
|
+
if (ongoing === true) {
|
|
3791
|
+
return this.reattachRedrive(conv, sessionId, message, ocId);
|
|
3792
|
+
}
|
|
3793
|
+
if (ongoing === false) {
|
|
3794
|
+
this.clearRedriveUnresolved(message.id);
|
|
3795
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3796
|
+
return "dispatch";
|
|
3797
|
+
}
|
|
3798
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3799
|
+
}
|
|
3800
|
+
this.clearRedriveUnresolved(message.id);
|
|
3801
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3802
|
+
return "dispatch";
|
|
3803
|
+
}
|
|
3804
|
+
/**
|
|
3805
|
+
* The `reattached` outcome (Task 3.3): the prior turn is STILL ONGOING per
|
|
3806
|
+
* opencode's own status map — undo the false reclaim instead of starting a
|
|
3807
|
+
* second turn.
|
|
3808
|
+
*/
|
|
3809
|
+
async reattachRedrive(conv, sessionId, message, ocId) {
|
|
3810
|
+
let anchorMs;
|
|
3811
|
+
const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
|
|
3812
|
+
if (!Number.isNaN(parsed)) {
|
|
3813
|
+
anchorMs = parsed;
|
|
3814
|
+
} else {
|
|
3815
|
+
anchorMs = this.now();
|
|
3816
|
+
this.log({
|
|
3817
|
+
level: "error",
|
|
3818
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} has null/unparseable processing_started_at (${String(message.processing_started_at)}) \u2014 anchoring the watcher's absolute-age ceiling to now (defensive)`,
|
|
3819
|
+
conversation_id: conv.id,
|
|
3820
|
+
message_id: message.id
|
|
3821
|
+
});
|
|
3822
|
+
}
|
|
3823
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3824
|
+
try {
|
|
3825
|
+
await this.markProcessing(conv.id, message.id, sessionId, ocId, title);
|
|
3826
|
+
} catch (err) {
|
|
3827
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3828
|
+
this.log({
|
|
3829
|
+
level: "warn",
|
|
3830
|
+
message: `Re-drive: failed to restore message ${message.id.slice(0, 8)} to processing (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3831
|
+
conversation_id: conv.id,
|
|
3832
|
+
message_id: message.id
|
|
3833
|
+
});
|
|
3834
|
+
return "unresolved";
|
|
3835
|
+
}
|
|
3836
|
+
this.clearRedriveUnresolved(message.id);
|
|
3837
|
+
this.registerReadopted(conv, sessionId, message, ocId ?? "", anchorMs);
|
|
3838
|
+
this.dispatched.add(message.id);
|
|
3839
|
+
this.readopted.add(message.id);
|
|
3840
|
+
this.ensureWatcherRunning(sessionId);
|
|
3841
|
+
const watchedForMs = this.now() - anchorMs;
|
|
3842
|
+
void this.postSignal(conv.id, message.id, "redrive_reattached", {
|
|
3843
|
+
watched_for_ms: watchedForMs
|
|
3844
|
+
});
|
|
3845
|
+
this.log({
|
|
3846
|
+
level: "warn",
|
|
3847
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) was wrongly reclaimed to pending while its turn was still running (watched ${watchedForMs}ms) \u2014 restored to processing instead of re-dispatching`,
|
|
3848
|
+
conversation_id: conv.id,
|
|
3849
|
+
message_id: message.id
|
|
3850
|
+
});
|
|
3851
|
+
return "reattached";
|
|
3852
|
+
}
|
|
3853
|
+
/**
|
|
3854
|
+
* The `settled` outcome (Task 3.2): the prior turn already finished (or
|
|
3855
|
+
* errored) while nobody was watching — deliver/report it instead of re-running.
|
|
3856
|
+
* Mirrors `readoptOne`'s `done`/`failed` branches' error discipline, simplified
|
|
3857
|
+
* (no `doneUndeliverable` park: a terminal PATCH failure here just retries next
|
|
3858
|
+
* drain, same as any other non-auth failure).
|
|
3859
|
+
*/
|
|
3860
|
+
async settleRedrive(conv, sessionId, message, ocId, messages, state) {
|
|
3861
|
+
try {
|
|
3862
|
+
if (state === "done") {
|
|
3863
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3864
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3865
|
+
this.log({
|
|
3866
|
+
level: "info",
|
|
3867
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} completed while its row was wrongly reclaimed to pending \u2014 marking done instead of re-dispatching`,
|
|
2406
3868
|
conversation_id: conv.id,
|
|
2407
3869
|
message_id: message.id
|
|
2408
3870
|
});
|
|
2409
|
-
|
|
2410
|
-
|
|
2411
|
-
|
|
2412
|
-
|
|
2413
|
-
);
|
|
2414
|
-
} catch (err) {
|
|
2415
|
-
if (err instanceof ChannelAuthError) throw err;
|
|
2416
|
-
this.dispatched.delete(message.id);
|
|
2417
|
-
if (await sessionExists(this.port, sessionId) === false) {
|
|
2418
|
-
this.sessions.delete(conv.id);
|
|
2419
|
-
this.log({
|
|
2420
|
-
level: "warn",
|
|
2421
|
-
message: `Message ${message.id.slice(0, 8)} dispatch hit a session (${sessionId.slice(0, 8)}) that was deleted mid-dispatch (cleanup race) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick (recreated then, in order). Already-dispatched turns keep their watcher.`,
|
|
2422
|
-
conversation_id: conv.id,
|
|
2423
|
-
message_id: message.id
|
|
2424
|
-
});
|
|
2425
|
-
break;
|
|
2426
|
-
}
|
|
2427
|
-
await this.markFailed(conv.id, message.id).catch(() => {
|
|
2428
|
-
});
|
|
3871
|
+
await this.markDone(conv.id, message.id, sessionId, ocId, title, usage);
|
|
3872
|
+
} else {
|
|
3873
|
+
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3874
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3875
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
2429
3876
|
this.log({
|
|
2430
3877
|
level: "error",
|
|
2431
|
-
message: `
|
|
2432
|
-
conversation_id: conv.id,
|
|
2433
|
-
message_id: message.id
|
|
2434
|
-
});
|
|
2435
|
-
continue;
|
|
2436
|
-
}
|
|
2437
|
-
if (opencodeMessageId === null) {
|
|
2438
|
-
this.log({
|
|
2439
|
-
level: "warn",
|
|
2440
|
-
message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next tick`,
|
|
3878
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} errored while its row was wrongly reclaimed to pending \u2014 marking failed instead of re-dispatching: ${error2 ?? "(no error text)"}`,
|
|
2441
3879
|
conversation_id: conv.id,
|
|
2442
3880
|
message_id: message.id
|
|
2443
3881
|
});
|
|
2444
|
-
|
|
3882
|
+
await this.markFailed(conv.id, message.id, sessionId, error2, usage, failure);
|
|
2445
3883
|
}
|
|
2446
|
-
|
|
2447
|
-
|
|
2448
|
-
dispatched += 1;
|
|
2449
|
-
void this.postSignal(conv.id, message.id, "dispatched");
|
|
2450
|
-
}
|
|
2451
|
-
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3884
|
+
} catch (err) {
|
|
3885
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
2452
3886
|
this.log({
|
|
2453
3887
|
level: "warn",
|
|
2454
|
-
message: `
|
|
2455
|
-
conversation_id: conv.id
|
|
3888
|
+
message: `Re-drive: failed to report message ${message.id.slice(0, 8)} ${state} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3889
|
+
conversation_id: conv.id,
|
|
3890
|
+
message_id: message.id
|
|
2456
3891
|
});
|
|
3892
|
+
return "unresolved";
|
|
3893
|
+
}
|
|
3894
|
+
this.clearRedriveUnresolved(message.id);
|
|
3895
|
+
void this.postSignal(conv.id, message.id, "redrive_settled");
|
|
3896
|
+
return "settled";
|
|
3897
|
+
}
|
|
3898
|
+
/**
|
|
3899
|
+
* The bounded `unresolved` outcome (Task 3.4): opencode's state could not be
|
|
3900
|
+
* observed (snapshot unreadable/empty, or `isSessionOngoing` returned `null`).
|
|
3901
|
+
* A `pending` row is invisible to every cron arm (all require `status =
|
|
3902
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
3903
|
+
* nothing driving it — bound it to the existing `pausedMaxWaitMs` window
|
|
3904
|
+
* (reusing the knob, not a new constant) and take `dispatch` once elapsed.
|
|
3905
|
+
*/
|
|
3906
|
+
resolveRedriveUnresolved(conv, message) {
|
|
3907
|
+
const now = this.now();
|
|
3908
|
+
const since = this.redriveUnresolvedSince.get(message.id);
|
|
3909
|
+
if (since !== void 0 && now - since >= this.pausedMaxWaitMs) {
|
|
3910
|
+
this.clearRedriveUnresolved(message.id);
|
|
3911
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3912
|
+
return "dispatch";
|
|
3913
|
+
}
|
|
3914
|
+
if (since === void 0) {
|
|
3915
|
+
this.redriveUnresolvedSince.set(message.id, now);
|
|
3916
|
+
}
|
|
3917
|
+
if (!this.redriveUnresolvedSignalled.has(message.id)) {
|
|
3918
|
+
this.redriveUnresolvedSignalled.add(message.id);
|
|
3919
|
+
void this.postSignal(conv.id, message.id, "redrive_unresolved");
|
|
3920
|
+
}
|
|
3921
|
+
return "unresolved";
|
|
3922
|
+
}
|
|
3923
|
+
/** Clear both `unresolved`-bound trackers for a row (any non-`unresolved` outcome). */
|
|
3924
|
+
clearRedriveUnresolved(messageId) {
|
|
3925
|
+
this.redriveUnresolvedSince.delete(messageId);
|
|
3926
|
+
this.redriveUnresolvedSignalled.delete(messageId);
|
|
3927
|
+
}
|
|
3928
|
+
/**
|
|
3929
|
+
* Record that `sessionId` is no longer a valid binding for `conversationId`
|
|
3930
|
+
* (#553). Keyed by conversation and hard-capped, so it cannot grow with the
|
|
3931
|
+
* number of failures — see the `supersededSessions` field doc.
|
|
3932
|
+
*/
|
|
3933
|
+
supersede(conversationId, sessionId) {
|
|
3934
|
+
this.supersededSessions.delete(conversationId);
|
|
3935
|
+
this.supersededSessions.set(conversationId, sessionId);
|
|
3936
|
+
while (this.supersededSessions.size > MAX_SUPERSEDED_CONVERSATIONS) {
|
|
3937
|
+
const oldest = this.supersededSessions.keys().next().value;
|
|
3938
|
+
if (oldest === void 0) return;
|
|
3939
|
+
this.supersededSessions.delete(oldest);
|
|
2457
3940
|
}
|
|
2458
|
-
this.ensureWatcherRunning(sessionId);
|
|
2459
|
-
return dispatched;
|
|
2460
3941
|
}
|
|
3942
|
+
/** Whether `sessionId` is the session this conversation has abandoned (#553). */
|
|
3943
|
+
isSuperseded(conversationId, sessionId) {
|
|
3944
|
+
return this.supersededSessions.get(conversationId) === sessionId;
|
|
3945
|
+
}
|
|
3946
|
+
/**
|
|
3947
|
+
* Resolve the opencode session to run this conversation's turns in.
|
|
3948
|
+
*
|
|
3949
|
+
* `refusedSessionId` is set when the #553 guard fired — i.e. the persisted
|
|
3950
|
+
* binding was an id this runner had abandoned, so a resurrection genuinely
|
|
3951
|
+
* happened and a fresh session was bound instead. The caller reports it.
|
|
3952
|
+
*/
|
|
2461
3953
|
async ensureSession(conv) {
|
|
2462
3954
|
const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
|
|
3955
|
+
if (bound && this.isSuperseded(conv.id, bound)) {
|
|
3956
|
+
this.log({
|
|
3957
|
+
level: "warn",
|
|
3958
|
+
message: `OpenCode session ${bound.slice(0, 8)} was abandoned for conversation ${conv.id.slice(0, 8)} after a failed dispatch but is still bound to it (the persisted id was written back by a turn already in flight) \u2014 ignoring it and binding a fresh session.`,
|
|
3959
|
+
conversation_id: conv.id
|
|
3960
|
+
});
|
|
3961
|
+
this.sessions.delete(conv.id);
|
|
3962
|
+
return { sessionId: await this.createAndBindSession(conv.id), refusedSessionId: bound };
|
|
3963
|
+
}
|
|
2463
3964
|
if (bound) {
|
|
2464
3965
|
const exists = await sessionExists(this.port, bound);
|
|
2465
3966
|
if (exists === false) {
|
|
@@ -2469,12 +3970,12 @@ var ChannelDriver = class {
|
|
|
2469
3970
|
conversation_id: conv.id
|
|
2470
3971
|
});
|
|
2471
3972
|
this.sessions.delete(conv.id);
|
|
2472
|
-
return this.createAndBindSession(conv.id);
|
|
3973
|
+
return { sessionId: await this.createAndBindSession(conv.id) };
|
|
2473
3974
|
}
|
|
2474
3975
|
this.sessions.set(conv.id, bound);
|
|
2475
|
-
return bound;
|
|
3976
|
+
return { sessionId: bound };
|
|
2476
3977
|
}
|
|
2477
|
-
return this.createAndBindSession(conv.id);
|
|
3978
|
+
return { sessionId: await this.createAndBindSession(conv.id) };
|
|
2478
3979
|
}
|
|
2479
3980
|
/**
|
|
2480
3981
|
* Create a fresh OpenCode session for a conversation, cache the binding, and
|
|
@@ -2485,7 +3986,12 @@ var ChannelDriver = class {
|
|
|
2485
3986
|
const directory = await this.resolveOpenCodeDirectory();
|
|
2486
3987
|
const sessionId = await createOpenCodeSession(this.port, directory);
|
|
2487
3988
|
this.sessions.set(conversationId, sessionId);
|
|
2488
|
-
await this.persistSession(conversationId, sessionId).catch(() => {
|
|
3989
|
+
await this.persistSession(conversationId, sessionId).catch((err) => {
|
|
3990
|
+
this.log({
|
|
3991
|
+
level: "warn",
|
|
3992
|
+
message: `Persisting the OpenCode session binding ${sessionId.slice(0, 8)} for conversation ${conversationId.slice(0, 8)} failed (best-effort, not retried) \u2014 the completion PATCH also carries opencode_session_id, so the binding is repaired when the turn finishes: ${err instanceof Error ? err.message : String(err)}`,
|
|
3993
|
+
conversation_id: conversationId
|
|
3994
|
+
});
|
|
2489
3995
|
});
|
|
2490
3996
|
return sessionId;
|
|
2491
3997
|
}
|
|
@@ -2562,7 +4068,11 @@ var ChannelDriver = class {
|
|
|
2562
4068
|
* (not-owned / out-of-range / deleted-at-source / workspace gone) / 413
|
|
2563
4069
|
* (over-cap). On ANY non-2xx or thrown failure we return `null` so the caller
|
|
2564
4070
|
* OMITS that one image and the text turn still sends — NEVER throws the turn.
|
|
2565
|
-
*
|
|
4071
|
+
* A 404 body carrying `{ reason: 'needs_reauth' }` (#547 — the server CONFIRMED
|
|
4072
|
+
* a Slack `files:read` scope problem via `files.info`) instead resolves the
|
|
4073
|
+
* `AttachmentFetchNeedsReauth` sentinel, so the in-thread note can steer the
|
|
4074
|
+
* user to reconnect Slack instead of a generic "unavailable". Failures are
|
|
4075
|
+
* logged with context (no silent swallow).
|
|
2566
4076
|
*/
|
|
2567
4077
|
async fetchAttachmentDataUrl(messageId, index, mime) {
|
|
2568
4078
|
try {
|
|
@@ -2571,6 +4081,25 @@ var ChannelDriver = class {
|
|
|
2571
4081
|
{ headers: { Authorization: this.getAuthHeader() } }
|
|
2572
4082
|
);
|
|
2573
4083
|
if (!res.ok) {
|
|
4084
|
+
let reason;
|
|
4085
|
+
try {
|
|
4086
|
+
const body = await res.json();
|
|
4087
|
+
if (body && typeof body.reason === "string") reason = body.reason;
|
|
4088
|
+
} catch (parseErr) {
|
|
4089
|
+
this.log({
|
|
4090
|
+
level: "debug",
|
|
4091
|
+
message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index}: error body was not JSON (${parseErr instanceof Error ? parseErr.message : String(parseErr)}) \u2014 treating as a plain failure`,
|
|
4092
|
+
message_id: messageId
|
|
4093
|
+
});
|
|
4094
|
+
}
|
|
4095
|
+
if (reason === "needs_reauth") {
|
|
4096
|
+
this.log({
|
|
4097
|
+
level: "error",
|
|
4098
|
+
message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} returned HTTP ${res.status} \u2014 server confirmed a Slack reauth/scope problem \u2014 omitting this image (text turn proceeds)`,
|
|
4099
|
+
message_id: messageId
|
|
4100
|
+
});
|
|
4101
|
+
return { needsReauth: true };
|
|
4102
|
+
}
|
|
2574
4103
|
this.log({
|
|
2575
4104
|
level: "error",
|
|
2576
4105
|
message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} returned HTTP ${res.status} \u2014 omitting this image (text turn proceeds)`,
|
|
@@ -2610,6 +4139,9 @@ var ChannelDriver = class {
|
|
|
2610
4139
|
if (this.attachmentsSkippedSignalled.has(messageId)) return;
|
|
2611
4140
|
this.attachmentsSkippedSignalled.add(messageId);
|
|
2612
4141
|
const skippedReason = capabilityUnknown ? "unknown" : "unsupported";
|
|
4142
|
+
const failedReason = outcomes.some(
|
|
4143
|
+
(o) => o.status === "failed" && o.reason === "needs_reauth"
|
|
4144
|
+
) ? "needs_reauth" : void 0;
|
|
2613
4145
|
this.log({
|
|
2614
4146
|
level: "info",
|
|
2615
4147
|
message: `Message ${messageId.slice(0, 8)}: ${skipped} image(s) skipped (${capabilityUnknown ? "capability was unreadable \u2014 failed open to text-only" : "model not attachment-capable"}), ${failed} image(s) unavailable (deleted-at-source or fetch failure) \u2014 noting to Evident`,
|
|
@@ -2619,7 +4151,8 @@ var ChannelDriver = class {
|
|
|
2619
4151
|
void this.postSignal(conversationId, messageId, "attachments_skipped", {
|
|
2620
4152
|
skipped,
|
|
2621
4153
|
failed,
|
|
2622
|
-
...skipped > 0 ? { skipped_reason: skippedReason } : {}
|
|
4154
|
+
...skipped > 0 ? { skipped_reason: skippedReason } : {},
|
|
4155
|
+
...failedReason ? { failed_reason: failedReason } : {}
|
|
2623
4156
|
});
|
|
2624
4157
|
}
|
|
2625
4158
|
/** Register a freshly-dispatched message with its session's watcher state. */
|
|
@@ -2650,12 +4183,17 @@ var ChannelDriver = class {
|
|
|
2650
4183
|
stuckReported: false,
|
|
2651
4184
|
lastAliveAt: 0,
|
|
2652
4185
|
aliveInFlight: false,
|
|
4186
|
+
titleSynced: false,
|
|
4187
|
+
titleSyncInFlight: false,
|
|
2653
4188
|
awaitingHumanLatched: false,
|
|
2654
4189
|
pausedOnQuestion: false,
|
|
2655
4190
|
pausedOnPermission: false,
|
|
2656
4191
|
pausedClearConfirmed: false,
|
|
2657
4192
|
pausedInFlight: false,
|
|
2658
|
-
deliveryDeadlineAnchored: false
|
|
4193
|
+
deliveryDeadlineAnchored: false,
|
|
4194
|
+
b2PinnedSinceMs: 0,
|
|
4195
|
+
b2LastDescendantCheckMs: 0,
|
|
4196
|
+
b2AbandonedSignalled: false
|
|
2659
4197
|
});
|
|
2660
4198
|
}
|
|
2661
4199
|
/**
|
|
@@ -2670,9 +4208,10 @@ var ChannelDriver = class {
|
|
|
2670
4208
|
* opencode reports ACTIVELY `running` is watched to completion (its liveness
|
|
2671
4209
|
* heartbeat keeps the cron off its row), while a re-adopted turn that is paused
|
|
2672
4210
|
* awaiting a human — or queued/unreachable — is still bounded by `deadline` and
|
|
2673
|
-
* handed to the cron.
|
|
2674
|
-
*
|
|
2675
|
-
*
|
|
4211
|
+
* handed to the cron. Real invariant (#965): the cron MAY reclaim a row this
|
|
4212
|
+
* runner still holds; a reclaimed row that already ran is never re-dispatched
|
|
4213
|
+
* while opencode reports its turn ongoing (readopt's own gate here, and the
|
|
4214
|
+
* `pending`-row re-drive fence, `resolveRedrive`). `dispatchedAt` stays `now`
|
|
2676
4215
|
* (only the appear-guard uses it).
|
|
2677
4216
|
*
|
|
2678
4217
|
* `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
|
|
@@ -2723,12 +4262,17 @@ var ChannelDriver = class {
|
|
|
2723
4262
|
// with no extra `re_adopted` signal needed (folds old WI-6).
|
|
2724
4263
|
lastAliveAt: 0,
|
|
2725
4264
|
aliveInFlight: false,
|
|
4265
|
+
titleSynced: false,
|
|
4266
|
+
titleSyncInFlight: false,
|
|
2726
4267
|
awaitingHumanLatched: false,
|
|
2727
4268
|
pausedOnQuestion: false,
|
|
2728
4269
|
pausedOnPermission: false,
|
|
2729
4270
|
pausedClearConfirmed: false,
|
|
2730
4271
|
pausedInFlight: false,
|
|
2731
|
-
deliveryDeadlineAnchored: false
|
|
4272
|
+
deliveryDeadlineAnchored: false,
|
|
4273
|
+
b2PinnedSinceMs: 0,
|
|
4274
|
+
b2LastDescendantCheckMs: 0,
|
|
4275
|
+
b2AbandonedSignalled: false
|
|
2732
4276
|
});
|
|
2733
4277
|
}
|
|
2734
4278
|
/**
|
|
@@ -2890,58 +4434,7 @@ var ChannelDriver = class {
|
|
|
2890
4434
|
}
|
|
2891
4435
|
}
|
|
2892
4436
|
if (state === "done") {
|
|
2893
|
-
this.
|
|
2894
|
-
if (!inFlight.done) {
|
|
2895
|
-
this.log({
|
|
2896
|
-
level: "info",
|
|
2897
|
-
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
2898
|
-
conversation_id: conv.id,
|
|
2899
|
-
message_id: inFlight.evidentMessageId
|
|
2900
|
-
});
|
|
2901
|
-
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
2902
|
-
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
2903
|
-
try {
|
|
2904
|
-
await this.markDone(
|
|
2905
|
-
conv.id,
|
|
2906
|
-
inFlight.evidentMessageId,
|
|
2907
|
-
sessionId,
|
|
2908
|
-
inFlight.opencodeMessageId,
|
|
2909
|
-
title,
|
|
2910
|
-
usage
|
|
2911
|
-
);
|
|
2912
|
-
} catch (err) {
|
|
2913
|
-
if (err instanceof ChannelAuthError) throw err;
|
|
2914
|
-
if (err instanceof ChannelTerminalError) {
|
|
2915
|
-
this.log({
|
|
2916
|
-
level: "warn",
|
|
2917
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
2918
|
-
conversation_id: conv.id,
|
|
2919
|
-
message_id: inFlight.evidentMessageId
|
|
2920
|
-
});
|
|
2921
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2922
|
-
return;
|
|
2923
|
-
}
|
|
2924
|
-
if (this.now() >= inFlight.deadline) {
|
|
2925
|
-
this.log({
|
|
2926
|
-
level: "warn",
|
|
2927
|
-
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)}`,
|
|
2928
|
-
conversation_id: conv.id,
|
|
2929
|
-
message_id: inFlight.evidentMessageId
|
|
2930
|
-
});
|
|
2931
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2932
|
-
return;
|
|
2933
|
-
}
|
|
2934
|
-
this.log({
|
|
2935
|
-
level: "warn",
|
|
2936
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
2937
|
-
conversation_id: conv.id,
|
|
2938
|
-
message_id: inFlight.evidentMessageId
|
|
2939
|
-
});
|
|
2940
|
-
return;
|
|
2941
|
-
}
|
|
2942
|
-
inFlight.done = true;
|
|
2943
|
-
}
|
|
2944
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4437
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
2945
4438
|
return;
|
|
2946
4439
|
}
|
|
2947
4440
|
if (state === "failed") {
|
|
@@ -2955,8 +4448,16 @@ var ChannelDriver = class {
|
|
|
2955
4448
|
message_id: inFlight.evidentMessageId
|
|
2956
4449
|
});
|
|
2957
4450
|
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4451
|
+
const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
|
|
2958
4452
|
try {
|
|
2959
|
-
await this.markFailed(
|
|
4453
|
+
await this.markFailed(
|
|
4454
|
+
conv.id,
|
|
4455
|
+
inFlight.evidentMessageId,
|
|
4456
|
+
sessionId,
|
|
4457
|
+
error2,
|
|
4458
|
+
usage,
|
|
4459
|
+
failure
|
|
4460
|
+
);
|
|
2960
4461
|
} catch (err) {
|
|
2961
4462
|
if (err instanceof ChannelAuthError) throw err;
|
|
2962
4463
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -3001,6 +4502,44 @@ var ChannelDriver = class {
|
|
|
3001
4502
|
});
|
|
3002
4503
|
}
|
|
3003
4504
|
const activelyRunning = state === "running" && !awaitingHuman;
|
|
4505
|
+
const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
|
|
4506
|
+
const snapshotReadable = messages != null && messages.length > 0;
|
|
4507
|
+
if (!pinnedNow) {
|
|
4508
|
+
if (snapshotReadable) {
|
|
4509
|
+
inFlight.b2PinnedSinceMs = 0;
|
|
4510
|
+
inFlight.b2LastDescendantCheckMs = 0;
|
|
4511
|
+
inFlight.b2AbandonedSignalled = false;
|
|
4512
|
+
}
|
|
4513
|
+
} else {
|
|
4514
|
+
if (inFlight.b2AbandonedSignalled) {
|
|
4515
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4516
|
+
return;
|
|
4517
|
+
}
|
|
4518
|
+
if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
|
|
4519
|
+
const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
|
|
4520
|
+
if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
|
|
4521
|
+
inFlight.b2LastDescendantCheckMs = this.now();
|
|
4522
|
+
const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
|
|
4523
|
+
if (isB2AbandonmentConfirmed({
|
|
4524
|
+
pinnedForMs,
|
|
4525
|
+
minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
|
|
4526
|
+
descendantOngoing
|
|
4527
|
+
})) {
|
|
4528
|
+
inFlight.b2AbandonedSignalled = true;
|
|
4529
|
+
this.log({
|
|
4530
|
+
level: "warn",
|
|
4531
|
+
message: `Message ${id.slice(0, 8)} b2-pinned for ${Math.round(pinnedForMs / 1e3)}s with no ongoing descendant sub-agent session (status-map confirmed) \u2014 treating the delegated/tool turn as abandoned, resolving done`,
|
|
4532
|
+
conversation_id: conv.id,
|
|
4533
|
+
message_id: id
|
|
4534
|
+
});
|
|
4535
|
+
void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
|
|
4536
|
+
watched_for_ms: pinnedForMs
|
|
4537
|
+
});
|
|
4538
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4539
|
+
return;
|
|
4540
|
+
}
|
|
4541
|
+
}
|
|
4542
|
+
}
|
|
3004
4543
|
if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
|
|
3005
4544
|
this.log({
|
|
3006
4545
|
level: "warn",
|
|
@@ -3020,6 +4559,18 @@ var ChannelDriver = class {
|
|
|
3020
4559
|
inFlight.aliveInFlight = false;
|
|
3021
4560
|
if (ok) inFlight.lastAliveAt = this.now();
|
|
3022
4561
|
});
|
|
4562
|
+
if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
|
|
4563
|
+
inFlight.titleSyncInFlight = true;
|
|
4564
|
+
void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
|
|
4565
|
+
if (!title) {
|
|
4566
|
+
inFlight.titleSyncInFlight = false;
|
|
4567
|
+
return;
|
|
4568
|
+
}
|
|
4569
|
+
const ok = await this.patchConversationTitle(conv.id, title);
|
|
4570
|
+
inFlight.titleSyncInFlight = false;
|
|
4571
|
+
if (ok) inFlight.titleSynced = true;
|
|
4572
|
+
});
|
|
4573
|
+
}
|
|
3023
4574
|
}
|
|
3024
4575
|
if (awaitingHuman) {
|
|
3025
4576
|
if (!inFlight.awaitingHumanLatched) {
|
|
@@ -3057,6 +4608,70 @@ var ChannelDriver = class {
|
|
|
3057
4608
|
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3058
4609
|
}
|
|
3059
4610
|
}
|
|
4611
|
+
/**
|
|
4612
|
+
* Settle a message whose run-state has resolved `'done'` — extracted verbatim
|
|
4613
|
+
* (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
|
|
4614
|
+
* inline `state === 'done'` branch body, so a SECOND caller (the #721
|
|
4615
|
+
* b2-abandonment resolution) can reach the exact same completion behavior
|
|
4616
|
+
* (delivery-deadline anchoring, title resolution, usage extraction, and
|
|
4617
|
+
* `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
|
|
4618
|
+
* and risking the two copies silently drifting apart.
|
|
4619
|
+
*/
|
|
4620
|
+
async settleMessageDone(sessionId, watcher, inFlight, messages) {
|
|
4621
|
+
const conv = watcher.conv;
|
|
4622
|
+
this.anchorDeliveryDeadline(inFlight);
|
|
4623
|
+
if (!inFlight.done) {
|
|
4624
|
+
this.log({
|
|
4625
|
+
level: "info",
|
|
4626
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
4627
|
+
conversation_id: conv.id,
|
|
4628
|
+
message_id: inFlight.evidentMessageId
|
|
4629
|
+
});
|
|
4630
|
+
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
4631
|
+
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4632
|
+
try {
|
|
4633
|
+
await this.markDone(
|
|
4634
|
+
conv.id,
|
|
4635
|
+
inFlight.evidentMessageId,
|
|
4636
|
+
sessionId,
|
|
4637
|
+
inFlight.opencodeMessageId,
|
|
4638
|
+
title,
|
|
4639
|
+
usage
|
|
4640
|
+
);
|
|
4641
|
+
} catch (err) {
|
|
4642
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
4643
|
+
if (err instanceof ChannelTerminalError) {
|
|
4644
|
+
this.log({
|
|
4645
|
+
level: "warn",
|
|
4646
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
4647
|
+
conversation_id: conv.id,
|
|
4648
|
+
message_id: inFlight.evidentMessageId
|
|
4649
|
+
});
|
|
4650
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4651
|
+
return;
|
|
4652
|
+
}
|
|
4653
|
+
if (this.now() >= inFlight.deadline) {
|
|
4654
|
+
this.log({
|
|
4655
|
+
level: "warn",
|
|
4656
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
|
|
4657
|
+
conversation_id: conv.id,
|
|
4658
|
+
message_id: inFlight.evidentMessageId
|
|
4659
|
+
});
|
|
4660
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4661
|
+
return;
|
|
4662
|
+
}
|
|
4663
|
+
this.log({
|
|
4664
|
+
level: "warn",
|
|
4665
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
4666
|
+
conversation_id: conv.id,
|
|
4667
|
+
message_id: inFlight.evidentMessageId
|
|
4668
|
+
});
|
|
4669
|
+
return;
|
|
4670
|
+
}
|
|
4671
|
+
inFlight.done = true;
|
|
4672
|
+
}
|
|
4673
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4674
|
+
}
|
|
3060
4675
|
// Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
|
|
3061
4676
|
/**
|
|
3062
4677
|
* Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
|
|
@@ -3223,6 +4838,7 @@ var ChannelDriver = class {
|
|
|
3223
4838
|
if (state === "failed") {
|
|
3224
4839
|
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3225
4840
|
const usage = messageUsage(messages, ocId ?? "");
|
|
4841
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3226
4842
|
this.log({
|
|
3227
4843
|
level: "error",
|
|
3228
4844
|
message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
|
|
@@ -3230,7 +4846,7 @@ var ChannelDriver = class {
|
|
|
3230
4846
|
message_id: row.id
|
|
3231
4847
|
});
|
|
3232
4848
|
try {
|
|
3233
|
-
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
|
|
4849
|
+
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
|
|
3234
4850
|
} catch (err) {
|
|
3235
4851
|
if (err instanceof ChannelAuthError) throw err;
|
|
3236
4852
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -3497,7 +5113,8 @@ var ChannelDriver = class {
|
|
|
3497
5113
|
opencode_model: row.opencode_model,
|
|
3498
5114
|
source_message_id: row.source_message_id,
|
|
3499
5115
|
slack_user_id: row.slack_user_id,
|
|
3500
|
-
attachments: row.attachments ?? null
|
|
5116
|
+
attachments: row.attachments ?? null,
|
|
5117
|
+
opencode_message_id: row.opencode_message_id
|
|
3501
5118
|
};
|
|
3502
5119
|
}
|
|
3503
5120
|
/**
|
|
@@ -3636,6 +5253,47 @@ var ChannelDriver = class {
|
|
|
3636
5253
|
}
|
|
3637
5254
|
return false;
|
|
3638
5255
|
}
|
|
5256
|
+
/**
|
|
5257
|
+
* Tri-state variant of the upward parentID membership walk (#721), used ONLY
|
|
5258
|
+
* by `isAnyDescendantSessionOngoing`. Walks the SAME cached
|
|
5259
|
+
* `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
|
|
5260
|
+
* `sessionBelongsTo`, which deliberately collapses "confirmed not a
|
|
5261
|
+
* descendant" and "the walk's fetch failed" into the same `false` (safe for
|
|
5262
|
+
* its OTHER callers: interaction attribution and the recovery-path
|
|
5263
|
+
* `isAnyDescendantSessionAlive`, both of which just retry next tick with no
|
|
5264
|
+
* safety consequence either way) — this variant keeps those two outcomes
|
|
5265
|
+
* SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
|
|
5266
|
+
* (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
|
|
5267
|
+
* not ongoing".
|
|
5268
|
+
*
|
|
5269
|
+
* Return contract:
|
|
5270
|
+
* - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
|
|
5271
|
+
* - `false` → the walk reached a definitive, parent-less root session
|
|
5272
|
+
* WITHOUT ever matching `rootSessionId` — `sessionId` is
|
|
5273
|
+
* CONFIRMED NOT a descendant of it.
|
|
5274
|
+
* - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
|
|
5275
|
+
* through the walk (`resolveSessionParent` returned `undefined`),
|
|
5276
|
+
* or the depth cap (32) was hit without a definitive answer (a
|
|
5277
|
+
* pathological/cyclic chain proves nothing either way). NEVER
|
|
5278
|
+
* treat this the same as `false` — see `sessionBelongsTo`'s own
|
|
5279
|
+
* doc comment above for why that collapse is safe THERE but not
|
|
5280
|
+
* here.
|
|
5281
|
+
*
|
|
5282
|
+
* `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
|
|
5283
|
+
* to the live-path descendant check, not a modification of shared code used
|
|
5284
|
+
* by interaction attribution or the recovery path.
|
|
5285
|
+
*/
|
|
5286
|
+
async resolveSessionMembership(sessionId, rootSessionId) {
|
|
5287
|
+
let current = sessionId;
|
|
5288
|
+
for (let depth = 0; current && depth < 32; depth++) {
|
|
5289
|
+
if (current === rootSessionId) return true;
|
|
5290
|
+
const parent = await this.resolveSessionParent(current);
|
|
5291
|
+
if (parent === void 0) return null;
|
|
5292
|
+
if (parent === null) return false;
|
|
5293
|
+
current = parent;
|
|
5294
|
+
}
|
|
5295
|
+
return null;
|
|
5296
|
+
}
|
|
3639
5297
|
/**
|
|
3640
5298
|
* Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
|
|
3641
5299
|
* `null` for a root session (no parent) and `undefined` when opencode is
|
|
@@ -3658,19 +5316,36 @@ var ChannelDriver = class {
|
|
|
3658
5316
|
if (parent !== void 0) this.sessionParents.set(sessionId, parent);
|
|
3659
5317
|
return parent;
|
|
3660
5318
|
}
|
|
5319
|
+
/**
|
|
5320
|
+
* OpenCode's synchronous default session title (e.g.
|
|
5321
|
+
* `"New session - 1737800000000"`), assigned immediately when a session is
|
|
5322
|
+
* created — before OpenCode's async LLM-based auto-titling later renames it
|
|
5323
|
+
* mid-turn (#549). Matched by this literal, case-sensitive prefix only; the
|
|
5324
|
+
* timestamp suffix's exact format is deliberately NOT matched, since the prefix
|
|
5325
|
+
* alone is the stable, cheap signal and over-anchoring on the timestamp
|
|
5326
|
+
* representation risks silently breaking if OpenCode ever changes it. Accepted
|
|
5327
|
+
* trade-off: a genuine LLM-assigned title that happens to literally start with
|
|
5328
|
+
* this prefix would also fail to latch (see `resolveSessionTitle`) —
|
|
5329
|
+
* vanishingly unlikely in practice, and deliberately not engineered around.
|
|
5330
|
+
*/
|
|
5331
|
+
static OPENCODE_DEFAULT_TITLE_PREFIX = /^New session - /;
|
|
3661
5332
|
/**
|
|
3662
5333
|
* Resolve (and cache in `sessionTitles`) the OpenCode session TITLE (#310) so the
|
|
3663
5334
|
* status PATCH can carry it into the "Live sessions" list. Driver-level cache so
|
|
3664
5335
|
* BOTH the watcher completion path and the restart-recovery re-adopt path (which
|
|
3665
5336
|
* has no watcher) can use it. `conversationId` is passed only for log context.
|
|
3666
5337
|
* Best-effort:
|
|
3667
|
-
* - a resolved NON-EMPTY title
|
|
5338
|
+
* - a resolved NON-EMPTY title that does NOT match
|
|
5339
|
+
* `OPENCODE_DEFAULT_TITLE_PREFIX` is cached and terminal (a real session name
|
|
3668
5340
|
* won't later un-name), so we do NOT re-GET `/session/:id` every tick;
|
|
3669
|
-
* - while the title is still absent
|
|
3670
|
-
*
|
|
3671
|
-
*
|
|
3672
|
-
*
|
|
3673
|
-
*
|
|
5341
|
+
* - while the title is still absent, empty, or matches the OpenCode
|
|
5342
|
+
* placeholder prefix (#549) we do NOT latch it — OpenCode names sessions
|
|
5343
|
+
* asynchronously mid-turn, so an early call (e.g. at `processing`) must leave
|
|
5344
|
+
* the cache unresolved and re-fetch on the next need so a later call (e.g. at
|
|
5345
|
+
* `done`) picks up the name assigned in the meantime. Such a call returns
|
|
5346
|
+
* `null` (omit the title on THIS PATCH) without caching. If a session is
|
|
5347
|
+
* never renamed, the title is omitted forever rather than ever persisting
|
|
5348
|
+
* the placeholder as a last resort;
|
|
3674
5349
|
* - a failed request likewise leaves the cache unresolved (retry next need)
|
|
3675
5350
|
* and returns `null` — it must NEVER throw or block completion.
|
|
3676
5351
|
* A failure is logged with agent/session context (no silent catch).
|
|
@@ -3683,7 +5358,7 @@ var ChannelDriver = class {
|
|
|
3683
5358
|
if (res.ok) {
|
|
3684
5359
|
const body = await res.json();
|
|
3685
5360
|
const title = body && typeof body.title === "string" ? body.title.trim() : "";
|
|
3686
|
-
if (title.length > 0) {
|
|
5361
|
+
if (title.length > 0 && !_ChannelDriver.OPENCODE_DEFAULT_TITLE_PREFIX.test(title)) {
|
|
3687
5362
|
this.sessionTitles.set(sessionId, title);
|
|
3688
5363
|
return title;
|
|
3689
5364
|
}
|
|
@@ -3703,6 +5378,54 @@ var ChannelDriver = class {
|
|
|
3703
5378
|
}
|
|
3704
5379
|
return null;
|
|
3705
5380
|
}
|
|
5381
|
+
/**
|
|
5382
|
+
* Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
|
|
5383
|
+
* session title onto the conversation via the PLAIN conversation-update
|
|
5384
|
+
* endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
|
|
5385
|
+
* message-status endpoint `markProcessing`/`markDone` use. Deliberately a
|
|
5386
|
+
* separate, lighter call: it carries no `status`, so it cannot re-trigger the
|
|
5387
|
+
* `processing`/`done` transition side effects (Slack notices, activity-log
|
|
5388
|
+
* rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
|
|
5389
|
+
* ever touches `conversations.title`. That route (`routes/conversations.ts`)
|
|
5390
|
+
* skips a title write matching the stored value, so a redundant call with the
|
|
5391
|
+
* same title is a real no-op — it does not bump `updated_at`, which the
|
|
5392
|
+
* conversation list sorts and paginates on. (Note this is a DIFFERENT guard
|
|
5393
|
+
* from `threads.ts`'s "non-empty AND changed" one, which only covers the
|
|
5394
|
+
* message-status PATCH; the non-empty half is enforced here instead, by
|
|
5395
|
+
* `resolveSessionTitle` never returning an empty/placeholder title.)
|
|
5396
|
+
*
|
|
5397
|
+
* Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
|
|
5398
|
+
* is logged and the title is simply retried on the next heartbeat tick (the
|
|
5399
|
+
* caller only latches `titleSynced` on `true`).
|
|
5400
|
+
*/
|
|
5401
|
+
async patchConversationTitle(conversationId, title) {
|
|
5402
|
+
try {
|
|
5403
|
+
const res = await this.fetchImpl(
|
|
5404
|
+
`${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
|
|
5405
|
+
{
|
|
5406
|
+
method: "PATCH",
|
|
5407
|
+
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
5408
|
+
body: JSON.stringify({ title })
|
|
5409
|
+
}
|
|
5410
|
+
);
|
|
5411
|
+
if (!res.ok) {
|
|
5412
|
+
this.log({
|
|
5413
|
+
level: "debug",
|
|
5414
|
+
message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
|
|
5415
|
+
conversation_id: conversationId
|
|
5416
|
+
});
|
|
5417
|
+
return false;
|
|
5418
|
+
}
|
|
5419
|
+
return true;
|
|
5420
|
+
} catch (err) {
|
|
5421
|
+
this.log({
|
|
5422
|
+
level: "debug",
|
|
5423
|
+
message: `Best-effort mid-turn title sync PATCH failed for conversation ${conversationId.slice(0, 8)} (will retry next heartbeat): ${err instanceof Error ? err.message : String(err)}`,
|
|
5424
|
+
conversation_id: conversationId
|
|
5425
|
+
});
|
|
5426
|
+
return false;
|
|
5427
|
+
}
|
|
5428
|
+
}
|
|
3706
5429
|
/**
|
|
3707
5430
|
* DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
|
|
3708
5431
|
* (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
|
|
@@ -3761,6 +5484,84 @@ var ChannelDriver = class {
|
|
|
3761
5484
|
}
|
|
3762
5485
|
return false;
|
|
3763
5486
|
}
|
|
5487
|
+
/**
|
|
5488
|
+
* LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
|
|
5489
|
+
* sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
|
|
5490
|
+
* in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
|
|
5491
|
+
*
|
|
5492
|
+
* Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
|
|
5493
|
+
* cross-check above): that method judges liveness from the child's OWN
|
|
5494
|
+
* TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
|
|
5495
|
+
* on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
|
|
5496
|
+
* path the local opencode server IS running, so its in-memory status map is
|
|
5497
|
+
* live and authoritative — and per ADR-0047 §4a ("the child has its own entry
|
|
5498
|
+
* [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
|
|
5499
|
+
* ENTIRE turn (including any tool call it is itself executing), not a
|
|
5500
|
+
* per-message transcript snapshot. This sidesteps the "child's own tool is
|
|
5501
|
+
* executing, between its step's completion and the next generation step"
|
|
5502
|
+
* transcript gap that a transcript-based check would need a second,
|
|
5503
|
+
* sustained-window bound to guard against — it is simply not derived from
|
|
5504
|
+
* message timestamps at all.
|
|
5505
|
+
*
|
|
5506
|
+
* Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
|
|
5507
|
+
* status, as the recovery path does per §4a)? Because on the LIVE path the
|
|
5508
|
+
* root session can be shared: a SECOND, unrelated user message can land on the
|
|
5509
|
+
* SAME session (issue #721's own root cause) and keep the root `busy` for a
|
|
5510
|
+
* reason that has nothing to do with THIS message's delegation. A `task`
|
|
5511
|
+
* descendant session is spawned for exactly one delegated turn and never
|
|
5512
|
+
* reused, so its OWN status-map entry is unambiguous evidence about that one
|
|
5513
|
+
* delegation — which the root's status is not.
|
|
5514
|
+
*
|
|
5515
|
+
* Why membership is checked via `resolveSessionMembership`, NOT
|
|
5516
|
+
* `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
|
|
5517
|
+
* `GET /session/:id` fetch failure into "not a descendant", which would
|
|
5518
|
+
* silently drop a genuinely-live candidate from consideration on the one
|
|
5519
|
+
* unlucky tick its membership-walk fetch hiccups (#721).
|
|
5520
|
+
* `resolveSessionMembership` keeps that failure mode as a distinct `null`
|
|
5521
|
+
* (indeterminate) so it is folded into THIS method's own `indeterminate` flag
|
|
5522
|
+
* instead.
|
|
5523
|
+
*
|
|
5524
|
+
* Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
|
|
5525
|
+
* - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
|
|
5526
|
+
* - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
|
|
5527
|
+
* confirmed either way (`resolveSessionMembership` never
|
|
5528
|
+
* returned `null`), and every CONFIRMED descendant's status read
|
|
5529
|
+
* succeeded and is not ongoing (includes "no descendant session
|
|
5530
|
+
* exists at all" — e.g. a plain, non-`task` tool call).
|
|
5531
|
+
* - `null` → INDETERMINATE: `listSessions` failed, OR at least one
|
|
5532
|
+
* candidate's MEMBERSHIP could not be confirmed
|
|
5533
|
+
* (`resolveSessionMembership` returned `null` — a fetch failure
|
|
5534
|
+
* or pathological chain partway through the parent walk), OR at
|
|
5535
|
+
* least one CONFIRMED descendant's `isSessionOngoing` read
|
|
5536
|
+
* failed — and no OTHER candidate was already confirmed `true`.
|
|
5537
|
+
* The caller MUST NOT treat `null` the same as `false` here
|
|
5538
|
+
* (unlike the recovery cross-check's contract) — see
|
|
5539
|
+
* `isB2AbandonmentConfirmed`.
|
|
5540
|
+
*/
|
|
5541
|
+
async isAnyDescendantSessionOngoing(rootSessionId) {
|
|
5542
|
+
const sessions = await listSessions(this.port);
|
|
5543
|
+
if (!sessions) {
|
|
5544
|
+
this.log({
|
|
5545
|
+
level: "warn",
|
|
5546
|
+
message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
|
|
5547
|
+
});
|
|
5548
|
+
return null;
|
|
5549
|
+
}
|
|
5550
|
+
let indeterminate = false;
|
|
5551
|
+
for (const candidate of sessions) {
|
|
5552
|
+
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
5553
|
+
const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
|
|
5554
|
+
if (membership === null) {
|
|
5555
|
+
indeterminate = true;
|
|
5556
|
+
continue;
|
|
5557
|
+
}
|
|
5558
|
+
if (membership === false) continue;
|
|
5559
|
+
const ongoing = await isSessionOngoing(this.port, candidate.id);
|
|
5560
|
+
if (ongoing === true) return true;
|
|
5561
|
+
if (ongoing === null) indeterminate = true;
|
|
5562
|
+
}
|
|
5563
|
+
return indeterminate ? null : false;
|
|
5564
|
+
}
|
|
3764
5565
|
/**
|
|
3765
5566
|
* Cheap decision-telemetry label for a running row's LAST correlated reply
|
|
3766
5567
|
* (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
|
|
@@ -3889,6 +5690,32 @@ var ChannelDriver = class {
|
|
|
3889
5690
|
}
|
|
3890
5691
|
return messages;
|
|
3891
5692
|
}
|
|
5693
|
+
/**
|
|
5694
|
+
* The `opencode_session_id` fragment of a status PATCH body — `{}` when this
|
|
5695
|
+
* conversation has ABANDONED that session (#553). The field is optional
|
|
5696
|
+
* server-side and an absent one leaves the persisted binding untouched, so
|
|
5697
|
+
* omitting it is how a routine status write stops resurrecting it.
|
|
5698
|
+
*
|
|
5699
|
+
* ONLY for writes whose sole cost is a lost deep link. The `processing` notice
|
|
5700
|
+
* degrades to no "View in Evident" link (the reaction swap still fires) and the
|
|
5701
|
+
* turn-failure notice is built from the PATCH's own `error` text with a link off
|
|
5702
|
+
* the persisted row — neither loses content the user came for. `markDone`
|
|
5703
|
+
* deliberately does NOT use this helper: the server fetches the reply text
|
|
5704
|
+
* THROUGH the session id it is given, so suppressing there would replace the
|
|
5705
|
+
* agent's answer with a bare "✅ Done!" (the #183/#187 failure). The
|
|
5706
|
+
* `ensureSession` guard, not this suppression, is what makes the self-heal
|
|
5707
|
+
* stick.
|
|
5708
|
+
*/
|
|
5709
|
+
sessionIdBody(sessionId, conversationId, messageId, status2) {
|
|
5710
|
+
if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
|
|
5711
|
+
this.log({
|
|
5712
|
+
level: "debug",
|
|
5713
|
+
message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${status2}' update for message ${messageId.slice(0, 8)} so it is not re-bound to conversation ${conversationId.slice(0, 8)}`,
|
|
5714
|
+
conversation_id: conversationId,
|
|
5715
|
+
message_id: messageId
|
|
5716
|
+
});
|
|
5717
|
+
return {};
|
|
5718
|
+
}
|
|
3892
5719
|
/**
|
|
3893
5720
|
* EXISTING combinedAuth route — now fired by the watcher on queued→running
|
|
3894
5721
|
* (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
|
|
@@ -3918,7 +5745,7 @@ var ChannelDriver = class {
|
|
|
3918
5745
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
3919
5746
|
body: JSON.stringify({
|
|
3920
5747
|
status: "processing",
|
|
3921
|
-
|
|
5748
|
+
...this.sessionIdBody(sessionId, conversationId, messageId, "processing"),
|
|
3922
5749
|
...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
|
|
3923
5750
|
...title ? { title } : {}
|
|
3924
5751
|
})
|
|
@@ -3967,6 +5794,11 @@ var ChannelDriver = class {
|
|
|
3967
5794
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
3968
5795
|
body: JSON.stringify({
|
|
3969
5796
|
status: "done",
|
|
5797
|
+
// ALWAYS sent, even for a session this conversation has abandoned
|
|
5798
|
+
// (#553): the server reads the reply text back out of THIS session id
|
|
5799
|
+
// to deliver it. Omitting it would leave the user with "✅ Done!"
|
|
5800
|
+
// instead of the answer — a worse regression than the resurrection it
|
|
5801
|
+
// would prevent, which `ensureSession`'s guard handles anyway.
|
|
3970
5802
|
opencode_session_id: sessionId,
|
|
3971
5803
|
...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
|
|
3972
5804
|
...title ? { title } : {},
|
|
@@ -3983,16 +5815,31 @@ var ChannelDriver = class {
|
|
|
3983
5815
|
}
|
|
3984
5816
|
/**
|
|
3985
5817
|
* Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
|
|
3986
|
-
* when provided (issue #182)
|
|
3987
|
-
* `
|
|
3988
|
-
*
|
|
3989
|
-
*
|
|
5818
|
+
* when provided (issue #182). Three states for `sessionId`:
|
|
5819
|
+
* - omitted (`undefined`) → don't send the field, leave the persisted
|
|
5820
|
+
* session untouched (unused today; kept for API symmetry).
|
|
5821
|
+
* - a real id (`string`) → send it, update the persisted session (the
|
|
5822
|
+
* turn-failure call sites: an errored OpenCode turn).
|
|
5823
|
+
* - explicit `null` → send it, CLEAR the persisted session (issue
|
|
5824
|
+
* #485's dispatch-handoff-failure call site: the session id still
|
|
5825
|
+
* exists but is wedged, so the next attempt must get a fresh one
|
|
5826
|
+
* instead of reusing it — see WI-1's server-side null-clearing PATCH).
|
|
3990
5827
|
*/
|
|
3991
|
-
async markFailed(conversationId, messageId, sessionId, error2, usage) {
|
|
5828
|
+
async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
|
|
3992
5829
|
const body = { status: "failed" };
|
|
3993
|
-
if (sessionId
|
|
5830
|
+
if (sessionId === null) {
|
|
5831
|
+
body.opencode_session_id = null;
|
|
5832
|
+
} else if (sessionId !== void 0) {
|
|
5833
|
+
Object.assign(body, this.sessionIdBody(sessionId, conversationId, messageId, "failed"));
|
|
5834
|
+
}
|
|
3994
5835
|
if (error2 !== void 0) body.error = error2;
|
|
3995
5836
|
if (usage) Object.assign(body, usage);
|
|
5837
|
+
if (failure) {
|
|
5838
|
+
body.failure_kind = failure.kind;
|
|
5839
|
+
body.failure_provider_id = failure.providerId;
|
|
5840
|
+
body.failure_model_id = failure.modelId;
|
|
5841
|
+
body.failure_reason = failure.reason;
|
|
5842
|
+
}
|
|
3996
5843
|
await this.callWithRetry(
|
|
3997
5844
|
"marking message as failed",
|
|
3998
5845
|
() => this.fetchImpl(
|
|
@@ -4005,6 +5852,29 @@ var ChannelDriver = class {
|
|
|
4005
5852
|
)
|
|
4006
5853
|
);
|
|
4007
5854
|
}
|
|
5855
|
+
/**
|
|
5856
|
+
* Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
|
|
5857
|
+
*
|
|
5858
|
+
* `messageFailure` alone (structured OpenCode error → `model_auth`) covers
|
|
5859
|
+
* most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
|
|
5860
|
+
* to the P1-2b zero-provider check — one extra loopback call to
|
|
5861
|
+
* `hasAnyConfiguredProvider`, only reached when the structured classifier
|
|
5862
|
+
* couldn't place it. Fails open (never throws): a fallback probe failure
|
|
5863
|
+
* (`null`/indeterminate) leaves the classification `null`, which produces
|
|
5864
|
+
* today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
|
|
5865
|
+
*/
|
|
5866
|
+
async classifyModelAuthFailure(messages, userMessageId) {
|
|
5867
|
+
const classified = messageFailure(messages, userMessageId);
|
|
5868
|
+
if (classified != null) return classified;
|
|
5869
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
5870
|
+
const hasProvider = await hasAnyConfiguredProvider(this.port);
|
|
5871
|
+
return applyZeroProviderFallback(
|
|
5872
|
+
classified,
|
|
5873
|
+
hasProvider,
|
|
5874
|
+
reply?.info?.providerID ?? null,
|
|
5875
|
+
reply?.info?.modelID ?? null
|
|
5876
|
+
);
|
|
5877
|
+
}
|
|
4008
5878
|
/**
|
|
4009
5879
|
* Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
|
|
4010
5880
|
* (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
|
|
@@ -4157,10 +6027,16 @@ var ChannelDriver = class {
|
|
|
4157
6027
|
import chalk5 from "chalk";
|
|
4158
6028
|
import ora2 from "ora";
|
|
4159
6029
|
import { select as select2 } from "@inquirer/prompts";
|
|
6030
|
+
var INTERACTIVE_START_TIMEOUT_MS = 3e4;
|
|
4160
6031
|
async function ensureOpenCodeRunning(ctx) {
|
|
4161
6032
|
const healthCheck = await checkOpenCodeHealth(ctx.port);
|
|
4162
6033
|
if (healthCheck.healthy) {
|
|
4163
|
-
return {
|
|
6034
|
+
return {
|
|
6035
|
+
port: ctx.port,
|
|
6036
|
+
process: null,
|
|
6037
|
+
version: healthCheck.version ?? null,
|
|
6038
|
+
notReadyReason: null
|
|
6039
|
+
};
|
|
4164
6040
|
}
|
|
4165
6041
|
const runningInstances = await findHealthyOpenCodeInstances();
|
|
4166
6042
|
if (runningInstances.length > 0) {
|
|
@@ -4181,7 +6057,7 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4181
6057
|
console.log(chalk5.yellow("Tip: Run with the correct port:"));
|
|
4182
6058
|
console.log(
|
|
4183
6059
|
chalk5.dim(
|
|
4184
|
-
` ${getCliName()} run --
|
|
6060
|
+
` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
|
|
4185
6061
|
)
|
|
4186
6062
|
);
|
|
4187
6063
|
}
|
|
@@ -4201,14 +6077,22 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4201
6077
|
if (!ctx.interactive) {
|
|
4202
6078
|
ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
|
|
4203
6079
|
const proc = await startOpenCode(ctx.port);
|
|
4204
|
-
const health = await waitForOpenCodeHealth(ctx.port,
|
|
6080
|
+
const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
|
|
4205
6081
|
if (!health.healthy) {
|
|
4206
|
-
|
|
4207
|
-
|
|
4208
|
-
|
|
6082
|
+
return {
|
|
6083
|
+
port: ctx.port,
|
|
6084
|
+
process: proc,
|
|
6085
|
+
version: null,
|
|
6086
|
+
notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
|
|
6087
|
+
};
|
|
4209
6088
|
}
|
|
4210
6089
|
ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
|
|
4211
|
-
return {
|
|
6090
|
+
return {
|
|
6091
|
+
port: ctx.port,
|
|
6092
|
+
process: proc,
|
|
6093
|
+
version: health.version ?? null,
|
|
6094
|
+
notReadyReason: null
|
|
6095
|
+
};
|
|
4212
6096
|
}
|
|
4213
6097
|
let port = ctx.port;
|
|
4214
6098
|
if (isPortInUse(port)) {
|
|
@@ -4261,122 +6145,15 @@ Port ${port} is already in use.`));
|
|
|
4261
6145
|
if (action === "start") {
|
|
4262
6146
|
const spinner = ora2("Starting OpenCode...").start();
|
|
4263
6147
|
const proc = await startOpenCode(port);
|
|
4264
|
-
const health = await waitForOpenCodeHealth(port,
|
|
6148
|
+
const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
|
|
4265
6149
|
if (!health.healthy) {
|
|
4266
6150
|
spinner.fail("Failed to start OpenCode");
|
|
4267
6151
|
throw new Error("OpenCode failed to start");
|
|
4268
6152
|
}
|
|
4269
6153
|
spinner.stop();
|
|
4270
|
-
return { port, process: proc, version: health.version ?? null };
|
|
4271
|
-
}
|
|
4272
|
-
return { port, process: null, version: null };
|
|
4273
|
-
}
|
|
4274
|
-
|
|
4275
|
-
// src/commands/agent-lookup.ts
|
|
4276
|
-
async function readErrorMessage(response) {
|
|
4277
|
-
const text = await response.text().catch(() => "");
|
|
4278
|
-
if (!text) return response.statusText || void 0;
|
|
4279
|
-
try {
|
|
4280
|
-
const data = JSON.parse(text);
|
|
4281
|
-
const message = data.message ?? data.error;
|
|
4282
|
-
if (typeof message === "string" && message.trim()) {
|
|
4283
|
-
return message;
|
|
4284
|
-
}
|
|
4285
|
-
} catch {
|
|
4286
|
-
}
|
|
4287
|
-
return text.trim() || response.statusText || void 0;
|
|
4288
|
-
}
|
|
4289
|
-
function authFailureHint(apiUrl, serverMessage) {
|
|
4290
|
-
const reason = serverMessage ? `: ${serverMessage}` : "";
|
|
4291
|
-
return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
|
|
4292
|
-
}
|
|
4293
|
-
async function resolveAgentIdFromKey(authHeader) {
|
|
4294
|
-
const apiUrl = getApiUrlConfig();
|
|
4295
|
-
try {
|
|
4296
|
-
const response = await fetch(`${apiUrl}/me`, {
|
|
4297
|
-
headers: { Authorization: authHeader }
|
|
4298
|
-
});
|
|
4299
|
-
if (response.status === 401) {
|
|
4300
|
-
const serverMessage = await readErrorMessage(response);
|
|
4301
|
-
return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
4302
|
-
}
|
|
4303
|
-
if (!response.ok) {
|
|
4304
|
-
const serverMessage = await readErrorMessage(response);
|
|
4305
|
-
return {
|
|
4306
|
-
error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
4307
|
-
};
|
|
4308
|
-
}
|
|
4309
|
-
const data = await response.json();
|
|
4310
|
-
if (data.auth_type === "agent_key" && data.agent_id) {
|
|
4311
|
-
return { agent_id: data.agent_id };
|
|
4312
|
-
}
|
|
4313
|
-
return {
|
|
4314
|
-
error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
|
|
4315
|
-
};
|
|
4316
|
-
} catch (error2) {
|
|
4317
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
4318
|
-
return { error: `Failed to resolve runner from key: ${message}` };
|
|
4319
|
-
}
|
|
4320
|
-
}
|
|
4321
|
-
async function notifyAgentDisconnected(agentId, authHeader) {
|
|
4322
|
-
const apiUrl = getApiUrlConfig();
|
|
4323
|
-
try {
|
|
4324
|
-
const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
|
|
4325
|
-
method: "POST",
|
|
4326
|
-
headers: { Authorization: authHeader }
|
|
4327
|
-
});
|
|
4328
|
-
if (!response.ok) {
|
|
4329
|
-
const serverMessage = await readErrorMessage(response);
|
|
4330
|
-
return {
|
|
4331
|
-
ok: false,
|
|
4332
|
-
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
4333
|
-
};
|
|
4334
|
-
}
|
|
4335
|
-
return { ok: true };
|
|
4336
|
-
} catch (error2) {
|
|
4337
|
-
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
4338
|
-
}
|
|
4339
|
-
}
|
|
4340
|
-
async function getAgentInfo(agentId, authHeader) {
|
|
4341
|
-
const apiUrl = getApiUrlConfig();
|
|
4342
|
-
try {
|
|
4343
|
-
const response = await fetch(`${apiUrl}/runners/${agentId}`, {
|
|
4344
|
-
headers: { Authorization: authHeader }
|
|
4345
|
-
});
|
|
4346
|
-
if (response.status === 401) {
|
|
4347
|
-
const serverMessage = await readErrorMessage(response);
|
|
4348
|
-
return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
4349
|
-
}
|
|
4350
|
-
if (response.status === 403) {
|
|
4351
|
-
const serverMessage = await readErrorMessage(response);
|
|
4352
|
-
return {
|
|
4353
|
-
valid: false,
|
|
4354
|
-
error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
|
|
4355
|
-
};
|
|
4356
|
-
}
|
|
4357
|
-
if (response.status === 404) {
|
|
4358
|
-
const serverMessage = await readErrorMessage(response);
|
|
4359
|
-
return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
|
|
4360
|
-
}
|
|
4361
|
-
if (!response.ok) {
|
|
4362
|
-
const serverMessage = await readErrorMessage(response);
|
|
4363
|
-
return {
|
|
4364
|
-
valid: false,
|
|
4365
|
-
error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
4366
|
-
};
|
|
4367
|
-
}
|
|
4368
|
-
const agent = await response.json();
|
|
4369
|
-
if (agent.agent_type !== "local") {
|
|
4370
|
-
return {
|
|
4371
|
-
valid: false,
|
|
4372
|
-
error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
|
|
4373
|
-
};
|
|
4374
|
-
}
|
|
4375
|
-
return { valid: true, agent };
|
|
4376
|
-
} catch (error2) {
|
|
4377
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
4378
|
-
return { valid: false, error: `Failed to validate runner: ${message}` };
|
|
6154
|
+
return { port, process: proc, version: health.version ?? null, notReadyReason: null };
|
|
4379
6155
|
}
|
|
6156
|
+
return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
|
|
4380
6157
|
}
|
|
4381
6158
|
|
|
4382
6159
|
// src/commands/run.ts
|
|
@@ -4384,6 +6161,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
|
|
|
4384
6161
|
var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
|
|
4385
6162
|
var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
|
|
4386
6163
|
var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
|
|
6164
|
+
var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
|
|
4387
6165
|
function resolveLogLevel(options) {
|
|
4388
6166
|
const accepted = Object.keys(LOG_LEVELS);
|
|
4389
6167
|
const validate = (value, source) => {
|
|
@@ -4407,6 +6185,63 @@ function resolveLogLevel(options) {
|
|
|
4407
6185
|
}
|
|
4408
6186
|
return "info";
|
|
4409
6187
|
}
|
|
6188
|
+
function resolveFileSyncDirectories(raw, homeDir) {
|
|
6189
|
+
const directories = [];
|
|
6190
|
+
for (const entry of raw ?? []) {
|
|
6191
|
+
const trimmed = entry.trim();
|
|
6192
|
+
if (trimmed === "") {
|
|
6193
|
+
throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
|
|
6194
|
+
}
|
|
6195
|
+
const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join3(homeDir, trimmed.slice(2)) : trimmed;
|
|
6196
|
+
if (!isAbsolute2(expanded)) {
|
|
6197
|
+
throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
|
|
6198
|
+
}
|
|
6199
|
+
const normalized = resolvePath(expanded);
|
|
6200
|
+
if (parse(normalized).root === normalized) {
|
|
6201
|
+
throw new Error(
|
|
6202
|
+
`--enable-file-sync-to will not allow-list the filesystem root ("${entry}"); name the specific directory the credentials belong in (for example ~/.claude)`
|
|
6203
|
+
);
|
|
6204
|
+
}
|
|
6205
|
+
if (!directories.includes(normalized)) {
|
|
6206
|
+
directories.push(normalized);
|
|
6207
|
+
}
|
|
6208
|
+
}
|
|
6209
|
+
if (directories.length > MAX_FILE_SYNC_DIRECTORIES) {
|
|
6210
|
+
throw new Error(
|
|
6211
|
+
`--enable-file-sync-to accepts at most ${MAX_FILE_SYNC_DIRECTORIES} directories; got ${directories.length}`
|
|
6212
|
+
);
|
|
6213
|
+
}
|
|
6214
|
+
return directories;
|
|
6215
|
+
}
|
|
6216
|
+
var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
|
|
6217
|
+
var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
|
|
6218
|
+
var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
|
|
6219
|
+
function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
|
|
6220
|
+
const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
|
|
6221
|
+
let raw;
|
|
6222
|
+
let source;
|
|
6223
|
+
if (options.opencodeStartTimeout !== void 0) {
|
|
6224
|
+
raw = options.opencodeStartTimeout;
|
|
6225
|
+
source = "--opencode-start-timeout";
|
|
6226
|
+
} else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
|
|
6227
|
+
raw = env[OPENCODE_START_TIMEOUT_ENV];
|
|
6228
|
+
source = OPENCODE_START_TIMEOUT_ENV;
|
|
6229
|
+
} else {
|
|
6230
|
+
return { timeoutMs: defaultMs, warnings: [] };
|
|
6231
|
+
}
|
|
6232
|
+
const trimmed = raw.trim();
|
|
6233
|
+
const seconds = Number(trimmed);
|
|
6234
|
+
const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
|
|
6235
|
+
if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
|
|
6236
|
+
return {
|
|
6237
|
+
timeoutMs: defaultMs,
|
|
6238
|
+
warnings: [
|
|
6239
|
+
`Ignoring invalid ${source} "${raw}": expected a positive integer number of seconds (at most ${MAX_OPENCODE_START_TIMEOUT_SECONDS}); using the default ${DEFAULT_OPENCODE_START_TIMEOUT_SECONDS}s`
|
|
6240
|
+
]
|
|
6241
|
+
};
|
|
6242
|
+
}
|
|
6243
|
+
return { timeoutMs: seconds * 1e3, warnings: [] };
|
|
6244
|
+
}
|
|
4410
6245
|
function meetsThreshold(state, level) {
|
|
4411
6246
|
return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
|
|
4412
6247
|
}
|
|
@@ -4428,6 +6263,10 @@ function log2(state, message, level = "info") {
|
|
|
4428
6263
|
function logActivity(state, entry) {
|
|
4429
6264
|
const level = entry.level ?? (entry.type === "error" ? "error" : "info");
|
|
4430
6265
|
if (!meetsThreshold(state, level)) return;
|
|
6266
|
+
forwardRunnerActivity(
|
|
6267
|
+
{ level, message: entry.message, error: entry.error },
|
|
6268
|
+
{ agentId: state.agentId, authHeader: state.authHeader }
|
|
6269
|
+
);
|
|
4431
6270
|
const fullEntry = {
|
|
4432
6271
|
...entry,
|
|
4433
6272
|
level,
|
|
@@ -4527,23 +6366,46 @@ async function handleAuthError(state, error2) {
|
|
|
4527
6366
|
}
|
|
4528
6367
|
async function driveChannels(state, driver) {
|
|
4529
6368
|
let idlePolls = 0;
|
|
6369
|
+
let idleMs = 0;
|
|
6370
|
+
let consecutiveDrainFailures = 0;
|
|
6371
|
+
let unreachableMs = 0;
|
|
4530
6372
|
let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
6373
|
+
let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
4531
6374
|
while (state.running) {
|
|
6375
|
+
const cycleStartedAtMs = performance.now();
|
|
6376
|
+
let idleThisCycle = false;
|
|
6377
|
+
let unreachableThisCycle = false;
|
|
4532
6378
|
if (state.connection?.reconnecting && state.connection.reconnectPromise) {
|
|
4533
6379
|
logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
|
|
4534
6380
|
if (state.interactive) displayStatus(state);
|
|
4535
6381
|
await state.connection.reconnectPromise;
|
|
4536
6382
|
}
|
|
6383
|
+
const carriedOverFileSync = driver.fileSyncActivity().inFlight;
|
|
6384
|
+
void driver.syncPendingFiles().catch(
|
|
6385
|
+
(error2) => logActivity(state, {
|
|
6386
|
+
type: "error",
|
|
6387
|
+
error: `Runner file sync failed: ${error2 instanceof Error ? error2.message : String(error2)}`
|
|
6388
|
+
})
|
|
6389
|
+
);
|
|
4537
6390
|
try {
|
|
4538
6391
|
const processed = await driver.drainPending();
|
|
6392
|
+
consecutiveDrainFailures = 0;
|
|
6393
|
+
unreachableMs = 0;
|
|
4539
6394
|
state.messageCount += processed;
|
|
4540
6395
|
const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
|
|
4541
6396
|
lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
4542
|
-
|
|
6397
|
+
const appliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
6398
|
+
const filesApplied = appliedFiles !== lastSeenAppliedFiles;
|
|
6399
|
+
const fileActivity = carriedOverFileSync || filesApplied;
|
|
6400
|
+
lastSeenAppliedFiles = appliedFiles;
|
|
6401
|
+
if (filesApplied) state.claudeUsageRearm?.();
|
|
6402
|
+
if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
|
|
4543
6403
|
idlePolls = 0;
|
|
6404
|
+
idleMs = 0;
|
|
4544
6405
|
if (processed > 0 && state.interactive) displayStatus(state);
|
|
4545
6406
|
} else if (state.idleTimeout !== null) {
|
|
4546
6407
|
idlePolls++;
|
|
6408
|
+
idleThisCycle = true;
|
|
4547
6409
|
if (idlePolls === 1) {
|
|
4548
6410
|
logActivity(state, {
|
|
4549
6411
|
type: "info",
|
|
@@ -4567,21 +6429,44 @@ async function driveChannels(state, driver) {
|
|
|
4567
6429
|
const errorMessage = error2 instanceof Error ? error2.message : String(error2);
|
|
4568
6430
|
logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
|
|
4569
6431
|
if (state.interactive) displayStatus(state);
|
|
4570
|
-
|
|
4571
|
-
|
|
4572
|
-
|
|
4573
|
-
|
|
4574
|
-
|
|
4575
|
-
|
|
4576
|
-
if (
|
|
4577
|
-
|
|
6432
|
+
if (driver.hasInFlightWatchers()) {
|
|
6433
|
+
consecutiveDrainFailures = 0;
|
|
6434
|
+
unreachableMs = 0;
|
|
6435
|
+
} else if (state.idleTimeout !== null) {
|
|
6436
|
+
consecutiveDrainFailures++;
|
|
6437
|
+
unreachableThisCycle = true;
|
|
6438
|
+
if (consecutiveDrainFailures === 1) {
|
|
6439
|
+
logActivity(state, {
|
|
6440
|
+
type: "info",
|
|
6441
|
+
message: `Cannot reach Evident, will exit if this persists past the idle timeout (timeout: ${state.idleTimeout}s)...`
|
|
6442
|
+
});
|
|
6443
|
+
if (state.interactive) displayStatus(state);
|
|
6444
|
+
}
|
|
4578
6445
|
}
|
|
4579
6446
|
}
|
|
6447
|
+
await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
|
|
6448
|
+
const cycleMs = performance.now() - cycleStartedAtMs;
|
|
6449
|
+
if (idleThisCycle) idleMs += cycleMs;
|
|
6450
|
+
if (unreachableThisCycle) unreachableMs += cycleMs;
|
|
6451
|
+
if (state.idleTimeout !== null && consecutiveDrainFailures >= 2 && unreachableMs > state.idleTimeout * 1e3) {
|
|
6452
|
+
logActivity(state, {
|
|
6453
|
+
type: "info",
|
|
6454
|
+
level: "warn",
|
|
6455
|
+
message: `Exiting: could not reach Evident for ${consecutiveDrainFailures} consecutive polls (${Math.round(unreachableMs / 1e3)}s)`
|
|
6456
|
+
});
|
|
6457
|
+
if (state.interactive) displayStatus(state);
|
|
6458
|
+
break;
|
|
6459
|
+
}
|
|
6460
|
+
if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
|
|
6461
|
+
logActivity(state, { type: "info", message: "Idle timeout reached" });
|
|
6462
|
+
if (state.interactive) displayStatus(state);
|
|
6463
|
+
break;
|
|
6464
|
+
}
|
|
4580
6465
|
}
|
|
4581
6466
|
}
|
|
4582
6467
|
var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
|
|
4583
|
-
async function runSweep(state, driver,
|
|
4584
|
-
const mode = `age=${
|
|
6468
|
+
async function runSweep(state, driver, config) {
|
|
6469
|
+
const mode = `age=${config.maxAgeMs ?? "\u2014"} count=${config.maxCount ?? "\u2014"}`;
|
|
4585
6470
|
try {
|
|
4586
6471
|
const sessions = await listSessions(state.port);
|
|
4587
6472
|
if (sessions === null) {
|
|
@@ -4594,8 +6479,8 @@ async function runSweep(state, driver, config2) {
|
|
|
4594
6479
|
const toDelete = selectSessionsToDelete(
|
|
4595
6480
|
sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
|
|
4596
6481
|
{
|
|
4597
|
-
maxAgeMs:
|
|
4598
|
-
maxCount:
|
|
6482
|
+
maxAgeMs: config.maxAgeMs,
|
|
6483
|
+
maxCount: config.maxCount,
|
|
4599
6484
|
nowMs: Date.now(),
|
|
4600
6485
|
protectedIds: driver.protectedSessionIds()
|
|
4601
6486
|
}
|
|
@@ -4631,7 +6516,7 @@ async function runSweep(state, driver, config2) {
|
|
|
4631
6516
|
}
|
|
4632
6517
|
}
|
|
4633
6518
|
function scheduleSessionCleanup(state, driver, options) {
|
|
4634
|
-
const
|
|
6519
|
+
const config = resolveSessionCleanupConfig(
|
|
4635
6520
|
{
|
|
4636
6521
|
maxAge: options.sessionCleanupMaxAge,
|
|
4637
6522
|
maxCount: options.sessionCleanupMaxCount,
|
|
@@ -4639,21 +6524,129 @@ function scheduleSessionCleanup(state, driver, options) {
|
|
|
4639
6524
|
},
|
|
4640
6525
|
process.env
|
|
4641
6526
|
);
|
|
4642
|
-
for (const warning2 of
|
|
6527
|
+
for (const warning2 of config.warnings) {
|
|
4643
6528
|
logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
|
|
4644
6529
|
}
|
|
4645
|
-
if (!
|
|
6530
|
+
if (!config.enabled) return;
|
|
4646
6531
|
logActivity(state, {
|
|
4647
6532
|
type: "info",
|
|
4648
|
-
message: `Session cleanup enabled (age=${
|
|
6533
|
+
message: `Session cleanup enabled (age=${config.maxAgeMs ?? "\u2014"}, count=${config.maxCount ?? "\u2014"}, interval=${config.intervalMs}ms)`
|
|
4649
6534
|
});
|
|
4650
|
-
const interval = setInterval(() => void runSweep(state, driver,
|
|
6535
|
+
const interval = setInterval(() => void runSweep(state, driver, config), config.intervalMs);
|
|
4651
6536
|
const firstSweep = setTimeout(
|
|
4652
|
-
() => void runSweep(state, driver,
|
|
6537
|
+
() => void runSweep(state, driver, config),
|
|
4653
6538
|
SESSION_CLEANUP_FIRST_SWEEP_MS
|
|
4654
6539
|
);
|
|
4655
6540
|
state.sessionCleanupTimers.push(interval, firstSweep);
|
|
4656
6541
|
}
|
|
6542
|
+
function claudeUsageFailureStreakSuffix(consecutiveFailures) {
|
|
6543
|
+
return consecutiveFailures > 1 ? ` (${consecutiveFailures} consecutive failures)` : "";
|
|
6544
|
+
}
|
|
6545
|
+
function scheduleClaudeUsageReporting(state, options) {
|
|
6546
|
+
const { mode, warnings } = resolveClaudeUsageReportingMode(
|
|
6547
|
+
options.claudeUsageReporting,
|
|
6548
|
+
process.env
|
|
6549
|
+
);
|
|
6550
|
+
for (const warning2 of warnings) {
|
|
6551
|
+
logActivity(state, {
|
|
6552
|
+
type: "info",
|
|
6553
|
+
level: "warn",
|
|
6554
|
+
message: `Claude usage reporting: ${warning2}`
|
|
6555
|
+
});
|
|
6556
|
+
}
|
|
6557
|
+
if (mode === "off") {
|
|
6558
|
+
logActivity(state, {
|
|
6559
|
+
type: "info",
|
|
6560
|
+
level: "debug",
|
|
6561
|
+
message: "Claude usage reporting is off (--claude-usage-reporting off)"
|
|
6562
|
+
});
|
|
6563
|
+
return null;
|
|
6564
|
+
}
|
|
6565
|
+
let consecutiveFailures = 0;
|
|
6566
|
+
let armed = false;
|
|
6567
|
+
let rearmRequested = false;
|
|
6568
|
+
const scheduleNextTick = () => {
|
|
6569
|
+
armed = true;
|
|
6570
|
+
rearmRequested = false;
|
|
6571
|
+
state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
|
|
6572
|
+
};
|
|
6573
|
+
const rearm = () => {
|
|
6574
|
+
if (armed) {
|
|
6575
|
+
rearmRequested = true;
|
|
6576
|
+
return;
|
|
6577
|
+
}
|
|
6578
|
+
rearmRequested = false;
|
|
6579
|
+
armed = true;
|
|
6580
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6581
|
+
};
|
|
6582
|
+
const tick = async (isProbe) => {
|
|
6583
|
+
try {
|
|
6584
|
+
const usage = await getClaudeUsage();
|
|
6585
|
+
const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
|
|
6586
|
+
if (result.ok) {
|
|
6587
|
+
if (consecutiveFailures > 0) {
|
|
6588
|
+
logActivity(state, {
|
|
6589
|
+
type: "info",
|
|
6590
|
+
level: "info",
|
|
6591
|
+
message: "Claude usage reporting recovered"
|
|
6592
|
+
});
|
|
6593
|
+
}
|
|
6594
|
+
consecutiveFailures = 0;
|
|
6595
|
+
logActivity(state, {
|
|
6596
|
+
type: "info",
|
|
6597
|
+
level: "debug",
|
|
6598
|
+
message: "Reported Claude usage to Evident"
|
|
6599
|
+
});
|
|
6600
|
+
} else {
|
|
6601
|
+
consecutiveFailures++;
|
|
6602
|
+
logActivity(state, {
|
|
6603
|
+
type: "info",
|
|
6604
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6605
|
+
message: `Failed to report Claude usage: ${result.error}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6606
|
+
});
|
|
6607
|
+
}
|
|
6608
|
+
scheduleNextTick();
|
|
6609
|
+
} catch (error2) {
|
|
6610
|
+
if (error2 instanceof ClaudeUsageError && isLocalCredentialProblem(error2)) {
|
|
6611
|
+
if (mode === "on") {
|
|
6612
|
+
logActivity(state, {
|
|
6613
|
+
type: "info",
|
|
6614
|
+
level: "warn",
|
|
6615
|
+
message: "Claude usage reporting is forced on but no usable Claude Code login was found \u2014 run `claude` to sign in; reporting will keep retrying"
|
|
6616
|
+
});
|
|
6617
|
+
scheduleNextTick();
|
|
6618
|
+
} else if (isProbe) {
|
|
6619
|
+
logActivity(state, {
|
|
6620
|
+
type: "info",
|
|
6621
|
+
level: "debug",
|
|
6622
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6623
|
+
});
|
|
6624
|
+
armed = false;
|
|
6625
|
+
if (rearmRequested) rearm();
|
|
6626
|
+
} else {
|
|
6627
|
+
logActivity(state, {
|
|
6628
|
+
type: "info",
|
|
6629
|
+
level: "debug",
|
|
6630
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6631
|
+
});
|
|
6632
|
+
scheduleNextTick();
|
|
6633
|
+
}
|
|
6634
|
+
} else {
|
|
6635
|
+
consecutiveFailures++;
|
|
6636
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
6637
|
+
logActivity(state, {
|
|
6638
|
+
type: "info",
|
|
6639
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6640
|
+
message: `Claude usage reporting failed: ${message}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6641
|
+
});
|
|
6642
|
+
scheduleNextTick();
|
|
6643
|
+
}
|
|
6644
|
+
}
|
|
6645
|
+
};
|
|
6646
|
+
armed = true;
|
|
6647
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6648
|
+
return rearm;
|
|
6649
|
+
}
|
|
4657
6650
|
async function notifyOffline(state) {
|
|
4658
6651
|
if (!state.agentId || !state.authHeader) return;
|
|
4659
6652
|
if (!state.connected) {
|
|
@@ -4671,13 +6664,29 @@ async function notifyOffline(state) {
|
|
|
4671
6664
|
if (state.interactive) displayStatus(state);
|
|
4672
6665
|
}
|
|
4673
6666
|
}
|
|
6667
|
+
async function timeShutdownPhase(state, durations, name, run2) {
|
|
6668
|
+
const startedAt = Date.now();
|
|
6669
|
+
try {
|
|
6670
|
+
return await run2();
|
|
6671
|
+
} finally {
|
|
6672
|
+
const elapsedMs = Date.now() - startedAt;
|
|
6673
|
+
durations[name] = elapsedMs;
|
|
6674
|
+
log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
|
|
6675
|
+
}
|
|
6676
|
+
}
|
|
4674
6677
|
async function cleanup(state, opts = {}) {
|
|
6678
|
+
const durations = {};
|
|
4675
6679
|
state.running = false;
|
|
4676
6680
|
for (const timer of state.sessionCleanupTimers) {
|
|
4677
6681
|
clearInterval(timer);
|
|
4678
6682
|
clearTimeout(timer);
|
|
4679
6683
|
}
|
|
4680
6684
|
state.sessionCleanupTimers = [];
|
|
6685
|
+
if (state.claudeUsageTimer) {
|
|
6686
|
+
clearTimeout(state.claudeUsageTimer);
|
|
6687
|
+
state.claudeUsageTimer = null;
|
|
6688
|
+
}
|
|
6689
|
+
state.claudeUsageRearm = null;
|
|
4681
6690
|
if (opts.graceful && state.channelDriver) {
|
|
4682
6691
|
state.channelDriver.stop();
|
|
4683
6692
|
log2(state, "Draining in-flight channel work before shutdown...");
|
|
@@ -4685,7 +6694,13 @@ async function cleanup(state, opts = {}) {
|
|
|
4685
6694
|
logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
|
|
4686
6695
|
displayStatus(state);
|
|
4687
6696
|
}
|
|
4688
|
-
const
|
|
6697
|
+
const driver = state.channelDriver;
|
|
6698
|
+
const settled = await timeShutdownPhase(
|
|
6699
|
+
state,
|
|
6700
|
+
durations,
|
|
6701
|
+
"drain",
|
|
6702
|
+
() => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
|
|
6703
|
+
);
|
|
4689
6704
|
if (!settled) {
|
|
4690
6705
|
logActivity(state, {
|
|
4691
6706
|
type: "info",
|
|
@@ -4694,13 +6709,15 @@ async function cleanup(state, opts = {}) {
|
|
|
4694
6709
|
if (state.interactive) displayStatus(state);
|
|
4695
6710
|
}
|
|
4696
6711
|
}
|
|
4697
|
-
await notifyOffline(state);
|
|
6712
|
+
await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
|
|
4698
6713
|
if (state.connection) {
|
|
4699
|
-
state.connection
|
|
6714
|
+
const connection = state.connection;
|
|
6715
|
+
await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
|
|
4700
6716
|
state.connection = null;
|
|
4701
6717
|
}
|
|
4702
6718
|
if (state.opencodeProcess) {
|
|
4703
|
-
|
|
6719
|
+
const opencodeProcess = state.opencodeProcess;
|
|
6720
|
+
await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
|
|
4704
6721
|
if (state.interactive) {
|
|
4705
6722
|
logActivity(state, { type: "info", message: "Stopped OpenCode process" });
|
|
4706
6723
|
displayStatus(state);
|
|
@@ -4709,12 +6726,15 @@ async function cleanup(state, opts = {}) {
|
|
|
4709
6726
|
}
|
|
4710
6727
|
state.opencodeProcess = null;
|
|
4711
6728
|
}
|
|
6729
|
+
return durations;
|
|
4712
6730
|
}
|
|
4713
6731
|
async function run(options) {
|
|
4714
6732
|
const interactive = isInteractive(options.json);
|
|
4715
6733
|
let logLevel;
|
|
6734
|
+
let fileSyncDirectories;
|
|
4716
6735
|
try {
|
|
4717
6736
|
logLevel = resolveLogLevel(options);
|
|
6737
|
+
fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir3());
|
|
4718
6738
|
} catch (error2) {
|
|
4719
6739
|
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
4720
6740
|
if (options.json) {
|
|
@@ -4747,8 +6767,16 @@ async function run(options) {
|
|
|
4747
6767
|
messageCount: 0,
|
|
4748
6768
|
lastProxiedActivityAt: null,
|
|
4749
6769
|
sessionCleanupTimers: [],
|
|
6770
|
+
claudeUsageTimer: null,
|
|
6771
|
+
claudeUsageRearm: null,
|
|
4750
6772
|
authHeader: ""
|
|
4751
6773
|
};
|
|
6774
|
+
setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
|
|
6775
|
+
if (fileSyncDirectories.length > 0) {
|
|
6776
|
+
log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
|
|
6777
|
+
} else {
|
|
6778
|
+
log2(state, "File sync is disabled (no --enable-file-sync-to given)", "debug");
|
|
6779
|
+
}
|
|
4752
6780
|
if (!options.runner && options.agent) {
|
|
4753
6781
|
telemetry.info(
|
|
4754
6782
|
EventTypes.DEPRECATED_AGENT_FLAG_USED,
|
|
@@ -4772,14 +6800,38 @@ async function run(options) {
|
|
|
4772
6800
|
const handleSignal = async () => {
|
|
4773
6801
|
if (state.shuttingDown) return;
|
|
4774
6802
|
state.shuttingDown = true;
|
|
6803
|
+
const shutdownStartedAt = Date.now();
|
|
4775
6804
|
if (state.interactive) {
|
|
4776
6805
|
logActivity(state, { type: "info", message: "Shutting down..." });
|
|
4777
6806
|
displayStatus(state);
|
|
4778
6807
|
} else {
|
|
4779
6808
|
log2(state, "Shutting down...");
|
|
4780
6809
|
}
|
|
4781
|
-
await cleanup(state, { graceful: true });
|
|
4782
|
-
|
|
6810
|
+
const durations = await cleanup(state, { graceful: true });
|
|
6811
|
+
const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
|
|
6812
|
+
await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
|
|
6813
|
+
let timer;
|
|
6814
|
+
const flushed = shutdownTelemetry().then(
|
|
6815
|
+
() => true,
|
|
6816
|
+
(error2) => {
|
|
6817
|
+
log2(
|
|
6818
|
+
state,
|
|
6819
|
+
`Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
|
|
6820
|
+
"warn"
|
|
6821
|
+
);
|
|
6822
|
+
return true;
|
|
6823
|
+
}
|
|
6824
|
+
);
|
|
6825
|
+
const timedOut = new Promise((resolve3) => {
|
|
6826
|
+
timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
|
|
6827
|
+
});
|
|
6828
|
+
if (!await Promise.race([flushed, timedOut])) {
|
|
6829
|
+
log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
|
|
6830
|
+
}
|
|
6831
|
+
clearTimeout(timer);
|
|
6832
|
+
});
|
|
6833
|
+
const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
|
|
6834
|
+
log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
|
|
4783
6835
|
process.exit(0);
|
|
4784
6836
|
};
|
|
4785
6837
|
process.on("SIGINT", handleSignal);
|
|
@@ -4796,6 +6848,7 @@ async function run(options) {
|
|
|
4796
6848
|
console.log(chalk6.dim("Or run `evident login` for interactive authentication"));
|
|
4797
6849
|
blank();
|
|
4798
6850
|
process.exit(1);
|
|
6851
|
+
return;
|
|
4799
6852
|
}
|
|
4800
6853
|
blank();
|
|
4801
6854
|
console.log(chalk6.yellow("You are not logged in to Evident."));
|
|
@@ -4840,6 +6893,7 @@ async function run(options) {
|
|
|
4840
6893
|
} else {
|
|
4841
6894
|
printError(resolved.error || "Failed to resolve runner ID from key");
|
|
4842
6895
|
process.exit(1);
|
|
6896
|
+
return;
|
|
4843
6897
|
}
|
|
4844
6898
|
} else {
|
|
4845
6899
|
printError(
|
|
@@ -4853,6 +6907,7 @@ async function run(options) {
|
|
|
4853
6907
|
);
|
|
4854
6908
|
blank();
|
|
4855
6909
|
process.exit(1);
|
|
6910
|
+
return;
|
|
4856
6911
|
}
|
|
4857
6912
|
}
|
|
4858
6913
|
telemetry.info(
|
|
@@ -4893,40 +6948,67 @@ async function run(options) {
|
|
|
4893
6948
|
}
|
|
4894
6949
|
spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
|
|
4895
6950
|
state.agentName = validation.agent.name;
|
|
6951
|
+
const microvmId = process.env.MICROVM_ID?.trim();
|
|
6952
|
+
if (microvmId) {
|
|
6953
|
+
const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
|
|
6954
|
+
if (reported.ok) {
|
|
6955
|
+
log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
|
|
6956
|
+
} else {
|
|
6957
|
+
const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
|
|
6958
|
+
log2(state, message, "warn");
|
|
6959
|
+
if (state.interactive && !state.json) {
|
|
6960
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
6961
|
+
}
|
|
6962
|
+
}
|
|
6963
|
+
} else {
|
|
6964
|
+
log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
|
|
6965
|
+
}
|
|
6966
|
+
const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
|
|
6967
|
+
for (const warning2 of opencodeStartTimeoutWarnings) {
|
|
6968
|
+
logActivity(state, { type: "info", level: "warn", message: warning2 });
|
|
6969
|
+
}
|
|
4896
6970
|
const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
|
|
4897
6971
|
try {
|
|
4898
6972
|
const oc = await ensureOpenCodeRunning({
|
|
4899
6973
|
port: state.port,
|
|
4900
6974
|
interactive: state.interactive,
|
|
4901
6975
|
agentId: state.agentId,
|
|
4902
|
-
log: (message) => log2(state, message)
|
|
6976
|
+
log: (message) => log2(state, message),
|
|
6977
|
+
startTimeoutMs: opencodeStartTimeoutMs
|
|
4903
6978
|
});
|
|
4904
6979
|
state.port = oc.port;
|
|
4905
6980
|
state.opencodeProcess = oc.process;
|
|
4906
6981
|
state.opencodeVersion = oc.version;
|
|
4907
|
-
state.opencodeConnected = oc.
|
|
6982
|
+
state.opencodeConnected = oc.notReadyReason === null;
|
|
4908
6983
|
const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
|
|
4909
6984
|
ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
|
|
4910
|
-
|
|
4911
|
-
|
|
4912
|
-
|
|
4913
|
-
|
|
4914
|
-
|
|
6985
|
+
if (!state.interactive && oc.notReadyReason !== null) {
|
|
6986
|
+
const message = `OpenCode is not ready on port ${state.port}: ${oc.notReadyReason}. The runner will still come online, but messages will fail until opencode answers \u2014 raise the wait with --opencode-start-timeout <seconds> (env ${OPENCODE_START_TIMEOUT_ENV}).`;
|
|
6987
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
6988
|
+
} else {
|
|
6989
|
+
const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
|
|
6990
|
+
if (versionWarning) {
|
|
6991
|
+
log2(state, versionWarning, "warn");
|
|
6992
|
+
if (state.interactive && !state.json) {
|
|
6993
|
+
logActivity(state, { type: "info", level: "warn", message: versionWarning });
|
|
6994
|
+
}
|
|
4915
6995
|
}
|
|
4916
|
-
|
|
4917
|
-
|
|
4918
|
-
|
|
4919
|
-
|
|
4920
|
-
|
|
4921
|
-
|
|
4922
|
-
|
|
4923
|
-
|
|
4924
|
-
|
|
4925
|
-
|
|
4926
|
-
|
|
4927
|
-
|
|
4928
|
-
|
|
4929
|
-
|
|
6996
|
+
const noProviderWarning = buildNoProviderWarning(
|
|
6997
|
+
await hasAnyConfiguredProvider(state.port)
|
|
6998
|
+
);
|
|
6999
|
+
if (noProviderWarning) {
|
|
7000
|
+
log2(state, noProviderWarning, "warn");
|
|
7001
|
+
if (state.interactive && !state.json) {
|
|
7002
|
+
logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
|
|
7003
|
+
blank();
|
|
7004
|
+
console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
|
|
7005
|
+
console.log(
|
|
7006
|
+
chalk6.dim(
|
|
7007
|
+
`Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
|
|
7008
|
+
)
|
|
7009
|
+
);
|
|
7010
|
+
blank();
|
|
7011
|
+
}
|
|
4930
7012
|
}
|
|
4931
7013
|
}
|
|
4932
7014
|
} catch (error2) {
|
|
@@ -4941,6 +7023,10 @@ async function run(options) {
|
|
|
4941
7023
|
getAuthHeader: () => state.authHeader,
|
|
4942
7024
|
conversationFilter: state.conversationFilter,
|
|
4943
7025
|
stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
|
|
7026
|
+
// #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
|
|
7027
|
+
// REJECTED with `file_sync_disabled` on the ack, not silently ignored.
|
|
7028
|
+
fileSyncDirectories,
|
|
7029
|
+
homeDir: homedir3(),
|
|
4944
7030
|
log: (entry) => (
|
|
4945
7031
|
// Thread the driver's real level straight through so `debug`/`warn`
|
|
4946
7032
|
// survive the sink filter (they no longer collapse to info). `type`
|
|
@@ -4967,6 +7053,18 @@ async function run(options) {
|
|
|
4967
7053
|
type: "info",
|
|
4968
7054
|
message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
|
|
4969
7055
|
});
|
|
7056
|
+
if (options.tunnelReadyFile) {
|
|
7057
|
+
const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
|
|
7058
|
+
if (marker.ok) {
|
|
7059
|
+
log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
|
|
7060
|
+
} else {
|
|
7061
|
+
log2(
|
|
7062
|
+
state,
|
|
7063
|
+
`Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
|
|
7064
|
+
"error"
|
|
7065
|
+
);
|
|
7066
|
+
}
|
|
7067
|
+
}
|
|
4970
7068
|
emitAgentConnected(state.agentId, {
|
|
4971
7069
|
port: state.port,
|
|
4972
7070
|
cli_version: getCliVersion(),
|
|
@@ -5022,6 +7120,12 @@ async function run(options) {
|
|
|
5022
7120
|
onDrainPing: () => {
|
|
5023
7121
|
if (!state.running) return;
|
|
5024
7122
|
logActivity(state, { type: "info", message: "Drain ping received \u2014 draining" });
|
|
7123
|
+
void channelDriver.syncPendingFiles().catch(
|
|
7124
|
+
(error2) => logActivity(state, {
|
|
7125
|
+
type: "error",
|
|
7126
|
+
error: `Runner file sync failed on ping: ${error2 instanceof Error ? error2.message : String(error2)}`
|
|
7127
|
+
})
|
|
7128
|
+
);
|
|
5025
7129
|
channelDriver.drainPending().then((processed) => {
|
|
5026
7130
|
if (processed > 0) {
|
|
5027
7131
|
state.messageCount += processed;
|
|
@@ -5051,6 +7155,7 @@ async function run(options) {
|
|
|
5051
7155
|
throw error2;
|
|
5052
7156
|
}
|
|
5053
7157
|
scheduleSessionCleanup(state, channelDriver, options);
|
|
7158
|
+
state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
|
|
5054
7159
|
if (!interactive || state.json) {
|
|
5055
7160
|
log2(state, "Driving channel messages...");
|
|
5056
7161
|
}
|
|
@@ -5105,10 +7210,18 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
|
|
|
5105
7210
|
program.command("login").description("Authenticate with Evident").option("--token", "Use token-based authentication (for CI/CD)").option("--no-browser", "Do not open the browser automatically").action(login);
|
|
5106
7211
|
program.command("logout").description("Remove stored credentials for the current endpoint").option("--all", "Remove stored credentials for all endpoints").action((options) => logout({ all: options.all }));
|
|
5107
7212
|
program.command("whoami").description("Show the currently logged in user").action(whoami);
|
|
5108
|
-
program.command("
|
|
7213
|
+
program.command("status").description("Check whether the configured credentials can reach Evident").option("--json", "Output in JSON format").action((options) => status({ json: options.json }));
|
|
7214
|
+
program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
|
|
7215
|
+
program.command("run").description("Connect to Evident and process messages").option("--runner [id]", "Runner ID to connect to (optional when EVIDENT_RUNNER_KEY is set)").option(
|
|
7216
|
+
"-a, --agent [id]",
|
|
7217
|
+
"Deprecated alias for --runner (still supported; --runner wins if both are given)"
|
|
7218
|
+
).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
|
|
5109
7219
|
"--log-level <level>",
|
|
5110
7220
|
"Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
|
|
5111
|
-
).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option(
|
|
7221
|
+
).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option(
|
|
7222
|
+
"--opencode-start-timeout <seconds>",
|
|
7223
|
+
"Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
|
|
7224
|
+
).option("--json", "Output in JSON format").option(
|
|
5112
7225
|
"--session-cleanup-max-age <duration>",
|
|
5113
7226
|
"Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
|
|
5114
7227
|
).option(
|
|
@@ -5117,6 +7230,17 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5117
7230
|
).option(
|
|
5118
7231
|
"--session-cleanup-interval <duration>",
|
|
5119
7232
|
"How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
|
|
7233
|
+
).option(
|
|
7234
|
+
"--claude-usage-reporting <mode>",
|
|
7235
|
+
"Report Claude subscription usage to Evident: auto | on | off (default: auto). Env: EVIDENT_CLAUDE_USAGE_REPORTING"
|
|
7236
|
+
).option(
|
|
7237
|
+
"--enable-file-sync-to <dir>",
|
|
7238
|
+
"Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
|
|
7239
|
+
(value, previous) => previous.concat([value]),
|
|
7240
|
+
[]
|
|
7241
|
+
).option(
|
|
7242
|
+
"--tunnel-ready-file <path>",
|
|
7243
|
+
"Path to write once the tunnel is connected (opt-in; unused unless an operator/image sets it)"
|
|
5120
7244
|
).action(
|
|
5121
7245
|
(options) => {
|
|
5122
7246
|
run({
|
|
@@ -5129,11 +7253,21 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5129
7253
|
verbose: options.verbose,
|
|
5130
7254
|
conversation: options.conversation,
|
|
5131
7255
|
idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
|
|
7256
|
+
// Raw string — validation/precedence is single-sourced in run.ts's
|
|
7257
|
+
// resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
|
|
7258
|
+
opencodeStartTimeout: options.opencodeStartTimeout,
|
|
5132
7259
|
json: options.json,
|
|
5133
7260
|
// Raw strings — the resolver in run.ts single-sources parsing (M1).
|
|
5134
7261
|
sessionCleanupMaxAge: options.sessionCleanupMaxAge,
|
|
5135
7262
|
sessionCleanupMaxCount: options.sessionCleanupMaxCount,
|
|
5136
|
-
sessionCleanupInterval: options.sessionCleanupInterval
|
|
7263
|
+
sessionCleanupInterval: options.sessionCleanupInterval,
|
|
7264
|
+
// Raw string — the resolver in run.ts single-sources parsing
|
|
7265
|
+
// (resolveClaudeUsageReportingMode).
|
|
7266
|
+
claudeUsageReporting: options.claudeUsageReporting,
|
|
7267
|
+
// Raw values — expansion/validation is single-sourced in run.ts's
|
|
7268
|
+
// resolveFileSyncDirectories.
|
|
7269
|
+
enableFileSyncTo: options.enableFileSyncTo,
|
|
7270
|
+
tunnelReadyFile: options.tunnelReadyFile
|
|
5137
7271
|
});
|
|
5138
7272
|
}
|
|
5139
7273
|
);
|