@evident-ai/cli 3.1.1-dev.993411a → 3.1.1-dev.9deb35f
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +60 -11
- package/dist/index.js +3004 -575
- 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 ${response.status}` };
|
|
1294
|
+
}
|
|
1295
|
+
const data = await response.json().catch(() => ({}));
|
|
1296
|
+
return { healthy: true, version: data.version };
|
|
1297
|
+
} catch (error2) {
|
|
1298
|
+
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
1299
|
+
return { healthy: false, error: message };
|
|
1300
|
+
}
|
|
1301
|
+
}
|
|
1302
|
+
async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
|
|
1303
|
+
const startTime = Date.now();
|
|
1304
|
+
while (Date.now() - startTime < timeoutMs) {
|
|
1305
|
+
const health = await checkOpenCodeHealth(port);
|
|
1306
|
+
if (health.healthy) {
|
|
1307
|
+
return health;
|
|
1308
|
+
}
|
|
1309
|
+
await new Promise((resolve3) => setTimeout(resolve3, 1e3));
|
|
1310
|
+
}
|
|
1311
|
+
return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
|
|
1312
|
+
}
|
|
1313
|
+
|
|
1314
|
+
// src/lib/opencode/opencode-version-gate.ts
|
|
1315
|
+
var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11", "1.18.3"];
|
|
1316
|
+
function isQueueValidatedVersion(version2) {
|
|
730
1317
|
if (!version2) return false;
|
|
731
1318
|
return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version2);
|
|
732
1319
|
}
|
|
@@ -734,7 +1321,7 @@ function buildOpenCodeVersionWarning(version2) {
|
|
|
734
1321
|
if (isQueueValidatedVersion(version2)) return null;
|
|
735
1322
|
const detected = version2 ? `v${version2}` : "unknown";
|
|
736
1323
|
const validated = QUEUE_VALIDATED_OPENCODE_VERSIONS.map((v) => `v${v}`).join(", ");
|
|
737
|
-
return `Warning: opencode ${detected} is not a queue-validated version (validated: ${validated}). Native message queuing \u2014 which channel (Slack
|
|
1324
|
+
return `Warning: opencode ${detected} is not a queue-validated version (validated: ${validated}). Native message queuing \u2014 which channel (Slack) message handling relies on \u2014 is unverified on this version; queued/follow-up messages may behave unexpectedly. Continuing anyway. Bumping the validated set requires re-running the queue validation.`;
|
|
738
1325
|
}
|
|
739
1326
|
|
|
740
1327
|
// src/lib/opencode/process.ts
|
|
@@ -1026,6 +1613,12 @@ async function promptOpenCodeInstall(interactive) {
|
|
|
1026
1613
|
return action;
|
|
1027
1614
|
}
|
|
1028
1615
|
|
|
1616
|
+
// src/lib/opencode/provider-check.ts
|
|
1617
|
+
function buildNoProviderWarning(hasProvider) {
|
|
1618
|
+
if (hasProvider !== false) return null;
|
|
1619
|
+
return "Warning: opencode has no authenticated model provider configured, so it won't be able to answer prompts. Run `opencode auth login` to set one up (see https://opencode.ai for details).";
|
|
1620
|
+
}
|
|
1621
|
+
|
|
1029
1622
|
// src/lib/opencode/session.ts
|
|
1030
1623
|
function opencodeBase(port) {
|
|
1031
1624
|
return `http://127.0.0.1:${port}`;
|
|
@@ -1228,6 +1821,11 @@ async function getModelAttachmentCapability(port, model) {
|
|
|
1228
1821
|
}
|
|
1229
1822
|
const entry = provider.models[modelId];
|
|
1230
1823
|
if (!entry || typeof entry !== "object") return null;
|
|
1824
|
+
if (entry.capabilities && typeof entry.capabilities === "object") {
|
|
1825
|
+
if (typeof entry.capabilities.attachment === "boolean") {
|
|
1826
|
+
return entry.capabilities.attachment;
|
|
1827
|
+
}
|
|
1828
|
+
}
|
|
1231
1829
|
return typeof entry.attachment === "boolean" ? entry.attachment : null;
|
|
1232
1830
|
} catch (err) {
|
|
1233
1831
|
console.error(
|
|
@@ -1256,6 +1854,16 @@ async function buildFileParts(attachments, capable) {
|
|
|
1256
1854
|
);
|
|
1257
1855
|
dataUrl = null;
|
|
1258
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
|
+
}
|
|
1259
1867
|
if (dataUrl == null) {
|
|
1260
1868
|
outcomes.push({ index: a.index, mime: a.mime, filename: a.filename, status: "failed" });
|
|
1261
1869
|
continue;
|
|
@@ -1337,7 +1945,7 @@ async function sendPromptAsync(port, sessionId, content, options, attachments) {
|
|
|
1337
1945
|
}
|
|
1338
1946
|
}
|
|
1339
1947
|
if (attempt < READ_BACK_ATTEMPTS - 1) {
|
|
1340
|
-
await new Promise((
|
|
1948
|
+
await new Promise((resolve3) => setTimeout(resolve3, READ_BACK_DELAY_MS));
|
|
1341
1949
|
}
|
|
1342
1950
|
}
|
|
1343
1951
|
return null;
|
|
@@ -1465,6 +2073,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
|
|
|
1465
2073
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1466
2074
|
return completedOf(reply) != null && finishOf(reply) === "tool-calls";
|
|
1467
2075
|
}
|
|
2076
|
+
function isB2AbandonmentConfirmed(params) {
|
|
2077
|
+
return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
|
|
2078
|
+
}
|
|
1468
2079
|
function messageError(messages, userMessageId) {
|
|
1469
2080
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1470
2081
|
const error2 = errorOf(reply);
|
|
@@ -1478,12 +2089,94 @@ function messageError(messages, userMessageId) {
|
|
|
1478
2089
|
}
|
|
1479
2090
|
return "The agent run failed.";
|
|
1480
2091
|
}
|
|
2092
|
+
function isAbortedTerminalReply(messages, userMessageId) {
|
|
2093
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
2094
|
+
const error2 = errorOf(reply);
|
|
2095
|
+
if (error2 == null) return false;
|
|
2096
|
+
if (typeof error2 === "string") return error2.trim() === "Aborted";
|
|
2097
|
+
if (typeof error2 === "object") {
|
|
2098
|
+
const e = error2;
|
|
2099
|
+
if (e.name === "MessageAbortedError") return true;
|
|
2100
|
+
if (e.name === "AbortError") return true;
|
|
2101
|
+
const dataMessage = e.data?.message;
|
|
2102
|
+
const rendered = typeof dataMessage === "string" ? dataMessage : typeof e.message === "string" ? e.message : null;
|
|
2103
|
+
return rendered != null && rendered.trim() === "Aborted";
|
|
2104
|
+
}
|
|
2105
|
+
return false;
|
|
2106
|
+
}
|
|
2107
|
+
function messageFailure(messages, userMessageId) {
|
|
2108
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
2109
|
+
const error2 = errorOf(reply);
|
|
2110
|
+
if (error2 == null || typeof error2 !== "object") return null;
|
|
2111
|
+
const e = error2;
|
|
2112
|
+
const replyProviderId = reply?.info?.providerID ?? null;
|
|
2113
|
+
const replyModelId = reply?.info?.modelID ?? null;
|
|
2114
|
+
if (e.name === "ProviderAuthError") {
|
|
2115
|
+
const data = e.data;
|
|
2116
|
+
const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
|
|
2117
|
+
return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
|
|
2118
|
+
}
|
|
2119
|
+
if (e.name === "APIError") {
|
|
2120
|
+
const data = e.data;
|
|
2121
|
+
const statusCode = data?.statusCode;
|
|
2122
|
+
if (statusCode === 401 || statusCode === 403) {
|
|
2123
|
+
return {
|
|
2124
|
+
kind: "model_auth",
|
|
2125
|
+
providerId: replyProviderId,
|
|
2126
|
+
modelId: replyModelId,
|
|
2127
|
+
reason: "rejected"
|
|
2128
|
+
};
|
|
2129
|
+
}
|
|
2130
|
+
}
|
|
2131
|
+
return null;
|
|
2132
|
+
}
|
|
2133
|
+
function applyZeroProviderFallback(classified, hasConfiguredProvider, replyProviderId, replyModelId = null) {
|
|
2134
|
+
if (classified != null) return classified;
|
|
2135
|
+
if (hasConfiguredProvider !== false) return null;
|
|
2136
|
+
return {
|
|
2137
|
+
kind: "model_auth",
|
|
2138
|
+
providerId: replyProviderId,
|
|
2139
|
+
modelId: replyModelId,
|
|
2140
|
+
reason: "missing"
|
|
2141
|
+
};
|
|
2142
|
+
}
|
|
1481
2143
|
function hasRunningAssistantExcept(messages, exceptUserMessageId) {
|
|
1482
2144
|
if (!messages || messages.length === 0) return false;
|
|
1483
2145
|
return messages.some(
|
|
1484
2146
|
(m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
|
|
1485
2147
|
);
|
|
1486
2148
|
}
|
|
2149
|
+
async function hasAnyConfiguredProvider(port) {
|
|
2150
|
+
try {
|
|
2151
|
+
const res = await fetch(`${opencodeBase(port)}/config/providers`);
|
|
2152
|
+
if (!res.ok) {
|
|
2153
|
+
console.error(
|
|
2154
|
+
`[hasAnyConfiguredProvider] GET /config/providers returned HTTP ${res.status} (port ${port})`
|
|
2155
|
+
);
|
|
2156
|
+
return null;
|
|
2157
|
+
}
|
|
2158
|
+
const body = await res.json();
|
|
2159
|
+
if (!body || typeof body !== "object" || Array.isArray(body)) {
|
|
2160
|
+
console.error(
|
|
2161
|
+
`[hasAnyConfiguredProvider] GET /config/providers body was not a plain object (port ${port})`
|
|
2162
|
+
);
|
|
2163
|
+
return null;
|
|
2164
|
+
}
|
|
2165
|
+
const defaults2 = body.default;
|
|
2166
|
+
if (!defaults2 || typeof defaults2 !== "object" || Array.isArray(defaults2)) {
|
|
2167
|
+
console.error(
|
|
2168
|
+
`[hasAnyConfiguredProvider] GET /config/providers body had no \`default\` object (port ${port})`
|
|
2169
|
+
);
|
|
2170
|
+
return null;
|
|
2171
|
+
}
|
|
2172
|
+
return Object.keys(defaults2).length > 0;
|
|
2173
|
+
} catch (err) {
|
|
2174
|
+
console.error(
|
|
2175
|
+
`[hasAnyConfiguredProvider] GET /config/providers failed (port ${port}): ${err instanceof Error ? err.message : String(err)}`
|
|
2176
|
+
);
|
|
2177
|
+
return null;
|
|
2178
|
+
}
|
|
2179
|
+
}
|
|
1487
2180
|
|
|
1488
2181
|
// src/lib/opencode/session-cleanup.ts
|
|
1489
2182
|
var DURATION_UNIT_MS = {
|
|
@@ -1642,10 +2335,11 @@ var StreamForwarder = class {
|
|
|
1642
2335
|
* Abort every in-flight stream (e.g. on WebSocket close).
|
|
1643
2336
|
*/
|
|
1644
2337
|
abortAll() {
|
|
1645
|
-
for (const stream of this.inflight.
|
|
2338
|
+
for (const [sid, stream] of this.inflight.entries()) {
|
|
1646
2339
|
try {
|
|
1647
2340
|
stream.abort();
|
|
1648
|
-
} catch {
|
|
2341
|
+
} catch (err) {
|
|
2342
|
+
log("error", "forwarder_abort_failed", { sid, ...errorFields(err) });
|
|
1649
2343
|
}
|
|
1650
2344
|
}
|
|
1651
2345
|
this.inflight.clear();
|
|
@@ -1679,12 +2373,12 @@ var StreamForwarder = class {
|
|
|
1679
2373
|
let endBody;
|
|
1680
2374
|
if (has_body) {
|
|
1681
2375
|
const chunks = [];
|
|
1682
|
-
bodyPromise = new Promise((
|
|
2376
|
+
bodyPromise = new Promise((resolve3) => {
|
|
1683
2377
|
pushBody = (buf) => {
|
|
1684
2378
|
chunks.push(buf);
|
|
1685
2379
|
};
|
|
1686
2380
|
endBody = () => {
|
|
1687
|
-
|
|
2381
|
+
resolve3(Buffer.concat(chunks));
|
|
1688
2382
|
};
|
|
1689
2383
|
});
|
|
1690
2384
|
}
|
|
@@ -1795,31 +2489,20 @@ function connectTunnel(options) {
|
|
|
1795
2489
|
onConnected,
|
|
1796
2490
|
onDisconnected,
|
|
1797
2491
|
onError,
|
|
1798
|
-
onRequest,
|
|
1799
2492
|
onResponse,
|
|
1800
2493
|
onInfo,
|
|
1801
2494
|
onDrainPing
|
|
1802
2495
|
} = options;
|
|
1803
2496
|
const tunnelUrl = getTunnelUrlConfig();
|
|
1804
2497
|
const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
|
|
1805
|
-
return new Promise((
|
|
2498
|
+
return new Promise((resolve3, reject) => {
|
|
1806
2499
|
const ws = new WebSocket2(url, {
|
|
1807
2500
|
headers: {
|
|
1808
2501
|
Authorization: authHeader
|
|
1809
2502
|
}
|
|
1810
2503
|
});
|
|
1811
|
-
const streamStartTimes = /* @__PURE__ */ new Map();
|
|
1812
2504
|
const forwarder = new StreamForwarder(ws, port, {
|
|
1813
|
-
|
|
1814
|
-
if (path === TUNNEL_DRAIN_PING_PATH) return;
|
|
1815
|
-
streamStartTimes.set(sid, Date.now());
|
|
1816
|
-
onRequest?.(method, path, sid);
|
|
1817
|
-
},
|
|
1818
|
-
onHead: (sid, status) => {
|
|
1819
|
-
const startedAt = streamStartTimes.get(sid);
|
|
1820
|
-
streamStartTimes.delete(sid);
|
|
1821
|
-
onResponse?.(status, startedAt ? Date.now() - startedAt : 0, sid);
|
|
1822
|
-
},
|
|
2505
|
+
onHead: () => onResponse?.(),
|
|
1823
2506
|
onDrainPing: () => onDrainPing?.()
|
|
1824
2507
|
});
|
|
1825
2508
|
const connectionTimeout = setTimeout(() => {
|
|
@@ -1867,7 +2550,7 @@ function connectTunnel(options) {
|
|
|
1867
2550
|
clearTimeout(connectionTimeout);
|
|
1868
2551
|
const connectedAgentId = message.agent_id ?? agentId;
|
|
1869
2552
|
onConnected?.(connectedAgentId);
|
|
1870
|
-
|
|
2553
|
+
resolve3({
|
|
1871
2554
|
ws,
|
|
1872
2555
|
close: () => ws.close(1e3, "CLI shutdown")
|
|
1873
2556
|
});
|
|
@@ -1895,7 +2578,6 @@ function connectTunnel(options) {
|
|
|
1895
2578
|
ws.on("close", (code, reason) => {
|
|
1896
2579
|
const reasonStr = reason.toString() || upgradeRejection || (code === 1006 ? "abnormal closure" : "No reason provided");
|
|
1897
2580
|
forwarder.abortAll();
|
|
1898
|
-
streamStartTimes.clear();
|
|
1899
2581
|
onDisconnected?.(code, reasonStr);
|
|
1900
2582
|
});
|
|
1901
2583
|
});
|
|
@@ -1930,57 +2612,502 @@ var RunnerConnection = class {
|
|
|
1930
2612
|
if (this.connection) {
|
|
1931
2613
|
try {
|
|
1932
2614
|
this.connection.close();
|
|
1933
|
-
} catch {
|
|
2615
|
+
} catch (err) {
|
|
2616
|
+
log("error", "runner_connection_close_failed", {
|
|
2617
|
+
agent_id: this.resolvedAgentId,
|
|
2618
|
+
...errorFields(err)
|
|
2619
|
+
});
|
|
2620
|
+
}
|
|
2621
|
+
this.connection = null;
|
|
2622
|
+
}
|
|
2623
|
+
}
|
|
2624
|
+
async connectWithRetry(isReconnect) {
|
|
2625
|
+
if (isReconnect && this.reconnecting) return;
|
|
2626
|
+
this.reconnecting = true;
|
|
2627
|
+
this.close();
|
|
2628
|
+
const { events } = this.opts;
|
|
2629
|
+
while (this.opts.isRunning()) {
|
|
2630
|
+
try {
|
|
2631
|
+
this.connection = await connectTunnel({
|
|
2632
|
+
agentId: this.resolvedAgentId,
|
|
2633
|
+
authHeader: this.opts.getAuthHeader(),
|
|
2634
|
+
port: this.opts.port,
|
|
2635
|
+
onConnected: (agentId) => {
|
|
2636
|
+
this.reconnectAttempt = 0;
|
|
2637
|
+
this.reconnecting = false;
|
|
2638
|
+
this.resolvedAgentId = agentId;
|
|
2639
|
+
events.onConnected(agentId, isReconnect);
|
|
2640
|
+
},
|
|
2641
|
+
onDisconnected: (code, reason) => {
|
|
2642
|
+
events.onDisconnected(code, reason);
|
|
2643
|
+
if (this.opts.isRunning() && code !== 1e3 && !this.reconnecting) {
|
|
2644
|
+
this.reconnectPromise = this.connectWithRetry(true).catch((err) => {
|
|
2645
|
+
events.onError?.(`Reconnection failed: ${err.message}`);
|
|
2646
|
+
});
|
|
2647
|
+
}
|
|
2648
|
+
},
|
|
2649
|
+
onError: (error2) => events.onError?.(error2),
|
|
2650
|
+
onResponse: () => events.onResponse?.(),
|
|
2651
|
+
onDrainPing: () => events.onDrainPing?.(),
|
|
2652
|
+
onInfo: (message) => events.onInfo?.(message)
|
|
2653
|
+
});
|
|
2654
|
+
return;
|
|
2655
|
+
} catch (error2) {
|
|
2656
|
+
this.reconnectAttempt++;
|
|
2657
|
+
if (error2.message === "Unauthorized") {
|
|
2658
|
+
this.reconnecting = false;
|
|
2659
|
+
throw error2;
|
|
2660
|
+
}
|
|
2661
|
+
const delay = getReconnectDelay(this.reconnectAttempt);
|
|
2662
|
+
events.onReconnecting?.(this.reconnectAttempt);
|
|
2663
|
+
events.onError?.(`Connection failed, retrying in ${Math.round(delay / 1e3)}s...`);
|
|
2664
|
+
await this.sleep(delay);
|
|
2665
|
+
}
|
|
2666
|
+
}
|
|
2667
|
+
this.reconnecting = false;
|
|
2668
|
+
}
|
|
2669
|
+
};
|
|
2670
|
+
|
|
2671
|
+
// src/lib/tunnel/ready-marker.ts
|
|
2672
|
+
import { writeFileSync } from "fs";
|
|
2673
|
+
function writeTunnelReadyMarker(path, agentId) {
|
|
2674
|
+
try {
|
|
2675
|
+
writeFileSync(path, `${agentId}
|
|
2676
|
+
`);
|
|
2677
|
+
return { ok: true };
|
|
2678
|
+
} catch (error2) {
|
|
2679
|
+
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
2680
|
+
}
|
|
2681
|
+
}
|
|
2682
|
+
|
|
2683
|
+
// src/lib/claude-usage-reporting.ts
|
|
2684
|
+
var VALID_MODES = ["auto", "on", "off"];
|
|
2685
|
+
function resolveClaudeUsageReportingMode(flagValue, env) {
|
|
2686
|
+
const raw = flagValue ?? env.EVIDENT_CLAUDE_USAGE_REPORTING;
|
|
2687
|
+
if (raw === void 0 || raw === "") {
|
|
2688
|
+
return { mode: "auto", warnings: [] };
|
|
2689
|
+
}
|
|
2690
|
+
const normalized = raw.trim().toLowerCase();
|
|
2691
|
+
if (VALID_MODES.includes(normalized)) {
|
|
2692
|
+
return { mode: normalized, warnings: [] };
|
|
2693
|
+
}
|
|
2694
|
+
const source = flagValue !== void 0 ? "--claude-usage-reporting" : "EVIDENT_CLAUDE_USAGE_REPORTING";
|
|
2695
|
+
return {
|
|
2696
|
+
mode: "auto",
|
|
2697
|
+
warnings: [
|
|
2698
|
+
`Ignoring invalid ${source} "${raw}": expected one of ${VALID_MODES.join(", ")}; using auto`
|
|
2699
|
+
]
|
|
2700
|
+
};
|
|
2701
|
+
}
|
|
2702
|
+
var BASE_REPORT_DELAY_MS = 10 * 6e4;
|
|
2703
|
+
var REPORT_DELAY_JITTER_FRACTION = 0.2;
|
|
2704
|
+
function nextReportDelayMs(random = Math.random) {
|
|
2705
|
+
const jitterRangeMs = BASE_REPORT_DELAY_MS * REPORT_DELAY_JITTER_FRACTION;
|
|
2706
|
+
return BASE_REPORT_DELAY_MS - jitterRangeMs + random() * (2 * jitterRangeMs);
|
|
2707
|
+
}
|
|
2708
|
+
var FIRST_REPORT_DELAY_MS = 5e3 + Math.random() * 1e4;
|
|
2709
|
+
var CLAUDE_USAGE_FAILURE_REESCALATION_TICKS = 6;
|
|
2710
|
+
function claudeUsageFailureLogLevel(consecutiveFailures) {
|
|
2711
|
+
return consecutiveFailures === 1 || consecutiveFailures % CLAUDE_USAGE_FAILURE_REESCALATION_TICKS === 0 ? "warn" : "debug";
|
|
2712
|
+
}
|
|
2713
|
+
|
|
2714
|
+
// src/lib/channels/driver.ts
|
|
2715
|
+
import { homedir as homedir2 } from "os";
|
|
2716
|
+
|
|
2717
|
+
// src/lib/file-push.ts
|
|
2718
|
+
import { randomUUID } from "crypto";
|
|
2719
|
+
import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
|
|
2720
|
+
import { basename, dirname as dirname2, isAbsolute, join as join2, relative, resolve as resolve2, sep } from "path";
|
|
2721
|
+
var FILE_MODE = 384;
|
|
2722
|
+
var DIRECTORY_MODE = 448;
|
|
2723
|
+
async function writePushedFile(request) {
|
|
2724
|
+
const { requestedPath, content, allowedDirectories, homeDir } = request;
|
|
2725
|
+
const bytes = content.byteLength;
|
|
2726
|
+
if (allowedDirectories.length === 0) {
|
|
2727
|
+
return refuse("file_sync_disabled", "File sync is not enabled on this runner.", {
|
|
2728
|
+
path: requestedPath,
|
|
2729
|
+
bytes
|
|
2730
|
+
});
|
|
2731
|
+
}
|
|
2732
|
+
if (bytes > MAX_FILE_PUSH_BYTES) {
|
|
2733
|
+
return refuse(
|
|
2734
|
+
"file_too_large",
|
|
2735
|
+
`File is ${bytes} bytes; the limit is ${MAX_FILE_PUSH_BYTES}.`,
|
|
2736
|
+
{
|
|
2737
|
+
path: requestedPath,
|
|
2738
|
+
bytes
|
|
2739
|
+
}
|
|
2740
|
+
);
|
|
2741
|
+
}
|
|
2742
|
+
const candidate = expandAndValidate(requestedPath, homeDir);
|
|
2743
|
+
if (candidate === null) {
|
|
2744
|
+
return refuse("invalid_path", "The requested path is not a valid absolute file path.", {
|
|
2745
|
+
path: requestedPath,
|
|
2746
|
+
bytes
|
|
2747
|
+
});
|
|
2748
|
+
}
|
|
2749
|
+
try {
|
|
2750
|
+
const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
|
|
2751
|
+
dirname2(candidate)
|
|
2752
|
+
);
|
|
2753
|
+
const realTarget = join2(existingAncestor, ...missingSegments, basename(candidate));
|
|
2754
|
+
const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
|
|
2755
|
+
if (allowedDirectory === null) {
|
|
2756
|
+
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
2757
|
+
path: realTarget,
|
|
2758
|
+
bytes
|
|
2759
|
+
});
|
|
2760
|
+
}
|
|
2761
|
+
if (missingSegments.length > 0) {
|
|
2762
|
+
await createMissingDirectories(existingAncestor, missingSegments);
|
|
2763
|
+
const realParent = await realpath(dirname2(realTarget));
|
|
2764
|
+
if (realParent !== dirname2(realTarget) || !contains(allowedDirectory, realTarget)) {
|
|
2765
|
+
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
2766
|
+
path: realTarget,
|
|
2767
|
+
bytes,
|
|
2768
|
+
reason: "parent_changed_after_create"
|
|
2769
|
+
});
|
|
2770
|
+
}
|
|
2771
|
+
}
|
|
2772
|
+
await writeAtomically(realTarget, content);
|
|
2773
|
+
log("info", "file_push_written", { path: realTarget, bytes });
|
|
2774
|
+
return { ok: true, path: realTarget };
|
|
2775
|
+
} catch (err) {
|
|
2776
|
+
const errno = err.code ?? "UNKNOWN";
|
|
2777
|
+
return refuse("write_failed", `The runner could not write the file (${errno}).`, {
|
|
2778
|
+
path: candidate,
|
|
2779
|
+
bytes,
|
|
2780
|
+
errno,
|
|
2781
|
+
...errorFields(err)
|
|
2782
|
+
});
|
|
2783
|
+
}
|
|
2784
|
+
}
|
|
2785
|
+
function expandAndValidate(requestedPath, homeDir) {
|
|
2786
|
+
if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
|
|
2787
|
+
return null;
|
|
2788
|
+
}
|
|
2789
|
+
const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join2(homeDir, requestedPath.slice(2)) : requestedPath;
|
|
2790
|
+
if (expanded.split(/[/\\]/).includes("..")) {
|
|
2791
|
+
return null;
|
|
2792
|
+
}
|
|
2793
|
+
if (!isAbsolute(expanded)) {
|
|
2794
|
+
return null;
|
|
2795
|
+
}
|
|
2796
|
+
const candidate = resolve2(expanded);
|
|
2797
|
+
const name = basename(candidate);
|
|
2798
|
+
return name === "" || name === "." || name === ".." ? null : candidate;
|
|
2799
|
+
}
|
|
2800
|
+
async function resolveNearestExistingAncestor(directory) {
|
|
2801
|
+
const missingSegments = [];
|
|
2802
|
+
let current = directory;
|
|
2803
|
+
for (; ; ) {
|
|
2804
|
+
try {
|
|
2805
|
+
return { existingAncestor: await realpath(current), missingSegments };
|
|
2806
|
+
} catch (err) {
|
|
2807
|
+
const parent = dirname2(current);
|
|
2808
|
+
if (err.code !== "ENOENT" || parent === current) {
|
|
2809
|
+
throw err;
|
|
2810
|
+
}
|
|
2811
|
+
missingSegments.unshift(basename(current));
|
|
2812
|
+
current = parent;
|
|
2813
|
+
}
|
|
2814
|
+
}
|
|
2815
|
+
}
|
|
2816
|
+
async function findContainingAllowedDirectory(allowedDirectories, realTarget) {
|
|
2817
|
+
for (const directory of allowedDirectories) {
|
|
2818
|
+
if (!isAbsolute(directory)) {
|
|
2819
|
+
log("warn", "file_push_allowed_directory_skipped", { directory, reason: "not_absolute" });
|
|
2820
|
+
continue;
|
|
2821
|
+
}
|
|
2822
|
+
const realDirectory = await realpathCreatingIfMissing(directory);
|
|
2823
|
+
if (realDirectory !== null && contains(realDirectory, realTarget)) {
|
|
2824
|
+
return realDirectory;
|
|
2825
|
+
}
|
|
2826
|
+
}
|
|
2827
|
+
return null;
|
|
2828
|
+
}
|
|
2829
|
+
async function realpathCreatingIfMissing(directory) {
|
|
2830
|
+
try {
|
|
2831
|
+
return await realpath(directory);
|
|
2832
|
+
} catch (err) {
|
|
2833
|
+
if (err.code !== "ENOENT") {
|
|
2834
|
+
log("warn", "file_push_allowed_directory_skipped", {
|
|
2835
|
+
directory,
|
|
2836
|
+
reason: "unresolvable",
|
|
2837
|
+
...errorFields(err)
|
|
2838
|
+
});
|
|
2839
|
+
return null;
|
|
2840
|
+
}
|
|
2841
|
+
}
|
|
2842
|
+
try {
|
|
2843
|
+
await mkdir(directory, { recursive: true, mode: DIRECTORY_MODE });
|
|
2844
|
+
await chmod(directory, DIRECTORY_MODE);
|
|
2845
|
+
return await realpath(directory);
|
|
2846
|
+
} catch (err) {
|
|
2847
|
+
log("warn", "file_push_allowed_directory_skipped", {
|
|
2848
|
+
directory,
|
|
2849
|
+
reason: "create_failed",
|
|
2850
|
+
...errorFields(err)
|
|
2851
|
+
});
|
|
2852
|
+
return null;
|
|
2853
|
+
}
|
|
2854
|
+
}
|
|
2855
|
+
function contains(realDirectory, realTarget) {
|
|
2856
|
+
const rel = relative(realDirectory, realTarget);
|
|
2857
|
+
return rel !== "" && rel !== ".." && !rel.startsWith(`..${sep}`) && !isAbsolute(rel);
|
|
2858
|
+
}
|
|
2859
|
+
async function createMissingDirectories(existingAncestor, missingSegments) {
|
|
2860
|
+
let current = existingAncestor;
|
|
2861
|
+
for (const segment of missingSegments) {
|
|
2862
|
+
current = join2(current, segment);
|
|
2863
|
+
await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
|
|
2864
|
+
await chmod(current, DIRECTORY_MODE);
|
|
2865
|
+
}
|
|
2866
|
+
}
|
|
2867
|
+
async function writeAtomically(realTarget, content) {
|
|
2868
|
+
const temporaryPath = join2(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
|
|
2869
|
+
let handle;
|
|
2870
|
+
try {
|
|
2871
|
+
handle = await open2(temporaryPath, "wx", FILE_MODE);
|
|
2872
|
+
await handle.writeFile(content);
|
|
2873
|
+
await handle.chmod(FILE_MODE);
|
|
2874
|
+
await handle.close();
|
|
2875
|
+
handle = void 0;
|
|
2876
|
+
await rename(temporaryPath, realTarget);
|
|
2877
|
+
} catch (err) {
|
|
2878
|
+
await discardTemporaryFile(temporaryPath, handle);
|
|
2879
|
+
throw err;
|
|
2880
|
+
}
|
|
2881
|
+
}
|
|
2882
|
+
async function discardTemporaryFile(temporaryPath, handle) {
|
|
2883
|
+
try {
|
|
2884
|
+
await handle?.close();
|
|
2885
|
+
} catch (err) {
|
|
2886
|
+
log("warn", "file_push_temp_close_failed", { path: temporaryPath, ...errorFields(err) });
|
|
2887
|
+
}
|
|
2888
|
+
try {
|
|
2889
|
+
await unlink(temporaryPath);
|
|
2890
|
+
} catch (err) {
|
|
2891
|
+
const errno = err.code;
|
|
2892
|
+
if (errno !== "ENOENT" && errno !== "ENOTDIR") {
|
|
2893
|
+
log("warn", "file_push_temp_cleanup_failed", { path: temporaryPath, ...errorFields(err) });
|
|
2894
|
+
}
|
|
2895
|
+
}
|
|
2896
|
+
}
|
|
2897
|
+
function refuse(code, message, fields) {
|
|
2898
|
+
log(code === "write_failed" ? "error" : "warn", "file_push_refused", { code, ...fields });
|
|
2899
|
+
return { ok: false, code, message };
|
|
2900
|
+
}
|
|
2901
|
+
|
|
2902
|
+
// src/lib/runner-file-sync.ts
|
|
2903
|
+
var MAX_ACK_ATTEMPTS = 5;
|
|
2904
|
+
async function syncPendingRunnerFiles(options) {
|
|
2905
|
+
const pending = await listPendingFiles(options);
|
|
2906
|
+
const pendingIds = new Set(pending.map((file) => file.id));
|
|
2907
|
+
for (const id of options.ackFailures.keys()) {
|
|
2908
|
+
if (!pendingIds.has(id)) options.ackFailures.delete(id);
|
|
2909
|
+
}
|
|
2910
|
+
if (pending.length === 0) return 0;
|
|
2911
|
+
options.log({
|
|
2912
|
+
level: "info",
|
|
2913
|
+
message: `Runner file sync: ${pending.length} file(s) queued for this runner`
|
|
2914
|
+
});
|
|
2915
|
+
let applied = 0;
|
|
2916
|
+
for (const file of pending) {
|
|
2917
|
+
if ((options.ackFailures.get(file.id) ?? 0) >= MAX_ACK_ATTEMPTS) continue;
|
|
2918
|
+
if (await applyOne(options, file)) applied += 1;
|
|
2919
|
+
}
|
|
2920
|
+
return applied;
|
|
2921
|
+
}
|
|
2922
|
+
async function listPendingFiles(options) {
|
|
2923
|
+
let res;
|
|
2924
|
+
try {
|
|
2925
|
+
res = await options.fetchImpl(`${options.apiUrl}/runners/${options.agentId}/files/pending`, {
|
|
2926
|
+
headers: { Authorization: options.getAuthHeader() }
|
|
2927
|
+
});
|
|
2928
|
+
} catch (err) {
|
|
2929
|
+
options.log({
|
|
2930
|
+
level: "warn",
|
|
2931
|
+
message: `Could not list pending runner files \u2014 retrying on the next drain: ${describe(err)}`
|
|
2932
|
+
});
|
|
2933
|
+
return [];
|
|
2934
|
+
}
|
|
2935
|
+
if (!res.ok) {
|
|
2936
|
+
options.log({
|
|
2937
|
+
level: res.status === 404 ? "debug" : "warn",
|
|
2938
|
+
message: `Listing pending runner files returned HTTP ${res.status} \u2014 retrying on the next drain`
|
|
2939
|
+
});
|
|
2940
|
+
return [];
|
|
2941
|
+
}
|
|
2942
|
+
let body;
|
|
2943
|
+
try {
|
|
2944
|
+
body = await res.json();
|
|
2945
|
+
} catch (err) {
|
|
2946
|
+
options.log({
|
|
2947
|
+
level: "warn",
|
|
2948
|
+
message: `Pending runner file list was not readable JSON \u2014 retrying on the next drain: ${describe(err)}`
|
|
2949
|
+
});
|
|
2950
|
+
return [];
|
|
2951
|
+
}
|
|
2952
|
+
if (!Array.isArray(body)) {
|
|
2953
|
+
options.log({
|
|
2954
|
+
level: "warn",
|
|
2955
|
+
message: "Pending runner file list was not an array \u2014 ignoring it for this drain"
|
|
2956
|
+
});
|
|
2957
|
+
return [];
|
|
2958
|
+
}
|
|
2959
|
+
const files = [];
|
|
2960
|
+
for (const entry of body) {
|
|
2961
|
+
const file = asPendingFile(entry);
|
|
2962
|
+
if (file === null) {
|
|
2963
|
+
options.log({
|
|
2964
|
+
level: "warn",
|
|
2965
|
+
message: "Ignoring a malformed pending runner file entry (expected id, path and size)"
|
|
2966
|
+
});
|
|
2967
|
+
continue;
|
|
2968
|
+
}
|
|
2969
|
+
files.push(file);
|
|
2970
|
+
}
|
|
2971
|
+
return files;
|
|
2972
|
+
}
|
|
2973
|
+
function asPendingFile(entry) {
|
|
2974
|
+
if (entry === null || typeof entry !== "object") return null;
|
|
2975
|
+
const { id, path, size } = entry;
|
|
2976
|
+
if (typeof id !== "string" || id === "") return null;
|
|
2977
|
+
if (typeof path !== "string" || path === "") return null;
|
|
2978
|
+
if (typeof size !== "number" || !Number.isFinite(size) || size < 0) return null;
|
|
2979
|
+
return { id, path, size };
|
|
2980
|
+
}
|
|
2981
|
+
async function applyOne(options, file) {
|
|
2982
|
+
const label = `${file.id.slice(0, 8)} (${file.path})`;
|
|
2983
|
+
if (options.allowedDirectories.length === 0) {
|
|
2984
|
+
options.log({
|
|
2985
|
+
level: "warn",
|
|
2986
|
+
message: `Runner file ${label} rejected: file sync is not enabled on this runner (start it with --enable-file-sync-to)`
|
|
2987
|
+
});
|
|
2988
|
+
await ack(options, file, "rejected", "file_sync_disabled");
|
|
2989
|
+
return false;
|
|
2990
|
+
}
|
|
2991
|
+
if (file.size > MAX_FILE_PUSH_BYTES) {
|
|
2992
|
+
options.log({
|
|
2993
|
+
level: "warn",
|
|
2994
|
+
message: `Runner file ${label} rejected: declared ${file.size} bytes, the limit is ${MAX_FILE_PUSH_BYTES}`
|
|
2995
|
+
});
|
|
2996
|
+
await ack(options, file, "rejected", "file_too_large");
|
|
2997
|
+
return false;
|
|
2998
|
+
}
|
|
2999
|
+
const download = await downloadContent(options, file, label);
|
|
3000
|
+
if (!download.ok) {
|
|
3001
|
+
if (download.terminal) await ack(options, file, "rejected", download.code);
|
|
3002
|
+
return false;
|
|
3003
|
+
}
|
|
3004
|
+
let outcome;
|
|
3005
|
+
try {
|
|
3006
|
+
outcome = await writePushedFile({
|
|
3007
|
+
requestedPath: file.path,
|
|
3008
|
+
content: download.content,
|
|
3009
|
+
allowedDirectories: options.allowedDirectories,
|
|
3010
|
+
homeDir: options.homeDir
|
|
3011
|
+
});
|
|
3012
|
+
} catch (err) {
|
|
3013
|
+
options.log({
|
|
3014
|
+
level: "error",
|
|
3015
|
+
message: `Runner file ${label} could not be written: ${describe(err)}`
|
|
3016
|
+
});
|
|
3017
|
+
await ack(options, file, "rejected", "write_failed");
|
|
3018
|
+
return false;
|
|
3019
|
+
}
|
|
3020
|
+
if (!outcome.ok) {
|
|
3021
|
+
options.log({
|
|
3022
|
+
level: "warn",
|
|
3023
|
+
message: `Runner file ${label} rejected (${outcome.code}): ${outcome.message}`
|
|
3024
|
+
});
|
|
3025
|
+
await ack(options, file, "rejected", outcome.code);
|
|
3026
|
+
return false;
|
|
3027
|
+
}
|
|
3028
|
+
options.log({
|
|
3029
|
+
level: "info",
|
|
3030
|
+
message: `Runner file ${label} applied (${download.content.byteLength} bytes)`
|
|
3031
|
+
});
|
|
3032
|
+
await ack(options, file, "applied");
|
|
3033
|
+
return true;
|
|
3034
|
+
}
|
|
3035
|
+
function durableDownloadCode(status2) {
|
|
3036
|
+
return status2 === 413 ? "file_too_large" : "write_failed";
|
|
3037
|
+
}
|
|
3038
|
+
async function downloadContent(options, file, label) {
|
|
3039
|
+
try {
|
|
3040
|
+
const res = await options.fetchImpl(
|
|
3041
|
+
`${options.apiUrl}/runners/${options.agentId}/files/${file.id}/content`,
|
|
3042
|
+
{ headers: { Authorization: options.getAuthHeader() } }
|
|
3043
|
+
);
|
|
3044
|
+
if (!res.ok) {
|
|
3045
|
+
const terminal = res.status >= 400 && res.status < 500 && res.status !== 401 && res.status !== 403 && res.status !== 408 && res.status !== 429;
|
|
3046
|
+
if (!terminal) {
|
|
3047
|
+
options.log({
|
|
3048
|
+
level: "warn",
|
|
3049
|
+
message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 retrying on the next drain`
|
|
3050
|
+
});
|
|
3051
|
+
return { ok: false, terminal: false };
|
|
1934
3052
|
}
|
|
1935
|
-
|
|
3053
|
+
const code = durableDownloadCode(res.status);
|
|
3054
|
+
options.log({
|
|
3055
|
+
level: "error",
|
|
3056
|
+
message: `Downloading runner file ${label} returned HTTP ${res.status} \u2014 rejecting it as ${code} (the bytes never reached the writer)`
|
|
3057
|
+
});
|
|
3058
|
+
return { ok: false, terminal: true, code };
|
|
1936
3059
|
}
|
|
3060
|
+
return { ok: true, content: Buffer.from(await res.arrayBuffer()) };
|
|
3061
|
+
} catch (err) {
|
|
3062
|
+
options.log({
|
|
3063
|
+
level: "warn",
|
|
3064
|
+
message: `Downloading runner file ${label} failed \u2014 retrying on the next drain: ${describe(err)}`
|
|
3065
|
+
});
|
|
3066
|
+
return { ok: false, terminal: false };
|
|
1937
3067
|
}
|
|
1938
|
-
|
|
1939
|
-
|
|
1940
|
-
|
|
1941
|
-
|
|
1942
|
-
const
|
|
1943
|
-
|
|
1944
|
-
|
|
1945
|
-
|
|
1946
|
-
|
|
1947
|
-
|
|
1948
|
-
|
|
1949
|
-
|
|
1950
|
-
|
|
1951
|
-
this.reconnecting = false;
|
|
1952
|
-
this.resolvedAgentId = agentId;
|
|
1953
|
-
events.onConnected(agentId, isReconnect);
|
|
1954
|
-
},
|
|
1955
|
-
onDisconnected: (code, reason) => {
|
|
1956
|
-
events.onDisconnected(code, reason);
|
|
1957
|
-
if (this.opts.isRunning() && code !== 1e3 && !this.reconnecting) {
|
|
1958
|
-
this.reconnectPromise = this.connectWithRetry(true).catch((err) => {
|
|
1959
|
-
events.onError?.(`Reconnection failed: ${err.message}`);
|
|
1960
|
-
});
|
|
1961
|
-
}
|
|
1962
|
-
},
|
|
1963
|
-
onError: (error2) => events.onError?.(error2),
|
|
1964
|
-
onResponse: () => events.onResponse?.(),
|
|
1965
|
-
onDrainPing: () => events.onDrainPing?.(),
|
|
1966
|
-
onInfo: (message) => events.onInfo?.(message)
|
|
1967
|
-
});
|
|
1968
|
-
return;
|
|
1969
|
-
} catch (error2) {
|
|
1970
|
-
this.reconnectAttempt++;
|
|
1971
|
-
if (error2.message === "Unauthorized") {
|
|
1972
|
-
this.reconnecting = false;
|
|
1973
|
-
throw error2;
|
|
1974
|
-
}
|
|
1975
|
-
const delay = getReconnectDelay(this.reconnectAttempt);
|
|
1976
|
-
events.onReconnecting?.(this.reconnectAttempt);
|
|
1977
|
-
events.onError?.(`Connection failed, retrying in ${Math.round(delay / 1e3)}s...`);
|
|
1978
|
-
await this.sleep(delay);
|
|
3068
|
+
}
|
|
3069
|
+
async function ack(options, file, status2, reason) {
|
|
3070
|
+
const outcome = `${status2}${reason ? ` (${reason})` : ""}`;
|
|
3071
|
+
try {
|
|
3072
|
+
const res = await options.fetchImpl(
|
|
3073
|
+
`${options.apiUrl}/runners/${options.agentId}/files/${file.id}/ack`,
|
|
3074
|
+
{
|
|
3075
|
+
method: "POST",
|
|
3076
|
+
headers: {
|
|
3077
|
+
Authorization: options.getAuthHeader(),
|
|
3078
|
+
"Content-Type": "application/json"
|
|
3079
|
+
},
|
|
3080
|
+
body: JSON.stringify(reason ? { status: status2, reason } : { status: status2 })
|
|
1979
3081
|
}
|
|
3082
|
+
);
|
|
3083
|
+
if (!res.ok) {
|
|
3084
|
+
recordAckFailure(
|
|
3085
|
+
options,
|
|
3086
|
+
file,
|
|
3087
|
+
`Acking runner file ${file.id.slice(0, 8)} as ${outcome} returned HTTP ${res.status}`
|
|
3088
|
+
);
|
|
3089
|
+
return;
|
|
1980
3090
|
}
|
|
1981
|
-
|
|
3091
|
+
options.ackFailures.delete(file.id);
|
|
3092
|
+
} catch (err) {
|
|
3093
|
+
recordAckFailure(
|
|
3094
|
+
options,
|
|
3095
|
+
file,
|
|
3096
|
+
`Acking runner file ${file.id.slice(0, 8)} as ${outcome} failed: ${describe(err)}`
|
|
3097
|
+
);
|
|
1982
3098
|
}
|
|
1983
|
-
}
|
|
3099
|
+
}
|
|
3100
|
+
function recordAckFailure(options, file, what) {
|
|
3101
|
+
const attempts = (options.ackFailures.get(file.id) ?? 0) + 1;
|
|
3102
|
+
options.ackFailures.set(file.id, attempts);
|
|
3103
|
+
options.log({
|
|
3104
|
+
level: "error",
|
|
3105
|
+
message: attempts >= MAX_ACK_ATTEMPTS ? `${what} \u2014 giving up after ${attempts} attempts. It stays pending until the server expires it; restart the runner to retry.` : `${what} \u2014 it stays pending until a later drain re-acks it (attempt ${attempts} of ${MAX_ACK_ATTEMPTS})`
|
|
3106
|
+
});
|
|
3107
|
+
}
|
|
3108
|
+
function describe(err) {
|
|
3109
|
+
return err instanceof Error ? err.message : String(err);
|
|
3110
|
+
}
|
|
1984
3111
|
|
|
1985
3112
|
// src/lib/channels/driver.ts
|
|
1986
3113
|
function messageIdOf(m) {
|
|
@@ -2010,7 +3137,11 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
|
|
|
2010
3137
|
var DEFAULT_STUCK_QUEUED_MS = 6e4;
|
|
2011
3138
|
var HEARTBEAT_MS = 6e4;
|
|
2012
3139
|
var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
|
|
3140
|
+
var B2_ABANDONMENT_MIN_PINNED_MS = 3 * 6e4;
|
|
3141
|
+
var B2_ABANDONMENT_RECHECK_MS = HEARTBEAT_MS;
|
|
2013
3142
|
var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
|
|
3143
|
+
var MAX_SUPERSEDED_CONVERSATIONS = 256;
|
|
3144
|
+
var MAX_IDENTICAL_REDRIVE_POLL_FAILURES = 5;
|
|
2014
3145
|
var ChannelAuthError = class extends Error {
|
|
2015
3146
|
constructor(message) {
|
|
2016
3147
|
super(message);
|
|
@@ -2019,10 +3150,10 @@ var ChannelAuthError = class extends Error {
|
|
|
2019
3150
|
};
|
|
2020
3151
|
var ChannelTerminalError = class extends Error {
|
|
2021
3152
|
status;
|
|
2022
|
-
constructor(message,
|
|
3153
|
+
constructor(message, status2) {
|
|
2023
3154
|
super(message);
|
|
2024
3155
|
this.name = "ChannelTerminalError";
|
|
2025
|
-
this.status =
|
|
3156
|
+
this.status = status2;
|
|
2026
3157
|
}
|
|
2027
3158
|
};
|
|
2028
3159
|
function backoffDelay(attempt, policy) {
|
|
@@ -2030,10 +3161,14 @@ function backoffDelay(attempt, policy) {
|
|
|
2030
3161
|
const capped = Math.min(policy.maxDelayMs, exp);
|
|
2031
3162
|
return Math.floor(Math.random() * capped);
|
|
2032
3163
|
}
|
|
2033
|
-
function isRetryableStatus(
|
|
2034
|
-
return
|
|
3164
|
+
function isRetryableStatus(status2) {
|
|
3165
|
+
return status2 === 429 || status2 >= 500 && status2 <= 599;
|
|
3166
|
+
}
|
|
3167
|
+
var VOLATILE_BODY_FIELD_PATTERN = /("(?:ref|requestId|request_id|traceId|trace_id)"\s*:\s*)"[^"]*"/gi;
|
|
3168
|
+
function normalizeRedrivePollFailureBody(body) {
|
|
3169
|
+
return body.replace(VOLATILE_BODY_FIELD_PATTERN, '$1"<redacted>"').replace(/\s+/g, " ").trim().slice(0, 200);
|
|
2035
3170
|
}
|
|
2036
|
-
var ChannelDriver = class {
|
|
3171
|
+
var ChannelDriver = class _ChannelDriver {
|
|
2037
3172
|
agentId;
|
|
2038
3173
|
port;
|
|
2039
3174
|
apiUrl;
|
|
@@ -2047,8 +3182,38 @@ var ChannelDriver = class {
|
|
|
2047
3182
|
pausedMaxWaitMs;
|
|
2048
3183
|
stuckQueuedMs;
|
|
2049
3184
|
now;
|
|
3185
|
+
fileSyncDirectories;
|
|
3186
|
+
homeDir;
|
|
2050
3187
|
/** Cache of conversationId → opencode sessionId. */
|
|
2051
3188
|
sessions = /* @__PURE__ */ new Map();
|
|
3189
|
+
/**
|
|
3190
|
+
* conversationId → the opencode session this runner has ABANDONED as that
|
|
3191
|
+
* conversation's binding (#553), after a genuine (`sessionExists === true`)
|
|
3192
|
+
* dispatch failure: the session still exists but is wedged, so #485's self-heal
|
|
3193
|
+
* must bind a fresh one.
|
|
3194
|
+
*
|
|
3195
|
+
* Dropping the local binding + clearing the server row is not enough on its own:
|
|
3196
|
+
* a SIBLING message dispatched earlier in the same drain is still in-flight under
|
|
3197
|
+
* the same session, and its watcher's routine status writes carry
|
|
3198
|
+
* `opencode_session_id`, RESURRECTING the wedged id server-side after the clear —
|
|
3199
|
+
* and `ensureSession`'s persisted-id fallback then reuses it, defeating the
|
|
3200
|
+
* self-heal. This map makes the runner authoritative instead of racing those
|
|
3201
|
+
* writes: *`ensureSession` never reuses an abandoned id for that conversation,
|
|
3202
|
+
* whatever the server row says* — which holds even when the resurrecting write
|
|
3203
|
+
* is one we deliberately keep (see `markDone`).
|
|
3204
|
+
*
|
|
3205
|
+
* Bounded by construction, on both axes: keyed by CONVERSATION, so N failures on
|
|
3206
|
+
* one conversation hold ONE entry (the newest abandonment replaces the older), and
|
|
3207
|
+
* hard-capped at `MAX_SUPERSEDED_CONVERSATIONS` with FIFO eviction. Only the
|
|
3208
|
+
* NEWEST abandoned id per conversation is guarded: after a second abandonment a
|
|
3209
|
+
* late sibling of the FIRST session can write that id back and `ensureSession`
|
|
3210
|
+
* will reuse it — costing ONE repeat failure, which re-supersedes it. Deliberately
|
|
3211
|
+
* NOT dropped when the session's watcher tears down: `markDone` still writes the
|
|
3212
|
+
* abandoned id back (it must, or the reply is lost), so the guard has to outlive
|
|
3213
|
+
* the turn that resurrects it. In-memory only — a restart forgets it, at the same
|
|
3214
|
+
* bounded cost.
|
|
3215
|
+
*/
|
|
3216
|
+
supersededSessions = /* @__PURE__ */ new Map();
|
|
2052
3217
|
/**
|
|
2053
3218
|
* Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
|
|
2054
3219
|
* longer idempotent (no caller-supplied `messageID`), and its read-back picks
|
|
@@ -2116,6 +3281,58 @@ var ChannelDriver = class {
|
|
|
2116
3281
|
* processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
|
|
2117
3282
|
*/
|
|
2118
3283
|
readoptPollUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
3284
|
+
/**
|
|
3285
|
+
* "Already emitted `redrive_unresolved` for this row" (#965). Mirrors
|
|
3286
|
+
* `readoptPollUnresolvedSignalled`: `resolveRedrive`'s `unresolved` leaf recurs
|
|
3287
|
+
* every ~2s drain until opencode's status becomes readable, but the
|
|
3288
|
+
* server-visible signal is an OUTCOME, so it fires at most once per row. Cleared
|
|
3289
|
+
* on any non-`unresolved` outcome so the set cannot grow beyond the currently
|
|
3290
|
+
* unresolvable rows.
|
|
3291
|
+
*/
|
|
3292
|
+
redriveUnresolvedSignalled = /* @__PURE__ */ new Set();
|
|
3293
|
+
/**
|
|
3294
|
+
* First `now()` a `pending` row's re-drive was observed `unresolved` (#965). A
|
|
3295
|
+
* `pending` row is invisible to every cron arm (all require `status =
|
|
3296
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
3297
|
+
* nothing driving it. Once `now - since >= pausedMaxWaitMs`, `resolveRedrive`
|
|
3298
|
+
* takes `dispatch` instead of `unresolved` (reusing the existing knob — see
|
|
3299
|
+
* ADR-0047's own "unreachable ⇒ bounded" rule). Cleared on any other outcome.
|
|
3300
|
+
*/
|
|
3301
|
+
redriveUnresolvedSince = /* @__PURE__ */ new Map();
|
|
3302
|
+
/**
|
|
3303
|
+
* Consecutive-identical-poll-failure streak for the re-drive fence (#1348),
|
|
3304
|
+
* keyed by Evident **message id** (not session) so `clearRedriveUnresolved`
|
|
3305
|
+
* can drop it with the other two trackers and it cannot leak. `sessionId` is
|
|
3306
|
+
* carried inside the entry, not the key: a session change is a different
|
|
3307
|
+
* situation and resets the streak, which gives the `(sessionId, message.id)`
|
|
3308
|
+
* pairing #1348 asks for without a composite map key.
|
|
3309
|
+
*/
|
|
3310
|
+
redrivePollFailures = /* @__PURE__ */ new Map();
|
|
3311
|
+
/**
|
|
3312
|
+
* "Already emitted `redrive_outcome_unreported` for THIS (message, outcome)
|
|
3313
|
+
* streak" (Class B, #1340: the runner DECIDED reattach/settle/fail_permanent
|
|
3314
|
+
* but its own PATCH to record it failed — distinct from Class A's
|
|
3315
|
+
* `redrive_poll_failed`, where opencode itself can't be observed). Keyed by
|
|
3316
|
+
* message id, valued by the outcome currently failing to report, so a
|
|
3317
|
+
* change of outcome starts a fresh signal. Cleared by
|
|
3318
|
+
* `clearRedriveUnresolved` the instant either PATCH succeeds.
|
|
3319
|
+
*/
|
|
3320
|
+
redriveOutcomeUnreportedSignalled = /* @__PURE__ */ new Map();
|
|
3321
|
+
/**
|
|
3322
|
+
* Consecutive-UNCONFIRMED-dispatch streak for a `pending` row with NO stored
|
|
3323
|
+
* `opencode_message_id` yet — i.e. one that has never even reached the
|
|
3324
|
+
* re-drive fence above. `sendPromptAsync`'s POST may 2xx, but its own
|
|
3325
|
+
* read-back retries can never confirm the assigned id when the session's
|
|
3326
|
+
* message list is PERMANENTLY unreadable (e.g. a corrupted local opencode
|
|
3327
|
+
* SQLite DB, #1345/#1348's exact fault, just hit BEFORE the row is ever
|
|
3328
|
+
* dispatched instead of after). Unlike an already-dispatched row, THIS row has
|
|
3329
|
+
* no other safety net at all: the lifecycle cron only reclaims `status =
|
|
3330
|
+
* 'processing'` rows, and a row stuck here never reaches `processing`. Keyed
|
|
3331
|
+
* by message id, carrying `sessionId` so a session change (a fresh one bound
|
|
3332
|
+
* after abandonment) starts a new streak rather than inheriting the old
|
|
3333
|
+
* session's count — same shape as `redrivePollFailures` above.
|
|
3334
|
+
*/
|
|
3335
|
+
unconfirmedDispatchFailures = /* @__PURE__ */ new Map();
|
|
2119
3336
|
/**
|
|
2120
3337
|
* "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
|
|
2121
3338
|
* Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
|
|
@@ -2158,9 +3375,12 @@ var ChannelDriver = class {
|
|
|
2158
3375
|
sessionParents = /* @__PURE__ */ new Map();
|
|
2159
3376
|
/**
|
|
2160
3377
|
* Per-session OpenCode title cache (#310), keyed by sessionId. Only a resolved
|
|
2161
|
-
* NON-EMPTY name is stored (terminal — a real session name
|
|
2162
|
-
* so we do NOT re-GET `/session/:id` every tick.
|
|
2163
|
-
*
|
|
3378
|
+
* NON-EMPTY, non-placeholder name is stored (terminal — a real session name
|
|
3379
|
+
* won't later un-name), so we do NOT re-GET `/session/:id` every tick. "Non-empty"
|
|
3380
|
+
* excludes OpenCode's synchronous default title (see
|
|
3381
|
+
* `OPENCODE_DEFAULT_TITLE_PREFIX`, #549) — that placeholder is treated the same
|
|
3382
|
+
* as an empty title so it never latches. A missing entry = not yet resolved OR
|
|
3383
|
+
* resolved-but-still-empty/placeholder → re-fetch on next need, since OpenCode
|
|
2164
3384
|
* names sessions asynchronously mid-turn. Driver-level (not per-watcher) so both
|
|
2165
3385
|
* the watcher completion path AND the restart-recovery re-adopt path (which has
|
|
2166
3386
|
* no watcher) can resolve the title.
|
|
@@ -2168,6 +3388,24 @@ var ChannelDriver = class {
|
|
|
2168
3388
|
sessionTitles = /* @__PURE__ */ new Map();
|
|
2169
3389
|
/** Serialises drains so a reconnect during a drain doesn't double-process. */
|
|
2170
3390
|
draining = false;
|
|
3391
|
+
/**
|
|
3392
|
+
* Serialises runner-file syncs (#559) so the ~2s poll tick and a concurrent
|
|
3393
|
+
* drain ping don't download, write and ack the same file twice.
|
|
3394
|
+
*/
|
|
3395
|
+
syncingFiles = false;
|
|
3396
|
+
/**
|
|
3397
|
+
* Consecutive failed acks per pending file (#559). Lives on the driver so it
|
|
3398
|
+
* survives across drains — without it, a file whose ack keeps failing is
|
|
3399
|
+
* re-downloaded and re-written every ~2s until the server expires it.
|
|
3400
|
+
*/
|
|
3401
|
+
fileAckFailures = /* @__PURE__ */ new Map();
|
|
3402
|
+
/**
|
|
3403
|
+
* Monotonic count of files this runner has pulled and written (#559). Only
|
|
3404
|
+
* ever increases, so `run.ts` detects work by comparing it against the value
|
|
3405
|
+
* it saw on the previous cycle — including work that landed mid-sleep, the
|
|
3406
|
+
* same trick `lastProxiedActivityAt` uses.
|
|
3407
|
+
*/
|
|
3408
|
+
appliedFileCount = 0;
|
|
2171
3409
|
/**
|
|
2172
3410
|
* The currently-executing `drainPending()` promise, or null when idle. Lets a
|
|
2173
3411
|
* graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
|
|
@@ -2183,21 +3421,23 @@ var ChannelDriver = class {
|
|
|
2183
3421
|
* and stops opencode.
|
|
2184
3422
|
*/
|
|
2185
3423
|
stopped = false;
|
|
2186
|
-
constructor(
|
|
2187
|
-
this.agentId =
|
|
2188
|
-
this.port =
|
|
2189
|
-
this.apiUrl =
|
|
2190
|
-
this.getAuthHeader =
|
|
2191
|
-
this.conversationFilter =
|
|
2192
|
-
this.retry = { ...DEFAULT_RETRY_POLICY, ...
|
|
2193
|
-
this.log =
|
|
3424
|
+
constructor(config) {
|
|
3425
|
+
this.agentId = config.agentId;
|
|
3426
|
+
this.port = config.port;
|
|
3427
|
+
this.apiUrl = config.apiUrl.replace(/\/$/, "");
|
|
3428
|
+
this.getAuthHeader = config.getAuthHeader;
|
|
3429
|
+
this.conversationFilter = config.conversationFilter ?? null;
|
|
3430
|
+
this.retry = { ...DEFAULT_RETRY_POLICY, ...config.retry };
|
|
3431
|
+
this.log = config.log ?? (() => {
|
|
2194
3432
|
});
|
|
2195
|
-
this.fetchImpl =
|
|
2196
|
-
this.sleep =
|
|
2197
|
-
this.pausedPollIntervalMs =
|
|
2198
|
-
this.pausedMaxWaitMs =
|
|
2199
|
-
this.stuckQueuedMs =
|
|
2200
|
-
this.now =
|
|
3433
|
+
this.fetchImpl = config.fetchImpl ?? fetch;
|
|
3434
|
+
this.sleep = config.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
|
|
3435
|
+
this.pausedPollIntervalMs = config.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
|
|
3436
|
+
this.pausedMaxWaitMs = config.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
|
|
3437
|
+
this.stuckQueuedMs = config.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
|
|
3438
|
+
this.now = config.now ?? (() => Date.now());
|
|
3439
|
+
this.fileSyncDirectories = config.fileSyncDirectories ?? [];
|
|
3440
|
+
this.homeDir = config.homeDir ?? homedir2();
|
|
2201
3441
|
}
|
|
2202
3442
|
/** The IPv4-loopback base URL for the local `opencode serve`. */
|
|
2203
3443
|
get opencodeBase() {
|
|
@@ -2225,6 +3465,47 @@ var ChannelDriver = class {
|
|
|
2225
3465
|
);
|
|
2226
3466
|
return run2;
|
|
2227
3467
|
}
|
|
3468
|
+
/**
|
|
3469
|
+
* Pull-and-apply any files Evident has queued for this runner (#559), riding
|
|
3470
|
+
* the EXISTING drain cycle — `run.ts` calls it from the same ~2s channel poll
|
|
3471
|
+
* and drain ping that call `drainPending()`. There is deliberately no channel,
|
|
3472
|
+
* control frame or poll loop of its own: worst-case latency is one poll tick.
|
|
3473
|
+
*
|
|
3474
|
+
* NEVER throws and never surfaces a `ChannelAuthError`: a file failure must not
|
|
3475
|
+
* cost a conversation turn. Failures are logged and either acked as a terminal
|
|
3476
|
+
* outcome or left pending for the next drain (see `runner-file-sync.ts`).
|
|
3477
|
+
*
|
|
3478
|
+
* Re-entrant calls are skipped (the poll tick and a drain ping can overlap).
|
|
3479
|
+
*
|
|
3480
|
+
* @returns the number of files written to disk.
|
|
3481
|
+
*/
|
|
3482
|
+
async syncPendingFiles() {
|
|
3483
|
+
if (this.stopped) return 0;
|
|
3484
|
+
if (this.syncingFiles) return 0;
|
|
3485
|
+
this.syncingFiles = true;
|
|
3486
|
+
try {
|
|
3487
|
+
const applied = await syncPendingRunnerFiles({
|
|
3488
|
+
agentId: this.agentId,
|
|
3489
|
+
apiUrl: this.apiUrl,
|
|
3490
|
+
getAuthHeader: this.getAuthHeader,
|
|
3491
|
+
fetchImpl: this.fetchImpl,
|
|
3492
|
+
allowedDirectories: this.fileSyncDirectories,
|
|
3493
|
+
homeDir: this.homeDir,
|
|
3494
|
+
ackFailures: this.fileAckFailures,
|
|
3495
|
+
log: this.log
|
|
3496
|
+
});
|
|
3497
|
+
this.appliedFileCount += applied;
|
|
3498
|
+
return applied;
|
|
3499
|
+
} catch (err) {
|
|
3500
|
+
this.log({
|
|
3501
|
+
level: "error",
|
|
3502
|
+
message: `Runner file sync failed unexpectedly (message processing is unaffected): ${err instanceof Error ? err.message : String(err)}`
|
|
3503
|
+
});
|
|
3504
|
+
return 0;
|
|
3505
|
+
} finally {
|
|
3506
|
+
this.syncingFiles = false;
|
|
3507
|
+
}
|
|
3508
|
+
}
|
|
2228
3509
|
async runDrain() {
|
|
2229
3510
|
let dispatched = 0;
|
|
2230
3511
|
try {
|
|
@@ -2258,6 +3539,28 @@ var ChannelDriver = class {
|
|
|
2258
3539
|
}
|
|
2259
3540
|
return false;
|
|
2260
3541
|
}
|
|
3542
|
+
/**
|
|
3543
|
+
* File-pull work, for `run.ts`'s idle accounting (#559).
|
|
3544
|
+
*
|
|
3545
|
+
* Pulling a file is real work that `drainPending()` knows nothing about, so
|
|
3546
|
+
* without this a near-idle runner counts a credential pull as an empty tick
|
|
3547
|
+
* and `--idle-timeout` can `process.exit` mid-pull — leaving a
|
|
3548
|
+
* `.evident-push-*.tmp` behind — or immediately after the write, before the
|
|
3549
|
+
* browser has run the authorize/callback that activates it (the user then sees
|
|
3550
|
+
* `saved_not_activated` for a runner that was fine).
|
|
3551
|
+
*
|
|
3552
|
+
* Two signals because one cannot cover both cases: `inFlight` is the pull
|
|
3553
|
+
* happening RIGHT NOW (it may outlive the tick that started it), and
|
|
3554
|
+
* `appliedFiles` is monotonic so a pull that started AND finished between two
|
|
3555
|
+
* idle checks still shows up as an advance.
|
|
3556
|
+
*
|
|
3557
|
+
* CALLER CONTRACT: sample `inFlight` BEFORE calling `syncPendingFiles()` for
|
|
3558
|
+
* the cycle. `syncPendingFiles` sets the flag synchronously, so a caller that
|
|
3559
|
+
* samples afterwards reads `true` every single cycle and can never idle out.
|
|
3560
|
+
*/
|
|
3561
|
+
fileSyncActivity() {
|
|
3562
|
+
return { appliedFiles: this.appliedFileCount, inFlight: this.syncingFiles };
|
|
3563
|
+
}
|
|
2261
3564
|
/**
|
|
2262
3565
|
* OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
|
|
2263
3566
|
* exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
|
|
@@ -2319,7 +3622,7 @@ var ChannelDriver = class {
|
|
|
2319
3622
|
await this.sleep(step);
|
|
2320
3623
|
}
|
|
2321
3624
|
}
|
|
2322
|
-
while (this.hasInFlightWatchers()) {
|
|
3625
|
+
while (this.hasInFlightWatchers() || this.syncingFiles) {
|
|
2323
3626
|
if (this.now() >= deadline) return false;
|
|
2324
3627
|
await this.sleep(step);
|
|
2325
3628
|
}
|
|
@@ -2354,16 +3657,27 @@ var ChannelDriver = class {
|
|
|
2354
3657
|
* @returns the count of messages NEWLY dispatched (not already in-flight).
|
|
2355
3658
|
*/
|
|
2356
3659
|
async processConversation(conv) {
|
|
2357
|
-
const sessionId = await this.ensureSession(conv);
|
|
3660
|
+
const { sessionId, refusedSessionId } = await this.ensureSession(conv);
|
|
2358
3661
|
const messages = await this.getPendingMessages(conv.id);
|
|
2359
3662
|
let dispatched = 0;
|
|
2360
3663
|
let skippedAlreadyDispatched = 0;
|
|
3664
|
+
if (refusedSessionId && messages.length > 0) {
|
|
3665
|
+
void this.postSignal(conv.id, messages[0].id, "session_superseded", {
|
|
3666
|
+
superseded_session_id: refusedSessionId
|
|
3667
|
+
});
|
|
3668
|
+
}
|
|
2361
3669
|
for (const message of messages) {
|
|
2362
3670
|
if (this.stopped) break;
|
|
2363
3671
|
if (this.dispatched.has(message.id)) {
|
|
2364
3672
|
skippedAlreadyDispatched += 1;
|
|
2365
3673
|
continue;
|
|
2366
3674
|
}
|
|
3675
|
+
if (message.opencode_message_id) {
|
|
3676
|
+
const outcome = await this.resolveRedrive(conv, sessionId, message, refusedSessionId);
|
|
3677
|
+
if (outcome !== "dispatch") {
|
|
3678
|
+
break;
|
|
3679
|
+
}
|
|
3680
|
+
}
|
|
2367
3681
|
const options = {
|
|
2368
3682
|
agent: message.opencode_agent ?? void 0,
|
|
2369
3683
|
model: message.opencode_model ?? void 0
|
|
@@ -2384,7 +3698,8 @@ var ChannelDriver = class {
|
|
|
2384
3698
|
} catch (err) {
|
|
2385
3699
|
if (err instanceof ChannelAuthError) throw err;
|
|
2386
3700
|
this.dispatched.delete(message.id);
|
|
2387
|
-
|
|
3701
|
+
const exists = await sessionExists(this.port, sessionId);
|
|
3702
|
+
if (exists === false) {
|
|
2388
3703
|
this.sessions.delete(conv.id);
|
|
2389
3704
|
this.log({
|
|
2390
3705
|
level: "warn",
|
|
@@ -2394,42 +3709,472 @@ var ChannelDriver = class {
|
|
|
2394
3709
|
});
|
|
2395
3710
|
break;
|
|
2396
3711
|
}
|
|
2397
|
-
|
|
3712
|
+
if (exists === null) {
|
|
3713
|
+
this.log({
|
|
3714
|
+
level: "warn",
|
|
3715
|
+
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.`,
|
|
3716
|
+
conversation_id: conv.id,
|
|
3717
|
+
message_id: message.id
|
|
3718
|
+
});
|
|
3719
|
+
break;
|
|
3720
|
+
}
|
|
3721
|
+
const errorMessage = err instanceof Error ? err.message : String(err);
|
|
3722
|
+
this.sessions.delete(conv.id);
|
|
3723
|
+
this.supersede(conv.id, sessionId);
|
|
3724
|
+
this.log({
|
|
3725
|
+
level: "warn",
|
|
3726
|
+
message: `Abandoning OpenCode session ${sessionId.slice(0, 8)} as the binding for conversation ${conv.id.slice(0, 8)} (it exists but failed to run a turn) \u2014 a fresh session is created on the next tick, whatever the persisted binding says by then.`,
|
|
3727
|
+
conversation_id: conv.id,
|
|
3728
|
+
message_id: message.id
|
|
3729
|
+
});
|
|
3730
|
+
await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
|
|
3731
|
+
this.log({
|
|
3732
|
+
level: "warn",
|
|
3733
|
+
message: `markFailed PATCH for message ${message.id.slice(0, 8)} (conversation ${conv.id.slice(0, 8)}) failed (best-effort, not retried): ${markErr instanceof Error ? markErr.message : String(markErr)}`,
|
|
3734
|
+
conversation_id: conv.id,
|
|
3735
|
+
message_id: message.id
|
|
3736
|
+
});
|
|
2398
3737
|
});
|
|
2399
3738
|
this.log({
|
|
2400
3739
|
level: "error",
|
|
2401
|
-
message: `Message ${message.id.slice(0, 8)} dispatch failed: ${
|
|
3740
|
+
message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
|
|
2402
3741
|
conversation_id: conv.id,
|
|
2403
3742
|
message_id: message.id
|
|
2404
3743
|
});
|
|
2405
|
-
|
|
3744
|
+
break;
|
|
2406
3745
|
}
|
|
2407
3746
|
if (opencodeMessageId === null) {
|
|
3747
|
+
const streak = this.recordUnconfirmedDispatch(message.id, sessionId);
|
|
3748
|
+
if (streak < MAX_IDENTICAL_REDRIVE_POLL_FAILURES) {
|
|
3749
|
+
this.log({
|
|
3750
|
+
level: "warn",
|
|
3751
|
+
message: `Message ${message.id.slice(0, 8)} dispatched but its opencode id could not be read back (${streak}/${MAX_IDENTICAL_REDRIVE_POLL_FAILURES}) \u2014 leaving un-tracked to retry next tick`,
|
|
3752
|
+
conversation_id: conv.id,
|
|
3753
|
+
message_id: message.id
|
|
3754
|
+
});
|
|
3755
|
+
continue;
|
|
3756
|
+
}
|
|
3757
|
+
this.unconfirmedDispatchFailures.delete(message.id);
|
|
3758
|
+
this.sessions.delete(conv.id);
|
|
3759
|
+
this.supersede(conv.id, sessionId);
|
|
3760
|
+
const errorMessage = `OpenCode accepted this message ${streak} times in a row but never confirmed its assigned id (session ${sessionId.slice(0, 8)}'s message list could not be read back) \u2014 the session was abandoned; a fresh one is used for further messages.`;
|
|
2408
3761
|
this.log({
|
|
2409
|
-
level: "
|
|
2410
|
-
message:
|
|
3762
|
+
level: "error",
|
|
3763
|
+
message: errorMessage,
|
|
2411
3764
|
conversation_id: conv.id,
|
|
2412
3765
|
message_id: message.id
|
|
2413
3766
|
});
|
|
2414
|
-
|
|
3767
|
+
await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
|
|
3768
|
+
this.log({
|
|
3769
|
+
level: "warn",
|
|
3770
|
+
message: `markFailed PATCH for message ${message.id.slice(0, 8)} (conversation ${conv.id.slice(0, 8)}) failed (best-effort, not retried): ${markErr instanceof Error ? markErr.message : String(markErr)}`,
|
|
3771
|
+
conversation_id: conv.id,
|
|
3772
|
+
message_id: message.id
|
|
3773
|
+
});
|
|
3774
|
+
});
|
|
3775
|
+
break;
|
|
2415
3776
|
}
|
|
3777
|
+
this.unconfirmedDispatchFailures.delete(message.id);
|
|
2416
3778
|
this.dispatched.add(message.id);
|
|
2417
3779
|
this.registerInFlight(conv, sessionId, message, opencodeMessageId);
|
|
2418
3780
|
dispatched += 1;
|
|
2419
3781
|
void this.postSignal(conv.id, message.id, "dispatched");
|
|
2420
3782
|
}
|
|
2421
|
-
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3783
|
+
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3784
|
+
this.log({
|
|
3785
|
+
level: "warn",
|
|
3786
|
+
message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
|
|
3787
|
+
conversation_id: conv.id
|
|
3788
|
+
});
|
|
3789
|
+
}
|
|
3790
|
+
this.ensureWatcherRunning(sessionId);
|
|
3791
|
+
return dispatched;
|
|
3792
|
+
}
|
|
3793
|
+
/**
|
|
3794
|
+
* Poll a session's message list for the re-drive fence (#965), via the
|
|
3795
|
+
* INJECTED `fetchImpl` — NOT the imported `getSessionMessages` helper, which
|
|
3796
|
+
* hits the global `fetch` and would bypass the same override every other
|
|
3797
|
+
* opencode poll in this file respects. Mirrors `readoptProcessing`'s own
|
|
3798
|
+
* snapshot fetch (`:3081-3111`).
|
|
3799
|
+
*
|
|
3800
|
+
* Returns `{ ok: true, messages }` on a readable snapshot, or
|
|
3801
|
+
* `{ ok: false, signature }` on failure — `signature` is a string that
|
|
3802
|
+
* repeats across attempts for the SAME underlying fault (used by the
|
|
3803
|
+
* consecutive-identical-failure bound, #1348), or `null` for a thrown
|
|
3804
|
+
* exception, which is NOT countable toward that bound (a network blip / an
|
|
3805
|
+
* opencode restart also throws identically every tick, and must keep
|
|
3806
|
+
* retrying unbounded rather than ever being treated as permanent).
|
|
3807
|
+
*/
|
|
3808
|
+
async pollSessionMessagesForRedrive(conv, message, sessionId) {
|
|
3809
|
+
try {
|
|
3810
|
+
const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
|
|
3811
|
+
if (!res.ok) {
|
|
3812
|
+
const rawBody = await res.text();
|
|
3813
|
+
const normalized = normalizeRedrivePollFailureBody(rawBody);
|
|
3814
|
+
this.log({
|
|
3815
|
+
level: "warn",
|
|
3816
|
+
message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned HTTP ${res.status}${normalized ? `: ${normalized}` : ""} \u2014 treating as unreadable this tick`,
|
|
3817
|
+
conversation_id: conv.id,
|
|
3818
|
+
message_id: message.id
|
|
3819
|
+
});
|
|
3820
|
+
return { ok: false, signature: `HTTP ${res.status}${normalized ? `: ${normalized}` : ""}` };
|
|
3821
|
+
}
|
|
3822
|
+
const body = await res.json();
|
|
3823
|
+
if (!Array.isArray(body)) {
|
|
3824
|
+
this.log({
|
|
3825
|
+
level: "warn",
|
|
3826
|
+
message: `Re-drive: polling session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} returned a non-array message body \u2014 treating as unreadable this tick`,
|
|
3827
|
+
conversation_id: conv.id,
|
|
3828
|
+
message_id: message.id
|
|
3829
|
+
});
|
|
3830
|
+
return { ok: false, signature: "non-array message body" };
|
|
3831
|
+
}
|
|
3832
|
+
return { ok: true, messages: body };
|
|
3833
|
+
} catch (err) {
|
|
3834
|
+
this.log({
|
|
3835
|
+
level: "warn",
|
|
3836
|
+
message: `Re-drive: failed to poll session ${sessionId.slice(0, 8)} for message ${message.id.slice(0, 8)} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3837
|
+
conversation_id: conv.id,
|
|
3838
|
+
message_id: message.id
|
|
3839
|
+
});
|
|
3840
|
+
return { ok: false, signature: null };
|
|
3841
|
+
}
|
|
3842
|
+
}
|
|
3843
|
+
/**
|
|
3844
|
+
* The re-drive fence for a `pending` row that already carries a stored
|
|
3845
|
+
* `opencode_message_id` (#965) — i.e. it has already been handed to opencode at
|
|
3846
|
+
* least once (see the invariant at `QueuedMessage.opencode_message_id`'s doc).
|
|
3847
|
+
* The lifecycle cron can falsely reclaim a `processing` row back to `pending`
|
|
3848
|
+
* mid-turn (a 5-minute liveness-staleness check racing a still-running turn);
|
|
3849
|
+
* without this fence the drain loop would re-`prompt_async` the SAME turn a
|
|
3850
|
+
* second time against live GitHub state. Mirrors `readoptOne`'s job for the
|
|
3851
|
+
* `processing` re-adopt path, but simpler: no b1/b2 preamble cross-check is
|
|
3852
|
+
* needed here because `refusedSessionId` already handles the one case
|
|
3853
|
+
* (#553 abandoned session) that path exists for.
|
|
3854
|
+
*
|
|
3855
|
+
* Only `ChannelAuthError` propagates. A poll that fails identically
|
|
3856
|
+
* `MAX_IDENTICAL_REDRIVE_POLL_FAILURES` times in a row reports the message
|
|
3857
|
+
* failed instead of retrying it (#1348) — SEPARATE from, not a replacement
|
|
3858
|
+
* for, `resolveRedriveUnresolved`'s own `pausedMaxWaitMs` bound below. Every
|
|
3859
|
+
* other failure resolves to `unresolved` and is retried whole on the next
|
|
3860
|
+
* ~2s drain tick.
|
|
3861
|
+
*/
|
|
3862
|
+
async resolveRedrive(conv, sessionId, message, refusedSessionId) {
|
|
3863
|
+
const ocId = message.opencode_message_id ?? null;
|
|
3864
|
+
if (refusedSessionId) {
|
|
3865
|
+
this.clearRedriveUnresolved(message.id);
|
|
3866
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3867
|
+
return "dispatch";
|
|
3868
|
+
}
|
|
3869
|
+
const polled = await this.pollSessionMessagesForRedrive(conv, message, sessionId);
|
|
3870
|
+
if (!polled.ok) {
|
|
3871
|
+
const streak = this.recordRedrivePollFailure(message.id, sessionId, polled.signature);
|
|
3872
|
+
if (streak >= MAX_IDENTICAL_REDRIVE_POLL_FAILURES && polled.signature !== null) {
|
|
3873
|
+
return this.failRedrivePollPermanent(conv, sessionId, message, polled.signature, streak);
|
|
3874
|
+
}
|
|
3875
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3876
|
+
}
|
|
3877
|
+
this.redrivePollFailures.delete(message.id);
|
|
3878
|
+
const messages = polled.messages;
|
|
3879
|
+
if (messages.length === 0) {
|
|
3880
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3881
|
+
}
|
|
3882
|
+
const state = messageRunState(messages, ocId ?? "");
|
|
3883
|
+
if (state === "failed" && isAbortedTerminalReply(messages, ocId ?? "")) {
|
|
3884
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3885
|
+
if (ongoing === false) {
|
|
3886
|
+
this.log({
|
|
3887
|
+
level: "info",
|
|
3888
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} carries an abort-shaped terminal error and session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status \u2014 restart orphan, re-dispatching instead of marking it permanently failed`,
|
|
3889
|
+
conversation_id: conv.id,
|
|
3890
|
+
message_id: message.id
|
|
3891
|
+
});
|
|
3892
|
+
this.clearRedriveUnresolved(message.id);
|
|
3893
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3894
|
+
return "dispatch";
|
|
3895
|
+
}
|
|
3896
|
+
}
|
|
3897
|
+
if (state === "done" || state === "failed") {
|
|
3898
|
+
return this.settleRedrive(conv, sessionId, message, ocId, messages, state);
|
|
3899
|
+
}
|
|
3900
|
+
if (state === "running" || state === "queued") {
|
|
3901
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3902
|
+
if (ongoing === true) {
|
|
3903
|
+
return this.reattachRedrive(conv, sessionId, message, ocId);
|
|
3904
|
+
}
|
|
3905
|
+
if (ongoing === false) {
|
|
3906
|
+
this.clearRedriveUnresolved(message.id);
|
|
3907
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3908
|
+
return "dispatch";
|
|
3909
|
+
}
|
|
3910
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3911
|
+
}
|
|
3912
|
+
this.clearRedriveUnresolved(message.id);
|
|
3913
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3914
|
+
return "dispatch";
|
|
3915
|
+
}
|
|
3916
|
+
/**
|
|
3917
|
+
* The `reattached` outcome (Task 3.3): the prior turn is STILL ONGOING per
|
|
3918
|
+
* opencode's own status map — undo the false reclaim instead of starting a
|
|
3919
|
+
* second turn.
|
|
3920
|
+
*/
|
|
3921
|
+
async reattachRedrive(conv, sessionId, message, ocId) {
|
|
3922
|
+
let anchorMs;
|
|
3923
|
+
const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
|
|
3924
|
+
if (!Number.isNaN(parsed)) {
|
|
3925
|
+
anchorMs = parsed;
|
|
3926
|
+
} else {
|
|
3927
|
+
anchorMs = this.now();
|
|
3928
|
+
this.log({
|
|
3929
|
+
level: "error",
|
|
3930
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} has null/unparseable processing_started_at (${String(message.processing_started_at)}) \u2014 anchoring the watcher's absolute-age ceiling to now (defensive)`,
|
|
3931
|
+
conversation_id: conv.id,
|
|
3932
|
+
message_id: message.id
|
|
3933
|
+
});
|
|
3934
|
+
}
|
|
3935
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3936
|
+
try {
|
|
3937
|
+
await this.markProcessing(conv.id, message.id, sessionId, ocId, title);
|
|
3938
|
+
} catch (err) {
|
|
3939
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3940
|
+
this.log({
|
|
3941
|
+
level: "warn",
|
|
3942
|
+
message: `Re-drive: failed to restore message ${message.id.slice(0, 8)} to processing (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3943
|
+
conversation_id: conv.id,
|
|
3944
|
+
message_id: message.id
|
|
3945
|
+
});
|
|
3946
|
+
this.signalRedriveOutcomeUnreported(conv, message, "reattach");
|
|
3947
|
+
return "unresolved";
|
|
3948
|
+
}
|
|
3949
|
+
this.clearRedriveUnresolved(message.id);
|
|
3950
|
+
this.registerReadopted(conv, sessionId, message, ocId ?? "", anchorMs);
|
|
3951
|
+
this.dispatched.add(message.id);
|
|
3952
|
+
this.readopted.add(message.id);
|
|
3953
|
+
this.ensureWatcherRunning(sessionId);
|
|
3954
|
+
const watchedForMs = this.now() - anchorMs;
|
|
3955
|
+
void this.postSignal(conv.id, message.id, "redrive_reattached", {
|
|
3956
|
+
watched_for_ms: watchedForMs
|
|
3957
|
+
});
|
|
3958
|
+
this.log({
|
|
3959
|
+
level: "warn",
|
|
3960
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) was wrongly reclaimed to pending while its turn was still running (watched ${watchedForMs}ms) \u2014 restored to processing instead of re-dispatching`,
|
|
3961
|
+
conversation_id: conv.id,
|
|
3962
|
+
message_id: message.id
|
|
3963
|
+
});
|
|
3964
|
+
return "reattached";
|
|
3965
|
+
}
|
|
3966
|
+
/**
|
|
3967
|
+
* The `settled` outcome (Task 3.2): the prior turn already finished (or
|
|
3968
|
+
* errored) while nobody was watching — deliver/report it instead of re-running.
|
|
3969
|
+
* Mirrors `readoptOne`'s `done`/`failed` branches' error discipline, simplified
|
|
3970
|
+
* (no `doneUndeliverable` park: a terminal PATCH failure here just retries next
|
|
3971
|
+
* drain, same as any other non-auth failure). The restart-abort carve-out that
|
|
3972
|
+
* keeps the two in step for `failed` lives in the caller (`resolveRedrive`, #1310),
|
|
3973
|
+
* so a row reaching this `failed` branch is a GENUINE failure.
|
|
3974
|
+
*/
|
|
3975
|
+
async settleRedrive(conv, sessionId, message, ocId, messages, state) {
|
|
3976
|
+
try {
|
|
3977
|
+
if (state === "done") {
|
|
3978
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3979
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3980
|
+
this.log({
|
|
3981
|
+
level: "info",
|
|
3982
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} completed while its row was wrongly reclaimed to pending \u2014 marking done instead of re-dispatching`,
|
|
3983
|
+
conversation_id: conv.id,
|
|
3984
|
+
message_id: message.id
|
|
3985
|
+
});
|
|
3986
|
+
await this.markDone(conv.id, message.id, sessionId, ocId, title, usage);
|
|
3987
|
+
} else {
|
|
3988
|
+
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3989
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3990
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3991
|
+
this.log({
|
|
3992
|
+
level: "error",
|
|
3993
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} errored while its row was wrongly reclaimed to pending \u2014 marking failed instead of re-dispatching: ${error2 ?? "(no error text)"}`,
|
|
3994
|
+
conversation_id: conv.id,
|
|
3995
|
+
message_id: message.id
|
|
3996
|
+
});
|
|
3997
|
+
await this.markFailed(conv.id, message.id, sessionId, error2, usage, failure);
|
|
3998
|
+
}
|
|
3999
|
+
} catch (err) {
|
|
4000
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
4001
|
+
this.log({
|
|
4002
|
+
level: "warn",
|
|
4003
|
+
message: `Re-drive: failed to report message ${message.id.slice(0, 8)} ${state} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
4004
|
+
conversation_id: conv.id,
|
|
4005
|
+
message_id: message.id
|
|
4006
|
+
});
|
|
4007
|
+
this.signalRedriveOutcomeUnreported(conv, message, "settle");
|
|
4008
|
+
return "unresolved";
|
|
4009
|
+
}
|
|
4010
|
+
this.clearRedriveUnresolved(message.id);
|
|
4011
|
+
void this.postSignal(conv.id, message.id, "redrive_settled");
|
|
4012
|
+
return "settled";
|
|
4013
|
+
}
|
|
4014
|
+
/**
|
|
4015
|
+
* The permanent-failure outcome (#1348): the fence's own poll of this session
|
|
4016
|
+
* failed with the SAME opencode-answered signature
|
|
4017
|
+
* `MAX_IDENTICAL_REDRIVE_POLL_FAILURES` times in a row — a transient blip
|
|
4018
|
+
* would have varied or eventually cleared (see `pollSessionMessagesForRedrive`
|
|
4019
|
+
* and `recordRedrivePollFailure`), so this is a durable fault (e.g. #1345's
|
|
4020
|
+
* corrupted opencode session) rather than something worth retrying forever.
|
|
4021
|
+
* Mirrors `settleRedrive`'s error discipline: no `usage`/`failure` args to
|
|
4022
|
+
* `markFailed` (no opencode snapshot to extract them from — this poll never
|
|
4023
|
+
* got a readable one).
|
|
4024
|
+
*/
|
|
4025
|
+
async failRedrivePollPermanent(conv, sessionId, message, signature, streak) {
|
|
4026
|
+
this.log({
|
|
4027
|
+
level: "error",
|
|
4028
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) failed to poll with the identical signature "${signature}" ${streak} times in a row \u2014 reporting the message failed instead of retrying forever`,
|
|
4029
|
+
conversation_id: conv.id,
|
|
4030
|
+
message_id: message.id
|
|
4031
|
+
});
|
|
4032
|
+
try {
|
|
4033
|
+
await this.markFailed(
|
|
4034
|
+
conv.id,
|
|
4035
|
+
message.id,
|
|
4036
|
+
sessionId,
|
|
4037
|
+
`The runner could not read this conversation's state from OpenCode (${signature}). The same failure repeated ${streak} times in a row, so the message was not retried further.`
|
|
4038
|
+
);
|
|
4039
|
+
} catch (err) {
|
|
4040
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
2422
4041
|
this.log({
|
|
2423
4042
|
level: "warn",
|
|
2424
|
-
message: `
|
|
2425
|
-
conversation_id: conv.id
|
|
4043
|
+
message: `Re-drive: failed to report message ${message.id.slice(0, 8)} permanently failed (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
4044
|
+
conversation_id: conv.id,
|
|
4045
|
+
message_id: message.id
|
|
2426
4046
|
});
|
|
4047
|
+
this.signalRedriveOutcomeUnreported(conv, message, "fail_permanent");
|
|
4048
|
+
return "unresolved";
|
|
2427
4049
|
}
|
|
2428
|
-
this.
|
|
2429
|
-
|
|
4050
|
+
this.clearRedriveUnresolved(message.id);
|
|
4051
|
+
void this.postSignal(conv.id, message.id, "redrive_poll_failed");
|
|
4052
|
+
return "settled";
|
|
4053
|
+
}
|
|
4054
|
+
/**
|
|
4055
|
+
* The bounded `unresolved` outcome (Task 3.4): opencode's state could not be
|
|
4056
|
+
* observed (snapshot unreadable/empty, or `isSessionOngoing` returned `null`).
|
|
4057
|
+
* A `pending` row is invisible to every cron arm (all require `status =
|
|
4058
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
4059
|
+
* nothing driving it — bound it to the existing `pausedMaxWaitMs` window
|
|
4060
|
+
* (reusing the knob, not a new constant) and take `dispatch` once elapsed.
|
|
4061
|
+
*/
|
|
4062
|
+
resolveRedriveUnresolved(conv, message) {
|
|
4063
|
+
const now = this.now();
|
|
4064
|
+
const since = this.redriveUnresolvedSince.get(message.id);
|
|
4065
|
+
if (since !== void 0 && now - since >= this.pausedMaxWaitMs) {
|
|
4066
|
+
this.clearRedriveUnresolved(message.id);
|
|
4067
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
4068
|
+
return "dispatch";
|
|
4069
|
+
}
|
|
4070
|
+
if (since === void 0) {
|
|
4071
|
+
this.redriveUnresolvedSince.set(message.id, now);
|
|
4072
|
+
}
|
|
4073
|
+
if (!this.redriveUnresolvedSignalled.has(message.id)) {
|
|
4074
|
+
this.redriveUnresolvedSignalled.add(message.id);
|
|
4075
|
+
void this.postSignal(conv.id, message.id, "redrive_unresolved");
|
|
4076
|
+
}
|
|
4077
|
+
return "unresolved";
|
|
4078
|
+
}
|
|
4079
|
+
/** Clear all `unresolved`/failure-streak trackers for a row (any non-`unresolved` outcome). */
|
|
4080
|
+
clearRedriveUnresolved(messageId) {
|
|
4081
|
+
this.redriveUnresolvedSince.delete(messageId);
|
|
4082
|
+
this.redriveUnresolvedSignalled.delete(messageId);
|
|
4083
|
+
this.redrivePollFailures.delete(messageId);
|
|
4084
|
+
this.redriveOutcomeUnreportedSignalled.delete(messageId);
|
|
4085
|
+
}
|
|
4086
|
+
/**
|
|
4087
|
+
* Class B (#1340): the runner DECIDED an outcome (reattach/settle/fail_permanent)
|
|
4088
|
+
* but its own PATCH to record it failed. Deliberately NOT bounded like
|
|
4089
|
+
* `resolveRedriveUnresolved`, for two load-bearing reasons: (1) the only
|
|
4090
|
+
* remedy is a PATCH to the server — the very call that is failing, so a bound
|
|
4091
|
+
* would act by retrying it; (2) this class is self-clearing by construction
|
|
4092
|
+
* (the next successful PATCH resolves it via `clearRedriveUnresolved`),
|
|
4093
|
+
* unlike #1340's deterministic, permanent shape. Fires at most once per
|
|
4094
|
+
* (message, outcome) streak.
|
|
4095
|
+
*/
|
|
4096
|
+
signalRedriveOutcomeUnreported(conv, message, outcome) {
|
|
4097
|
+
if (this.redriveOutcomeUnreportedSignalled.get(message.id) === outcome) return;
|
|
4098
|
+
this.redriveOutcomeUnreportedSignalled.set(message.id, outcome);
|
|
4099
|
+
void this.postSignal(conv.id, message.id, "redrive_outcome_unreported", {
|
|
4100
|
+
attempted_outcome: outcome
|
|
4101
|
+
});
|
|
4102
|
+
}
|
|
4103
|
+
/**
|
|
4104
|
+
* Record one poll outcome toward the re-drive fence's consecutive-identical-
|
|
4105
|
+
* failure streak (#1348) and return the resulting count. `signature === null`
|
|
4106
|
+
* (a thrown exception, H1) always clears the streak and returns `0` — it is
|
|
4107
|
+
* never countable. Otherwise the streak continues only when BOTH the session
|
|
4108
|
+
* and the signature match the previous failure; anything else (a different
|
|
4109
|
+
* session, or the same session failing a DIFFERENT way) starts a fresh streak
|
|
4110
|
+
* at `1`.
|
|
4111
|
+
*/
|
|
4112
|
+
recordRedrivePollFailure(messageId, sessionId, signature) {
|
|
4113
|
+
if (signature === null) {
|
|
4114
|
+
this.redrivePollFailures.delete(messageId);
|
|
4115
|
+
return 0;
|
|
4116
|
+
}
|
|
4117
|
+
const existing = this.redrivePollFailures.get(messageId);
|
|
4118
|
+
if (existing && existing.sessionId === sessionId && existing.signature === signature) {
|
|
4119
|
+
existing.count += 1;
|
|
4120
|
+
return existing.count;
|
|
4121
|
+
}
|
|
4122
|
+
this.redrivePollFailures.set(messageId, { sessionId, signature, count: 1 });
|
|
4123
|
+
return 1;
|
|
4124
|
+
}
|
|
4125
|
+
/**
|
|
4126
|
+
* Record one UNCONFIRMED-dispatch outcome (a `pending` row with no stored
|
|
4127
|
+
* `opencode_message_id` whose `sendPromptAsync` returned `null`) toward the
|
|
4128
|
+
* bound in `processConversation`'s dispatch loop, and return the resulting
|
|
4129
|
+
* count. Mirrors `recordRedrivePollFailure`'s session-scoping: a session
|
|
4130
|
+
* change starts a fresh streak at `1` rather than inheriting the old one's
|
|
4131
|
+
* count, since a new session is a genuinely different attempt.
|
|
4132
|
+
*/
|
|
4133
|
+
recordUnconfirmedDispatch(messageId, sessionId) {
|
|
4134
|
+
const existing = this.unconfirmedDispatchFailures.get(messageId);
|
|
4135
|
+
if (existing && existing.sessionId === sessionId) {
|
|
4136
|
+
existing.count += 1;
|
|
4137
|
+
return existing.count;
|
|
4138
|
+
}
|
|
4139
|
+
this.unconfirmedDispatchFailures.set(messageId, { sessionId, count: 1 });
|
|
4140
|
+
return 1;
|
|
4141
|
+
}
|
|
4142
|
+
/**
|
|
4143
|
+
* Record that `sessionId` is no longer a valid binding for `conversationId`
|
|
4144
|
+
* (#553). Keyed by conversation and hard-capped, so it cannot grow with the
|
|
4145
|
+
* number of failures — see the `supersededSessions` field doc.
|
|
4146
|
+
*/
|
|
4147
|
+
supersede(conversationId, sessionId) {
|
|
4148
|
+
this.supersededSessions.delete(conversationId);
|
|
4149
|
+
this.supersededSessions.set(conversationId, sessionId);
|
|
4150
|
+
while (this.supersededSessions.size > MAX_SUPERSEDED_CONVERSATIONS) {
|
|
4151
|
+
const oldest = this.supersededSessions.keys().next().value;
|
|
4152
|
+
if (oldest === void 0) return;
|
|
4153
|
+
this.supersededSessions.delete(oldest);
|
|
4154
|
+
}
|
|
4155
|
+
}
|
|
4156
|
+
/** Whether `sessionId` is the session this conversation has abandoned (#553). */
|
|
4157
|
+
isSuperseded(conversationId, sessionId) {
|
|
4158
|
+
return this.supersededSessions.get(conversationId) === sessionId;
|
|
2430
4159
|
}
|
|
4160
|
+
/**
|
|
4161
|
+
* Resolve the opencode session to run this conversation's turns in.
|
|
4162
|
+
*
|
|
4163
|
+
* `refusedSessionId` is set when the #553 guard fired — i.e. the persisted
|
|
4164
|
+
* binding was an id this runner had abandoned, so a resurrection genuinely
|
|
4165
|
+
* happened and a fresh session was bound instead. The caller reports it.
|
|
4166
|
+
*/
|
|
2431
4167
|
async ensureSession(conv) {
|
|
2432
4168
|
const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
|
|
4169
|
+
if (bound && this.isSuperseded(conv.id, bound)) {
|
|
4170
|
+
this.log({
|
|
4171
|
+
level: "warn",
|
|
4172
|
+
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.`,
|
|
4173
|
+
conversation_id: conv.id
|
|
4174
|
+
});
|
|
4175
|
+
this.sessions.delete(conv.id);
|
|
4176
|
+
return { sessionId: await this.createAndBindSession(conv.id), refusedSessionId: bound };
|
|
4177
|
+
}
|
|
2433
4178
|
if (bound) {
|
|
2434
4179
|
const exists = await sessionExists(this.port, bound);
|
|
2435
4180
|
if (exists === false) {
|
|
@@ -2439,12 +4184,12 @@ var ChannelDriver = class {
|
|
|
2439
4184
|
conversation_id: conv.id
|
|
2440
4185
|
});
|
|
2441
4186
|
this.sessions.delete(conv.id);
|
|
2442
|
-
return this.createAndBindSession(conv.id);
|
|
4187
|
+
return { sessionId: await this.createAndBindSession(conv.id) };
|
|
2443
4188
|
}
|
|
2444
4189
|
this.sessions.set(conv.id, bound);
|
|
2445
|
-
return bound;
|
|
4190
|
+
return { sessionId: bound };
|
|
2446
4191
|
}
|
|
2447
|
-
return this.createAndBindSession(conv.id);
|
|
4192
|
+
return { sessionId: await this.createAndBindSession(conv.id) };
|
|
2448
4193
|
}
|
|
2449
4194
|
/**
|
|
2450
4195
|
* Create a fresh OpenCode session for a conversation, cache the binding, and
|
|
@@ -2455,7 +4200,12 @@ var ChannelDriver = class {
|
|
|
2455
4200
|
const directory = await this.resolveOpenCodeDirectory();
|
|
2456
4201
|
const sessionId = await createOpenCodeSession(this.port, directory);
|
|
2457
4202
|
this.sessions.set(conversationId, sessionId);
|
|
2458
|
-
await this.persistSession(conversationId, sessionId).catch(() => {
|
|
4203
|
+
await this.persistSession(conversationId, sessionId).catch((err) => {
|
|
4204
|
+
this.log({
|
|
4205
|
+
level: "warn",
|
|
4206
|
+
message: `Persisting the OpenCode session binding ${sessionId.slice(0, 8)} for conversation ${conversationId.slice(0, 8)} failed (best-effort, not retried) \u2014 the completion PATCH also carries opencode_session_id, so the binding is repaired when the turn finishes: ${err instanceof Error ? err.message : String(err)}`,
|
|
4207
|
+
conversation_id: conversationId
|
|
4208
|
+
});
|
|
2459
4209
|
});
|
|
2460
4210
|
return sessionId;
|
|
2461
4211
|
}
|
|
@@ -2532,7 +4282,11 @@ var ChannelDriver = class {
|
|
|
2532
4282
|
* (not-owned / out-of-range / deleted-at-source / workspace gone) / 413
|
|
2533
4283
|
* (over-cap). On ANY non-2xx or thrown failure we return `null` so the caller
|
|
2534
4284
|
* OMITS that one image and the text turn still sends — NEVER throws the turn.
|
|
2535
|
-
*
|
|
4285
|
+
* A 404 body carrying `{ reason: 'needs_reauth' }` (#547 — the server CONFIRMED
|
|
4286
|
+
* a Slack `files:read` scope problem via `files.info`) instead resolves the
|
|
4287
|
+
* `AttachmentFetchNeedsReauth` sentinel, so the in-thread note can steer the
|
|
4288
|
+
* user to reconnect Slack instead of a generic "unavailable". Failures are
|
|
4289
|
+
* logged with context (no silent swallow).
|
|
2536
4290
|
*/
|
|
2537
4291
|
async fetchAttachmentDataUrl(messageId, index, mime) {
|
|
2538
4292
|
try {
|
|
@@ -2541,6 +4295,25 @@ var ChannelDriver = class {
|
|
|
2541
4295
|
{ headers: { Authorization: this.getAuthHeader() } }
|
|
2542
4296
|
);
|
|
2543
4297
|
if (!res.ok) {
|
|
4298
|
+
let reason;
|
|
4299
|
+
try {
|
|
4300
|
+
const body = await res.json();
|
|
4301
|
+
if (body && typeof body.reason === "string") reason = body.reason;
|
|
4302
|
+
} catch (parseErr) {
|
|
4303
|
+
this.log({
|
|
4304
|
+
level: "debug",
|
|
4305
|
+
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`,
|
|
4306
|
+
message_id: messageId
|
|
4307
|
+
});
|
|
4308
|
+
}
|
|
4309
|
+
if (reason === "needs_reauth") {
|
|
4310
|
+
this.log({
|
|
4311
|
+
level: "error",
|
|
4312
|
+
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)`,
|
|
4313
|
+
message_id: messageId
|
|
4314
|
+
});
|
|
4315
|
+
return { needsReauth: true };
|
|
4316
|
+
}
|
|
2544
4317
|
this.log({
|
|
2545
4318
|
level: "error",
|
|
2546
4319
|
message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} returned HTTP ${res.status} \u2014 omitting this image (text turn proceeds)`,
|
|
@@ -2580,6 +4353,9 @@ var ChannelDriver = class {
|
|
|
2580
4353
|
if (this.attachmentsSkippedSignalled.has(messageId)) return;
|
|
2581
4354
|
this.attachmentsSkippedSignalled.add(messageId);
|
|
2582
4355
|
const skippedReason = capabilityUnknown ? "unknown" : "unsupported";
|
|
4356
|
+
const failedReason = outcomes.some(
|
|
4357
|
+
(o) => o.status === "failed" && o.reason === "needs_reauth"
|
|
4358
|
+
) ? "needs_reauth" : void 0;
|
|
2583
4359
|
this.log({
|
|
2584
4360
|
level: "info",
|
|
2585
4361
|
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`,
|
|
@@ -2589,7 +4365,8 @@ var ChannelDriver = class {
|
|
|
2589
4365
|
void this.postSignal(conversationId, messageId, "attachments_skipped", {
|
|
2590
4366
|
skipped,
|
|
2591
4367
|
failed,
|
|
2592
|
-
...skipped > 0 ? { skipped_reason: skippedReason } : {}
|
|
4368
|
+
...skipped > 0 ? { skipped_reason: skippedReason } : {},
|
|
4369
|
+
...failedReason ? { failed_reason: failedReason } : {}
|
|
2593
4370
|
});
|
|
2594
4371
|
}
|
|
2595
4372
|
/** Register a freshly-dispatched message with its session's watcher state. */
|
|
@@ -2620,12 +4397,17 @@ var ChannelDriver = class {
|
|
|
2620
4397
|
stuckReported: false,
|
|
2621
4398
|
lastAliveAt: 0,
|
|
2622
4399
|
aliveInFlight: false,
|
|
4400
|
+
titleSynced: false,
|
|
4401
|
+
titleSyncInFlight: false,
|
|
2623
4402
|
awaitingHumanLatched: false,
|
|
2624
4403
|
pausedOnQuestion: false,
|
|
2625
4404
|
pausedOnPermission: false,
|
|
2626
4405
|
pausedClearConfirmed: false,
|
|
2627
4406
|
pausedInFlight: false,
|
|
2628
|
-
deliveryDeadlineAnchored: false
|
|
4407
|
+
deliveryDeadlineAnchored: false,
|
|
4408
|
+
b2PinnedSinceMs: 0,
|
|
4409
|
+
b2LastDescendantCheckMs: 0,
|
|
4410
|
+
b2AbandonedSignalled: false
|
|
2629
4411
|
});
|
|
2630
4412
|
}
|
|
2631
4413
|
/**
|
|
@@ -2640,9 +4422,10 @@ var ChannelDriver = class {
|
|
|
2640
4422
|
* opencode reports ACTIVELY `running` is watched to completion (its liveness
|
|
2641
4423
|
* heartbeat keeps the cron off its row), while a re-adopted turn that is paused
|
|
2642
4424
|
* awaiting a human — or queued/unreachable — is still bounded by `deadline` and
|
|
2643
|
-
* handed to the cron.
|
|
2644
|
-
*
|
|
2645
|
-
*
|
|
4425
|
+
* handed to the cron. Real invariant (#965): the cron MAY reclaim a row this
|
|
4426
|
+
* runner still holds; a reclaimed row that already ran is never re-dispatched
|
|
4427
|
+
* while opencode reports its turn ongoing (readopt's own gate here, and the
|
|
4428
|
+
* `pending`-row re-drive fence, `resolveRedrive`). `dispatchedAt` stays `now`
|
|
2646
4429
|
* (only the appear-guard uses it).
|
|
2647
4430
|
*
|
|
2648
4431
|
* `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
|
|
@@ -2693,12 +4476,17 @@ var ChannelDriver = class {
|
|
|
2693
4476
|
// with no extra `re_adopted` signal needed (folds old WI-6).
|
|
2694
4477
|
lastAliveAt: 0,
|
|
2695
4478
|
aliveInFlight: false,
|
|
4479
|
+
titleSynced: false,
|
|
4480
|
+
titleSyncInFlight: false,
|
|
2696
4481
|
awaitingHumanLatched: false,
|
|
2697
4482
|
pausedOnQuestion: false,
|
|
2698
4483
|
pausedOnPermission: false,
|
|
2699
4484
|
pausedClearConfirmed: false,
|
|
2700
4485
|
pausedInFlight: false,
|
|
2701
|
-
deliveryDeadlineAnchored: false
|
|
4486
|
+
deliveryDeadlineAnchored: false,
|
|
4487
|
+
b2PinnedSinceMs: 0,
|
|
4488
|
+
b2LastDescendantCheckMs: 0,
|
|
4489
|
+
b2AbandonedSignalled: false
|
|
2702
4490
|
});
|
|
2703
4491
|
}
|
|
2704
4492
|
/**
|
|
@@ -2860,58 +4648,7 @@ var ChannelDriver = class {
|
|
|
2860
4648
|
}
|
|
2861
4649
|
}
|
|
2862
4650
|
if (state === "done") {
|
|
2863
|
-
this.
|
|
2864
|
-
if (!inFlight.done) {
|
|
2865
|
-
this.log({
|
|
2866
|
-
level: "info",
|
|
2867
|
-
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
2868
|
-
conversation_id: conv.id,
|
|
2869
|
-
message_id: inFlight.evidentMessageId
|
|
2870
|
-
});
|
|
2871
|
-
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
2872
|
-
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
2873
|
-
try {
|
|
2874
|
-
await this.markDone(
|
|
2875
|
-
conv.id,
|
|
2876
|
-
inFlight.evidentMessageId,
|
|
2877
|
-
sessionId,
|
|
2878
|
-
inFlight.opencodeMessageId,
|
|
2879
|
-
title,
|
|
2880
|
-
usage
|
|
2881
|
-
);
|
|
2882
|
-
} catch (err) {
|
|
2883
|
-
if (err instanceof ChannelAuthError) throw err;
|
|
2884
|
-
if (err instanceof ChannelTerminalError) {
|
|
2885
|
-
this.log({
|
|
2886
|
-
level: "warn",
|
|
2887
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
2888
|
-
conversation_id: conv.id,
|
|
2889
|
-
message_id: inFlight.evidentMessageId
|
|
2890
|
-
});
|
|
2891
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2892
|
-
return;
|
|
2893
|
-
}
|
|
2894
|
-
if (this.now() >= inFlight.deadline) {
|
|
2895
|
-
this.log({
|
|
2896
|
-
level: "warn",
|
|
2897
|
-
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)}`,
|
|
2898
|
-
conversation_id: conv.id,
|
|
2899
|
-
message_id: inFlight.evidentMessageId
|
|
2900
|
-
});
|
|
2901
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2902
|
-
return;
|
|
2903
|
-
}
|
|
2904
|
-
this.log({
|
|
2905
|
-
level: "warn",
|
|
2906
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
2907
|
-
conversation_id: conv.id,
|
|
2908
|
-
message_id: inFlight.evidentMessageId
|
|
2909
|
-
});
|
|
2910
|
-
return;
|
|
2911
|
-
}
|
|
2912
|
-
inFlight.done = true;
|
|
2913
|
-
}
|
|
2914
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4651
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
2915
4652
|
return;
|
|
2916
4653
|
}
|
|
2917
4654
|
if (state === "failed") {
|
|
@@ -2925,8 +4662,16 @@ var ChannelDriver = class {
|
|
|
2925
4662
|
message_id: inFlight.evidentMessageId
|
|
2926
4663
|
});
|
|
2927
4664
|
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4665
|
+
const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
|
|
2928
4666
|
try {
|
|
2929
|
-
await this.markFailed(
|
|
4667
|
+
await this.markFailed(
|
|
4668
|
+
conv.id,
|
|
4669
|
+
inFlight.evidentMessageId,
|
|
4670
|
+
sessionId,
|
|
4671
|
+
error2,
|
|
4672
|
+
usage,
|
|
4673
|
+
failure
|
|
4674
|
+
);
|
|
2930
4675
|
} catch (err) {
|
|
2931
4676
|
if (err instanceof ChannelAuthError) throw err;
|
|
2932
4677
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -2971,6 +4716,44 @@ var ChannelDriver = class {
|
|
|
2971
4716
|
});
|
|
2972
4717
|
}
|
|
2973
4718
|
const activelyRunning = state === "running" && !awaitingHuman;
|
|
4719
|
+
const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
|
|
4720
|
+
const snapshotReadable = messages != null && messages.length > 0;
|
|
4721
|
+
if (!pinnedNow) {
|
|
4722
|
+
if (snapshotReadable) {
|
|
4723
|
+
inFlight.b2PinnedSinceMs = 0;
|
|
4724
|
+
inFlight.b2LastDescendantCheckMs = 0;
|
|
4725
|
+
inFlight.b2AbandonedSignalled = false;
|
|
4726
|
+
}
|
|
4727
|
+
} else {
|
|
4728
|
+
if (inFlight.b2AbandonedSignalled) {
|
|
4729
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4730
|
+
return;
|
|
4731
|
+
}
|
|
4732
|
+
if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
|
|
4733
|
+
const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
|
|
4734
|
+
if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
|
|
4735
|
+
inFlight.b2LastDescendantCheckMs = this.now();
|
|
4736
|
+
const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
|
|
4737
|
+
if (isB2AbandonmentConfirmed({
|
|
4738
|
+
pinnedForMs,
|
|
4739
|
+
minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
|
|
4740
|
+
descendantOngoing
|
|
4741
|
+
})) {
|
|
4742
|
+
inFlight.b2AbandonedSignalled = true;
|
|
4743
|
+
this.log({
|
|
4744
|
+
level: "warn",
|
|
4745
|
+
message: `Message ${id.slice(0, 8)} b2-pinned for ${Math.round(pinnedForMs / 1e3)}s with no ongoing descendant sub-agent session (status-map confirmed) \u2014 treating the delegated/tool turn as abandoned, resolving done`,
|
|
4746
|
+
conversation_id: conv.id,
|
|
4747
|
+
message_id: id
|
|
4748
|
+
});
|
|
4749
|
+
void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
|
|
4750
|
+
watched_for_ms: pinnedForMs
|
|
4751
|
+
});
|
|
4752
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4753
|
+
return;
|
|
4754
|
+
}
|
|
4755
|
+
}
|
|
4756
|
+
}
|
|
2974
4757
|
if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
|
|
2975
4758
|
this.log({
|
|
2976
4759
|
level: "warn",
|
|
@@ -2990,6 +4773,18 @@ var ChannelDriver = class {
|
|
|
2990
4773
|
inFlight.aliveInFlight = false;
|
|
2991
4774
|
if (ok) inFlight.lastAliveAt = this.now();
|
|
2992
4775
|
});
|
|
4776
|
+
if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
|
|
4777
|
+
inFlight.titleSyncInFlight = true;
|
|
4778
|
+
void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
|
|
4779
|
+
if (!title) {
|
|
4780
|
+
inFlight.titleSyncInFlight = false;
|
|
4781
|
+
return;
|
|
4782
|
+
}
|
|
4783
|
+
const ok = await this.patchConversationTitle(conv.id, title);
|
|
4784
|
+
inFlight.titleSyncInFlight = false;
|
|
4785
|
+
if (ok) inFlight.titleSynced = true;
|
|
4786
|
+
});
|
|
4787
|
+
}
|
|
2993
4788
|
}
|
|
2994
4789
|
if (awaitingHuman) {
|
|
2995
4790
|
if (!inFlight.awaitingHumanLatched) {
|
|
@@ -3027,6 +4822,70 @@ var ChannelDriver = class {
|
|
|
3027
4822
|
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3028
4823
|
}
|
|
3029
4824
|
}
|
|
4825
|
+
/**
|
|
4826
|
+
* Settle a message whose run-state has resolved `'done'` — extracted verbatim
|
|
4827
|
+
* (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
|
|
4828
|
+
* inline `state === 'done'` branch body, so a SECOND caller (the #721
|
|
4829
|
+
* b2-abandonment resolution) can reach the exact same completion behavior
|
|
4830
|
+
* (delivery-deadline anchoring, title resolution, usage extraction, and
|
|
4831
|
+
* `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
|
|
4832
|
+
* and risking the two copies silently drifting apart.
|
|
4833
|
+
*/
|
|
4834
|
+
async settleMessageDone(sessionId, watcher, inFlight, messages) {
|
|
4835
|
+
const conv = watcher.conv;
|
|
4836
|
+
this.anchorDeliveryDeadline(inFlight);
|
|
4837
|
+
if (!inFlight.done) {
|
|
4838
|
+
this.log({
|
|
4839
|
+
level: "info",
|
|
4840
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
4841
|
+
conversation_id: conv.id,
|
|
4842
|
+
message_id: inFlight.evidentMessageId
|
|
4843
|
+
});
|
|
4844
|
+
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
4845
|
+
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4846
|
+
try {
|
|
4847
|
+
await this.markDone(
|
|
4848
|
+
conv.id,
|
|
4849
|
+
inFlight.evidentMessageId,
|
|
4850
|
+
sessionId,
|
|
4851
|
+
inFlight.opencodeMessageId,
|
|
4852
|
+
title,
|
|
4853
|
+
usage
|
|
4854
|
+
);
|
|
4855
|
+
} catch (err) {
|
|
4856
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
4857
|
+
if (err instanceof ChannelTerminalError) {
|
|
4858
|
+
this.log({
|
|
4859
|
+
level: "warn",
|
|
4860
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
4861
|
+
conversation_id: conv.id,
|
|
4862
|
+
message_id: inFlight.evidentMessageId
|
|
4863
|
+
});
|
|
4864
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4865
|
+
return;
|
|
4866
|
+
}
|
|
4867
|
+
if (this.now() >= inFlight.deadline) {
|
|
4868
|
+
this.log({
|
|
4869
|
+
level: "warn",
|
|
4870
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done within the watch window \u2014 leaving for the cron safety net: ${err instanceof Error ? err.message : String(err)}`,
|
|
4871
|
+
conversation_id: conv.id,
|
|
4872
|
+
message_id: inFlight.evidentMessageId
|
|
4873
|
+
});
|
|
4874
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4875
|
+
return;
|
|
4876
|
+
}
|
|
4877
|
+
this.log({
|
|
4878
|
+
level: "warn",
|
|
4879
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
4880
|
+
conversation_id: conv.id,
|
|
4881
|
+
message_id: inFlight.evidentMessageId
|
|
4882
|
+
});
|
|
4883
|
+
return;
|
|
4884
|
+
}
|
|
4885
|
+
inFlight.done = true;
|
|
4886
|
+
}
|
|
4887
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4888
|
+
}
|
|
3030
4889
|
// Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
|
|
3031
4890
|
/**
|
|
3032
4891
|
* Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
|
|
@@ -3126,7 +4985,10 @@ var ChannelDriver = class {
|
|
|
3126
4985
|
* re-dispatched (at most once, see `forceReadoptRun`):
|
|
3127
4986
|
* - `done` → `markDone` now (guarded like the watcher's done branch);
|
|
3128
4987
|
* - `failed` → `markFailed` with the surfaced error (issue #182), so an
|
|
3129
|
-
* errored turn is reported failed on restart, NOT re-dispatched
|
|
4988
|
+
* errored turn is reported failed on restart, NOT re-dispatched —
|
|
4989
|
+
* EXCEPT a restart-ABORTED turn under a not-ongoing session,
|
|
4990
|
+
* which is a restart orphan wearing a terminal error and is
|
|
4991
|
+
* re-dispatched instead (issue #1310, see the branch below);
|
|
3130
4992
|
* - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
|
|
3131
4993
|
* tracking the stored id so the reply correlates by it;
|
|
3132
4994
|
* - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
|
|
@@ -3190,9 +5052,19 @@ var ChannelDriver = class {
|
|
|
3190
5052
|
void this.postSignal(row.conversation_id, row.id, "readopt_done");
|
|
3191
5053
|
return;
|
|
3192
5054
|
}
|
|
3193
|
-
|
|
5055
|
+
const restartAborted = state === "failed" && sessionOngoing === false && isAbortedTerminalReply(messages, ocId ?? "");
|
|
5056
|
+
if (restartAborted) {
|
|
5057
|
+
this.log({
|
|
5058
|
+
level: "info",
|
|
5059
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} carries an abort-shaped terminal error and session ${sessionId.slice(0, 8)} is not-ongoing per GET /session/status \u2014 restart orphan, re-dispatching instead of marking it permanently failed`,
|
|
5060
|
+
conversation_id: row.conversation_id,
|
|
5061
|
+
message_id: row.id
|
|
5062
|
+
});
|
|
5063
|
+
}
|
|
5064
|
+
if (state === "failed" && !restartAborted) {
|
|
3194
5065
|
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3195
5066
|
const usage = messageUsage(messages, ocId ?? "");
|
|
5067
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3196
5068
|
this.log({
|
|
3197
5069
|
level: "error",
|
|
3198
5070
|
message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
|
|
@@ -3200,7 +5072,7 @@ var ChannelDriver = class {
|
|
|
3200
5072
|
message_id: row.id
|
|
3201
5073
|
});
|
|
3202
5074
|
try {
|
|
3203
|
-
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
|
|
5075
|
+
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
|
|
3204
5076
|
} catch (err) {
|
|
3205
5077
|
if (err instanceof ChannelAuthError) throw err;
|
|
3206
5078
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -3403,15 +5275,39 @@ var ChannelDriver = class {
|
|
|
3403
5275
|
}
|
|
3404
5276
|
if (ocId === null) {
|
|
3405
5277
|
this.awaitingReadopt.delete(row.id);
|
|
5278
|
+
const streak = this.recordUnconfirmedDispatch(row.id, sessionId);
|
|
5279
|
+
if (streak >= MAX_IDENTICAL_REDRIVE_POLL_FAILURES) {
|
|
5280
|
+
this.unconfirmedDispatchFailures.delete(row.id);
|
|
5281
|
+
this.sessions.delete(readoptConv.id);
|
|
5282
|
+
this.supersede(readoptConv.id, sessionId);
|
|
5283
|
+
const errorMessage = `OpenCode accepted this re-dispatched message ${streak} times in a row but never confirmed its assigned id (session ${sessionId.slice(0, 8)}'s message list could not be read back) \u2014 the session was abandoned; a fresh one is used for further messages.`;
|
|
5284
|
+
this.log({
|
|
5285
|
+
level: "error",
|
|
5286
|
+
message: errorMessage,
|
|
5287
|
+
conversation_id: row.conversation_id,
|
|
5288
|
+
message_id: row.id
|
|
5289
|
+
});
|
|
5290
|
+
await this.markFailed(row.conversation_id, row.id, null, errorMessage).catch((markErr) => {
|
|
5291
|
+
this.log({
|
|
5292
|
+
level: "warn",
|
|
5293
|
+
message: `markFailed PATCH for message ${row.id.slice(0, 8)} (conversation ${row.conversation_id.slice(0, 8)}) failed (best-effort, not retried): ${markErr instanceof Error ? markErr.message : String(markErr)}`,
|
|
5294
|
+
conversation_id: row.conversation_id,
|
|
5295
|
+
message_id: row.id
|
|
5296
|
+
});
|
|
5297
|
+
});
|
|
5298
|
+
void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
|
|
5299
|
+
return;
|
|
5300
|
+
}
|
|
3406
5301
|
this.log({
|
|
3407
5302
|
level: "warn",
|
|
3408
|
-
message: `Re-adopt: message ${row.id.slice(0, 8)} re-dispatched but its opencode id could not be read back \u2014 leaving un-tracked to retry next drain`,
|
|
5303
|
+
message: `Re-adopt: message ${row.id.slice(0, 8)} re-dispatched but its opencode id could not be read back (${streak}/${MAX_IDENTICAL_REDRIVE_POLL_FAILURES}) \u2014 leaving un-tracked to retry next drain`,
|
|
3409
5304
|
conversation_id: row.conversation_id,
|
|
3410
5305
|
message_id: row.id
|
|
3411
5306
|
});
|
|
3412
5307
|
void this.postSignal(row.conversation_id, row.id, "readopt_orphan_unsent");
|
|
3413
5308
|
return;
|
|
3414
5309
|
}
|
|
5310
|
+
this.unconfirmedDispatchFailures.delete(row.id);
|
|
3415
5311
|
this.registerReadopted(readoptConv, sessionId, readoptMessage, ocId, this.processedAtMs(row));
|
|
3416
5312
|
this.dispatched.add(row.id);
|
|
3417
5313
|
this.readopted.add(row.id);
|
|
@@ -3467,7 +5363,8 @@ var ChannelDriver = class {
|
|
|
3467
5363
|
opencode_model: row.opencode_model,
|
|
3468
5364
|
source_message_id: row.source_message_id,
|
|
3469
5365
|
slack_user_id: row.slack_user_id,
|
|
3470
|
-
attachments: row.attachments ?? null
|
|
5366
|
+
attachments: row.attachments ?? null,
|
|
5367
|
+
opencode_message_id: row.opencode_message_id
|
|
3471
5368
|
};
|
|
3472
5369
|
}
|
|
3473
5370
|
/**
|
|
@@ -3606,6 +5503,47 @@ var ChannelDriver = class {
|
|
|
3606
5503
|
}
|
|
3607
5504
|
return false;
|
|
3608
5505
|
}
|
|
5506
|
+
/**
|
|
5507
|
+
* Tri-state variant of the upward parentID membership walk (#721), used ONLY
|
|
5508
|
+
* by `isAnyDescendantSessionOngoing`. Walks the SAME cached
|
|
5509
|
+
* `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
|
|
5510
|
+
* `sessionBelongsTo`, which deliberately collapses "confirmed not a
|
|
5511
|
+
* descendant" and "the walk's fetch failed" into the same `false` (safe for
|
|
5512
|
+
* its OTHER callers: interaction attribution and the recovery-path
|
|
5513
|
+
* `isAnyDescendantSessionAlive`, both of which just retry next tick with no
|
|
5514
|
+
* safety consequence either way) — this variant keeps those two outcomes
|
|
5515
|
+
* SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
|
|
5516
|
+
* (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
|
|
5517
|
+
* not ongoing".
|
|
5518
|
+
*
|
|
5519
|
+
* Return contract:
|
|
5520
|
+
* - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
|
|
5521
|
+
* - `false` → the walk reached a definitive, parent-less root session
|
|
5522
|
+
* WITHOUT ever matching `rootSessionId` — `sessionId` is
|
|
5523
|
+
* CONFIRMED NOT a descendant of it.
|
|
5524
|
+
* - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
|
|
5525
|
+
* through the walk (`resolveSessionParent` returned `undefined`),
|
|
5526
|
+
* or the depth cap (32) was hit without a definitive answer (a
|
|
5527
|
+
* pathological/cyclic chain proves nothing either way). NEVER
|
|
5528
|
+
* treat this the same as `false` — see `sessionBelongsTo`'s own
|
|
5529
|
+
* doc comment above for why that collapse is safe THERE but not
|
|
5530
|
+
* here.
|
|
5531
|
+
*
|
|
5532
|
+
* `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
|
|
5533
|
+
* to the live-path descendant check, not a modification of shared code used
|
|
5534
|
+
* by interaction attribution or the recovery path.
|
|
5535
|
+
*/
|
|
5536
|
+
async resolveSessionMembership(sessionId, rootSessionId) {
|
|
5537
|
+
let current = sessionId;
|
|
5538
|
+
for (let depth = 0; current && depth < 32; depth++) {
|
|
5539
|
+
if (current === rootSessionId) return true;
|
|
5540
|
+
const parent = await this.resolveSessionParent(current);
|
|
5541
|
+
if (parent === void 0) return null;
|
|
5542
|
+
if (parent === null) return false;
|
|
5543
|
+
current = parent;
|
|
5544
|
+
}
|
|
5545
|
+
return null;
|
|
5546
|
+
}
|
|
3609
5547
|
/**
|
|
3610
5548
|
* Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
|
|
3611
5549
|
* `null` for a root session (no parent) and `undefined` when opencode is
|
|
@@ -3628,19 +5566,36 @@ var ChannelDriver = class {
|
|
|
3628
5566
|
if (parent !== void 0) this.sessionParents.set(sessionId, parent);
|
|
3629
5567
|
return parent;
|
|
3630
5568
|
}
|
|
5569
|
+
/**
|
|
5570
|
+
* OpenCode's synchronous default session title (e.g.
|
|
5571
|
+
* `"New session - 1737800000000"`), assigned immediately when a session is
|
|
5572
|
+
* created — before OpenCode's async LLM-based auto-titling later renames it
|
|
5573
|
+
* mid-turn (#549). Matched by this literal, case-sensitive prefix only; the
|
|
5574
|
+
* timestamp suffix's exact format is deliberately NOT matched, since the prefix
|
|
5575
|
+
* alone is the stable, cheap signal and over-anchoring on the timestamp
|
|
5576
|
+
* representation risks silently breaking if OpenCode ever changes it. Accepted
|
|
5577
|
+
* trade-off: a genuine LLM-assigned title that happens to literally start with
|
|
5578
|
+
* this prefix would also fail to latch (see `resolveSessionTitle`) —
|
|
5579
|
+
* vanishingly unlikely in practice, and deliberately not engineered around.
|
|
5580
|
+
*/
|
|
5581
|
+
static OPENCODE_DEFAULT_TITLE_PREFIX = /^New session - /;
|
|
3631
5582
|
/**
|
|
3632
5583
|
* Resolve (and cache in `sessionTitles`) the OpenCode session TITLE (#310) so the
|
|
3633
5584
|
* status PATCH can carry it into the "Live sessions" list. Driver-level cache so
|
|
3634
5585
|
* BOTH the watcher completion path and the restart-recovery re-adopt path (which
|
|
3635
5586
|
* has no watcher) can use it. `conversationId` is passed only for log context.
|
|
3636
5587
|
* Best-effort:
|
|
3637
|
-
* - a resolved NON-EMPTY title
|
|
5588
|
+
* - a resolved NON-EMPTY title that does NOT match
|
|
5589
|
+
* `OPENCODE_DEFAULT_TITLE_PREFIX` is cached and terminal (a real session name
|
|
3638
5590
|
* won't later un-name), so we do NOT re-GET `/session/:id` every tick;
|
|
3639
|
-
* - while the title is still absent
|
|
3640
|
-
*
|
|
3641
|
-
*
|
|
3642
|
-
*
|
|
3643
|
-
*
|
|
5591
|
+
* - while the title is still absent, empty, or matches the OpenCode
|
|
5592
|
+
* placeholder prefix (#549) we do NOT latch it — OpenCode names sessions
|
|
5593
|
+
* asynchronously mid-turn, so an early call (e.g. at `processing`) must leave
|
|
5594
|
+
* the cache unresolved and re-fetch on the next need so a later call (e.g. at
|
|
5595
|
+
* `done`) picks up the name assigned in the meantime. Such a call returns
|
|
5596
|
+
* `null` (omit the title on THIS PATCH) without caching. If a session is
|
|
5597
|
+
* never renamed, the title is omitted forever rather than ever persisting
|
|
5598
|
+
* the placeholder as a last resort;
|
|
3644
5599
|
* - a failed request likewise leaves the cache unresolved (retry next need)
|
|
3645
5600
|
* and returns `null` — it must NEVER throw or block completion.
|
|
3646
5601
|
* A failure is logged with agent/session context (no silent catch).
|
|
@@ -3653,7 +5608,7 @@ var ChannelDriver = class {
|
|
|
3653
5608
|
if (res.ok) {
|
|
3654
5609
|
const body = await res.json();
|
|
3655
5610
|
const title = body && typeof body.title === "string" ? body.title.trim() : "";
|
|
3656
|
-
if (title.length > 0) {
|
|
5611
|
+
if (title.length > 0 && !_ChannelDriver.OPENCODE_DEFAULT_TITLE_PREFIX.test(title)) {
|
|
3657
5612
|
this.sessionTitles.set(sessionId, title);
|
|
3658
5613
|
return title;
|
|
3659
5614
|
}
|
|
@@ -3673,6 +5628,54 @@ var ChannelDriver = class {
|
|
|
3673
5628
|
}
|
|
3674
5629
|
return null;
|
|
3675
5630
|
}
|
|
5631
|
+
/**
|
|
5632
|
+
* Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
|
|
5633
|
+
* session title onto the conversation via the PLAIN conversation-update
|
|
5634
|
+
* endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
|
|
5635
|
+
* message-status endpoint `markProcessing`/`markDone` use. Deliberately a
|
|
5636
|
+
* separate, lighter call: it carries no `status`, so it cannot re-trigger the
|
|
5637
|
+
* `processing`/`done` transition side effects (Slack notices, activity-log
|
|
5638
|
+
* rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
|
|
5639
|
+
* ever touches `conversations.title`. That route (`routes/conversations.ts`)
|
|
5640
|
+
* skips a title write matching the stored value, so a redundant call with the
|
|
5641
|
+
* same title is a real no-op — it does not bump `updated_at`, which the
|
|
5642
|
+
* conversation list sorts and paginates on. (Note this is a DIFFERENT guard
|
|
5643
|
+
* from `threads.ts`'s "non-empty AND changed" one, which only covers the
|
|
5644
|
+
* message-status PATCH; the non-empty half is enforced here instead, by
|
|
5645
|
+
* `resolveSessionTitle` never returning an empty/placeholder title.)
|
|
5646
|
+
*
|
|
5647
|
+
* Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
|
|
5648
|
+
* is logged and the title is simply retried on the next heartbeat tick (the
|
|
5649
|
+
* caller only latches `titleSynced` on `true`).
|
|
5650
|
+
*/
|
|
5651
|
+
async patchConversationTitle(conversationId, title) {
|
|
5652
|
+
try {
|
|
5653
|
+
const res = await this.fetchImpl(
|
|
5654
|
+
`${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
|
|
5655
|
+
{
|
|
5656
|
+
method: "PATCH",
|
|
5657
|
+
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
5658
|
+
body: JSON.stringify({ title })
|
|
5659
|
+
}
|
|
5660
|
+
);
|
|
5661
|
+
if (!res.ok) {
|
|
5662
|
+
this.log({
|
|
5663
|
+
level: "debug",
|
|
5664
|
+
message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
|
|
5665
|
+
conversation_id: conversationId
|
|
5666
|
+
});
|
|
5667
|
+
return false;
|
|
5668
|
+
}
|
|
5669
|
+
return true;
|
|
5670
|
+
} catch (err) {
|
|
5671
|
+
this.log({
|
|
5672
|
+
level: "debug",
|
|
5673
|
+
message: `Best-effort mid-turn title sync PATCH failed for conversation ${conversationId.slice(0, 8)} (will retry next heartbeat): ${err instanceof Error ? err.message : String(err)}`,
|
|
5674
|
+
conversation_id: conversationId
|
|
5675
|
+
});
|
|
5676
|
+
return false;
|
|
5677
|
+
}
|
|
5678
|
+
}
|
|
3676
5679
|
/**
|
|
3677
5680
|
* DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
|
|
3678
5681
|
* (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
|
|
@@ -3731,6 +5734,84 @@ var ChannelDriver = class {
|
|
|
3731
5734
|
}
|
|
3732
5735
|
return false;
|
|
3733
5736
|
}
|
|
5737
|
+
/**
|
|
5738
|
+
* LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
|
|
5739
|
+
* sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
|
|
5740
|
+
* in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
|
|
5741
|
+
*
|
|
5742
|
+
* Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
|
|
5743
|
+
* cross-check above): that method judges liveness from the child's OWN
|
|
5744
|
+
* TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
|
|
5745
|
+
* on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
|
|
5746
|
+
* path the local opencode server IS running, so its in-memory status map is
|
|
5747
|
+
* live and authoritative — and per ADR-0047 §4a ("the child has its own entry
|
|
5748
|
+
* [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
|
|
5749
|
+
* ENTIRE turn (including any tool call it is itself executing), not a
|
|
5750
|
+
* per-message transcript snapshot. This sidesteps the "child's own tool is
|
|
5751
|
+
* executing, between its step's completion and the next generation step"
|
|
5752
|
+
* transcript gap that a transcript-based check would need a second,
|
|
5753
|
+
* sustained-window bound to guard against — it is simply not derived from
|
|
5754
|
+
* message timestamps at all.
|
|
5755
|
+
*
|
|
5756
|
+
* Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
|
|
5757
|
+
* status, as the recovery path does per §4a)? Because on the LIVE path the
|
|
5758
|
+
* root session can be shared: a SECOND, unrelated user message can land on the
|
|
5759
|
+
* SAME session (issue #721's own root cause) and keep the root `busy` for a
|
|
5760
|
+
* reason that has nothing to do with THIS message's delegation. A `task`
|
|
5761
|
+
* descendant session is spawned for exactly one delegated turn and never
|
|
5762
|
+
* reused, so its OWN status-map entry is unambiguous evidence about that one
|
|
5763
|
+
* delegation — which the root's status is not.
|
|
5764
|
+
*
|
|
5765
|
+
* Why membership is checked via `resolveSessionMembership`, NOT
|
|
5766
|
+
* `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
|
|
5767
|
+
* `GET /session/:id` fetch failure into "not a descendant", which would
|
|
5768
|
+
* silently drop a genuinely-live candidate from consideration on the one
|
|
5769
|
+
* unlucky tick its membership-walk fetch hiccups (#721).
|
|
5770
|
+
* `resolveSessionMembership` keeps that failure mode as a distinct `null`
|
|
5771
|
+
* (indeterminate) so it is folded into THIS method's own `indeterminate` flag
|
|
5772
|
+
* instead.
|
|
5773
|
+
*
|
|
5774
|
+
* Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
|
|
5775
|
+
* - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
|
|
5776
|
+
* - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
|
|
5777
|
+
* confirmed either way (`resolveSessionMembership` never
|
|
5778
|
+
* returned `null`), and every CONFIRMED descendant's status read
|
|
5779
|
+
* succeeded and is not ongoing (includes "no descendant session
|
|
5780
|
+
* exists at all" — e.g. a plain, non-`task` tool call).
|
|
5781
|
+
* - `null` → INDETERMINATE: `listSessions` failed, OR at least one
|
|
5782
|
+
* candidate's MEMBERSHIP could not be confirmed
|
|
5783
|
+
* (`resolveSessionMembership` returned `null` — a fetch failure
|
|
5784
|
+
* or pathological chain partway through the parent walk), OR at
|
|
5785
|
+
* least one CONFIRMED descendant's `isSessionOngoing` read
|
|
5786
|
+
* failed — and no OTHER candidate was already confirmed `true`.
|
|
5787
|
+
* The caller MUST NOT treat `null` the same as `false` here
|
|
5788
|
+
* (unlike the recovery cross-check's contract) — see
|
|
5789
|
+
* `isB2AbandonmentConfirmed`.
|
|
5790
|
+
*/
|
|
5791
|
+
async isAnyDescendantSessionOngoing(rootSessionId) {
|
|
5792
|
+
const sessions = await listSessions(this.port);
|
|
5793
|
+
if (!sessions) {
|
|
5794
|
+
this.log({
|
|
5795
|
+
level: "warn",
|
|
5796
|
+
message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
|
|
5797
|
+
});
|
|
5798
|
+
return null;
|
|
5799
|
+
}
|
|
5800
|
+
let indeterminate = false;
|
|
5801
|
+
for (const candidate of sessions) {
|
|
5802
|
+
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
5803
|
+
const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
|
|
5804
|
+
if (membership === null) {
|
|
5805
|
+
indeterminate = true;
|
|
5806
|
+
continue;
|
|
5807
|
+
}
|
|
5808
|
+
if (membership === false) continue;
|
|
5809
|
+
const ongoing = await isSessionOngoing(this.port, candidate.id);
|
|
5810
|
+
if (ongoing === true) return true;
|
|
5811
|
+
if (ongoing === null) indeterminate = true;
|
|
5812
|
+
}
|
|
5813
|
+
return indeterminate ? null : false;
|
|
5814
|
+
}
|
|
3734
5815
|
/**
|
|
3735
5816
|
* Cheap decision-telemetry label for a running row's LAST correlated reply
|
|
3736
5817
|
* (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
|
|
@@ -3859,6 +5940,32 @@ var ChannelDriver = class {
|
|
|
3859
5940
|
}
|
|
3860
5941
|
return messages;
|
|
3861
5942
|
}
|
|
5943
|
+
/**
|
|
5944
|
+
* The `opencode_session_id` fragment of a status PATCH body — `{}` when this
|
|
5945
|
+
* conversation has ABANDONED that session (#553). The field is optional
|
|
5946
|
+
* server-side and an absent one leaves the persisted binding untouched, so
|
|
5947
|
+
* omitting it is how a routine status write stops resurrecting it.
|
|
5948
|
+
*
|
|
5949
|
+
* ONLY for writes whose sole cost is a lost deep link. The `processing` notice
|
|
5950
|
+
* degrades to no "View in Evident" link (the reaction swap still fires) and the
|
|
5951
|
+
* turn-failure notice is built from the PATCH's own `error` text with a link off
|
|
5952
|
+
* the persisted row — neither loses content the user came for. `markDone`
|
|
5953
|
+
* deliberately does NOT use this helper: the server fetches the reply text
|
|
5954
|
+
* THROUGH the session id it is given, so suppressing there would replace the
|
|
5955
|
+
* agent's answer with a bare "✅ Done!" (the #183/#187 failure). The
|
|
5956
|
+
* `ensureSession` guard, not this suppression, is what makes the self-heal
|
|
5957
|
+
* stick.
|
|
5958
|
+
*/
|
|
5959
|
+
sessionIdBody(sessionId, conversationId, messageId, status2) {
|
|
5960
|
+
if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
|
|
5961
|
+
this.log({
|
|
5962
|
+
level: "debug",
|
|
5963
|
+
message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${status2}' update for message ${messageId.slice(0, 8)} so it is not re-bound to conversation ${conversationId.slice(0, 8)}`,
|
|
5964
|
+
conversation_id: conversationId,
|
|
5965
|
+
message_id: messageId
|
|
5966
|
+
});
|
|
5967
|
+
return {};
|
|
5968
|
+
}
|
|
3862
5969
|
/**
|
|
3863
5970
|
* EXISTING combinedAuth route — now fired by the watcher on queued→running
|
|
3864
5971
|
* (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
|
|
@@ -3888,7 +5995,7 @@ var ChannelDriver = class {
|
|
|
3888
5995
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
3889
5996
|
body: JSON.stringify({
|
|
3890
5997
|
status: "processing",
|
|
3891
|
-
|
|
5998
|
+
...this.sessionIdBody(sessionId, conversationId, messageId, "processing"),
|
|
3892
5999
|
...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
|
|
3893
6000
|
...title ? { title } : {}
|
|
3894
6001
|
})
|
|
@@ -3937,6 +6044,11 @@ var ChannelDriver = class {
|
|
|
3937
6044
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
3938
6045
|
body: JSON.stringify({
|
|
3939
6046
|
status: "done",
|
|
6047
|
+
// ALWAYS sent, even for a session this conversation has abandoned
|
|
6048
|
+
// (#553): the server reads the reply text back out of THIS session id
|
|
6049
|
+
// to deliver it. Omitting it would leave the user with "✅ Done!"
|
|
6050
|
+
// instead of the answer — a worse regression than the resurrection it
|
|
6051
|
+
// would prevent, which `ensureSession`'s guard handles anyway.
|
|
3940
6052
|
opencode_session_id: sessionId,
|
|
3941
6053
|
...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
|
|
3942
6054
|
...title ? { title } : {},
|
|
@@ -3953,16 +6065,31 @@ var ChannelDriver = class {
|
|
|
3953
6065
|
}
|
|
3954
6066
|
/**
|
|
3955
6067
|
* Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
|
|
3956
|
-
* when provided (issue #182)
|
|
3957
|
-
* `
|
|
3958
|
-
*
|
|
3959
|
-
*
|
|
6068
|
+
* when provided (issue #182). Three states for `sessionId`:
|
|
6069
|
+
* - omitted (`undefined`) → don't send the field, leave the persisted
|
|
6070
|
+
* session untouched (unused today; kept for API symmetry).
|
|
6071
|
+
* - a real id (`string`) → send it, update the persisted session (the
|
|
6072
|
+
* turn-failure call sites: an errored OpenCode turn).
|
|
6073
|
+
* - explicit `null` → send it, CLEAR the persisted session (issue
|
|
6074
|
+
* #485's dispatch-handoff-failure call site: the session id still
|
|
6075
|
+
* exists but is wedged, so the next attempt must get a fresh one
|
|
6076
|
+
* instead of reusing it — see WI-1's server-side null-clearing PATCH).
|
|
3960
6077
|
*/
|
|
3961
|
-
async markFailed(conversationId, messageId, sessionId, error2, usage) {
|
|
6078
|
+
async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
|
|
3962
6079
|
const body = { status: "failed" };
|
|
3963
|
-
if (sessionId
|
|
6080
|
+
if (sessionId === null) {
|
|
6081
|
+
body.opencode_session_id = null;
|
|
6082
|
+
} else if (sessionId !== void 0) {
|
|
6083
|
+
Object.assign(body, this.sessionIdBody(sessionId, conversationId, messageId, "failed"));
|
|
6084
|
+
}
|
|
3964
6085
|
if (error2 !== void 0) body.error = error2;
|
|
3965
6086
|
if (usage) Object.assign(body, usage);
|
|
6087
|
+
if (failure) {
|
|
6088
|
+
body.failure_kind = failure.kind;
|
|
6089
|
+
body.failure_provider_id = failure.providerId;
|
|
6090
|
+
body.failure_model_id = failure.modelId;
|
|
6091
|
+
body.failure_reason = failure.reason;
|
|
6092
|
+
}
|
|
3966
6093
|
await this.callWithRetry(
|
|
3967
6094
|
"marking message as failed",
|
|
3968
6095
|
() => this.fetchImpl(
|
|
@@ -3975,6 +6102,29 @@ var ChannelDriver = class {
|
|
|
3975
6102
|
)
|
|
3976
6103
|
);
|
|
3977
6104
|
}
|
|
6105
|
+
/**
|
|
6106
|
+
* Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
|
|
6107
|
+
*
|
|
6108
|
+
* `messageFailure` alone (structured OpenCode error → `model_auth`) covers
|
|
6109
|
+
* most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
|
|
6110
|
+
* to the P1-2b zero-provider check — one extra loopback call to
|
|
6111
|
+
* `hasAnyConfiguredProvider`, only reached when the structured classifier
|
|
6112
|
+
* couldn't place it. Fails open (never throws): a fallback probe failure
|
|
6113
|
+
* (`null`/indeterminate) leaves the classification `null`, which produces
|
|
6114
|
+
* today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
|
|
6115
|
+
*/
|
|
6116
|
+
async classifyModelAuthFailure(messages, userMessageId) {
|
|
6117
|
+
const classified = messageFailure(messages, userMessageId);
|
|
6118
|
+
if (classified != null) return classified;
|
|
6119
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
6120
|
+
const hasProvider = await hasAnyConfiguredProvider(this.port);
|
|
6121
|
+
return applyZeroProviderFallback(
|
|
6122
|
+
classified,
|
|
6123
|
+
hasProvider,
|
|
6124
|
+
reply?.info?.providerID ?? null,
|
|
6125
|
+
reply?.info?.modelID ?? null
|
|
6126
|
+
);
|
|
6127
|
+
}
|
|
3978
6128
|
/**
|
|
3979
6129
|
* Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
|
|
3980
6130
|
* (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
|
|
@@ -4127,10 +6277,16 @@ var ChannelDriver = class {
|
|
|
4127
6277
|
import chalk5 from "chalk";
|
|
4128
6278
|
import ora2 from "ora";
|
|
4129
6279
|
import { select as select2 } from "@inquirer/prompts";
|
|
6280
|
+
var INTERACTIVE_START_TIMEOUT_MS = 3e4;
|
|
4130
6281
|
async function ensureOpenCodeRunning(ctx) {
|
|
4131
6282
|
const healthCheck = await checkOpenCodeHealth(ctx.port);
|
|
4132
6283
|
if (healthCheck.healthy) {
|
|
4133
|
-
return {
|
|
6284
|
+
return {
|
|
6285
|
+
port: ctx.port,
|
|
6286
|
+
process: null,
|
|
6287
|
+
version: healthCheck.version ?? null,
|
|
6288
|
+
notReadyReason: null
|
|
6289
|
+
};
|
|
4134
6290
|
}
|
|
4135
6291
|
const runningInstances = await findHealthyOpenCodeInstances();
|
|
4136
6292
|
if (runningInstances.length > 0) {
|
|
@@ -4151,7 +6307,7 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4151
6307
|
console.log(chalk5.yellow("Tip: Run with the correct port:"));
|
|
4152
6308
|
console.log(
|
|
4153
6309
|
chalk5.dim(
|
|
4154
|
-
` ${getCliName()} run --
|
|
6310
|
+
` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
|
|
4155
6311
|
)
|
|
4156
6312
|
);
|
|
4157
6313
|
}
|
|
@@ -4171,14 +6327,22 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4171
6327
|
if (!ctx.interactive) {
|
|
4172
6328
|
ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
|
|
4173
6329
|
const proc = await startOpenCode(ctx.port);
|
|
4174
|
-
const health = await waitForOpenCodeHealth(ctx.port,
|
|
6330
|
+
const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
|
|
4175
6331
|
if (!health.healthy) {
|
|
4176
|
-
|
|
4177
|
-
|
|
4178
|
-
|
|
6332
|
+
return {
|
|
6333
|
+
port: ctx.port,
|
|
6334
|
+
process: proc,
|
|
6335
|
+
version: null,
|
|
6336
|
+
notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
|
|
6337
|
+
};
|
|
4179
6338
|
}
|
|
4180
6339
|
ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
|
|
4181
|
-
return {
|
|
6340
|
+
return {
|
|
6341
|
+
port: ctx.port,
|
|
6342
|
+
process: proc,
|
|
6343
|
+
version: health.version ?? null,
|
|
6344
|
+
notReadyReason: null
|
|
6345
|
+
};
|
|
4182
6346
|
}
|
|
4183
6347
|
let port = ctx.port;
|
|
4184
6348
|
if (isPortInUse(port)) {
|
|
@@ -4231,122 +6395,15 @@ Port ${port} is already in use.`));
|
|
|
4231
6395
|
if (action === "start") {
|
|
4232
6396
|
const spinner = ora2("Starting OpenCode...").start();
|
|
4233
6397
|
const proc = await startOpenCode(port);
|
|
4234
|
-
const health = await waitForOpenCodeHealth(port,
|
|
6398
|
+
const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
|
|
4235
6399
|
if (!health.healthy) {
|
|
4236
6400
|
spinner.fail("Failed to start OpenCode");
|
|
4237
6401
|
throw new Error("OpenCode failed to start");
|
|
4238
6402
|
}
|
|
4239
6403
|
spinner.stop();
|
|
4240
|
-
return { port, process: proc, version: health.version ?? null };
|
|
4241
|
-
}
|
|
4242
|
-
return { port, process: null, version: null };
|
|
4243
|
-
}
|
|
4244
|
-
|
|
4245
|
-
// src/commands/agent-lookup.ts
|
|
4246
|
-
async function readErrorMessage(response) {
|
|
4247
|
-
const text = await response.text().catch(() => "");
|
|
4248
|
-
if (!text) return response.statusText || void 0;
|
|
4249
|
-
try {
|
|
4250
|
-
const data = JSON.parse(text);
|
|
4251
|
-
const message = data.message ?? data.error;
|
|
4252
|
-
if (typeof message === "string" && message.trim()) {
|
|
4253
|
-
return message;
|
|
4254
|
-
}
|
|
4255
|
-
} catch {
|
|
4256
|
-
}
|
|
4257
|
-
return text.trim() || response.statusText || void 0;
|
|
4258
|
-
}
|
|
4259
|
-
function authFailureHint(apiUrl, serverMessage) {
|
|
4260
|
-
const reason = serverMessage ? `: ${serverMessage}` : "";
|
|
4261
|
-
return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
|
|
4262
|
-
}
|
|
4263
|
-
async function resolveAgentIdFromKey(authHeader) {
|
|
4264
|
-
const apiUrl = getApiUrlConfig();
|
|
4265
|
-
try {
|
|
4266
|
-
const response = await fetch(`${apiUrl}/me`, {
|
|
4267
|
-
headers: { Authorization: authHeader }
|
|
4268
|
-
});
|
|
4269
|
-
if (response.status === 401) {
|
|
4270
|
-
const serverMessage = await readErrorMessage(response);
|
|
4271
|
-
return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
4272
|
-
}
|
|
4273
|
-
if (!response.ok) {
|
|
4274
|
-
const serverMessage = await readErrorMessage(response);
|
|
4275
|
-
return {
|
|
4276
|
-
error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
4277
|
-
};
|
|
4278
|
-
}
|
|
4279
|
-
const data = await response.json();
|
|
4280
|
-
if (data.auth_type === "agent_key" && data.agent_id) {
|
|
4281
|
-
return { agent_id: data.agent_id };
|
|
4282
|
-
}
|
|
4283
|
-
return {
|
|
4284
|
-
error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
|
|
4285
|
-
};
|
|
4286
|
-
} catch (error2) {
|
|
4287
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
4288
|
-
return { error: `Failed to resolve runner from key: ${message}` };
|
|
4289
|
-
}
|
|
4290
|
-
}
|
|
4291
|
-
async function notifyAgentDisconnected(agentId, authHeader) {
|
|
4292
|
-
const apiUrl = getApiUrlConfig();
|
|
4293
|
-
try {
|
|
4294
|
-
const response = await fetch(`${apiUrl}/runners/${agentId}/disconnect`, {
|
|
4295
|
-
method: "POST",
|
|
4296
|
-
headers: { Authorization: authHeader }
|
|
4297
|
-
});
|
|
4298
|
-
if (!response.ok) {
|
|
4299
|
-
const serverMessage = await readErrorMessage(response);
|
|
4300
|
-
return {
|
|
4301
|
-
ok: false,
|
|
4302
|
-
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
4303
|
-
};
|
|
4304
|
-
}
|
|
4305
|
-
return { ok: true };
|
|
4306
|
-
} catch (error2) {
|
|
4307
|
-
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
4308
|
-
}
|
|
4309
|
-
}
|
|
4310
|
-
async function getAgentInfo(agentId, authHeader) {
|
|
4311
|
-
const apiUrl = getApiUrlConfig();
|
|
4312
|
-
try {
|
|
4313
|
-
const response = await fetch(`${apiUrl}/runners/${agentId}`, {
|
|
4314
|
-
headers: { Authorization: authHeader }
|
|
4315
|
-
});
|
|
4316
|
-
if (response.status === 401) {
|
|
4317
|
-
const serverMessage = await readErrorMessage(response);
|
|
4318
|
-
return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
4319
|
-
}
|
|
4320
|
-
if (response.status === 403) {
|
|
4321
|
-
const serverMessage = await readErrorMessage(response);
|
|
4322
|
-
return {
|
|
4323
|
-
valid: false,
|
|
4324
|
-
error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
|
|
4325
|
-
};
|
|
4326
|
-
}
|
|
4327
|
-
if (response.status === 404) {
|
|
4328
|
-
const serverMessage = await readErrorMessage(response);
|
|
4329
|
-
return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
|
|
4330
|
-
}
|
|
4331
|
-
if (!response.ok) {
|
|
4332
|
-
const serverMessage = await readErrorMessage(response);
|
|
4333
|
-
return {
|
|
4334
|
-
valid: false,
|
|
4335
|
-
error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
4336
|
-
};
|
|
4337
|
-
}
|
|
4338
|
-
const agent = await response.json();
|
|
4339
|
-
if (agent.agent_type !== "local") {
|
|
4340
|
-
return {
|
|
4341
|
-
valid: false,
|
|
4342
|
-
error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
|
|
4343
|
-
};
|
|
4344
|
-
}
|
|
4345
|
-
return { valid: true, agent };
|
|
4346
|
-
} catch (error2) {
|
|
4347
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
4348
|
-
return { valid: false, error: `Failed to validate runner: ${message}` };
|
|
6404
|
+
return { port, process: proc, version: health.version ?? null, notReadyReason: null };
|
|
4349
6405
|
}
|
|
6406
|
+
return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
|
|
4350
6407
|
}
|
|
4351
6408
|
|
|
4352
6409
|
// src/commands/run.ts
|
|
@@ -4354,6 +6411,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
|
|
|
4354
6411
|
var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
|
|
4355
6412
|
var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
|
|
4356
6413
|
var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
|
|
6414
|
+
var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
|
|
4357
6415
|
function resolveLogLevel(options) {
|
|
4358
6416
|
const accepted = Object.keys(LOG_LEVELS);
|
|
4359
6417
|
const validate = (value, source) => {
|
|
@@ -4377,6 +6435,63 @@ function resolveLogLevel(options) {
|
|
|
4377
6435
|
}
|
|
4378
6436
|
return "info";
|
|
4379
6437
|
}
|
|
6438
|
+
function resolveFileSyncDirectories(raw, homeDir) {
|
|
6439
|
+
const directories = [];
|
|
6440
|
+
for (const entry of raw ?? []) {
|
|
6441
|
+
const trimmed = entry.trim();
|
|
6442
|
+
if (trimmed === "") {
|
|
6443
|
+
throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
|
|
6444
|
+
}
|
|
6445
|
+
const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join3(homeDir, trimmed.slice(2)) : trimmed;
|
|
6446
|
+
if (!isAbsolute2(expanded)) {
|
|
6447
|
+
throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
|
|
6448
|
+
}
|
|
6449
|
+
const normalized = resolvePath(expanded);
|
|
6450
|
+
if (parse(normalized).root === normalized) {
|
|
6451
|
+
throw new Error(
|
|
6452
|
+
`--enable-file-sync-to will not allow-list the filesystem root ("${entry}"); name the specific directory the credentials belong in (for example ~/.claude)`
|
|
6453
|
+
);
|
|
6454
|
+
}
|
|
6455
|
+
if (!directories.includes(normalized)) {
|
|
6456
|
+
directories.push(normalized);
|
|
6457
|
+
}
|
|
6458
|
+
}
|
|
6459
|
+
if (directories.length > MAX_FILE_SYNC_DIRECTORIES) {
|
|
6460
|
+
throw new Error(
|
|
6461
|
+
`--enable-file-sync-to accepts at most ${MAX_FILE_SYNC_DIRECTORIES} directories; got ${directories.length}`
|
|
6462
|
+
);
|
|
6463
|
+
}
|
|
6464
|
+
return directories;
|
|
6465
|
+
}
|
|
6466
|
+
var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
|
|
6467
|
+
var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
|
|
6468
|
+
var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
|
|
6469
|
+
function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
|
|
6470
|
+
const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
|
|
6471
|
+
let raw;
|
|
6472
|
+
let source;
|
|
6473
|
+
if (options.opencodeStartTimeout !== void 0) {
|
|
6474
|
+
raw = options.opencodeStartTimeout;
|
|
6475
|
+
source = "--opencode-start-timeout";
|
|
6476
|
+
} else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
|
|
6477
|
+
raw = env[OPENCODE_START_TIMEOUT_ENV];
|
|
6478
|
+
source = OPENCODE_START_TIMEOUT_ENV;
|
|
6479
|
+
} else {
|
|
6480
|
+
return { timeoutMs: defaultMs, warnings: [] };
|
|
6481
|
+
}
|
|
6482
|
+
const trimmed = raw.trim();
|
|
6483
|
+
const seconds = Number(trimmed);
|
|
6484
|
+
const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
|
|
6485
|
+
if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
|
|
6486
|
+
return {
|
|
6487
|
+
timeoutMs: defaultMs,
|
|
6488
|
+
warnings: [
|
|
6489
|
+
`Ignoring invalid ${source} "${raw}": expected a positive integer number of seconds (at most ${MAX_OPENCODE_START_TIMEOUT_SECONDS}); using the default ${DEFAULT_OPENCODE_START_TIMEOUT_SECONDS}s`
|
|
6490
|
+
]
|
|
6491
|
+
};
|
|
6492
|
+
}
|
|
6493
|
+
return { timeoutMs: seconds * 1e3, warnings: [] };
|
|
6494
|
+
}
|
|
4380
6495
|
function meetsThreshold(state, level) {
|
|
4381
6496
|
return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
|
|
4382
6497
|
}
|
|
@@ -4398,6 +6513,10 @@ function log2(state, message, level = "info") {
|
|
|
4398
6513
|
function logActivity(state, entry) {
|
|
4399
6514
|
const level = entry.level ?? (entry.type === "error" ? "error" : "info");
|
|
4400
6515
|
if (!meetsThreshold(state, level)) return;
|
|
6516
|
+
forwardRunnerActivity(
|
|
6517
|
+
{ level, message: entry.message, error: entry.error },
|
|
6518
|
+
{ agentId: state.agentId, authHeader: state.authHeader }
|
|
6519
|
+
);
|
|
4401
6520
|
const fullEntry = {
|
|
4402
6521
|
...entry,
|
|
4403
6522
|
level,
|
|
@@ -4497,23 +6616,46 @@ async function handleAuthError(state, error2) {
|
|
|
4497
6616
|
}
|
|
4498
6617
|
async function driveChannels(state, driver) {
|
|
4499
6618
|
let idlePolls = 0;
|
|
6619
|
+
let idleMs = 0;
|
|
6620
|
+
let consecutiveDrainFailures = 0;
|
|
6621
|
+
let unreachableMs = 0;
|
|
4500
6622
|
let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
6623
|
+
let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
4501
6624
|
while (state.running) {
|
|
6625
|
+
const cycleStartedAtMs = performance.now();
|
|
6626
|
+
let idleThisCycle = false;
|
|
6627
|
+
let unreachableThisCycle = false;
|
|
4502
6628
|
if (state.connection?.reconnecting && state.connection.reconnectPromise) {
|
|
4503
6629
|
logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
|
|
4504
6630
|
if (state.interactive) displayStatus(state);
|
|
4505
6631
|
await state.connection.reconnectPromise;
|
|
4506
6632
|
}
|
|
6633
|
+
const carriedOverFileSync = driver.fileSyncActivity().inFlight;
|
|
6634
|
+
void driver.syncPendingFiles().catch(
|
|
6635
|
+
(error2) => logActivity(state, {
|
|
6636
|
+
type: "error",
|
|
6637
|
+
error: `Runner file sync failed: ${error2 instanceof Error ? error2.message : String(error2)}`
|
|
6638
|
+
})
|
|
6639
|
+
);
|
|
4507
6640
|
try {
|
|
4508
6641
|
const processed = await driver.drainPending();
|
|
6642
|
+
consecutiveDrainFailures = 0;
|
|
6643
|
+
unreachableMs = 0;
|
|
4509
6644
|
state.messageCount += processed;
|
|
4510
6645
|
const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
|
|
4511
6646
|
lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
4512
|
-
|
|
6647
|
+
const appliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
6648
|
+
const filesApplied = appliedFiles !== lastSeenAppliedFiles;
|
|
6649
|
+
const fileActivity = carriedOverFileSync || filesApplied;
|
|
6650
|
+
lastSeenAppliedFiles = appliedFiles;
|
|
6651
|
+
if (filesApplied) state.claudeUsageRearm?.();
|
|
6652
|
+
if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
|
|
4513
6653
|
idlePolls = 0;
|
|
6654
|
+
idleMs = 0;
|
|
4514
6655
|
if (processed > 0 && state.interactive) displayStatus(state);
|
|
4515
6656
|
} else if (state.idleTimeout !== null) {
|
|
4516
6657
|
idlePolls++;
|
|
6658
|
+
idleThisCycle = true;
|
|
4517
6659
|
if (idlePolls === 1) {
|
|
4518
6660
|
logActivity(state, {
|
|
4519
6661
|
type: "info",
|
|
@@ -4537,21 +6679,44 @@ async function driveChannels(state, driver) {
|
|
|
4537
6679
|
const errorMessage = error2 instanceof Error ? error2.message : String(error2);
|
|
4538
6680
|
logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
|
|
4539
6681
|
if (state.interactive) displayStatus(state);
|
|
4540
|
-
|
|
4541
|
-
|
|
4542
|
-
|
|
4543
|
-
|
|
4544
|
-
|
|
4545
|
-
|
|
4546
|
-
if (
|
|
4547
|
-
|
|
6682
|
+
if (driver.hasInFlightWatchers()) {
|
|
6683
|
+
consecutiveDrainFailures = 0;
|
|
6684
|
+
unreachableMs = 0;
|
|
6685
|
+
} else if (state.idleTimeout !== null) {
|
|
6686
|
+
consecutiveDrainFailures++;
|
|
6687
|
+
unreachableThisCycle = true;
|
|
6688
|
+
if (consecutiveDrainFailures === 1) {
|
|
6689
|
+
logActivity(state, {
|
|
6690
|
+
type: "info",
|
|
6691
|
+
message: `Cannot reach Evident, will exit if this persists past the idle timeout (timeout: ${state.idleTimeout}s)...`
|
|
6692
|
+
});
|
|
6693
|
+
if (state.interactive) displayStatus(state);
|
|
6694
|
+
}
|
|
4548
6695
|
}
|
|
4549
6696
|
}
|
|
6697
|
+
await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
|
|
6698
|
+
const cycleMs = performance.now() - cycleStartedAtMs;
|
|
6699
|
+
if (idleThisCycle) idleMs += cycleMs;
|
|
6700
|
+
if (unreachableThisCycle) unreachableMs += cycleMs;
|
|
6701
|
+
if (state.idleTimeout !== null && consecutiveDrainFailures >= 2 && unreachableMs > state.idleTimeout * 1e3) {
|
|
6702
|
+
logActivity(state, {
|
|
6703
|
+
type: "info",
|
|
6704
|
+
level: "warn",
|
|
6705
|
+
message: `Exiting: could not reach Evident for ${consecutiveDrainFailures} consecutive polls (${Math.round(unreachableMs / 1e3)}s)`
|
|
6706
|
+
});
|
|
6707
|
+
if (state.interactive) displayStatus(state);
|
|
6708
|
+
break;
|
|
6709
|
+
}
|
|
6710
|
+
if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
|
|
6711
|
+
logActivity(state, { type: "info", message: "Idle timeout reached" });
|
|
6712
|
+
if (state.interactive) displayStatus(state);
|
|
6713
|
+
break;
|
|
6714
|
+
}
|
|
4550
6715
|
}
|
|
4551
6716
|
}
|
|
4552
6717
|
var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
|
|
4553
|
-
async function runSweep(state, driver,
|
|
4554
|
-
const mode = `age=${
|
|
6718
|
+
async function runSweep(state, driver, config) {
|
|
6719
|
+
const mode = `age=${config.maxAgeMs ?? "\u2014"} count=${config.maxCount ?? "\u2014"}`;
|
|
4555
6720
|
try {
|
|
4556
6721
|
const sessions = await listSessions(state.port);
|
|
4557
6722
|
if (sessions === null) {
|
|
@@ -4564,8 +6729,8 @@ async function runSweep(state, driver, config2) {
|
|
|
4564
6729
|
const toDelete = selectSessionsToDelete(
|
|
4565
6730
|
sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
|
|
4566
6731
|
{
|
|
4567
|
-
maxAgeMs:
|
|
4568
|
-
maxCount:
|
|
6732
|
+
maxAgeMs: config.maxAgeMs,
|
|
6733
|
+
maxCount: config.maxCount,
|
|
4569
6734
|
nowMs: Date.now(),
|
|
4570
6735
|
protectedIds: driver.protectedSessionIds()
|
|
4571
6736
|
}
|
|
@@ -4601,7 +6766,7 @@ async function runSweep(state, driver, config2) {
|
|
|
4601
6766
|
}
|
|
4602
6767
|
}
|
|
4603
6768
|
function scheduleSessionCleanup(state, driver, options) {
|
|
4604
|
-
const
|
|
6769
|
+
const config = resolveSessionCleanupConfig(
|
|
4605
6770
|
{
|
|
4606
6771
|
maxAge: options.sessionCleanupMaxAge,
|
|
4607
6772
|
maxCount: options.sessionCleanupMaxCount,
|
|
@@ -4609,21 +6774,129 @@ function scheduleSessionCleanup(state, driver, options) {
|
|
|
4609
6774
|
},
|
|
4610
6775
|
process.env
|
|
4611
6776
|
);
|
|
4612
|
-
for (const warning2 of
|
|
6777
|
+
for (const warning2 of config.warnings) {
|
|
4613
6778
|
logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
|
|
4614
6779
|
}
|
|
4615
|
-
if (!
|
|
6780
|
+
if (!config.enabled) return;
|
|
4616
6781
|
logActivity(state, {
|
|
4617
6782
|
type: "info",
|
|
4618
|
-
message: `Session cleanup enabled (age=${
|
|
6783
|
+
message: `Session cleanup enabled (age=${config.maxAgeMs ?? "\u2014"}, count=${config.maxCount ?? "\u2014"}, interval=${config.intervalMs}ms)`
|
|
4619
6784
|
});
|
|
4620
|
-
const interval = setInterval(() => void runSweep(state, driver,
|
|
6785
|
+
const interval = setInterval(() => void runSweep(state, driver, config), config.intervalMs);
|
|
4621
6786
|
const firstSweep = setTimeout(
|
|
4622
|
-
() => void runSweep(state, driver,
|
|
6787
|
+
() => void runSweep(state, driver, config),
|
|
4623
6788
|
SESSION_CLEANUP_FIRST_SWEEP_MS
|
|
4624
6789
|
);
|
|
4625
6790
|
state.sessionCleanupTimers.push(interval, firstSweep);
|
|
4626
6791
|
}
|
|
6792
|
+
function claudeUsageFailureStreakSuffix(consecutiveFailures) {
|
|
6793
|
+
return consecutiveFailures > 1 ? ` (${consecutiveFailures} consecutive failures)` : "";
|
|
6794
|
+
}
|
|
6795
|
+
function scheduleClaudeUsageReporting(state, options) {
|
|
6796
|
+
const { mode, warnings } = resolveClaudeUsageReportingMode(
|
|
6797
|
+
options.claudeUsageReporting,
|
|
6798
|
+
process.env
|
|
6799
|
+
);
|
|
6800
|
+
for (const warning2 of warnings) {
|
|
6801
|
+
logActivity(state, {
|
|
6802
|
+
type: "info",
|
|
6803
|
+
level: "warn",
|
|
6804
|
+
message: `Claude usage reporting: ${warning2}`
|
|
6805
|
+
});
|
|
6806
|
+
}
|
|
6807
|
+
if (mode === "off") {
|
|
6808
|
+
logActivity(state, {
|
|
6809
|
+
type: "info",
|
|
6810
|
+
level: "debug",
|
|
6811
|
+
message: "Claude usage reporting is off (--claude-usage-reporting off)"
|
|
6812
|
+
});
|
|
6813
|
+
return null;
|
|
6814
|
+
}
|
|
6815
|
+
let consecutiveFailures = 0;
|
|
6816
|
+
let armed = false;
|
|
6817
|
+
let rearmRequested = false;
|
|
6818
|
+
const scheduleNextTick = () => {
|
|
6819
|
+
armed = true;
|
|
6820
|
+
rearmRequested = false;
|
|
6821
|
+
state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
|
|
6822
|
+
};
|
|
6823
|
+
const rearm = () => {
|
|
6824
|
+
if (armed) {
|
|
6825
|
+
rearmRequested = true;
|
|
6826
|
+
return;
|
|
6827
|
+
}
|
|
6828
|
+
rearmRequested = false;
|
|
6829
|
+
armed = true;
|
|
6830
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6831
|
+
};
|
|
6832
|
+
const tick = async (isProbe) => {
|
|
6833
|
+
try {
|
|
6834
|
+
const usage = await getClaudeUsage();
|
|
6835
|
+
const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
|
|
6836
|
+
if (result.ok) {
|
|
6837
|
+
if (consecutiveFailures > 0) {
|
|
6838
|
+
logActivity(state, {
|
|
6839
|
+
type: "info",
|
|
6840
|
+
level: "info",
|
|
6841
|
+
message: "Claude usage reporting recovered"
|
|
6842
|
+
});
|
|
6843
|
+
}
|
|
6844
|
+
consecutiveFailures = 0;
|
|
6845
|
+
logActivity(state, {
|
|
6846
|
+
type: "info",
|
|
6847
|
+
level: "debug",
|
|
6848
|
+
message: "Reported Claude usage to Evident"
|
|
6849
|
+
});
|
|
6850
|
+
} else {
|
|
6851
|
+
consecutiveFailures++;
|
|
6852
|
+
logActivity(state, {
|
|
6853
|
+
type: "info",
|
|
6854
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6855
|
+
message: `Failed to report Claude usage: ${result.error}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6856
|
+
});
|
|
6857
|
+
}
|
|
6858
|
+
scheduleNextTick();
|
|
6859
|
+
} catch (error2) {
|
|
6860
|
+
if (error2 instanceof ClaudeUsageError && isLocalCredentialProblem(error2)) {
|
|
6861
|
+
if (mode === "on") {
|
|
6862
|
+
logActivity(state, {
|
|
6863
|
+
type: "info",
|
|
6864
|
+
level: "warn",
|
|
6865
|
+
message: "Claude usage reporting is forced on but no usable Claude Code login was found \u2014 run `claude` to sign in; reporting will keep retrying"
|
|
6866
|
+
});
|
|
6867
|
+
scheduleNextTick();
|
|
6868
|
+
} else if (isProbe) {
|
|
6869
|
+
logActivity(state, {
|
|
6870
|
+
type: "info",
|
|
6871
|
+
level: "debug",
|
|
6872
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6873
|
+
});
|
|
6874
|
+
armed = false;
|
|
6875
|
+
if (rearmRequested) rearm();
|
|
6876
|
+
} else {
|
|
6877
|
+
logActivity(state, {
|
|
6878
|
+
type: "info",
|
|
6879
|
+
level: "debug",
|
|
6880
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6881
|
+
});
|
|
6882
|
+
scheduleNextTick();
|
|
6883
|
+
}
|
|
6884
|
+
} else {
|
|
6885
|
+
consecutiveFailures++;
|
|
6886
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
6887
|
+
logActivity(state, {
|
|
6888
|
+
type: "info",
|
|
6889
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6890
|
+
message: `Claude usage reporting failed: ${message}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6891
|
+
});
|
|
6892
|
+
scheduleNextTick();
|
|
6893
|
+
}
|
|
6894
|
+
}
|
|
6895
|
+
};
|
|
6896
|
+
armed = true;
|
|
6897
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6898
|
+
return rearm;
|
|
6899
|
+
}
|
|
4627
6900
|
async function notifyOffline(state) {
|
|
4628
6901
|
if (!state.agentId || !state.authHeader) return;
|
|
4629
6902
|
if (!state.connected) {
|
|
@@ -4641,13 +6914,29 @@ async function notifyOffline(state) {
|
|
|
4641
6914
|
if (state.interactive) displayStatus(state);
|
|
4642
6915
|
}
|
|
4643
6916
|
}
|
|
6917
|
+
async function timeShutdownPhase(state, durations, name, run2) {
|
|
6918
|
+
const startedAt = Date.now();
|
|
6919
|
+
try {
|
|
6920
|
+
return await run2();
|
|
6921
|
+
} finally {
|
|
6922
|
+
const elapsedMs = Date.now() - startedAt;
|
|
6923
|
+
durations[name] = elapsedMs;
|
|
6924
|
+
log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
|
|
6925
|
+
}
|
|
6926
|
+
}
|
|
4644
6927
|
async function cleanup(state, opts = {}) {
|
|
6928
|
+
const durations = {};
|
|
4645
6929
|
state.running = false;
|
|
4646
6930
|
for (const timer of state.sessionCleanupTimers) {
|
|
4647
6931
|
clearInterval(timer);
|
|
4648
6932
|
clearTimeout(timer);
|
|
4649
6933
|
}
|
|
4650
6934
|
state.sessionCleanupTimers = [];
|
|
6935
|
+
if (state.claudeUsageTimer) {
|
|
6936
|
+
clearTimeout(state.claudeUsageTimer);
|
|
6937
|
+
state.claudeUsageTimer = null;
|
|
6938
|
+
}
|
|
6939
|
+
state.claudeUsageRearm = null;
|
|
4651
6940
|
if (opts.graceful && state.channelDriver) {
|
|
4652
6941
|
state.channelDriver.stop();
|
|
4653
6942
|
log2(state, "Draining in-flight channel work before shutdown...");
|
|
@@ -4655,7 +6944,13 @@ async function cleanup(state, opts = {}) {
|
|
|
4655
6944
|
logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
|
|
4656
6945
|
displayStatus(state);
|
|
4657
6946
|
}
|
|
4658
|
-
const
|
|
6947
|
+
const driver = state.channelDriver;
|
|
6948
|
+
const settled = await timeShutdownPhase(
|
|
6949
|
+
state,
|
|
6950
|
+
durations,
|
|
6951
|
+
"drain",
|
|
6952
|
+
() => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
|
|
6953
|
+
);
|
|
4659
6954
|
if (!settled) {
|
|
4660
6955
|
logActivity(state, {
|
|
4661
6956
|
type: "info",
|
|
@@ -4664,13 +6959,15 @@ async function cleanup(state, opts = {}) {
|
|
|
4664
6959
|
if (state.interactive) displayStatus(state);
|
|
4665
6960
|
}
|
|
4666
6961
|
}
|
|
4667
|
-
await notifyOffline(state);
|
|
6962
|
+
await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
|
|
4668
6963
|
if (state.connection) {
|
|
4669
|
-
state.connection
|
|
6964
|
+
const connection = state.connection;
|
|
6965
|
+
await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
|
|
4670
6966
|
state.connection = null;
|
|
4671
6967
|
}
|
|
4672
6968
|
if (state.opencodeProcess) {
|
|
4673
|
-
|
|
6969
|
+
const opencodeProcess = state.opencodeProcess;
|
|
6970
|
+
await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
|
|
4674
6971
|
if (state.interactive) {
|
|
4675
6972
|
logActivity(state, { type: "info", message: "Stopped OpenCode process" });
|
|
4676
6973
|
displayStatus(state);
|
|
@@ -4679,12 +6976,15 @@ async function cleanup(state, opts = {}) {
|
|
|
4679
6976
|
}
|
|
4680
6977
|
state.opencodeProcess = null;
|
|
4681
6978
|
}
|
|
6979
|
+
return durations;
|
|
4682
6980
|
}
|
|
4683
6981
|
async function run(options) {
|
|
4684
6982
|
const interactive = isInteractive(options.json);
|
|
4685
6983
|
let logLevel;
|
|
6984
|
+
let fileSyncDirectories;
|
|
4686
6985
|
try {
|
|
4687
6986
|
logLevel = resolveLogLevel(options);
|
|
6987
|
+
fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir3());
|
|
4688
6988
|
} catch (error2) {
|
|
4689
6989
|
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
4690
6990
|
if (options.json) {
|
|
@@ -4717,8 +7017,16 @@ async function run(options) {
|
|
|
4717
7017
|
messageCount: 0,
|
|
4718
7018
|
lastProxiedActivityAt: null,
|
|
4719
7019
|
sessionCleanupTimers: [],
|
|
7020
|
+
claudeUsageTimer: null,
|
|
7021
|
+
claudeUsageRearm: null,
|
|
4720
7022
|
authHeader: ""
|
|
4721
7023
|
};
|
|
7024
|
+
setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
|
|
7025
|
+
if (fileSyncDirectories.length > 0) {
|
|
7026
|
+
log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
|
|
7027
|
+
} else {
|
|
7028
|
+
log2(state, "File sync is disabled (no --enable-file-sync-to given)", "debug");
|
|
7029
|
+
}
|
|
4722
7030
|
if (!options.runner && options.agent) {
|
|
4723
7031
|
telemetry.info(
|
|
4724
7032
|
EventTypes.DEPRECATED_AGENT_FLAG_USED,
|
|
@@ -4742,14 +7050,38 @@ async function run(options) {
|
|
|
4742
7050
|
const handleSignal = async () => {
|
|
4743
7051
|
if (state.shuttingDown) return;
|
|
4744
7052
|
state.shuttingDown = true;
|
|
7053
|
+
const shutdownStartedAt = Date.now();
|
|
4745
7054
|
if (state.interactive) {
|
|
4746
7055
|
logActivity(state, { type: "info", message: "Shutting down..." });
|
|
4747
7056
|
displayStatus(state);
|
|
4748
7057
|
} else {
|
|
4749
7058
|
log2(state, "Shutting down...");
|
|
4750
7059
|
}
|
|
4751
|
-
await cleanup(state, { graceful: true });
|
|
4752
|
-
|
|
7060
|
+
const durations = await cleanup(state, { graceful: true });
|
|
7061
|
+
const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
|
|
7062
|
+
await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
|
|
7063
|
+
let timer;
|
|
7064
|
+
const flushed = shutdownTelemetry().then(
|
|
7065
|
+
() => true,
|
|
7066
|
+
(error2) => {
|
|
7067
|
+
log2(
|
|
7068
|
+
state,
|
|
7069
|
+
`Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
|
|
7070
|
+
"warn"
|
|
7071
|
+
);
|
|
7072
|
+
return true;
|
|
7073
|
+
}
|
|
7074
|
+
);
|
|
7075
|
+
const timedOut = new Promise((resolve3) => {
|
|
7076
|
+
timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
|
|
7077
|
+
});
|
|
7078
|
+
if (!await Promise.race([flushed, timedOut])) {
|
|
7079
|
+
log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
|
|
7080
|
+
}
|
|
7081
|
+
clearTimeout(timer);
|
|
7082
|
+
});
|
|
7083
|
+
const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
|
|
7084
|
+
log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
|
|
4753
7085
|
process.exit(0);
|
|
4754
7086
|
};
|
|
4755
7087
|
process.on("SIGINT", handleSignal);
|
|
@@ -4766,6 +7098,7 @@ async function run(options) {
|
|
|
4766
7098
|
console.log(chalk6.dim("Or run `evident login` for interactive authentication"));
|
|
4767
7099
|
blank();
|
|
4768
7100
|
process.exit(1);
|
|
7101
|
+
return;
|
|
4769
7102
|
}
|
|
4770
7103
|
blank();
|
|
4771
7104
|
console.log(chalk6.yellow("You are not logged in to Evident."));
|
|
@@ -4810,6 +7143,7 @@ async function run(options) {
|
|
|
4810
7143
|
} else {
|
|
4811
7144
|
printError(resolved.error || "Failed to resolve runner ID from key");
|
|
4812
7145
|
process.exit(1);
|
|
7146
|
+
return;
|
|
4813
7147
|
}
|
|
4814
7148
|
} else {
|
|
4815
7149
|
printError(
|
|
@@ -4823,6 +7157,7 @@ async function run(options) {
|
|
|
4823
7157
|
);
|
|
4824
7158
|
blank();
|
|
4825
7159
|
process.exit(1);
|
|
7160
|
+
return;
|
|
4826
7161
|
}
|
|
4827
7162
|
}
|
|
4828
7163
|
telemetry.info(
|
|
@@ -4863,25 +7198,67 @@ async function run(options) {
|
|
|
4863
7198
|
}
|
|
4864
7199
|
spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
|
|
4865
7200
|
state.agentName = validation.agent.name;
|
|
7201
|
+
const microvmId = process.env.MICROVM_ID?.trim();
|
|
7202
|
+
if (microvmId) {
|
|
7203
|
+
const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
|
|
7204
|
+
if (reported.ok) {
|
|
7205
|
+
log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
|
|
7206
|
+
} else {
|
|
7207
|
+
const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
|
|
7208
|
+
log2(state, message, "warn");
|
|
7209
|
+
if (state.interactive && !state.json) {
|
|
7210
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
7211
|
+
}
|
|
7212
|
+
}
|
|
7213
|
+
} else {
|
|
7214
|
+
log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
|
|
7215
|
+
}
|
|
7216
|
+
const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
|
|
7217
|
+
for (const warning2 of opencodeStartTimeoutWarnings) {
|
|
7218
|
+
logActivity(state, { type: "info", level: "warn", message: warning2 });
|
|
7219
|
+
}
|
|
4866
7220
|
const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
|
|
4867
7221
|
try {
|
|
4868
7222
|
const oc = await ensureOpenCodeRunning({
|
|
4869
7223
|
port: state.port,
|
|
4870
7224
|
interactive: state.interactive,
|
|
4871
7225
|
agentId: state.agentId,
|
|
4872
|
-
log: (message) => log2(state, message)
|
|
7226
|
+
log: (message) => log2(state, message),
|
|
7227
|
+
startTimeoutMs: opencodeStartTimeoutMs
|
|
4873
7228
|
});
|
|
4874
7229
|
state.port = oc.port;
|
|
4875
7230
|
state.opencodeProcess = oc.process;
|
|
4876
7231
|
state.opencodeVersion = oc.version;
|
|
4877
|
-
state.opencodeConnected = oc.
|
|
7232
|
+
state.opencodeConnected = oc.notReadyReason === null;
|
|
4878
7233
|
const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
|
|
4879
7234
|
ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
|
|
4880
|
-
|
|
4881
|
-
|
|
4882
|
-
|
|
4883
|
-
|
|
4884
|
-
|
|
7235
|
+
if (!state.interactive && oc.notReadyReason !== null) {
|
|
7236
|
+
const message = `OpenCode is not ready on port ${state.port}: ${oc.notReadyReason}. The runner will still come online, but messages will fail until opencode answers \u2014 raise the wait with --opencode-start-timeout <seconds> (env ${OPENCODE_START_TIMEOUT_ENV}).`;
|
|
7237
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
7238
|
+
} else {
|
|
7239
|
+
const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
|
|
7240
|
+
if (versionWarning) {
|
|
7241
|
+
log2(state, versionWarning, "warn");
|
|
7242
|
+
if (state.interactive && !state.json) {
|
|
7243
|
+
logActivity(state, { type: "info", level: "warn", message: versionWarning });
|
|
7244
|
+
}
|
|
7245
|
+
}
|
|
7246
|
+
const noProviderWarning = buildNoProviderWarning(
|
|
7247
|
+
await hasAnyConfiguredProvider(state.port)
|
|
7248
|
+
);
|
|
7249
|
+
if (noProviderWarning) {
|
|
7250
|
+
log2(state, noProviderWarning, "warn");
|
|
7251
|
+
if (state.interactive && !state.json) {
|
|
7252
|
+
logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
|
|
7253
|
+
blank();
|
|
7254
|
+
console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
|
|
7255
|
+
console.log(
|
|
7256
|
+
chalk6.dim(
|
|
7257
|
+
`Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
|
|
7258
|
+
)
|
|
7259
|
+
);
|
|
7260
|
+
blank();
|
|
7261
|
+
}
|
|
4885
7262
|
}
|
|
4886
7263
|
}
|
|
4887
7264
|
} catch (error2) {
|
|
@@ -4896,6 +7273,10 @@ async function run(options) {
|
|
|
4896
7273
|
getAuthHeader: () => state.authHeader,
|
|
4897
7274
|
conversationFilter: state.conversationFilter,
|
|
4898
7275
|
stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
|
|
7276
|
+
// #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
|
|
7277
|
+
// REJECTED with `file_sync_disabled` on the ack, not silently ignored.
|
|
7278
|
+
fileSyncDirectories,
|
|
7279
|
+
homeDir: homedir3(),
|
|
4899
7280
|
log: (entry) => (
|
|
4900
7281
|
// Thread the driver's real level straight through so `debug`/`warn`
|
|
4901
7282
|
// survive the sink filter (they no longer collapse to info). `type`
|
|
@@ -4922,6 +7303,18 @@ async function run(options) {
|
|
|
4922
7303
|
type: "info",
|
|
4923
7304
|
message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
|
|
4924
7305
|
});
|
|
7306
|
+
if (options.tunnelReadyFile) {
|
|
7307
|
+
const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
|
|
7308
|
+
if (marker.ok) {
|
|
7309
|
+
log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
|
|
7310
|
+
} else {
|
|
7311
|
+
log2(
|
|
7312
|
+
state,
|
|
7313
|
+
`Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
|
|
7314
|
+
"error"
|
|
7315
|
+
);
|
|
7316
|
+
}
|
|
7317
|
+
}
|
|
4925
7318
|
emitAgentConnected(state.agentId, {
|
|
4926
7319
|
port: state.port,
|
|
4927
7320
|
cli_version: getCliVersion(),
|
|
@@ -4977,6 +7370,12 @@ async function run(options) {
|
|
|
4977
7370
|
onDrainPing: () => {
|
|
4978
7371
|
if (!state.running) return;
|
|
4979
7372
|
logActivity(state, { type: "info", message: "Drain ping received \u2014 draining" });
|
|
7373
|
+
void channelDriver.syncPendingFiles().catch(
|
|
7374
|
+
(error2) => logActivity(state, {
|
|
7375
|
+
type: "error",
|
|
7376
|
+
error: `Runner file sync failed on ping: ${error2 instanceof Error ? error2.message : String(error2)}`
|
|
7377
|
+
})
|
|
7378
|
+
);
|
|
4980
7379
|
channelDriver.drainPending().then((processed) => {
|
|
4981
7380
|
if (processed > 0) {
|
|
4982
7381
|
state.messageCount += processed;
|
|
@@ -5006,6 +7405,7 @@ async function run(options) {
|
|
|
5006
7405
|
throw error2;
|
|
5007
7406
|
}
|
|
5008
7407
|
scheduleSessionCleanup(state, channelDriver, options);
|
|
7408
|
+
state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
|
|
5009
7409
|
if (!interactive || state.json) {
|
|
5010
7410
|
log2(state, "Driving channel messages...");
|
|
5011
7411
|
}
|
|
@@ -5060,10 +7460,18 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
|
|
|
5060
7460
|
program.command("login").description("Authenticate with Evident").option("--token", "Use token-based authentication (for CI/CD)").option("--no-browser", "Do not open the browser automatically").action(login);
|
|
5061
7461
|
program.command("logout").description("Remove stored credentials for the current endpoint").option("--all", "Remove stored credentials for all endpoints").action((options) => logout({ all: options.all }));
|
|
5062
7462
|
program.command("whoami").description("Show the currently logged in user").action(whoami);
|
|
5063
|
-
program.command("
|
|
7463
|
+
program.command("status").description("Check whether the configured credentials can reach Evident").option("--json", "Output in JSON format").action((options) => status({ json: options.json }));
|
|
7464
|
+
program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
|
|
7465
|
+
program.command("run").description("Connect to Evident and process messages").option("--runner [id]", "Runner ID to connect to (optional when EVIDENT_RUNNER_KEY is set)").option(
|
|
7466
|
+
"-a, --agent [id]",
|
|
7467
|
+
"Deprecated alias for --runner (still supported; --runner wins if both are given)"
|
|
7468
|
+
).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
|
|
5064
7469
|
"--log-level <level>",
|
|
5065
7470
|
"Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
|
|
5066
|
-
).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option(
|
|
7471
|
+
).option("-v, --verbose", "Alias for --log-level debug (ignored if --log-level is set)").option("-c, --conversation <id>", "Process only this specific conversation").option("--idle-timeout <seconds>", "Exit after N seconds idle").option(
|
|
7472
|
+
"--opencode-start-timeout <seconds>",
|
|
7473
|
+
"Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
|
|
7474
|
+
).option("--json", "Output in JSON format").option(
|
|
5067
7475
|
"--session-cleanup-max-age <duration>",
|
|
5068
7476
|
"Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
|
|
5069
7477
|
).option(
|
|
@@ -5072,6 +7480,17 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5072
7480
|
).option(
|
|
5073
7481
|
"--session-cleanup-interval <duration>",
|
|
5074
7482
|
"How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
|
|
7483
|
+
).option(
|
|
7484
|
+
"--claude-usage-reporting <mode>",
|
|
7485
|
+
"Report Claude subscription usage to Evident: auto | on | off (default: auto). Env: EVIDENT_CLAUDE_USAGE_REPORTING"
|
|
7486
|
+
).option(
|
|
7487
|
+
"--enable-file-sync-to <dir>",
|
|
7488
|
+
"Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
|
|
7489
|
+
(value, previous) => previous.concat([value]),
|
|
7490
|
+
[]
|
|
7491
|
+
).option(
|
|
7492
|
+
"--tunnel-ready-file <path>",
|
|
7493
|
+
"Path to write once the tunnel is connected (opt-in; unused unless an operator/image sets it)"
|
|
5075
7494
|
).action(
|
|
5076
7495
|
(options) => {
|
|
5077
7496
|
run({
|
|
@@ -5084,11 +7503,21 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5084
7503
|
verbose: options.verbose,
|
|
5085
7504
|
conversation: options.conversation,
|
|
5086
7505
|
idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
|
|
7506
|
+
// Raw string — validation/precedence is single-sourced in run.ts's
|
|
7507
|
+
// resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
|
|
7508
|
+
opencodeStartTimeout: options.opencodeStartTimeout,
|
|
5087
7509
|
json: options.json,
|
|
5088
7510
|
// Raw strings — the resolver in run.ts single-sources parsing (M1).
|
|
5089
7511
|
sessionCleanupMaxAge: options.sessionCleanupMaxAge,
|
|
5090
7512
|
sessionCleanupMaxCount: options.sessionCleanupMaxCount,
|
|
5091
|
-
sessionCleanupInterval: options.sessionCleanupInterval
|
|
7513
|
+
sessionCleanupInterval: options.sessionCleanupInterval,
|
|
7514
|
+
// Raw string — the resolver in run.ts single-sources parsing
|
|
7515
|
+
// (resolveClaudeUsageReportingMode).
|
|
7516
|
+
claudeUsageReporting: options.claudeUsageReporting,
|
|
7517
|
+
// Raw values — expansion/validation is single-sourced in run.ts's
|
|
7518
|
+
// resolveFileSyncDirectories.
|
|
7519
|
+
enableFileSyncTo: options.enableFileSyncTo,
|
|
7520
|
+
tunnelReadyFile: options.tunnelReadyFile
|
|
5092
7521
|
});
|
|
5093
7522
|
}
|
|
5094
7523
|
);
|