@evident-ai/cli 3.1.1-dev.14c6359 → 3.1.1-dev.1640e9c
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 +68 -13
- package/dist/index.js +3003 -599
- 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();
|
|
@@ -466,193 +532,20 @@ 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
|
-
};
|
|
650
|
-
|
|
651
535
|
// src/lib/auth.ts
|
|
652
536
|
async function getAuthCredentials() {
|
|
537
|
+
const runnerKey = process.env.EVIDENT_RUNNER_KEY;
|
|
653
538
|
const agentKey = process.env.EVIDENT_AGENT_KEY;
|
|
539
|
+
if (runnerKey) {
|
|
540
|
+
return {
|
|
541
|
+
token: runnerKey,
|
|
542
|
+
authType: "agent_key",
|
|
543
|
+
keySource: "runner_key",
|
|
544
|
+
notice: agentKey ? "Both EVIDENT_RUNNER_KEY and EVIDENT_AGENT_KEY are set; using EVIDENT_RUNNER_KEY." : void 0
|
|
545
|
+
};
|
|
546
|
+
}
|
|
654
547
|
if (agentKey) {
|
|
655
|
-
return { token: agentKey, authType: "agent_key" };
|
|
548
|
+
return { token: agentKey, authType: "agent_key", keySource: "agent_key" };
|
|
656
549
|
}
|
|
657
550
|
const userToken = process.env.EVIDENT_TOKEN;
|
|
658
551
|
if (userToken) {
|
|
@@ -682,46 +575,753 @@ function isInteractive(jsonOutput) {
|
|
|
682
575
|
return true;
|
|
683
576
|
}
|
|
684
577
|
|
|
685
|
-
// src/
|
|
686
|
-
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;
|
|
687
582
|
try {
|
|
688
|
-
const
|
|
689
|
-
|
|
690
|
-
|
|
691
|
-
|
|
692
|
-
if (!response.ok) {
|
|
693
|
-
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;
|
|
694
587
|
}
|
|
695
|
-
|
|
696
|
-
return { healthy: true, version: data.version };
|
|
697
|
-
} catch (error2) {
|
|
698
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
699
|
-
return { healthy: false, error: message };
|
|
588
|
+
} catch {
|
|
700
589
|
}
|
|
590
|
+
return text.trim() || response.statusText || void 0;
|
|
701
591
|
}
|
|
702
|
-
|
|
703
|
-
const
|
|
704
|
-
|
|
705
|
-
const health = await checkOpenCodeHealth(port);
|
|
706
|
-
if (health.healthy) {
|
|
707
|
-
return health;
|
|
708
|
-
}
|
|
709
|
-
await new Promise((resolve2) => setTimeout(resolve2, 1e3));
|
|
710
|
-
}
|
|
711
|
-
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 ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
|
|
712
595
|
}
|
|
713
|
-
|
|
714
|
-
|
|
715
|
-
|
|
716
|
-
|
|
717
|
-
|
|
718
|
-
|
|
596
|
+
async function resolveAgentIdFromKey(authHeader) {
|
|
597
|
+
const apiUrl = getApiUrlConfig();
|
|
598
|
+
try {
|
|
599
|
+
const response = await fetch(`${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 ${response.status}` };
|
|
1294
|
+
}
|
|
1295
|
+
const data = await response.json().catch(() => ({}));
|
|
1296
|
+
return { healthy: true, version: data.version };
|
|
1297
|
+
} catch (error2) {
|
|
1298
|
+
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
1299
|
+
return { healthy: false, error: message };
|
|
1300
|
+
}
|
|
1301
|
+
}
|
|
1302
|
+
async function waitForOpenCodeHealth(port, timeoutMs = 3e4) {
|
|
1303
|
+
const startTime = Date.now();
|
|
1304
|
+
while (Date.now() - startTime < timeoutMs) {
|
|
1305
|
+
const health = await checkOpenCodeHealth(port);
|
|
1306
|
+
if (health.healthy) {
|
|
1307
|
+
return health;
|
|
1308
|
+
}
|
|
1309
|
+
await new Promise((resolve3) => setTimeout(resolve3, 1e3));
|
|
1310
|
+
}
|
|
1311
|
+
return { healthy: false, error: "Timeout waiting for OpenCode to be healthy" };
|
|
1312
|
+
}
|
|
1313
|
+
|
|
1314
|
+
// src/lib/opencode/opencode-version-gate.ts
|
|
1315
|
+
var QUEUE_VALIDATED_OPENCODE_VERSIONS = ["1.17.11", "1.18.3"];
|
|
1316
|
+
function isQueueValidatedVersion(version2) {
|
|
1317
|
+
if (!version2) return false;
|
|
1318
|
+
return QUEUE_VALIDATED_OPENCODE_VERSIONS.includes(version2);
|
|
719
1319
|
}
|
|
720
1320
|
function buildOpenCodeVersionWarning(version2) {
|
|
721
1321
|
if (isQueueValidatedVersion(version2)) return null;
|
|
722
1322
|
const detected = version2 ? `v${version2}` : "unknown";
|
|
723
1323
|
const validated = QUEUE_VALIDATED_OPENCODE_VERSIONS.map((v) => `v${v}`).join(", ");
|
|
724
|
-
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: ${validated}). Native message queuing \u2014 which channel (Slack) message handling relies on \u2014 is unverified on this version; queued/follow-up messages may behave unexpectedly. Continuing anyway. Bumping the validated set requires re-running the queue validation.`;
|
|
725
1325
|
}
|
|
726
1326
|
|
|
727
1327
|
// src/lib/opencode/process.ts
|
|
@@ -1013,6 +1613,12 @@ async function promptOpenCodeInstall(interactive) {
|
|
|
1013
1613
|
return action;
|
|
1014
1614
|
}
|
|
1015
1615
|
|
|
1616
|
+
// src/lib/opencode/provider-check.ts
|
|
1617
|
+
function buildNoProviderWarning(hasProvider) {
|
|
1618
|
+
if (hasProvider !== false) return null;
|
|
1619
|
+
return "Warning: opencode has no authenticated model provider configured, so it won't be able to answer prompts. Run `opencode auth login` to set one up (see https://opencode.ai for details).";
|
|
1620
|
+
}
|
|
1621
|
+
|
|
1016
1622
|
// src/lib/opencode/session.ts
|
|
1017
1623
|
function opencodeBase(port) {
|
|
1018
1624
|
return `http://127.0.0.1:${port}`;
|
|
@@ -1215,6 +1821,11 @@ async function getModelAttachmentCapability(port, model) {
|
|
|
1215
1821
|
}
|
|
1216
1822
|
const entry = provider.models[modelId];
|
|
1217
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
|
+
}
|
|
1218
1829
|
return typeof entry.attachment === "boolean" ? entry.attachment : null;
|
|
1219
1830
|
} catch (err) {
|
|
1220
1831
|
console.error(
|
|
@@ -1243,6 +1854,16 @@ async function buildFileParts(attachments, capable) {
|
|
|
1243
1854
|
);
|
|
1244
1855
|
dataUrl = null;
|
|
1245
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
|
+
}
|
|
1246
1867
|
if (dataUrl == null) {
|
|
1247
1868
|
outcomes.push({ index: a.index, mime: a.mime, filename: a.filename, status: "failed" });
|
|
1248
1869
|
continue;
|
|
@@ -1324,7 +1945,7 @@ async function sendPromptAsync(port, sessionId, content, options, attachments) {
|
|
|
1324
1945
|
}
|
|
1325
1946
|
}
|
|
1326
1947
|
if (attempt < READ_BACK_ATTEMPTS - 1) {
|
|
1327
|
-
await new Promise((
|
|
1948
|
+
await new Promise((resolve3) => setTimeout(resolve3, READ_BACK_DELAY_MS));
|
|
1328
1949
|
}
|
|
1329
1950
|
}
|
|
1330
1951
|
return null;
|
|
@@ -1452,6 +2073,9 @@ function isPreamblePinnedRunning(messages, userMessageId) {
|
|
|
1452
2073
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1453
2074
|
return completedOf(reply) != null && finishOf(reply) === "tool-calls";
|
|
1454
2075
|
}
|
|
2076
|
+
function isB2AbandonmentConfirmed(params) {
|
|
2077
|
+
return params.pinnedForMs >= params.minPinnedMs && params.descendantOngoing === false;
|
|
2078
|
+
}
|
|
1455
2079
|
function messageError(messages, userMessageId) {
|
|
1456
2080
|
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
1457
2081
|
const error2 = errorOf(reply);
|
|
@@ -1465,12 +2089,94 @@ function messageError(messages, userMessageId) {
|
|
|
1465
2089
|
}
|
|
1466
2090
|
return "The agent run failed.";
|
|
1467
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
|
+
}
|
|
1468
2143
|
function hasRunningAssistantExcept(messages, exceptUserMessageId) {
|
|
1469
2144
|
if (!messages || messages.length === 0) return false;
|
|
1470
2145
|
return messages.some(
|
|
1471
2146
|
(m) => roleOf(m) === "assistant" && parentIdOf(m) !== exceptUserMessageId && isAssistantInFlight(m)
|
|
1472
2147
|
);
|
|
1473
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}): ${err instanceof Error ? err.message : String(err)}`
|
|
2176
|
+
);
|
|
2177
|
+
return null;
|
|
2178
|
+
}
|
|
2179
|
+
}
|
|
1474
2180
|
|
|
1475
2181
|
// src/lib/opencode/session-cleanup.ts
|
|
1476
2182
|
var DURATION_UNIT_MS = {
|
|
@@ -1629,10 +2335,11 @@ var StreamForwarder = class {
|
|
|
1629
2335
|
* Abort every in-flight stream (e.g. on WebSocket close).
|
|
1630
2336
|
*/
|
|
1631
2337
|
abortAll() {
|
|
1632
|
-
for (const stream of this.inflight.
|
|
2338
|
+
for (const [sid, stream] of this.inflight.entries()) {
|
|
1633
2339
|
try {
|
|
1634
2340
|
stream.abort();
|
|
1635
|
-
} catch {
|
|
2341
|
+
} catch (err) {
|
|
2342
|
+
log("error", "forwarder_abort_failed", { sid, ...errorFields(err) });
|
|
1636
2343
|
}
|
|
1637
2344
|
}
|
|
1638
2345
|
this.inflight.clear();
|
|
@@ -1666,12 +2373,12 @@ var StreamForwarder = class {
|
|
|
1666
2373
|
let endBody;
|
|
1667
2374
|
if (has_body) {
|
|
1668
2375
|
const chunks = [];
|
|
1669
|
-
bodyPromise = new Promise((
|
|
2376
|
+
bodyPromise = new Promise((resolve3) => {
|
|
1670
2377
|
pushBody = (buf) => {
|
|
1671
2378
|
chunks.push(buf);
|
|
1672
2379
|
};
|
|
1673
2380
|
endBody = () => {
|
|
1674
|
-
|
|
2381
|
+
resolve3(Buffer.concat(chunks));
|
|
1675
2382
|
};
|
|
1676
2383
|
});
|
|
1677
2384
|
}
|
|
@@ -1782,31 +2489,20 @@ function connectTunnel(options) {
|
|
|
1782
2489
|
onConnected,
|
|
1783
2490
|
onDisconnected,
|
|
1784
2491
|
onError,
|
|
1785
|
-
onRequest,
|
|
1786
2492
|
onResponse,
|
|
1787
2493
|
onInfo,
|
|
1788
2494
|
onDrainPing
|
|
1789
2495
|
} = options;
|
|
1790
2496
|
const tunnelUrl = getTunnelUrlConfig();
|
|
1791
2497
|
const url = `${tunnelUrl}/tunnel/${agentId}/connect`;
|
|
1792
|
-
return new Promise((
|
|
2498
|
+
return new Promise((resolve3, reject) => {
|
|
1793
2499
|
const ws = new WebSocket2(url, {
|
|
1794
2500
|
headers: {
|
|
1795
2501
|
Authorization: authHeader
|
|
1796
2502
|
}
|
|
1797
2503
|
});
|
|
1798
|
-
const streamStartTimes = /* @__PURE__ */ new Map();
|
|
1799
2504
|
const forwarder = new StreamForwarder(ws, port, {
|
|
1800
|
-
|
|
1801
|
-
if (path === TUNNEL_DRAIN_PING_PATH) return;
|
|
1802
|
-
streamStartTimes.set(sid, Date.now());
|
|
1803
|
-
onRequest?.(method, path, sid);
|
|
1804
|
-
},
|
|
1805
|
-
onHead: (sid, status) => {
|
|
1806
|
-
const startedAt = streamStartTimes.get(sid);
|
|
1807
|
-
streamStartTimes.delete(sid);
|
|
1808
|
-
onResponse?.(status, startedAt ? Date.now() - startedAt : 0, sid);
|
|
1809
|
-
},
|
|
2505
|
+
onHead: () => onResponse?.(),
|
|
1810
2506
|
onDrainPing: () => onDrainPing?.()
|
|
1811
2507
|
});
|
|
1812
2508
|
const connectionTimeout = setTimeout(() => {
|
|
@@ -1854,7 +2550,7 @@ function connectTunnel(options) {
|
|
|
1854
2550
|
clearTimeout(connectionTimeout);
|
|
1855
2551
|
const connectedAgentId = message.agent_id ?? agentId;
|
|
1856
2552
|
onConnected?.(connectedAgentId);
|
|
1857
|
-
|
|
2553
|
+
resolve3({
|
|
1858
2554
|
ws,
|
|
1859
2555
|
close: () => ws.close(1e3, "CLI shutdown")
|
|
1860
2556
|
});
|
|
@@ -1882,7 +2578,6 @@ function connectTunnel(options) {
|
|
|
1882
2578
|
ws.on("close", (code, reason) => {
|
|
1883
2579
|
const reasonStr = reason.toString() || upgradeRejection || (code === 1006 ? "abnormal closure" : "No reason provided");
|
|
1884
2580
|
forwarder.abortAll();
|
|
1885
|
-
streamStartTimes.clear();
|
|
1886
2581
|
onDisconnected?.(code, reasonStr);
|
|
1887
2582
|
});
|
|
1888
2583
|
});
|
|
@@ -1908,66 +2603,511 @@ var RunnerConnection = class {
|
|
|
1908
2603
|
get agentId() {
|
|
1909
2604
|
return this.resolvedAgentId;
|
|
1910
2605
|
}
|
|
1911
|
-
/** Establish the initial tunnel connection (with retry/backoff). */
|
|
1912
|
-
async connect() {
|
|
1913
|
-
await this.connectWithRetry(false);
|
|
2606
|
+
/** Establish the initial tunnel connection (with retry/backoff). */
|
|
2607
|
+
async connect() {
|
|
2608
|
+
await this.connectWithRetry(false);
|
|
2609
|
+
}
|
|
2610
|
+
/** Close the active connection (idempotent). */
|
|
2611
|
+
close() {
|
|
2612
|
+
if (this.connection) {
|
|
2613
|
+
try {
|
|
2614
|
+
this.connection.close();
|
|
2615
|
+
} catch (err) {
|
|
2616
|
+
log("error", "runner_connection_close_failed", {
|
|
2617
|
+
agent_id: this.resolvedAgentId,
|
|
2618
|
+
...errorFields(err)
|
|
2619
|
+
});
|
|
2620
|
+
}
|
|
2621
|
+
this.connection = null;
|
|
2622
|
+
}
|
|
2623
|
+
}
|
|
2624
|
+
async connectWithRetry(isReconnect) {
|
|
2625
|
+
if (isReconnect && this.reconnecting) return;
|
|
2626
|
+
this.reconnecting = true;
|
|
2627
|
+
this.close();
|
|
2628
|
+
const { events } = this.opts;
|
|
2629
|
+
while (this.opts.isRunning()) {
|
|
2630
|
+
try {
|
|
2631
|
+
this.connection = await connectTunnel({
|
|
2632
|
+
agentId: this.resolvedAgentId,
|
|
2633
|
+
authHeader: this.opts.getAuthHeader(),
|
|
2634
|
+
port: this.opts.port,
|
|
2635
|
+
onConnected: (agentId) => {
|
|
2636
|
+
this.reconnectAttempt = 0;
|
|
2637
|
+
this.reconnecting = false;
|
|
2638
|
+
this.resolvedAgentId = agentId;
|
|
2639
|
+
events.onConnected(agentId, isReconnect);
|
|
2640
|
+
},
|
|
2641
|
+
onDisconnected: (code, reason) => {
|
|
2642
|
+
events.onDisconnected(code, reason);
|
|
2643
|
+
if (this.opts.isRunning() && code !== 1e3 && !this.reconnecting) {
|
|
2644
|
+
this.reconnectPromise = this.connectWithRetry(true).catch((err) => {
|
|
2645
|
+
events.onError?.(`Reconnection failed: ${err.message}`);
|
|
2646
|
+
});
|
|
2647
|
+
}
|
|
2648
|
+
},
|
|
2649
|
+
onError: (error2) => events.onError?.(error2),
|
|
2650
|
+
onResponse: () => events.onResponse?.(),
|
|
2651
|
+
onDrainPing: () => events.onDrainPing?.(),
|
|
2652
|
+
onInfo: (message) => events.onInfo?.(message)
|
|
2653
|
+
});
|
|
2654
|
+
return;
|
|
2655
|
+
} catch (error2) {
|
|
2656
|
+
this.reconnectAttempt++;
|
|
2657
|
+
if (error2.message === "Unauthorized") {
|
|
2658
|
+
this.reconnecting = false;
|
|
2659
|
+
throw error2;
|
|
2660
|
+
}
|
|
2661
|
+
const delay = getReconnectDelay(this.reconnectAttempt);
|
|
2662
|
+
events.onReconnecting?.(this.reconnectAttempt);
|
|
2663
|
+
events.onError?.(`Connection failed, retrying in ${Math.round(delay / 1e3)}s...`);
|
|
2664
|
+
await this.sleep(delay);
|
|
2665
|
+
}
|
|
2666
|
+
}
|
|
2667
|
+
this.reconnecting = false;
|
|
2668
|
+
}
|
|
2669
|
+
};
|
|
2670
|
+
|
|
2671
|
+
// src/lib/tunnel/ready-marker.ts
|
|
2672
|
+
import { writeFileSync } from "fs";
|
|
2673
|
+
function writeTunnelReadyMarker(path, agentId) {
|
|
2674
|
+
try {
|
|
2675
|
+
writeFileSync(path, `${agentId}
|
|
2676
|
+
`);
|
|
2677
|
+
return { ok: true };
|
|
2678
|
+
} catch (error2) {
|
|
2679
|
+
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
2680
|
+
}
|
|
2681
|
+
}
|
|
2682
|
+
|
|
2683
|
+
// src/lib/claude-usage-reporting.ts
|
|
2684
|
+
var VALID_MODES = ["auto", "on", "off"];
|
|
2685
|
+
function resolveClaudeUsageReportingMode(flagValue, env) {
|
|
2686
|
+
const raw = flagValue ?? env.EVIDENT_CLAUDE_USAGE_REPORTING;
|
|
2687
|
+
if (raw === void 0 || raw === "") {
|
|
2688
|
+
return { mode: "auto", warnings: [] };
|
|
2689
|
+
}
|
|
2690
|
+
const normalized = raw.trim().toLowerCase();
|
|
2691
|
+
if (VALID_MODES.includes(normalized)) {
|
|
2692
|
+
return { mode: normalized, warnings: [] };
|
|
2693
|
+
}
|
|
2694
|
+
const source = flagValue !== void 0 ? "--claude-usage-reporting" : "EVIDENT_CLAUDE_USAGE_REPORTING";
|
|
2695
|
+
return {
|
|
2696
|
+
mode: "auto",
|
|
2697
|
+
warnings: [
|
|
2698
|
+
`Ignoring invalid ${source} "${raw}": expected one of ${VALID_MODES.join(", ")}; using auto`
|
|
2699
|
+
]
|
|
2700
|
+
};
|
|
2701
|
+
}
|
|
2702
|
+
var BASE_REPORT_DELAY_MS = 10 * 6e4;
|
|
2703
|
+
var REPORT_DELAY_JITTER_FRACTION = 0.2;
|
|
2704
|
+
function nextReportDelayMs(random = Math.random) {
|
|
2705
|
+
const jitterRangeMs = BASE_REPORT_DELAY_MS * REPORT_DELAY_JITTER_FRACTION;
|
|
2706
|
+
return BASE_REPORT_DELAY_MS - jitterRangeMs + random() * (2 * jitterRangeMs);
|
|
2707
|
+
}
|
|
2708
|
+
var FIRST_REPORT_DELAY_MS = 5e3 + Math.random() * 1e4;
|
|
2709
|
+
var CLAUDE_USAGE_FAILURE_REESCALATION_TICKS = 6;
|
|
2710
|
+
function claudeUsageFailureLogLevel(consecutiveFailures) {
|
|
2711
|
+
return consecutiveFailures === 1 || consecutiveFailures % CLAUDE_USAGE_FAILURE_REESCALATION_TICKS === 0 ? "warn" : "debug";
|
|
2712
|
+
}
|
|
2713
|
+
|
|
2714
|
+
// src/lib/channels/driver.ts
|
|
2715
|
+
import { homedir as homedir2 } from "os";
|
|
2716
|
+
|
|
2717
|
+
// src/lib/file-push.ts
|
|
2718
|
+
import { randomUUID } from "crypto";
|
|
2719
|
+
import { chmod, mkdir, open as open2, realpath, rename, unlink } from "fs/promises";
|
|
2720
|
+
import { basename, dirname as dirname2, isAbsolute, join as join2, relative, resolve as resolve2, sep } from "path";
|
|
2721
|
+
var FILE_MODE = 384;
|
|
2722
|
+
var DIRECTORY_MODE = 448;
|
|
2723
|
+
async function writePushedFile(request) {
|
|
2724
|
+
const { requestedPath, content, allowedDirectories, homeDir } = request;
|
|
2725
|
+
const bytes = content.byteLength;
|
|
2726
|
+
if (allowedDirectories.length === 0) {
|
|
2727
|
+
return refuse("file_sync_disabled", "File sync is not enabled on this runner.", {
|
|
2728
|
+
path: requestedPath,
|
|
2729
|
+
bytes
|
|
2730
|
+
});
|
|
2731
|
+
}
|
|
2732
|
+
if (bytes > MAX_FILE_PUSH_BYTES) {
|
|
2733
|
+
return refuse(
|
|
2734
|
+
"file_too_large",
|
|
2735
|
+
`File is ${bytes} bytes; the limit is ${MAX_FILE_PUSH_BYTES}.`,
|
|
2736
|
+
{
|
|
2737
|
+
path: requestedPath,
|
|
2738
|
+
bytes
|
|
2739
|
+
}
|
|
2740
|
+
);
|
|
2741
|
+
}
|
|
2742
|
+
const candidate = expandAndValidate(requestedPath, homeDir);
|
|
2743
|
+
if (candidate === null) {
|
|
2744
|
+
return refuse("invalid_path", "The requested path is not a valid absolute file path.", {
|
|
2745
|
+
path: requestedPath,
|
|
2746
|
+
bytes
|
|
2747
|
+
});
|
|
2748
|
+
}
|
|
2749
|
+
try {
|
|
2750
|
+
const { existingAncestor, missingSegments } = await resolveNearestExistingAncestor(
|
|
2751
|
+
dirname2(candidate)
|
|
2752
|
+
);
|
|
2753
|
+
const realTarget = join2(existingAncestor, ...missingSegments, basename(candidate));
|
|
2754
|
+
const allowedDirectory = await findContainingAllowedDirectory(allowedDirectories, realTarget);
|
|
2755
|
+
if (allowedDirectory === null) {
|
|
2756
|
+
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
2757
|
+
path: realTarget,
|
|
2758
|
+
bytes
|
|
2759
|
+
});
|
|
2760
|
+
}
|
|
2761
|
+
if (missingSegments.length > 0) {
|
|
2762
|
+
await createMissingDirectories(existingAncestor, missingSegments);
|
|
2763
|
+
const realParent = await realpath(dirname2(realTarget));
|
|
2764
|
+
if (realParent !== dirname2(realTarget) || !contains(allowedDirectory, realTarget)) {
|
|
2765
|
+
return refuse("path_not_allowed", "The runner does not allow writing to that location.", {
|
|
2766
|
+
path: realTarget,
|
|
2767
|
+
bytes,
|
|
2768
|
+
reason: "parent_changed_after_create"
|
|
2769
|
+
});
|
|
2770
|
+
}
|
|
2771
|
+
}
|
|
2772
|
+
await writeAtomically(realTarget, content);
|
|
2773
|
+
log("info", "file_push_written", { path: realTarget, bytes });
|
|
2774
|
+
return { ok: true, path: realTarget };
|
|
2775
|
+
} catch (err) {
|
|
2776
|
+
const errno = err.code ?? "UNKNOWN";
|
|
2777
|
+
return refuse("write_failed", `The runner could not write the file (${errno}).`, {
|
|
2778
|
+
path: candidate,
|
|
2779
|
+
bytes,
|
|
2780
|
+
errno,
|
|
2781
|
+
...errorFields(err)
|
|
2782
|
+
});
|
|
2783
|
+
}
|
|
2784
|
+
}
|
|
2785
|
+
function expandAndValidate(requestedPath, homeDir) {
|
|
2786
|
+
if (requestedPath.trim() === "" || requestedPath.includes("\0")) {
|
|
2787
|
+
return null;
|
|
2788
|
+
}
|
|
2789
|
+
const expanded = requestedPath === "~" ? homeDir : requestedPath.startsWith("~/") ? join2(homeDir, requestedPath.slice(2)) : requestedPath;
|
|
2790
|
+
if (expanded.split(/[/\\]/).includes("..")) {
|
|
2791
|
+
return null;
|
|
2792
|
+
}
|
|
2793
|
+
if (!isAbsolute(expanded)) {
|
|
2794
|
+
return null;
|
|
2795
|
+
}
|
|
2796
|
+
const candidate = resolve2(expanded);
|
|
2797
|
+
const name = basename(candidate);
|
|
2798
|
+
return name === "" || name === "." || name === ".." ? null : candidate;
|
|
2799
|
+
}
|
|
2800
|
+
async function resolveNearestExistingAncestor(directory) {
|
|
2801
|
+
const missingSegments = [];
|
|
2802
|
+
let current = directory;
|
|
2803
|
+
for (; ; ) {
|
|
2804
|
+
try {
|
|
2805
|
+
return { existingAncestor: await realpath(current), missingSegments };
|
|
2806
|
+
} catch (err) {
|
|
2807
|
+
const parent = dirname2(current);
|
|
2808
|
+
if (err.code !== "ENOENT" || parent === current) {
|
|
2809
|
+
throw err;
|
|
2810
|
+
}
|
|
2811
|
+
missingSegments.unshift(basename(current));
|
|
2812
|
+
current = parent;
|
|
2813
|
+
}
|
|
2814
|
+
}
|
|
2815
|
+
}
|
|
2816
|
+
async function findContainingAllowedDirectory(allowedDirectories, realTarget) {
|
|
2817
|
+
for (const directory of allowedDirectories) {
|
|
2818
|
+
if (!isAbsolute(directory)) {
|
|
2819
|
+
log("warn", "file_push_allowed_directory_skipped", { directory, reason: "not_absolute" });
|
|
2820
|
+
continue;
|
|
2821
|
+
}
|
|
2822
|
+
const realDirectory = await realpathCreatingIfMissing(directory);
|
|
2823
|
+
if (realDirectory !== null && contains(realDirectory, realTarget)) {
|
|
2824
|
+
return realDirectory;
|
|
2825
|
+
}
|
|
2826
|
+
}
|
|
2827
|
+
return null;
|
|
2828
|
+
}
|
|
2829
|
+
async function realpathCreatingIfMissing(directory) {
|
|
2830
|
+
try {
|
|
2831
|
+
return await realpath(directory);
|
|
2832
|
+
} catch (err) {
|
|
2833
|
+
if (err.code !== "ENOENT") {
|
|
2834
|
+
log("warn", "file_push_allowed_directory_skipped", {
|
|
2835
|
+
directory,
|
|
2836
|
+
reason: "unresolvable",
|
|
2837
|
+
...errorFields(err)
|
|
2838
|
+
});
|
|
2839
|
+
return null;
|
|
2840
|
+
}
|
|
2841
|
+
}
|
|
2842
|
+
try {
|
|
2843
|
+
await mkdir(directory, { recursive: true, mode: DIRECTORY_MODE });
|
|
2844
|
+
await chmod(directory, DIRECTORY_MODE);
|
|
2845
|
+
return await realpath(directory);
|
|
2846
|
+
} catch (err) {
|
|
2847
|
+
log("warn", "file_push_allowed_directory_skipped", {
|
|
2848
|
+
directory,
|
|
2849
|
+
reason: "create_failed",
|
|
2850
|
+
...errorFields(err)
|
|
2851
|
+
});
|
|
2852
|
+
return null;
|
|
2853
|
+
}
|
|
2854
|
+
}
|
|
2855
|
+
function contains(realDirectory, realTarget) {
|
|
2856
|
+
const rel = relative(realDirectory, realTarget);
|
|
2857
|
+
return rel !== "" && rel !== ".." && !rel.startsWith(`..${sep}`) && !isAbsolute(rel);
|
|
2858
|
+
}
|
|
2859
|
+
async function createMissingDirectories(existingAncestor, missingSegments) {
|
|
2860
|
+
let current = existingAncestor;
|
|
2861
|
+
for (const segment of missingSegments) {
|
|
2862
|
+
current = join2(current, segment);
|
|
2863
|
+
await mkdir(current, { recursive: true, mode: DIRECTORY_MODE });
|
|
2864
|
+
await chmod(current, DIRECTORY_MODE);
|
|
2865
|
+
}
|
|
2866
|
+
}
|
|
2867
|
+
async function writeAtomically(realTarget, content) {
|
|
2868
|
+
const temporaryPath = join2(dirname2(realTarget), `.evident-push-${randomUUID()}.tmp`);
|
|
2869
|
+
let handle;
|
|
2870
|
+
try {
|
|
2871
|
+
handle = await open2(temporaryPath, "wx", FILE_MODE);
|
|
2872
|
+
await handle.writeFile(content);
|
|
2873
|
+
await handle.chmod(FILE_MODE);
|
|
2874
|
+
await handle.close();
|
|
2875
|
+
handle = void 0;
|
|
2876
|
+
await rename(temporaryPath, realTarget);
|
|
2877
|
+
} catch (err) {
|
|
2878
|
+
await discardTemporaryFile(temporaryPath, handle);
|
|
2879
|
+
throw err;
|
|
2880
|
+
}
|
|
2881
|
+
}
|
|
2882
|
+
async function discardTemporaryFile(temporaryPath, handle) {
|
|
2883
|
+
try {
|
|
2884
|
+
await handle?.close();
|
|
2885
|
+
} catch (err) {
|
|
2886
|
+
log("warn", "file_push_temp_close_failed", { path: temporaryPath, ...errorFields(err) });
|
|
2887
|
+
}
|
|
2888
|
+
try {
|
|
2889
|
+
await unlink(temporaryPath);
|
|
2890
|
+
} catch (err) {
|
|
2891
|
+
const errno = err.code;
|
|
2892
|
+
if (errno !== "ENOENT" && errno !== "ENOTDIR") {
|
|
2893
|
+
log("warn", "file_push_temp_cleanup_failed", { path: temporaryPath, ...errorFields(err) });
|
|
2894
|
+
}
|
|
2895
|
+
}
|
|
2896
|
+
}
|
|
2897
|
+
function refuse(code, message, fields) {
|
|
2898
|
+
log(code === "write_failed" ? "error" : "warn", "file_push_refused", { code, ...fields });
|
|
2899
|
+
return { ok: false, code, message };
|
|
2900
|
+
}
|
|
2901
|
+
|
|
2902
|
+
// src/lib/runner-file-sync.ts
|
|
2903
|
+
var MAX_ACK_ATTEMPTS = 5;
|
|
2904
|
+
async function syncPendingRunnerFiles(options) {
|
|
2905
|
+
const pending = await listPendingFiles(options);
|
|
2906
|
+
const pendingIds = new Set(pending.map((file) => file.id));
|
|
2907
|
+
for (const id of options.ackFailures.keys()) {
|
|
2908
|
+
if (!pendingIds.has(id)) options.ackFailures.delete(id);
|
|
2909
|
+
}
|
|
2910
|
+
if (pending.length === 0) return 0;
|
|
2911
|
+
options.log({
|
|
2912
|
+
level: "info",
|
|
2913
|
+
message: `Runner file sync: ${pending.length} file(s) queued for this runner`
|
|
2914
|
+
});
|
|
2915
|
+
let applied = 0;
|
|
2916
|
+
for (const file of pending) {
|
|
2917
|
+
if ((options.ackFailures.get(file.id) ?? 0) >= MAX_ACK_ATTEMPTS) continue;
|
|
2918
|
+
if (await applyOne(options, file)) applied += 1;
|
|
2919
|
+
}
|
|
2920
|
+
return applied;
|
|
2921
|
+
}
|
|
2922
|
+
async function listPendingFiles(options) {
|
|
2923
|
+
let res;
|
|
2924
|
+
try {
|
|
2925
|
+
res = await options.fetchImpl(`${options.apiUrl}/runners/${options.agentId}/files/pending`, {
|
|
2926
|
+
headers: { Authorization: options.getAuthHeader() }
|
|
2927
|
+
});
|
|
2928
|
+
} catch (err) {
|
|
2929
|
+
options.log({
|
|
2930
|
+
level: "warn",
|
|
2931
|
+
message: `Could not list pending runner files \u2014 retrying on the next drain: ${describe(err)}`
|
|
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;
|
|
1914
3019
|
}
|
|
1915
|
-
|
|
1916
|
-
|
|
1917
|
-
|
|
1918
|
-
|
|
1919
|
-
|
|
1920
|
-
|
|
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 };
|
|
1921
3052
|
}
|
|
1922
|
-
|
|
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 };
|
|
1923
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 };
|
|
1924
3067
|
}
|
|
1925
|
-
|
|
1926
|
-
|
|
1927
|
-
|
|
1928
|
-
|
|
1929
|
-
const
|
|
1930
|
-
|
|
1931
|
-
|
|
1932
|
-
|
|
1933
|
-
|
|
1934
|
-
|
|
1935
|
-
|
|
1936
|
-
|
|
1937
|
-
|
|
1938
|
-
this.reconnecting = false;
|
|
1939
|
-
this.resolvedAgentId = agentId;
|
|
1940
|
-
events.onConnected(agentId, isReconnect);
|
|
1941
|
-
},
|
|
1942
|
-
onDisconnected: (code, reason) => {
|
|
1943
|
-
events.onDisconnected(code, reason);
|
|
1944
|
-
if (this.opts.isRunning() && code !== 1e3 && !this.reconnecting) {
|
|
1945
|
-
this.reconnectPromise = this.connectWithRetry(true).catch((err) => {
|
|
1946
|
-
events.onError?.(`Reconnection failed: ${err.message}`);
|
|
1947
|
-
});
|
|
1948
|
-
}
|
|
1949
|
-
},
|
|
1950
|
-
onError: (error2) => events.onError?.(error2),
|
|
1951
|
-
onResponse: () => events.onResponse?.(),
|
|
1952
|
-
onDrainPing: () => events.onDrainPing?.(),
|
|
1953
|
-
onInfo: (message) => events.onInfo?.(message)
|
|
1954
|
-
});
|
|
1955
|
-
return;
|
|
1956
|
-
} catch (error2) {
|
|
1957
|
-
this.reconnectAttempt++;
|
|
1958
|
-
if (error2.message === "Unauthorized") {
|
|
1959
|
-
this.reconnecting = false;
|
|
1960
|
-
throw error2;
|
|
1961
|
-
}
|
|
1962
|
-
const delay = getReconnectDelay(this.reconnectAttempt);
|
|
1963
|
-
events.onReconnecting?.(this.reconnectAttempt);
|
|
1964
|
-
events.onError?.(`Connection failed, retrying in ${Math.round(delay / 1e3)}s...`);
|
|
1965
|
-
await this.sleep(delay);
|
|
3068
|
+
}
|
|
3069
|
+
async function ack(options, file, status2, reason) {
|
|
3070
|
+
const outcome = `${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 })
|
|
1966
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;
|
|
1967
3090
|
}
|
|
1968
|
-
|
|
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: ${describe(err)}`
|
|
3097
|
+
);
|
|
1969
3098
|
}
|
|
1970
|
-
}
|
|
3099
|
+
}
|
|
3100
|
+
function recordAckFailure(options, file, what) {
|
|
3101
|
+
const attempts = (options.ackFailures.get(file.id) ?? 0) + 1;
|
|
3102
|
+
options.ackFailures.set(file.id, attempts);
|
|
3103
|
+
options.log({
|
|
3104
|
+
level: "error",
|
|
3105
|
+
message: attempts >= MAX_ACK_ATTEMPTS ? `${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
|
+
}
|
|
1971
3111
|
|
|
1972
3112
|
// src/lib/channels/driver.ts
|
|
1973
3113
|
function messageIdOf(m) {
|
|
@@ -1997,7 +3137,11 @@ var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
|
|
|
1997
3137
|
var DEFAULT_STUCK_QUEUED_MS = 6e4;
|
|
1998
3138
|
var HEARTBEAT_MS = 6e4;
|
|
1999
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;
|
|
2000
3142
|
var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
|
|
3143
|
+
var MAX_SUPERSEDED_CONVERSATIONS = 256;
|
|
3144
|
+
var MAX_IDENTICAL_REDRIVE_POLL_FAILURES = 5;
|
|
2001
3145
|
var ChannelAuthError = class extends Error {
|
|
2002
3146
|
constructor(message) {
|
|
2003
3147
|
super(message);
|
|
@@ -2006,10 +3150,10 @@ var ChannelAuthError = class extends Error {
|
|
|
2006
3150
|
};
|
|
2007
3151
|
var ChannelTerminalError = class extends Error {
|
|
2008
3152
|
status;
|
|
2009
|
-
constructor(message,
|
|
3153
|
+
constructor(message, status2) {
|
|
2010
3154
|
super(message);
|
|
2011
3155
|
this.name = "ChannelTerminalError";
|
|
2012
|
-
this.status =
|
|
3156
|
+
this.status = status2;
|
|
2013
3157
|
}
|
|
2014
3158
|
};
|
|
2015
3159
|
function backoffDelay(attempt, policy) {
|
|
@@ -2017,10 +3161,14 @@ function backoffDelay(attempt, policy) {
|
|
|
2017
3161
|
const capped = Math.min(policy.maxDelayMs, exp);
|
|
2018
3162
|
return Math.floor(Math.random() * capped);
|
|
2019
3163
|
}
|
|
2020
|
-
function isRetryableStatus(
|
|
2021
|
-
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);
|
|
2022
3170
|
}
|
|
2023
|
-
var ChannelDriver = class {
|
|
3171
|
+
var ChannelDriver = class _ChannelDriver {
|
|
2024
3172
|
agentId;
|
|
2025
3173
|
port;
|
|
2026
3174
|
apiUrl;
|
|
@@ -2034,8 +3182,38 @@ var ChannelDriver = class {
|
|
|
2034
3182
|
pausedMaxWaitMs;
|
|
2035
3183
|
stuckQueuedMs;
|
|
2036
3184
|
now;
|
|
3185
|
+
fileSyncDirectories;
|
|
3186
|
+
homeDir;
|
|
2037
3187
|
/** Cache of conversationId → opencode sessionId. */
|
|
2038
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();
|
|
2039
3217
|
/**
|
|
2040
3218
|
* Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
|
|
2041
3219
|
* longer idempotent (no caller-supplied `messageID`), and its read-back picks
|
|
@@ -2103,6 +3281,43 @@ var ChannelDriver = class {
|
|
|
2103
3281
|
* processing list, exactly like `dontRedispatch`/`doneUndeliverable`.
|
|
2104
3282
|
*/
|
|
2105
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();
|
|
2106
3321
|
/**
|
|
2107
3322
|
* "A null-id re-adopt re-dispatch is in flight, awaiting its read-back" (WI-5
|
|
2108
3323
|
* Task 5.4, High-2). Since we no longer send a caller-supplied id, a re-dispatch
|
|
@@ -2145,9 +3360,12 @@ var ChannelDriver = class {
|
|
|
2145
3360
|
sessionParents = /* @__PURE__ */ new Map();
|
|
2146
3361
|
/**
|
|
2147
3362
|
* Per-session OpenCode title cache (#310), keyed by sessionId. Only a resolved
|
|
2148
|
-
* NON-EMPTY name is stored (terminal — a real session name
|
|
2149
|
-
* so we do NOT re-GET `/session/:id` every tick.
|
|
2150
|
-
*
|
|
3363
|
+
* NON-EMPTY, non-placeholder name is stored (terminal — a real session name
|
|
3364
|
+
* won't later un-name), so we do NOT re-GET `/session/:id` every tick. "Non-empty"
|
|
3365
|
+
* excludes OpenCode's synchronous default title (see
|
|
3366
|
+
* `OPENCODE_DEFAULT_TITLE_PREFIX`, #549) — that placeholder is treated the same
|
|
3367
|
+
* as an empty title so it never latches. A missing entry = not yet resolved OR
|
|
3368
|
+
* resolved-but-still-empty/placeholder → re-fetch on next need, since OpenCode
|
|
2151
3369
|
* names sessions asynchronously mid-turn. Driver-level (not per-watcher) so both
|
|
2152
3370
|
* the watcher completion path AND the restart-recovery re-adopt path (which has
|
|
2153
3371
|
* no watcher) can resolve the title.
|
|
@@ -2155,6 +3373,24 @@ var ChannelDriver = class {
|
|
|
2155
3373
|
sessionTitles = /* @__PURE__ */ new Map();
|
|
2156
3374
|
/** Serialises drains so a reconnect during a drain doesn't double-process. */
|
|
2157
3375
|
draining = false;
|
|
3376
|
+
/**
|
|
3377
|
+
* Serialises runner-file syncs (#559) so the ~2s poll tick and a concurrent
|
|
3378
|
+
* drain ping don't download, write and ack the same file twice.
|
|
3379
|
+
*/
|
|
3380
|
+
syncingFiles = false;
|
|
3381
|
+
/**
|
|
3382
|
+
* Consecutive failed acks per pending file (#559). Lives on the driver so it
|
|
3383
|
+
* survives across drains — without it, a file whose ack keeps failing is
|
|
3384
|
+
* re-downloaded and re-written every ~2s until the server expires it.
|
|
3385
|
+
*/
|
|
3386
|
+
fileAckFailures = /* @__PURE__ */ new Map();
|
|
3387
|
+
/**
|
|
3388
|
+
* Monotonic count of files this runner has pulled and written (#559). Only
|
|
3389
|
+
* ever increases, so `run.ts` detects work by comparing it against the value
|
|
3390
|
+
* it saw on the previous cycle — including work that landed mid-sleep, the
|
|
3391
|
+
* same trick `lastProxiedActivityAt` uses.
|
|
3392
|
+
*/
|
|
3393
|
+
appliedFileCount = 0;
|
|
2158
3394
|
/**
|
|
2159
3395
|
* The currently-executing `drainPending()` promise, or null when idle. Lets a
|
|
2160
3396
|
* graceful shutdown (`waitForInFlight`) await an in-progress drain so a turn it
|
|
@@ -2170,21 +3406,23 @@ var ChannelDriver = class {
|
|
|
2170
3406
|
* and stops opencode.
|
|
2171
3407
|
*/
|
|
2172
3408
|
stopped = false;
|
|
2173
|
-
constructor(
|
|
2174
|
-
this.agentId =
|
|
2175
|
-
this.port =
|
|
2176
|
-
this.apiUrl =
|
|
2177
|
-
this.getAuthHeader =
|
|
2178
|
-
this.conversationFilter =
|
|
2179
|
-
this.retry = { ...DEFAULT_RETRY_POLICY, ...
|
|
2180
|
-
this.log =
|
|
3409
|
+
constructor(config) {
|
|
3410
|
+
this.agentId = config.agentId;
|
|
3411
|
+
this.port = config.port;
|
|
3412
|
+
this.apiUrl = config.apiUrl.replace(/\/$/, "");
|
|
3413
|
+
this.getAuthHeader = config.getAuthHeader;
|
|
3414
|
+
this.conversationFilter = config.conversationFilter ?? null;
|
|
3415
|
+
this.retry = { ...DEFAULT_RETRY_POLICY, ...config.retry };
|
|
3416
|
+
this.log = config.log ?? (() => {
|
|
2181
3417
|
});
|
|
2182
|
-
this.fetchImpl =
|
|
2183
|
-
this.sleep =
|
|
2184
|
-
this.pausedPollIntervalMs =
|
|
2185
|
-
this.pausedMaxWaitMs =
|
|
2186
|
-
this.stuckQueuedMs =
|
|
2187
|
-
this.now =
|
|
3418
|
+
this.fetchImpl = config.fetchImpl ?? fetch;
|
|
3419
|
+
this.sleep = config.sleep ?? ((ms) => new Promise((r) => setTimeout(r, ms)));
|
|
3420
|
+
this.pausedPollIntervalMs = config.pausedPollIntervalMs ?? DEFAULT_PAUSED_POLL_INTERVAL_MS;
|
|
3421
|
+
this.pausedMaxWaitMs = config.pausedMaxWaitMs ?? DEFAULT_PAUSED_MAX_WAIT_MS;
|
|
3422
|
+
this.stuckQueuedMs = config.stuckQueuedMs ?? DEFAULT_STUCK_QUEUED_MS;
|
|
3423
|
+
this.now = config.now ?? (() => Date.now());
|
|
3424
|
+
this.fileSyncDirectories = config.fileSyncDirectories ?? [];
|
|
3425
|
+
this.homeDir = config.homeDir ?? homedir2();
|
|
2188
3426
|
}
|
|
2189
3427
|
/** The IPv4-loopback base URL for the local `opencode serve`. */
|
|
2190
3428
|
get opencodeBase() {
|
|
@@ -2212,6 +3450,47 @@ var ChannelDriver = class {
|
|
|
2212
3450
|
);
|
|
2213
3451
|
return run2;
|
|
2214
3452
|
}
|
|
3453
|
+
/**
|
|
3454
|
+
* Pull-and-apply any files Evident has queued for this runner (#559), riding
|
|
3455
|
+
* the EXISTING drain cycle — `run.ts` calls it from the same ~2s channel poll
|
|
3456
|
+
* and drain ping that call `drainPending()`. There is deliberately no channel,
|
|
3457
|
+
* control frame or poll loop of its own: worst-case latency is one poll tick.
|
|
3458
|
+
*
|
|
3459
|
+
* NEVER throws and never surfaces a `ChannelAuthError`: a file failure must not
|
|
3460
|
+
* cost a conversation turn. Failures are logged and either acked as a terminal
|
|
3461
|
+
* outcome or left pending for the next drain (see `runner-file-sync.ts`).
|
|
3462
|
+
*
|
|
3463
|
+
* Re-entrant calls are skipped (the poll tick and a drain ping can overlap).
|
|
3464
|
+
*
|
|
3465
|
+
* @returns the number of files written to disk.
|
|
3466
|
+
*/
|
|
3467
|
+
async syncPendingFiles() {
|
|
3468
|
+
if (this.stopped) return 0;
|
|
3469
|
+
if (this.syncingFiles) return 0;
|
|
3470
|
+
this.syncingFiles = true;
|
|
3471
|
+
try {
|
|
3472
|
+
const applied = await syncPendingRunnerFiles({
|
|
3473
|
+
agentId: this.agentId,
|
|
3474
|
+
apiUrl: this.apiUrl,
|
|
3475
|
+
getAuthHeader: this.getAuthHeader,
|
|
3476
|
+
fetchImpl: this.fetchImpl,
|
|
3477
|
+
allowedDirectories: this.fileSyncDirectories,
|
|
3478
|
+
homeDir: this.homeDir,
|
|
3479
|
+
ackFailures: this.fileAckFailures,
|
|
3480
|
+
log: this.log
|
|
3481
|
+
});
|
|
3482
|
+
this.appliedFileCount += applied;
|
|
3483
|
+
return applied;
|
|
3484
|
+
} catch (err) {
|
|
3485
|
+
this.log({
|
|
3486
|
+
level: "error",
|
|
3487
|
+
message: `Runner file sync failed unexpectedly (message processing is unaffected): ${err instanceof Error ? err.message : String(err)}`
|
|
3488
|
+
});
|
|
3489
|
+
return 0;
|
|
3490
|
+
} finally {
|
|
3491
|
+
this.syncingFiles = false;
|
|
3492
|
+
}
|
|
3493
|
+
}
|
|
2215
3494
|
async runDrain() {
|
|
2216
3495
|
let dispatched = 0;
|
|
2217
3496
|
try {
|
|
@@ -2245,6 +3524,28 @@ var ChannelDriver = class {
|
|
|
2245
3524
|
}
|
|
2246
3525
|
return false;
|
|
2247
3526
|
}
|
|
3527
|
+
/**
|
|
3528
|
+
* File-pull work, for `run.ts`'s idle accounting (#559).
|
|
3529
|
+
*
|
|
3530
|
+
* Pulling a file is real work that `drainPending()` knows nothing about, so
|
|
3531
|
+
* without this a near-idle runner counts a credential pull as an empty tick
|
|
3532
|
+
* and `--idle-timeout` can `process.exit` mid-pull — leaving a
|
|
3533
|
+
* `.evident-push-*.tmp` behind — or immediately after the write, before the
|
|
3534
|
+
* browser has run the authorize/callback that activates it (the user then sees
|
|
3535
|
+
* `saved_not_activated` for a runner that was fine).
|
|
3536
|
+
*
|
|
3537
|
+
* Two signals because one cannot cover both cases: `inFlight` is the pull
|
|
3538
|
+
* happening RIGHT NOW (it may outlive the tick that started it), and
|
|
3539
|
+
* `appliedFiles` is monotonic so a pull that started AND finished between two
|
|
3540
|
+
* idle checks still shows up as an advance.
|
|
3541
|
+
*
|
|
3542
|
+
* CALLER CONTRACT: sample `inFlight` BEFORE calling `syncPendingFiles()` for
|
|
3543
|
+
* the cycle. `syncPendingFiles` sets the flag synchronously, so a caller that
|
|
3544
|
+
* samples afterwards reads `true` every single cycle and can never idle out.
|
|
3545
|
+
*/
|
|
3546
|
+
fileSyncActivity() {
|
|
3547
|
+
return { appliedFiles: this.appliedFileCount, inFlight: this.syncingFiles };
|
|
3548
|
+
}
|
|
2248
3549
|
/**
|
|
2249
3550
|
* OpenCode session ids the session-cleanup sweep (issue #190) must NOT delete:
|
|
2250
3551
|
* exactly those with a live (dispatched-but-not-done / paused) turn, i.e. a
|
|
@@ -2306,7 +3607,7 @@ var ChannelDriver = class {
|
|
|
2306
3607
|
await this.sleep(step);
|
|
2307
3608
|
}
|
|
2308
3609
|
}
|
|
2309
|
-
while (this.hasInFlightWatchers()) {
|
|
3610
|
+
while (this.hasInFlightWatchers() || this.syncingFiles) {
|
|
2310
3611
|
if (this.now() >= deadline) return false;
|
|
2311
3612
|
await this.sleep(step);
|
|
2312
3613
|
}
|
|
@@ -2341,16 +3642,27 @@ var ChannelDriver = class {
|
|
|
2341
3642
|
* @returns the count of messages NEWLY dispatched (not already in-flight).
|
|
2342
3643
|
*/
|
|
2343
3644
|
async processConversation(conv) {
|
|
2344
|
-
const sessionId = await this.ensureSession(conv);
|
|
3645
|
+
const { sessionId, refusedSessionId } = await this.ensureSession(conv);
|
|
2345
3646
|
const messages = await this.getPendingMessages(conv.id);
|
|
2346
3647
|
let dispatched = 0;
|
|
2347
3648
|
let skippedAlreadyDispatched = 0;
|
|
3649
|
+
if (refusedSessionId && messages.length > 0) {
|
|
3650
|
+
void this.postSignal(conv.id, messages[0].id, "session_superseded", {
|
|
3651
|
+
superseded_session_id: refusedSessionId
|
|
3652
|
+
});
|
|
3653
|
+
}
|
|
2348
3654
|
for (const message of messages) {
|
|
2349
3655
|
if (this.stopped) break;
|
|
2350
3656
|
if (this.dispatched.has(message.id)) {
|
|
2351
3657
|
skippedAlreadyDispatched += 1;
|
|
2352
3658
|
continue;
|
|
2353
3659
|
}
|
|
3660
|
+
if (message.opencode_message_id) {
|
|
3661
|
+
const outcome = await this.resolveRedrive(conv, sessionId, message, refusedSessionId);
|
|
3662
|
+
if (outcome !== "dispatch") {
|
|
3663
|
+
break;
|
|
3664
|
+
}
|
|
3665
|
+
}
|
|
2354
3666
|
const options = {
|
|
2355
3667
|
agent: message.opencode_agent ?? void 0,
|
|
2356
3668
|
model: message.opencode_model ?? void 0
|
|
@@ -2371,7 +3683,8 @@ var ChannelDriver = class {
|
|
|
2371
3683
|
} catch (err) {
|
|
2372
3684
|
if (err instanceof ChannelAuthError) throw err;
|
|
2373
3685
|
this.dispatched.delete(message.id);
|
|
2374
|
-
|
|
3686
|
+
const exists = await sessionExists(this.port, sessionId);
|
|
3687
|
+
if (exists === false) {
|
|
2375
3688
|
this.sessions.delete(conv.id);
|
|
2376
3689
|
this.log({
|
|
2377
3690
|
level: "warn",
|
|
@@ -2381,15 +3694,39 @@ var ChannelDriver = class {
|
|
|
2381
3694
|
});
|
|
2382
3695
|
break;
|
|
2383
3696
|
}
|
|
2384
|
-
|
|
3697
|
+
if (exists === null) {
|
|
3698
|
+
this.log({
|
|
3699
|
+
level: "warn",
|
|
3700
|
+
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.`,
|
|
3701
|
+
conversation_id: conv.id,
|
|
3702
|
+
message_id: message.id
|
|
3703
|
+
});
|
|
3704
|
+
break;
|
|
3705
|
+
}
|
|
3706
|
+
const errorMessage = err instanceof Error ? err.message : String(err);
|
|
3707
|
+
this.sessions.delete(conv.id);
|
|
3708
|
+
this.supersede(conv.id, sessionId);
|
|
3709
|
+
this.log({
|
|
3710
|
+
level: "warn",
|
|
3711
|
+
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.`,
|
|
3712
|
+
conversation_id: conv.id,
|
|
3713
|
+
message_id: message.id
|
|
3714
|
+
});
|
|
3715
|
+
await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
|
|
3716
|
+
this.log({
|
|
3717
|
+
level: "warn",
|
|
3718
|
+
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)}`,
|
|
3719
|
+
conversation_id: conv.id,
|
|
3720
|
+
message_id: message.id
|
|
3721
|
+
});
|
|
2385
3722
|
});
|
|
2386
3723
|
this.log({
|
|
2387
3724
|
level: "error",
|
|
2388
|
-
message: `Message ${message.id.slice(0, 8)} dispatch failed: ${
|
|
3725
|
+
message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
|
|
2389
3726
|
conversation_id: conv.id,
|
|
2390
3727
|
message_id: message.id
|
|
2391
3728
|
});
|
|
2392
|
-
|
|
3729
|
+
break;
|
|
2393
3730
|
}
|
|
2394
3731
|
if (opencodeMessageId === null) {
|
|
2395
3732
|
this.log({
|
|
@@ -2398,25 +3735,391 @@ var ChannelDriver = class {
|
|
|
2398
3735
|
conversation_id: conv.id,
|
|
2399
3736
|
message_id: message.id
|
|
2400
3737
|
});
|
|
2401
|
-
continue;
|
|
3738
|
+
continue;
|
|
3739
|
+
}
|
|
3740
|
+
this.dispatched.add(message.id);
|
|
3741
|
+
this.registerInFlight(conv, sessionId, message, opencodeMessageId);
|
|
3742
|
+
dispatched += 1;
|
|
3743
|
+
void this.postSignal(conv.id, message.id, "dispatched");
|
|
3744
|
+
}
|
|
3745
|
+
if (messages.length > 0 && dispatched === 0 && skippedAlreadyDispatched === messages.length) {
|
|
3746
|
+
this.log({
|
|
3747
|
+
level: "warn",
|
|
3748
|
+
message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode this tick. If this repeats, a message may be stuck acknowledged-but-never-dispatched (its watcher never settled).`,
|
|
3749
|
+
conversation_id: conv.id
|
|
3750
|
+
});
|
|
3751
|
+
}
|
|
3752
|
+
this.ensureWatcherRunning(sessionId);
|
|
3753
|
+
return dispatched;
|
|
3754
|
+
}
|
|
3755
|
+
/**
|
|
3756
|
+
* Poll a session's message list for the re-drive fence (#965), via the
|
|
3757
|
+
* INJECTED `fetchImpl` — NOT the imported `getSessionMessages` helper, which
|
|
3758
|
+
* hits the global `fetch` and would bypass the same override every other
|
|
3759
|
+
* opencode poll in this file respects. Mirrors `readoptProcessing`'s own
|
|
3760
|
+
* snapshot fetch (`:3081-3111`).
|
|
3761
|
+
*
|
|
3762
|
+
* Returns `{ ok: true, messages }` on a readable snapshot, or
|
|
3763
|
+
* `{ ok: false, signature }` on failure — `signature` is a string that
|
|
3764
|
+
* repeats across attempts for the SAME underlying fault (used by the
|
|
3765
|
+
* consecutive-identical-failure bound, #1348), or `null` for a thrown
|
|
3766
|
+
* exception, which is NOT countable toward that bound (a network blip / an
|
|
3767
|
+
* opencode restart also throws identically every tick, and must keep
|
|
3768
|
+
* retrying unbounded rather than ever being treated as permanent).
|
|
3769
|
+
*/
|
|
3770
|
+
async pollSessionMessagesForRedrive(conv, message, sessionId) {
|
|
3771
|
+
try {
|
|
3772
|
+
const res = await this.fetchImpl(`${this.opencodeBase}/session/${sessionId}/message`);
|
|
3773
|
+
if (!res.ok) {
|
|
3774
|
+
const rawBody = await res.text();
|
|
3775
|
+
const normalized = normalizeRedrivePollFailureBody(rawBody);
|
|
3776
|
+
this.log({
|
|
3777
|
+
level: "warn",
|
|
3778
|
+
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`,
|
|
3779
|
+
conversation_id: conv.id,
|
|
3780
|
+
message_id: message.id
|
|
3781
|
+
});
|
|
3782
|
+
return { ok: false, signature: `HTTP ${res.status}${normalized ? `: ${normalized}` : ""}` };
|
|
3783
|
+
}
|
|
3784
|
+
const body = await res.json();
|
|
3785
|
+
if (!Array.isArray(body)) {
|
|
3786
|
+
this.log({
|
|
3787
|
+
level: "warn",
|
|
3788
|
+
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`,
|
|
3789
|
+
conversation_id: conv.id,
|
|
3790
|
+
message_id: message.id
|
|
3791
|
+
});
|
|
3792
|
+
return { ok: false, signature: "non-array message body" };
|
|
3793
|
+
}
|
|
3794
|
+
return { ok: true, messages: body };
|
|
3795
|
+
} catch (err) {
|
|
3796
|
+
this.log({
|
|
3797
|
+
level: "warn",
|
|
3798
|
+
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)}`,
|
|
3799
|
+
conversation_id: conv.id,
|
|
3800
|
+
message_id: message.id
|
|
3801
|
+
});
|
|
3802
|
+
return { ok: false, signature: null };
|
|
3803
|
+
}
|
|
3804
|
+
}
|
|
3805
|
+
/**
|
|
3806
|
+
* The re-drive fence for a `pending` row that already carries a stored
|
|
3807
|
+
* `opencode_message_id` (#965) — i.e. it has already been handed to opencode at
|
|
3808
|
+
* least once (see the invariant at `QueuedMessage.opencode_message_id`'s doc).
|
|
3809
|
+
* The lifecycle cron can falsely reclaim a `processing` row back to `pending`
|
|
3810
|
+
* mid-turn (a 5-minute liveness-staleness check racing a still-running turn);
|
|
3811
|
+
* without this fence the drain loop would re-`prompt_async` the SAME turn a
|
|
3812
|
+
* second time against live GitHub state. Mirrors `readoptOne`'s job for the
|
|
3813
|
+
* `processing` re-adopt path, but simpler: no b1/b2 preamble cross-check is
|
|
3814
|
+
* needed here because `refusedSessionId` already handles the one case
|
|
3815
|
+
* (#553 abandoned session) that path exists for.
|
|
3816
|
+
*
|
|
3817
|
+
* Only `ChannelAuthError` propagates. A poll that fails identically
|
|
3818
|
+
* `MAX_IDENTICAL_REDRIVE_POLL_FAILURES` times in a row reports the message
|
|
3819
|
+
* failed instead of retrying it (#1348) — SEPARATE from, not a replacement
|
|
3820
|
+
* for, `resolveRedriveUnresolved`'s own `pausedMaxWaitMs` bound below. Every
|
|
3821
|
+
* other failure resolves to `unresolved` and is retried whole on the next
|
|
3822
|
+
* ~2s drain tick.
|
|
3823
|
+
*/
|
|
3824
|
+
async resolveRedrive(conv, sessionId, message, refusedSessionId) {
|
|
3825
|
+
const ocId = message.opencode_message_id ?? null;
|
|
3826
|
+
if (refusedSessionId) {
|
|
3827
|
+
this.clearRedriveUnresolved(message.id);
|
|
3828
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3829
|
+
return "dispatch";
|
|
3830
|
+
}
|
|
3831
|
+
const polled = await this.pollSessionMessagesForRedrive(conv, message, sessionId);
|
|
3832
|
+
if (!polled.ok) {
|
|
3833
|
+
const streak = this.recordRedrivePollFailure(message.id, sessionId, polled.signature);
|
|
3834
|
+
if (streak >= MAX_IDENTICAL_REDRIVE_POLL_FAILURES && polled.signature !== null) {
|
|
3835
|
+
return this.failRedrivePollPermanent(conv, sessionId, message, polled.signature, streak);
|
|
3836
|
+
}
|
|
3837
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3838
|
+
}
|
|
3839
|
+
this.redrivePollFailures.delete(message.id);
|
|
3840
|
+
const messages = polled.messages;
|
|
3841
|
+
if (messages.length === 0) {
|
|
3842
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3843
|
+
}
|
|
3844
|
+
const state = messageRunState(messages, ocId ?? "");
|
|
3845
|
+
if (state === "failed" && isAbortedTerminalReply(messages, ocId ?? "")) {
|
|
3846
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3847
|
+
if (ongoing === false) {
|
|
3848
|
+
this.log({
|
|
3849
|
+
level: "info",
|
|
3850
|
+
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`,
|
|
3851
|
+
conversation_id: conv.id,
|
|
3852
|
+
message_id: message.id
|
|
3853
|
+
});
|
|
3854
|
+
this.clearRedriveUnresolved(message.id);
|
|
3855
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3856
|
+
return "dispatch";
|
|
3857
|
+
}
|
|
3858
|
+
}
|
|
3859
|
+
if (state === "done" || state === "failed") {
|
|
3860
|
+
return this.settleRedrive(conv, sessionId, message, ocId, messages, state);
|
|
3861
|
+
}
|
|
3862
|
+
if (state === "running" || state === "queued") {
|
|
3863
|
+
const ongoing = await isSessionOngoing(this.port, sessionId);
|
|
3864
|
+
if (ongoing === true) {
|
|
3865
|
+
return this.reattachRedrive(conv, sessionId, message, ocId);
|
|
3866
|
+
}
|
|
3867
|
+
if (ongoing === false) {
|
|
3868
|
+
this.clearRedriveUnresolved(message.id);
|
|
3869
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3870
|
+
return "dispatch";
|
|
3871
|
+
}
|
|
3872
|
+
return this.resolveRedriveUnresolved(conv, message);
|
|
3873
|
+
}
|
|
3874
|
+
this.clearRedriveUnresolved(message.id);
|
|
3875
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
3876
|
+
return "dispatch";
|
|
3877
|
+
}
|
|
3878
|
+
/**
|
|
3879
|
+
* The `reattached` outcome (Task 3.3): the prior turn is STILL ONGOING per
|
|
3880
|
+
* opencode's own status map — undo the false reclaim instead of starting a
|
|
3881
|
+
* second turn.
|
|
3882
|
+
*/
|
|
3883
|
+
async reattachRedrive(conv, sessionId, message, ocId) {
|
|
3884
|
+
let anchorMs;
|
|
3885
|
+
const parsed = message.processing_started_at ? Date.parse(message.processing_started_at) : NaN;
|
|
3886
|
+
if (!Number.isNaN(parsed)) {
|
|
3887
|
+
anchorMs = parsed;
|
|
3888
|
+
} else {
|
|
3889
|
+
anchorMs = this.now();
|
|
3890
|
+
this.log({
|
|
3891
|
+
level: "error",
|
|
3892
|
+
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)`,
|
|
3893
|
+
conversation_id: conv.id,
|
|
3894
|
+
message_id: message.id
|
|
3895
|
+
});
|
|
3896
|
+
}
|
|
3897
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3898
|
+
try {
|
|
3899
|
+
await this.markProcessing(conv.id, message.id, sessionId, ocId, title);
|
|
3900
|
+
} catch (err) {
|
|
3901
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3902
|
+
this.log({
|
|
3903
|
+
level: "warn",
|
|
3904
|
+
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)}`,
|
|
3905
|
+
conversation_id: conv.id,
|
|
3906
|
+
message_id: message.id
|
|
3907
|
+
});
|
|
3908
|
+
this.signalRedriveOutcomeUnreported(conv, message, "reattach");
|
|
3909
|
+
return "unresolved";
|
|
3910
|
+
}
|
|
3911
|
+
this.clearRedriveUnresolved(message.id);
|
|
3912
|
+
this.registerReadopted(conv, sessionId, message, ocId ?? "", anchorMs);
|
|
3913
|
+
this.dispatched.add(message.id);
|
|
3914
|
+
this.readopted.add(message.id);
|
|
3915
|
+
this.ensureWatcherRunning(sessionId);
|
|
3916
|
+
const watchedForMs = this.now() - anchorMs;
|
|
3917
|
+
void this.postSignal(conv.id, message.id, "redrive_reattached", {
|
|
3918
|
+
watched_for_ms: watchedForMs
|
|
3919
|
+
});
|
|
3920
|
+
this.log({
|
|
3921
|
+
level: "warn",
|
|
3922
|
+
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`,
|
|
3923
|
+
conversation_id: conv.id,
|
|
3924
|
+
message_id: message.id
|
|
3925
|
+
});
|
|
3926
|
+
return "reattached";
|
|
3927
|
+
}
|
|
3928
|
+
/**
|
|
3929
|
+
* The `settled` outcome (Task 3.2): the prior turn already finished (or
|
|
3930
|
+
* errored) while nobody was watching — deliver/report it instead of re-running.
|
|
3931
|
+
* Mirrors `readoptOne`'s `done`/`failed` branches' error discipline, simplified
|
|
3932
|
+
* (no `doneUndeliverable` park: a terminal PATCH failure here just retries next
|
|
3933
|
+
* drain, same as any other non-auth failure). The restart-abort carve-out that
|
|
3934
|
+
* keeps the two in step for `failed` lives in the caller (`resolveRedrive`, #1310),
|
|
3935
|
+
* so a row reaching this `failed` branch is a GENUINE failure.
|
|
3936
|
+
*/
|
|
3937
|
+
async settleRedrive(conv, sessionId, message, ocId, messages, state) {
|
|
3938
|
+
try {
|
|
3939
|
+
if (state === "done") {
|
|
3940
|
+
const title = await this.resolveSessionTitle(sessionId, conv.id);
|
|
3941
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3942
|
+
this.log({
|
|
3943
|
+
level: "info",
|
|
3944
|
+
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`,
|
|
3945
|
+
conversation_id: conv.id,
|
|
3946
|
+
message_id: message.id
|
|
3947
|
+
});
|
|
3948
|
+
await this.markDone(conv.id, message.id, sessionId, ocId, title, usage);
|
|
3949
|
+
} else {
|
|
3950
|
+
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3951
|
+
const usage = messageUsage(messages, ocId ?? "");
|
|
3952
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3953
|
+
this.log({
|
|
3954
|
+
level: "error",
|
|
3955
|
+
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)"}`,
|
|
3956
|
+
conversation_id: conv.id,
|
|
3957
|
+
message_id: message.id
|
|
3958
|
+
});
|
|
3959
|
+
await this.markFailed(conv.id, message.id, sessionId, error2, usage, failure);
|
|
2402
3960
|
}
|
|
2403
|
-
|
|
2404
|
-
|
|
2405
|
-
|
|
2406
|
-
|
|
3961
|
+
} catch (err) {
|
|
3962
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
3963
|
+
this.log({
|
|
3964
|
+
level: "warn",
|
|
3965
|
+
message: `Re-drive: failed to report message ${message.id.slice(0, 8)} ${state} (will retry next drain): ${err instanceof Error ? err.message : String(err)}`,
|
|
3966
|
+
conversation_id: conv.id,
|
|
3967
|
+
message_id: message.id
|
|
3968
|
+
});
|
|
3969
|
+
this.signalRedriveOutcomeUnreported(conv, message, "settle");
|
|
3970
|
+
return "unresolved";
|
|
2407
3971
|
}
|
|
2408
|
-
|
|
3972
|
+
this.clearRedriveUnresolved(message.id);
|
|
3973
|
+
void this.postSignal(conv.id, message.id, "redrive_settled");
|
|
3974
|
+
return "settled";
|
|
3975
|
+
}
|
|
3976
|
+
/**
|
|
3977
|
+
* The permanent-failure outcome (#1348): the fence's own poll of this session
|
|
3978
|
+
* failed with the SAME opencode-answered signature
|
|
3979
|
+
* `MAX_IDENTICAL_REDRIVE_POLL_FAILURES` times in a row — a transient blip
|
|
3980
|
+
* would have varied or eventually cleared (see `pollSessionMessagesForRedrive`
|
|
3981
|
+
* and `recordRedrivePollFailure`), so this is a durable fault (e.g. #1345's
|
|
3982
|
+
* corrupted opencode session) rather than something worth retrying forever.
|
|
3983
|
+
* Mirrors `settleRedrive`'s error discipline: no `usage`/`failure` args to
|
|
3984
|
+
* `markFailed` (no opencode snapshot to extract them from — this poll never
|
|
3985
|
+
* got a readable one).
|
|
3986
|
+
*/
|
|
3987
|
+
async failRedrivePollPermanent(conv, sessionId, message, signature, streak) {
|
|
3988
|
+
this.log({
|
|
3989
|
+
level: "error",
|
|
3990
|
+
message: `Re-drive: message ${message.id.slice(0, 8)} (session ${sessionId.slice(0, 8)}) failed to poll with the identical signature "${signature}" ${streak} times in a row \u2014 reporting the message failed instead of retrying forever`,
|
|
3991
|
+
conversation_id: conv.id,
|
|
3992
|
+
message_id: message.id
|
|
3993
|
+
});
|
|
3994
|
+
try {
|
|
3995
|
+
await this.markFailed(
|
|
3996
|
+
conv.id,
|
|
3997
|
+
message.id,
|
|
3998
|
+
sessionId,
|
|
3999
|
+
`The runner could not read this conversation's state from OpenCode (${signature}). The same failure repeated ${streak} times in a row, so the message was not retried further.`
|
|
4000
|
+
);
|
|
4001
|
+
} catch (err) {
|
|
4002
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
2409
4003
|
this.log({
|
|
2410
4004
|
level: "warn",
|
|
2411
|
-
message: `
|
|
2412
|
-
conversation_id: conv.id
|
|
4005
|
+
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)}`,
|
|
4006
|
+
conversation_id: conv.id,
|
|
4007
|
+
message_id: message.id
|
|
2413
4008
|
});
|
|
4009
|
+
this.signalRedriveOutcomeUnreported(conv, message, "fail_permanent");
|
|
4010
|
+
return "unresolved";
|
|
2414
4011
|
}
|
|
2415
|
-
this.
|
|
2416
|
-
|
|
4012
|
+
this.clearRedriveUnresolved(message.id);
|
|
4013
|
+
void this.postSignal(conv.id, message.id, "redrive_poll_failed");
|
|
4014
|
+
return "settled";
|
|
4015
|
+
}
|
|
4016
|
+
/**
|
|
4017
|
+
* The bounded `unresolved` outcome (Task 3.4): opencode's state could not be
|
|
4018
|
+
* observed (snapshot unreadable/empty, or `isSessionOngoing` returned `null`).
|
|
4019
|
+
* A `pending` row is invisible to every cron arm (all require `status =
|
|
4020
|
+
* 'processing'`), so an indefinitely-`unresolved` row would be stranded with
|
|
4021
|
+
* nothing driving it — bound it to the existing `pausedMaxWaitMs` window
|
|
4022
|
+
* (reusing the knob, not a new constant) and take `dispatch` once elapsed.
|
|
4023
|
+
*/
|
|
4024
|
+
resolveRedriveUnresolved(conv, message) {
|
|
4025
|
+
const now = this.now();
|
|
4026
|
+
const since = this.redriveUnresolvedSince.get(message.id);
|
|
4027
|
+
if (since !== void 0 && now - since >= this.pausedMaxWaitMs) {
|
|
4028
|
+
this.clearRedriveUnresolved(message.id);
|
|
4029
|
+
void this.postSignal(conv.id, message.id, "redrive_redispatched");
|
|
4030
|
+
return "dispatch";
|
|
4031
|
+
}
|
|
4032
|
+
if (since === void 0) {
|
|
4033
|
+
this.redriveUnresolvedSince.set(message.id, now);
|
|
4034
|
+
}
|
|
4035
|
+
if (!this.redriveUnresolvedSignalled.has(message.id)) {
|
|
4036
|
+
this.redriveUnresolvedSignalled.add(message.id);
|
|
4037
|
+
void this.postSignal(conv.id, message.id, "redrive_unresolved");
|
|
4038
|
+
}
|
|
4039
|
+
return "unresolved";
|
|
4040
|
+
}
|
|
4041
|
+
/** Clear all `unresolved`/failure-streak trackers for a row (any non-`unresolved` outcome). */
|
|
4042
|
+
clearRedriveUnresolved(messageId) {
|
|
4043
|
+
this.redriveUnresolvedSince.delete(messageId);
|
|
4044
|
+
this.redriveUnresolvedSignalled.delete(messageId);
|
|
4045
|
+
this.redrivePollFailures.delete(messageId);
|
|
4046
|
+
this.redriveOutcomeUnreportedSignalled.delete(messageId);
|
|
4047
|
+
}
|
|
4048
|
+
/**
|
|
4049
|
+
* Class B (#1340): the runner DECIDED an outcome (reattach/settle/fail_permanent)
|
|
4050
|
+
* but its own PATCH to record it failed. Deliberately NOT bounded like
|
|
4051
|
+
* `resolveRedriveUnresolved`, for two load-bearing reasons: (1) the only
|
|
4052
|
+
* remedy is a PATCH to the server — the very call that is failing, so a bound
|
|
4053
|
+
* would act by retrying it; (2) this class is self-clearing by construction
|
|
4054
|
+
* (the next successful PATCH resolves it via `clearRedriveUnresolved`),
|
|
4055
|
+
* unlike #1340's deterministic, permanent shape. Fires at most once per
|
|
4056
|
+
* (message, outcome) streak.
|
|
4057
|
+
*/
|
|
4058
|
+
signalRedriveOutcomeUnreported(conv, message, outcome) {
|
|
4059
|
+
if (this.redriveOutcomeUnreportedSignalled.get(message.id) === outcome) return;
|
|
4060
|
+
this.redriveOutcomeUnreportedSignalled.set(message.id, outcome);
|
|
4061
|
+
void this.postSignal(conv.id, message.id, "redrive_outcome_unreported", {
|
|
4062
|
+
attempted_outcome: outcome
|
|
4063
|
+
});
|
|
4064
|
+
}
|
|
4065
|
+
/**
|
|
4066
|
+
* Record one poll outcome toward the re-drive fence's consecutive-identical-
|
|
4067
|
+
* failure streak (#1348) and return the resulting count. `signature === null`
|
|
4068
|
+
* (a thrown exception, H1) always clears the streak and returns `0` — it is
|
|
4069
|
+
* never countable. Otherwise the streak continues only when BOTH the session
|
|
4070
|
+
* and the signature match the previous failure; anything else (a different
|
|
4071
|
+
* session, or the same session failing a DIFFERENT way) starts a fresh streak
|
|
4072
|
+
* at `1`.
|
|
4073
|
+
*/
|
|
4074
|
+
recordRedrivePollFailure(messageId, sessionId, signature) {
|
|
4075
|
+
if (signature === null) {
|
|
4076
|
+
this.redrivePollFailures.delete(messageId);
|
|
4077
|
+
return 0;
|
|
4078
|
+
}
|
|
4079
|
+
const existing = this.redrivePollFailures.get(messageId);
|
|
4080
|
+
if (existing && existing.sessionId === sessionId && existing.signature === signature) {
|
|
4081
|
+
existing.count += 1;
|
|
4082
|
+
return existing.count;
|
|
4083
|
+
}
|
|
4084
|
+
this.redrivePollFailures.set(messageId, { sessionId, signature, count: 1 });
|
|
4085
|
+
return 1;
|
|
4086
|
+
}
|
|
4087
|
+
/**
|
|
4088
|
+
* Record that `sessionId` is no longer a valid binding for `conversationId`
|
|
4089
|
+
* (#553). Keyed by conversation and hard-capped, so it cannot grow with the
|
|
4090
|
+
* number of failures — see the `supersededSessions` field doc.
|
|
4091
|
+
*/
|
|
4092
|
+
supersede(conversationId, sessionId) {
|
|
4093
|
+
this.supersededSessions.delete(conversationId);
|
|
4094
|
+
this.supersededSessions.set(conversationId, sessionId);
|
|
4095
|
+
while (this.supersededSessions.size > MAX_SUPERSEDED_CONVERSATIONS) {
|
|
4096
|
+
const oldest = this.supersededSessions.keys().next().value;
|
|
4097
|
+
if (oldest === void 0) return;
|
|
4098
|
+
this.supersededSessions.delete(oldest);
|
|
4099
|
+
}
|
|
4100
|
+
}
|
|
4101
|
+
/** Whether `sessionId` is the session this conversation has abandoned (#553). */
|
|
4102
|
+
isSuperseded(conversationId, sessionId) {
|
|
4103
|
+
return this.supersededSessions.get(conversationId) === sessionId;
|
|
2417
4104
|
}
|
|
4105
|
+
/**
|
|
4106
|
+
* Resolve the opencode session to run this conversation's turns in.
|
|
4107
|
+
*
|
|
4108
|
+
* `refusedSessionId` is set when the #553 guard fired — i.e. the persisted
|
|
4109
|
+
* binding was an id this runner had abandoned, so a resurrection genuinely
|
|
4110
|
+
* happened and a fresh session was bound instead. The caller reports it.
|
|
4111
|
+
*/
|
|
2418
4112
|
async ensureSession(conv) {
|
|
2419
4113
|
const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
|
|
4114
|
+
if (bound && this.isSuperseded(conv.id, bound)) {
|
|
4115
|
+
this.log({
|
|
4116
|
+
level: "warn",
|
|
4117
|
+
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.`,
|
|
4118
|
+
conversation_id: conv.id
|
|
4119
|
+
});
|
|
4120
|
+
this.sessions.delete(conv.id);
|
|
4121
|
+
return { sessionId: await this.createAndBindSession(conv.id), refusedSessionId: bound };
|
|
4122
|
+
}
|
|
2420
4123
|
if (bound) {
|
|
2421
4124
|
const exists = await sessionExists(this.port, bound);
|
|
2422
4125
|
if (exists === false) {
|
|
@@ -2426,12 +4129,12 @@ var ChannelDriver = class {
|
|
|
2426
4129
|
conversation_id: conv.id
|
|
2427
4130
|
});
|
|
2428
4131
|
this.sessions.delete(conv.id);
|
|
2429
|
-
return this.createAndBindSession(conv.id);
|
|
4132
|
+
return { sessionId: await this.createAndBindSession(conv.id) };
|
|
2430
4133
|
}
|
|
2431
4134
|
this.sessions.set(conv.id, bound);
|
|
2432
|
-
return bound;
|
|
4135
|
+
return { sessionId: bound };
|
|
2433
4136
|
}
|
|
2434
|
-
return this.createAndBindSession(conv.id);
|
|
4137
|
+
return { sessionId: await this.createAndBindSession(conv.id) };
|
|
2435
4138
|
}
|
|
2436
4139
|
/**
|
|
2437
4140
|
* Create a fresh OpenCode session for a conversation, cache the binding, and
|
|
@@ -2442,7 +4145,12 @@ var ChannelDriver = class {
|
|
|
2442
4145
|
const directory = await this.resolveOpenCodeDirectory();
|
|
2443
4146
|
const sessionId = await createOpenCodeSession(this.port, directory);
|
|
2444
4147
|
this.sessions.set(conversationId, sessionId);
|
|
2445
|
-
await this.persistSession(conversationId, sessionId).catch(() => {
|
|
4148
|
+
await this.persistSession(conversationId, sessionId).catch((err) => {
|
|
4149
|
+
this.log({
|
|
4150
|
+
level: "warn",
|
|
4151
|
+
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: ${err instanceof Error ? err.message : String(err)}`,
|
|
4152
|
+
conversation_id: conversationId
|
|
4153
|
+
});
|
|
2446
4154
|
});
|
|
2447
4155
|
return sessionId;
|
|
2448
4156
|
}
|
|
@@ -2511,7 +4219,7 @@ var ChannelDriver = class {
|
|
|
2511
4219
|
}
|
|
2512
4220
|
/**
|
|
2513
4221
|
* Fetch ONE inbound image's bytes through Evident's WI-6 endpoint
|
|
2514
|
-
* (`GET {apiUrl}/
|
|
4222
|
+
* (`GET {apiUrl}/runners/{agentId}/attachments/{messageId}/{index}`) using the
|
|
2515
4223
|
* existing authenticated fetch, and base64-encode into a
|
|
2516
4224
|
* `data:<mime>;base64,<…>` URL for the opencode `file` part's `url`.
|
|
2517
4225
|
*
|
|
@@ -2519,15 +4227,38 @@ var ChannelDriver = class {
|
|
|
2519
4227
|
* (not-owned / out-of-range / deleted-at-source / workspace gone) / 413
|
|
2520
4228
|
* (over-cap). On ANY non-2xx or thrown failure we return `null` so the caller
|
|
2521
4229
|
* OMITS that one image and the text turn still sends — NEVER throws the turn.
|
|
2522
|
-
*
|
|
4230
|
+
* A 404 body carrying `{ reason: 'needs_reauth' }` (#547 — the server CONFIRMED
|
|
4231
|
+
* a Slack `files:read` scope problem via `files.info`) instead resolves the
|
|
4232
|
+
* `AttachmentFetchNeedsReauth` sentinel, so the in-thread note can steer the
|
|
4233
|
+
* user to reconnect Slack instead of a generic "unavailable". Failures are
|
|
4234
|
+
* logged with context (no silent swallow).
|
|
2523
4235
|
*/
|
|
2524
4236
|
async fetchAttachmentDataUrl(messageId, index, mime) {
|
|
2525
4237
|
try {
|
|
2526
4238
|
const res = await this.fetchImpl(
|
|
2527
|
-
`${this.apiUrl}/
|
|
4239
|
+
`${this.apiUrl}/runners/${this.agentId}/attachments/${messageId}/${index}`,
|
|
2528
4240
|
{ headers: { Authorization: this.getAuthHeader() } }
|
|
2529
4241
|
);
|
|
2530
4242
|
if (!res.ok) {
|
|
4243
|
+
let reason;
|
|
4244
|
+
try {
|
|
4245
|
+
const body = await res.json();
|
|
4246
|
+
if (body && typeof body.reason === "string") reason = body.reason;
|
|
4247
|
+
} catch (parseErr) {
|
|
4248
|
+
this.log({
|
|
4249
|
+
level: "debug",
|
|
4250
|
+
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`,
|
|
4251
|
+
message_id: messageId
|
|
4252
|
+
});
|
|
4253
|
+
}
|
|
4254
|
+
if (reason === "needs_reauth") {
|
|
4255
|
+
this.log({
|
|
4256
|
+
level: "error",
|
|
4257
|
+
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)`,
|
|
4258
|
+
message_id: messageId
|
|
4259
|
+
});
|
|
4260
|
+
return { needsReauth: true };
|
|
4261
|
+
}
|
|
2531
4262
|
this.log({
|
|
2532
4263
|
level: "error",
|
|
2533
4264
|
message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} returned HTTP ${res.status} \u2014 omitting this image (text turn proceeds)`,
|
|
@@ -2567,6 +4298,9 @@ var ChannelDriver = class {
|
|
|
2567
4298
|
if (this.attachmentsSkippedSignalled.has(messageId)) return;
|
|
2568
4299
|
this.attachmentsSkippedSignalled.add(messageId);
|
|
2569
4300
|
const skippedReason = capabilityUnknown ? "unknown" : "unsupported";
|
|
4301
|
+
const failedReason = outcomes.some(
|
|
4302
|
+
(o) => o.status === "failed" && o.reason === "needs_reauth"
|
|
4303
|
+
) ? "needs_reauth" : void 0;
|
|
2570
4304
|
this.log({
|
|
2571
4305
|
level: "info",
|
|
2572
4306
|
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`,
|
|
@@ -2576,7 +4310,8 @@ var ChannelDriver = class {
|
|
|
2576
4310
|
void this.postSignal(conversationId, messageId, "attachments_skipped", {
|
|
2577
4311
|
skipped,
|
|
2578
4312
|
failed,
|
|
2579
|
-
...skipped > 0 ? { skipped_reason: skippedReason } : {}
|
|
4313
|
+
...skipped > 0 ? { skipped_reason: skippedReason } : {},
|
|
4314
|
+
...failedReason ? { failed_reason: failedReason } : {}
|
|
2580
4315
|
});
|
|
2581
4316
|
}
|
|
2582
4317
|
/** Register a freshly-dispatched message with its session's watcher state. */
|
|
@@ -2607,12 +4342,17 @@ var ChannelDriver = class {
|
|
|
2607
4342
|
stuckReported: false,
|
|
2608
4343
|
lastAliveAt: 0,
|
|
2609
4344
|
aliveInFlight: false,
|
|
4345
|
+
titleSynced: false,
|
|
4346
|
+
titleSyncInFlight: false,
|
|
2610
4347
|
awaitingHumanLatched: false,
|
|
2611
4348
|
pausedOnQuestion: false,
|
|
2612
4349
|
pausedOnPermission: false,
|
|
2613
4350
|
pausedClearConfirmed: false,
|
|
2614
4351
|
pausedInFlight: false,
|
|
2615
|
-
deliveryDeadlineAnchored: false
|
|
4352
|
+
deliveryDeadlineAnchored: false,
|
|
4353
|
+
b2PinnedSinceMs: 0,
|
|
4354
|
+
b2LastDescendantCheckMs: 0,
|
|
4355
|
+
b2AbandonedSignalled: false
|
|
2616
4356
|
});
|
|
2617
4357
|
}
|
|
2618
4358
|
/**
|
|
@@ -2627,9 +4367,10 @@ var ChannelDriver = class {
|
|
|
2627
4367
|
* opencode reports ACTIVELY `running` is watched to completion (its liveness
|
|
2628
4368
|
* heartbeat keeps the cron off its row), while a re-adopted turn that is paused
|
|
2629
4369
|
* awaiting a human — or queued/unreachable — is still bounded by `deadline` and
|
|
2630
|
-
* handed to the cron.
|
|
2631
|
-
*
|
|
2632
|
-
*
|
|
4370
|
+
* handed to the cron. Real invariant (#965): the cron MAY reclaim a row this
|
|
4371
|
+
* runner still holds; a reclaimed row that already ran is never re-dispatched
|
|
4372
|
+
* while opencode reports its turn ongoing (readopt's own gate here, and the
|
|
4373
|
+
* `pending`-row re-drive fence, `resolveRedrive`). `dispatchedAt` stays `now`
|
|
2633
4374
|
* (only the appear-guard uses it).
|
|
2634
4375
|
*
|
|
2635
4376
|
* `evidentMessageId` addresses the SERVER row (for markProcessing/markDone);
|
|
@@ -2680,12 +4421,17 @@ var ChannelDriver = class {
|
|
|
2680
4421
|
// with no extra `re_adopted` signal needed (folds old WI-6).
|
|
2681
4422
|
lastAliveAt: 0,
|
|
2682
4423
|
aliveInFlight: false,
|
|
4424
|
+
titleSynced: false,
|
|
4425
|
+
titleSyncInFlight: false,
|
|
2683
4426
|
awaitingHumanLatched: false,
|
|
2684
4427
|
pausedOnQuestion: false,
|
|
2685
4428
|
pausedOnPermission: false,
|
|
2686
4429
|
pausedClearConfirmed: false,
|
|
2687
4430
|
pausedInFlight: false,
|
|
2688
|
-
deliveryDeadlineAnchored: false
|
|
4431
|
+
deliveryDeadlineAnchored: false,
|
|
4432
|
+
b2PinnedSinceMs: 0,
|
|
4433
|
+
b2LastDescendantCheckMs: 0,
|
|
4434
|
+
b2AbandonedSignalled: false
|
|
2689
4435
|
});
|
|
2690
4436
|
}
|
|
2691
4437
|
/**
|
|
@@ -2847,58 +4593,7 @@ var ChannelDriver = class {
|
|
|
2847
4593
|
}
|
|
2848
4594
|
}
|
|
2849
4595
|
if (state === "done") {
|
|
2850
|
-
this.
|
|
2851
|
-
if (!inFlight.done) {
|
|
2852
|
-
this.log({
|
|
2853
|
-
level: "info",
|
|
2854
|
-
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
2855
|
-
conversation_id: conv.id,
|
|
2856
|
-
message_id: inFlight.evidentMessageId
|
|
2857
|
-
});
|
|
2858
|
-
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
2859
|
-
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
2860
|
-
try {
|
|
2861
|
-
await this.markDone(
|
|
2862
|
-
conv.id,
|
|
2863
|
-
inFlight.evidentMessageId,
|
|
2864
|
-
sessionId,
|
|
2865
|
-
inFlight.opencodeMessageId,
|
|
2866
|
-
title,
|
|
2867
|
-
usage
|
|
2868
|
-
);
|
|
2869
|
-
} catch (err) {
|
|
2870
|
-
if (err instanceof ChannelAuthError) throw err;
|
|
2871
|
-
if (err instanceof ChannelTerminalError) {
|
|
2872
|
-
this.log({
|
|
2873
|
-
level: "warn",
|
|
2874
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
2875
|
-
conversation_id: conv.id,
|
|
2876
|
-
message_id: inFlight.evidentMessageId
|
|
2877
|
-
});
|
|
2878
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2879
|
-
return;
|
|
2880
|
-
}
|
|
2881
|
-
if (this.now() >= inFlight.deadline) {
|
|
2882
|
-
this.log({
|
|
2883
|
-
level: "warn",
|
|
2884
|
-
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)}`,
|
|
2885
|
-
conversation_id: conv.id,
|
|
2886
|
-
message_id: inFlight.evidentMessageId
|
|
2887
|
-
});
|
|
2888
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
2889
|
-
return;
|
|
2890
|
-
}
|
|
2891
|
-
this.log({
|
|
2892
|
-
level: "warn",
|
|
2893
|
-
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
2894
|
-
conversation_id: conv.id,
|
|
2895
|
-
message_id: inFlight.evidentMessageId
|
|
2896
|
-
});
|
|
2897
|
-
return;
|
|
2898
|
-
}
|
|
2899
|
-
inFlight.done = true;
|
|
2900
|
-
}
|
|
2901
|
-
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4596
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
2902
4597
|
return;
|
|
2903
4598
|
}
|
|
2904
4599
|
if (state === "failed") {
|
|
@@ -2912,8 +4607,16 @@ var ChannelDriver = class {
|
|
|
2912
4607
|
message_id: inFlight.evidentMessageId
|
|
2913
4608
|
});
|
|
2914
4609
|
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4610
|
+
const failure = await this.classifyModelAuthFailure(messages, inFlight.opencodeMessageId);
|
|
2915
4611
|
try {
|
|
2916
|
-
await this.markFailed(
|
|
4612
|
+
await this.markFailed(
|
|
4613
|
+
conv.id,
|
|
4614
|
+
inFlight.evidentMessageId,
|
|
4615
|
+
sessionId,
|
|
4616
|
+
error2,
|
|
4617
|
+
usage,
|
|
4618
|
+
failure
|
|
4619
|
+
);
|
|
2917
4620
|
} catch (err) {
|
|
2918
4621
|
if (err instanceof ChannelAuthError) throw err;
|
|
2919
4622
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -2958,6 +4661,44 @@ var ChannelDriver = class {
|
|
|
2958
4661
|
});
|
|
2959
4662
|
}
|
|
2960
4663
|
const activelyRunning = state === "running" && !awaitingHuman;
|
|
4664
|
+
const pinnedNow = activelyRunning && isPreamblePinnedRunning(messages, inFlight.opencodeMessageId);
|
|
4665
|
+
const snapshotReadable = messages != null && messages.length > 0;
|
|
4666
|
+
if (!pinnedNow) {
|
|
4667
|
+
if (snapshotReadable) {
|
|
4668
|
+
inFlight.b2PinnedSinceMs = 0;
|
|
4669
|
+
inFlight.b2LastDescendantCheckMs = 0;
|
|
4670
|
+
inFlight.b2AbandonedSignalled = false;
|
|
4671
|
+
}
|
|
4672
|
+
} else {
|
|
4673
|
+
if (inFlight.b2AbandonedSignalled) {
|
|
4674
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4675
|
+
return;
|
|
4676
|
+
}
|
|
4677
|
+
if (inFlight.b2PinnedSinceMs === 0) inFlight.b2PinnedSinceMs = this.now();
|
|
4678
|
+
const pinnedForMs = this.now() - inFlight.b2PinnedSinceMs;
|
|
4679
|
+
if (pinnedForMs >= B2_ABANDONMENT_MIN_PINNED_MS && this.now() - inFlight.b2LastDescendantCheckMs >= B2_ABANDONMENT_RECHECK_MS) {
|
|
4680
|
+
inFlight.b2LastDescendantCheckMs = this.now();
|
|
4681
|
+
const descendantOngoing = await this.isAnyDescendantSessionOngoing(sessionId);
|
|
4682
|
+
if (isB2AbandonmentConfirmed({
|
|
4683
|
+
pinnedForMs,
|
|
4684
|
+
minPinnedMs: B2_ABANDONMENT_MIN_PINNED_MS,
|
|
4685
|
+
descendantOngoing
|
|
4686
|
+
})) {
|
|
4687
|
+
inFlight.b2AbandonedSignalled = true;
|
|
4688
|
+
this.log({
|
|
4689
|
+
level: "warn",
|
|
4690
|
+
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`,
|
|
4691
|
+
conversation_id: conv.id,
|
|
4692
|
+
message_id: id
|
|
4693
|
+
});
|
|
4694
|
+
void this.postSignal(conv.id, id, "b2_abandoned_resolved", {
|
|
4695
|
+
watched_for_ms: pinnedForMs
|
|
4696
|
+
});
|
|
4697
|
+
await this.settleMessageDone(sessionId, watcher, inFlight, messages);
|
|
4698
|
+
return;
|
|
4699
|
+
}
|
|
4700
|
+
}
|
|
4701
|
+
}
|
|
2961
4702
|
if (activelyRunning && this.now() - inFlight.processingAnchorMs >= ABSOLUTE_MAX_PROCESSING_MS) {
|
|
2962
4703
|
this.log({
|
|
2963
4704
|
level: "warn",
|
|
@@ -2977,6 +4718,18 @@ var ChannelDriver = class {
|
|
|
2977
4718
|
inFlight.aliveInFlight = false;
|
|
2978
4719
|
if (ok) inFlight.lastAliveAt = this.now();
|
|
2979
4720
|
});
|
|
4721
|
+
if (!inFlight.titleSynced && !inFlight.titleSyncInFlight) {
|
|
4722
|
+
inFlight.titleSyncInFlight = true;
|
|
4723
|
+
void this.resolveSessionTitle(sessionId, conv.id).then(async (title) => {
|
|
4724
|
+
if (!title) {
|
|
4725
|
+
inFlight.titleSyncInFlight = false;
|
|
4726
|
+
return;
|
|
4727
|
+
}
|
|
4728
|
+
const ok = await this.patchConversationTitle(conv.id, title);
|
|
4729
|
+
inFlight.titleSyncInFlight = false;
|
|
4730
|
+
if (ok) inFlight.titleSynced = true;
|
|
4731
|
+
});
|
|
4732
|
+
}
|
|
2980
4733
|
}
|
|
2981
4734
|
if (awaitingHuman) {
|
|
2982
4735
|
if (!inFlight.awaitingHumanLatched) {
|
|
@@ -3014,6 +4767,70 @@ var ChannelDriver = class {
|
|
|
3014
4767
|
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
3015
4768
|
}
|
|
3016
4769
|
}
|
|
4770
|
+
/**
|
|
4771
|
+
* Settle a message whose run-state has resolved `'done'` — extracted verbatim
|
|
4772
|
+
* (pure refactor, no behavior change) from `serviceInFlightMessage`'s former
|
|
4773
|
+
* inline `state === 'done'` branch body, so a SECOND caller (the #721
|
|
4774
|
+
* b2-abandonment resolution) can reach the exact same completion behavior
|
|
4775
|
+
* (delivery-deadline anchoring, title resolution, usage extraction, and
|
|
4776
|
+
* `markDone`'s auth/terminal/transient-retry discipline) without duplicating it
|
|
4777
|
+
* and risking the two copies silently drifting apart.
|
|
4778
|
+
*/
|
|
4779
|
+
async settleMessageDone(sessionId, watcher, inFlight, messages) {
|
|
4780
|
+
const conv = watcher.conv;
|
|
4781
|
+
this.anchorDeliveryDeadline(inFlight);
|
|
4782
|
+
if (!inFlight.done) {
|
|
4783
|
+
this.log({
|
|
4784
|
+
level: "info",
|
|
4785
|
+
message: `Message ${inFlight.evidentMessageId.slice(0, 8)} completed \u2014 marking done`,
|
|
4786
|
+
conversation_id: conv.id,
|
|
4787
|
+
message_id: inFlight.evidentMessageId
|
|
4788
|
+
});
|
|
4789
|
+
const title = await this.resolveSessionTitle(sessionId, watcher.conv.id);
|
|
4790
|
+
const usage = messageUsage(messages, inFlight.opencodeMessageId);
|
|
4791
|
+
try {
|
|
4792
|
+
await this.markDone(
|
|
4793
|
+
conv.id,
|
|
4794
|
+
inFlight.evidentMessageId,
|
|
4795
|
+
sessionId,
|
|
4796
|
+
inFlight.opencodeMessageId,
|
|
4797
|
+
title,
|
|
4798
|
+
usage
|
|
4799
|
+
);
|
|
4800
|
+
} catch (err) {
|
|
4801
|
+
if (err instanceof ChannelAuthError) throw err;
|
|
4802
|
+
if (err instanceof ChannelTerminalError) {
|
|
4803
|
+
this.log({
|
|
4804
|
+
level: "warn",
|
|
4805
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (terminal HTTP ${err.status}) \u2014 leaving for the cron safety net: ${err.message}`,
|
|
4806
|
+
conversation_id: conv.id,
|
|
4807
|
+
message_id: inFlight.evidentMessageId
|
|
4808
|
+
});
|
|
4809
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4810
|
+
return;
|
|
4811
|
+
}
|
|
4812
|
+
if (this.now() >= inFlight.deadline) {
|
|
4813
|
+
this.log({
|
|
4814
|
+
level: "warn",
|
|
4815
|
+
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)}`,
|
|
4816
|
+
conversation_id: conv.id,
|
|
4817
|
+
message_id: inFlight.evidentMessageId
|
|
4818
|
+
});
|
|
4819
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4820
|
+
return;
|
|
4821
|
+
}
|
|
4822
|
+
this.log({
|
|
4823
|
+
level: "warn",
|
|
4824
|
+
message: `Failed to mark message ${inFlight.evidentMessageId.slice(0, 8)} done (will retry next tick): ${err instanceof Error ? err.message : String(err)}`,
|
|
4825
|
+
conversation_id: conv.id,
|
|
4826
|
+
message_id: inFlight.evidentMessageId
|
|
4827
|
+
});
|
|
4828
|
+
return;
|
|
4829
|
+
}
|
|
4830
|
+
inFlight.done = true;
|
|
4831
|
+
}
|
|
4832
|
+
this.removeInFlight(watcher, inFlight.evidentMessageId);
|
|
4833
|
+
}
|
|
3017
4834
|
// Restart recovery: re-adopt `processing` messages (ADR-0046, WI-3/4/5)
|
|
3018
4835
|
/**
|
|
3019
4836
|
* Re-adopt this agent's `processing` messages on drain (ADR-0046 Decision §1).
|
|
@@ -3113,7 +4930,10 @@ var ChannelDriver = class {
|
|
|
3113
4930
|
* re-dispatched (at most once, see `forceReadoptRun`):
|
|
3114
4931
|
* - `done` → `markDone` now (guarded like the watcher's done branch);
|
|
3115
4932
|
* - `failed` → `markFailed` with the surfaced error (issue #182), so an
|
|
3116
|
-
* errored turn is reported failed on restart, NOT re-dispatched
|
|
4933
|
+
* errored turn is reported failed on restart, NOT re-dispatched —
|
|
4934
|
+
* EXCEPT a restart-ABORTED turn under a not-ongoing session,
|
|
4935
|
+
* which is a restart orphan wearing a terminal error and is
|
|
4936
|
+
* re-dispatched instead (issue #1310, see the branch below);
|
|
3117
4937
|
* - `running`/`queued` → re-attach a watcher via `registerReadopted` (no re-dispatch),
|
|
3118
4938
|
* tracking the stored id so the reply correlates by it;
|
|
3119
4939
|
* - `unknown`/null id → re-dispatch (opencode assigns a fresh id) + attach a watcher.
|
|
@@ -3177,9 +4997,19 @@ var ChannelDriver = class {
|
|
|
3177
4997
|
void this.postSignal(row.conversation_id, row.id, "readopt_done");
|
|
3178
4998
|
return;
|
|
3179
4999
|
}
|
|
3180
|
-
|
|
5000
|
+
const restartAborted = state === "failed" && sessionOngoing === false && isAbortedTerminalReply(messages, ocId ?? "");
|
|
5001
|
+
if (restartAborted) {
|
|
5002
|
+
this.log({
|
|
5003
|
+
level: "info",
|
|
5004
|
+
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`,
|
|
5005
|
+
conversation_id: row.conversation_id,
|
|
5006
|
+
message_id: row.id
|
|
5007
|
+
});
|
|
5008
|
+
}
|
|
5009
|
+
if (state === "failed" && !restartAborted) {
|
|
3181
5010
|
const error2 = messageError(messages, ocId ?? "") ?? void 0;
|
|
3182
5011
|
const usage = messageUsage(messages, ocId ?? "");
|
|
5012
|
+
const failure = await this.classifyModelAuthFailure(messages, ocId ?? "");
|
|
3183
5013
|
this.log({
|
|
3184
5014
|
level: "error",
|
|
3185
5015
|
message: `Re-adopt: message ${row.id.slice(0, 8)} errored while unwatched \u2014 marking failed: ${error2 ?? "(no error text)"}`,
|
|
@@ -3187,7 +5017,7 @@ var ChannelDriver = class {
|
|
|
3187
5017
|
message_id: row.id
|
|
3188
5018
|
});
|
|
3189
5019
|
try {
|
|
3190
|
-
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage);
|
|
5020
|
+
await this.markFailed(row.conversation_id, row.id, sessionId, error2, usage, failure);
|
|
3191
5021
|
} catch (err) {
|
|
3192
5022
|
if (err instanceof ChannelAuthError) throw err;
|
|
3193
5023
|
if (err instanceof ChannelTerminalError) {
|
|
@@ -3454,7 +5284,8 @@ var ChannelDriver = class {
|
|
|
3454
5284
|
opencode_model: row.opencode_model,
|
|
3455
5285
|
source_message_id: row.source_message_id,
|
|
3456
5286
|
slack_user_id: row.slack_user_id,
|
|
3457
|
-
attachments: row.attachments ?? null
|
|
5287
|
+
attachments: row.attachments ?? null,
|
|
5288
|
+
opencode_message_id: row.opencode_message_id
|
|
3458
5289
|
};
|
|
3459
5290
|
}
|
|
3460
5291
|
/**
|
|
@@ -3593,6 +5424,47 @@ var ChannelDriver = class {
|
|
|
3593
5424
|
}
|
|
3594
5425
|
return false;
|
|
3595
5426
|
}
|
|
5427
|
+
/**
|
|
5428
|
+
* Tri-state variant of the upward parentID membership walk (#721), used ONLY
|
|
5429
|
+
* by `isAnyDescendantSessionOngoing`. Walks the SAME cached
|
|
5430
|
+
* `resolveSessionParent` chain `sessionBelongsTo` uses above, but — unlike
|
|
5431
|
+
* `sessionBelongsTo`, which deliberately collapses "confirmed not a
|
|
5432
|
+
* descendant" and "the walk's fetch failed" into the same `false` (safe for
|
|
5433
|
+
* its OTHER callers: interaction attribution and the recovery-path
|
|
5434
|
+
* `isAnyDescendantSessionAlive`, both of which just retry next tick with no
|
|
5435
|
+
* safety consequence either way) — this variant keeps those two outcomes
|
|
5436
|
+
* SEPARATE, because `isAnyDescendantSessionOngoing`'s caller
|
|
5437
|
+
* (`isB2AbandonmentConfirmed`) must never treat "couldn't tell" as "confirmed
|
|
5438
|
+
* not ongoing".
|
|
5439
|
+
*
|
|
5440
|
+
* Return contract:
|
|
5441
|
+
* - `true` → the walk reached `rootSessionId` — `sessionId` IS a descendant.
|
|
5442
|
+
* - `false` → the walk reached a definitive, parent-less root session
|
|
5443
|
+
* WITHOUT ever matching `rootSessionId` — `sessionId` is
|
|
5444
|
+
* CONFIRMED NOT a descendant of it.
|
|
5445
|
+
* - `null` → INDETERMINATE: a `GET /session/:id` fetch failed partway
|
|
5446
|
+
* through the walk (`resolveSessionParent` returned `undefined`),
|
|
5447
|
+
* or the depth cap (32) was hit without a definitive answer (a
|
|
5448
|
+
* pathological/cyclic chain proves nothing either way). NEVER
|
|
5449
|
+
* treat this the same as `false` — see `sessionBelongsTo`'s own
|
|
5450
|
+
* doc comment above for why that collapse is safe THERE but not
|
|
5451
|
+
* here.
|
|
5452
|
+
*
|
|
5453
|
+
* `sessionBelongsTo` itself is UNCHANGED — this is an additive helper scoped
|
|
5454
|
+
* to the live-path descendant check, not a modification of shared code used
|
|
5455
|
+
* by interaction attribution or the recovery path.
|
|
5456
|
+
*/
|
|
5457
|
+
async resolveSessionMembership(sessionId, rootSessionId) {
|
|
5458
|
+
let current = sessionId;
|
|
5459
|
+
for (let depth = 0; current && depth < 32; depth++) {
|
|
5460
|
+
if (current === rootSessionId) return true;
|
|
5461
|
+
const parent = await this.resolveSessionParent(current);
|
|
5462
|
+
if (parent === void 0) return null;
|
|
5463
|
+
if (parent === null) return false;
|
|
5464
|
+
current = parent;
|
|
5465
|
+
}
|
|
5466
|
+
return null;
|
|
5467
|
+
}
|
|
3596
5468
|
/**
|
|
3597
5469
|
* Resolve (and cache) a session's `parentID` via `GET /session/:id`. Returns
|
|
3598
5470
|
* `null` for a root session (no parent) and `undefined` when opencode is
|
|
@@ -3615,19 +5487,36 @@ var ChannelDriver = class {
|
|
|
3615
5487
|
if (parent !== void 0) this.sessionParents.set(sessionId, parent);
|
|
3616
5488
|
return parent;
|
|
3617
5489
|
}
|
|
5490
|
+
/**
|
|
5491
|
+
* OpenCode's synchronous default session title (e.g.
|
|
5492
|
+
* `"New session - 1737800000000"`), assigned immediately when a session is
|
|
5493
|
+
* created — before OpenCode's async LLM-based auto-titling later renames it
|
|
5494
|
+
* mid-turn (#549). Matched by this literal, case-sensitive prefix only; the
|
|
5495
|
+
* timestamp suffix's exact format is deliberately NOT matched, since the prefix
|
|
5496
|
+
* alone is the stable, cheap signal and over-anchoring on the timestamp
|
|
5497
|
+
* representation risks silently breaking if OpenCode ever changes it. Accepted
|
|
5498
|
+
* trade-off: a genuine LLM-assigned title that happens to literally start with
|
|
5499
|
+
* this prefix would also fail to latch (see `resolveSessionTitle`) —
|
|
5500
|
+
* vanishingly unlikely in practice, and deliberately not engineered around.
|
|
5501
|
+
*/
|
|
5502
|
+
static OPENCODE_DEFAULT_TITLE_PREFIX = /^New session - /;
|
|
3618
5503
|
/**
|
|
3619
5504
|
* Resolve (and cache in `sessionTitles`) the OpenCode session TITLE (#310) so the
|
|
3620
5505
|
* status PATCH can carry it into the "Live sessions" list. Driver-level cache so
|
|
3621
5506
|
* BOTH the watcher completion path and the restart-recovery re-adopt path (which
|
|
3622
5507
|
* has no watcher) can use it. `conversationId` is passed only for log context.
|
|
3623
5508
|
* Best-effort:
|
|
3624
|
-
* - a resolved NON-EMPTY title
|
|
5509
|
+
* - a resolved NON-EMPTY title that does NOT match
|
|
5510
|
+
* `OPENCODE_DEFAULT_TITLE_PREFIX` is cached and terminal (a real session name
|
|
3625
5511
|
* won't later un-name), so we do NOT re-GET `/session/:id` every tick;
|
|
3626
|
-
* - while the title is still absent
|
|
3627
|
-
*
|
|
3628
|
-
*
|
|
3629
|
-
*
|
|
3630
|
-
*
|
|
5512
|
+
* - while the title is still absent, empty, or matches the OpenCode
|
|
5513
|
+
* placeholder prefix (#549) we do NOT latch it — OpenCode names sessions
|
|
5514
|
+
* asynchronously mid-turn, so an early call (e.g. at `processing`) must leave
|
|
5515
|
+
* the cache unresolved and re-fetch on the next need so a later call (e.g. at
|
|
5516
|
+
* `done`) picks up the name assigned in the meantime. Such a call returns
|
|
5517
|
+
* `null` (omit the title on THIS PATCH) without caching. If a session is
|
|
5518
|
+
* never renamed, the title is omitted forever rather than ever persisting
|
|
5519
|
+
* the placeholder as a last resort;
|
|
3631
5520
|
* - a failed request likewise leaves the cache unresolved (retry next need)
|
|
3632
5521
|
* and returns `null` — it must NEVER throw or block completion.
|
|
3633
5522
|
* A failure is logged with agent/session context (no silent catch).
|
|
@@ -3640,7 +5529,7 @@ var ChannelDriver = class {
|
|
|
3640
5529
|
if (res.ok) {
|
|
3641
5530
|
const body = await res.json();
|
|
3642
5531
|
const title = body && typeof body.title === "string" ? body.title.trim() : "";
|
|
3643
|
-
if (title.length > 0) {
|
|
5532
|
+
if (title.length > 0 && !_ChannelDriver.OPENCODE_DEFAULT_TITLE_PREFIX.test(title)) {
|
|
3644
5533
|
this.sessionTitles.set(sessionId, title);
|
|
3645
5534
|
return title;
|
|
3646
5535
|
}
|
|
@@ -3660,6 +5549,54 @@ var ChannelDriver = class {
|
|
|
3660
5549
|
}
|
|
3661
5550
|
return null;
|
|
3662
5551
|
}
|
|
5552
|
+
/**
|
|
5553
|
+
* Best-effort mid-turn title sync (#711 follow-up): PATCH a resolved OpenCode
|
|
5554
|
+
* session title onto the conversation via the PLAIN conversation-update
|
|
5555
|
+
* endpoint (`PATCH /runners/:agentId/conversations/:conversationId`) — NOT the
|
|
5556
|
+
* message-status endpoint `markProcessing`/`markDone` use. Deliberately a
|
|
5557
|
+
* separate, lighter call: it carries no `status`, so it cannot re-trigger the
|
|
5558
|
+
* `processing`/`done` transition side effects (Slack notices, activity-log
|
|
5559
|
+
* rows, delivery jobs) those PATCHes gate on `transitioned` — this call only
|
|
5560
|
+
* ever touches `conversations.title`. That route (`routes/conversations.ts`)
|
|
5561
|
+
* skips a title write matching the stored value, so a redundant call with the
|
|
5562
|
+
* same title is a real no-op — it does not bump `updated_at`, which the
|
|
5563
|
+
* conversation list sorts and paginates on. (Note this is a DIFFERENT guard
|
|
5564
|
+
* from `threads.ts`'s "non-empty AND changed" one, which only covers the
|
|
5565
|
+
* message-status PATCH; the non-empty half is enforced here instead, by
|
|
5566
|
+
* `resolveSessionTitle` never returning an empty/placeholder title.)
|
|
5567
|
+
*
|
|
5568
|
+
* Telemetry-only / never blocks the caller, mirroring `postSignal`: a failure
|
|
5569
|
+
* is logged and the title is simply retried on the next heartbeat tick (the
|
|
5570
|
+
* caller only latches `titleSynced` on `true`).
|
|
5571
|
+
*/
|
|
5572
|
+
async patchConversationTitle(conversationId, title) {
|
|
5573
|
+
try {
|
|
5574
|
+
const res = await this.fetchImpl(
|
|
5575
|
+
`${this.apiUrl}/runners/${this.agentId}/conversations/${conversationId}`,
|
|
5576
|
+
{
|
|
5577
|
+
method: "PATCH",
|
|
5578
|
+
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
5579
|
+
body: JSON.stringify({ title })
|
|
5580
|
+
}
|
|
5581
|
+
);
|
|
5582
|
+
if (!res.ok) {
|
|
5583
|
+
this.log({
|
|
5584
|
+
level: "debug",
|
|
5585
|
+
message: `Mid-turn title sync PATCH for conversation ${conversationId.slice(0, 8)} returned HTTP ${res.status} (best-effort, will retry next heartbeat)`,
|
|
5586
|
+
conversation_id: conversationId
|
|
5587
|
+
});
|
|
5588
|
+
return false;
|
|
5589
|
+
}
|
|
5590
|
+
return true;
|
|
5591
|
+
} catch (err) {
|
|
5592
|
+
this.log({
|
|
5593
|
+
level: "debug",
|
|
5594
|
+
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)}`,
|
|
5595
|
+
conversation_id: conversationId
|
|
5596
|
+
});
|
|
5597
|
+
return false;
|
|
5598
|
+
}
|
|
5599
|
+
}
|
|
3663
5600
|
/**
|
|
3664
5601
|
* DEFENSIVE cross-check for the restart-recovery path (WI-2): is any descendant
|
|
3665
5602
|
* (`task` sub-agent) session under `rootSessionId` still genuinely doing work?
|
|
@@ -3718,6 +5655,84 @@ var ChannelDriver = class {
|
|
|
3718
5655
|
}
|
|
3719
5656
|
return false;
|
|
3720
5657
|
}
|
|
5658
|
+
/**
|
|
5659
|
+
* LIVE-PATH descendant-liveness check (#721): is any descendant (`task`
|
|
5660
|
+
* sub-agent) session under `rootSessionId` currently ONGOING per OpenCode's own
|
|
5661
|
+
* in-memory status map (`isSessionOngoing` — `busy`/`retry`)?
|
|
5662
|
+
*
|
|
5663
|
+
* Deliberately NOT `isAnyDescendantSessionAlive` (the RECOVERY-path
|
|
5664
|
+
* cross-check above): that method judges liveness from the child's OWN
|
|
5665
|
+
* TRANSCRIPT (`isSessionActivelyGenerating`), which is the right (only) option
|
|
5666
|
+
* on the recovery path because a restart WIPES `SessionStatus`. On the LIVE
|
|
5667
|
+
* path the local opencode server IS running, so its in-memory status map is
|
|
5668
|
+
* live and authoritative — and per ADR-0047 §4a ("the child has its own entry
|
|
5669
|
+
* [in the map]"), a `task` descendant's OWN busy/retry entry reflects its
|
|
5670
|
+
* ENTIRE turn (including any tool call it is itself executing), not a
|
|
5671
|
+
* per-message transcript snapshot. This sidesteps the "child's own tool is
|
|
5672
|
+
* executing, between its step's completion and the next generation step"
|
|
5673
|
+
* transcript gap that a transcript-based check would need a second,
|
|
5674
|
+
* sustained-window bound to guard against — it is simply not derived from
|
|
5675
|
+
* message timestamps at all.
|
|
5676
|
+
*
|
|
5677
|
+
* Why not just check `isSessionOngoing(port, rootSessionId)` (the ROOT's own
|
|
5678
|
+
* status, as the recovery path does per §4a)? Because on the LIVE path the
|
|
5679
|
+
* root session can be shared: a SECOND, unrelated user message can land on the
|
|
5680
|
+
* SAME session (issue #721's own root cause) and keep the root `busy` for a
|
|
5681
|
+
* reason that has nothing to do with THIS message's delegation. A `task`
|
|
5682
|
+
* descendant session is spawned for exactly one delegated turn and never
|
|
5683
|
+
* reused, so its OWN status-map entry is unambiguous evidence about that one
|
|
5684
|
+
* delegation — which the root's status is not.
|
|
5685
|
+
*
|
|
5686
|
+
* Why membership is checked via `resolveSessionMembership`, NOT
|
|
5687
|
+
* `sessionBelongsTo`: `sessionBelongsTo` collapses a transient
|
|
5688
|
+
* `GET /session/:id` fetch failure into "not a descendant", which would
|
|
5689
|
+
* silently drop a genuinely-live candidate from consideration on the one
|
|
5690
|
+
* unlucky tick its membership-walk fetch hiccups (#721).
|
|
5691
|
+
* `resolveSessionMembership` keeps that failure mode as a distinct `null`
|
|
5692
|
+
* (indeterminate) so it is folded into THIS method's own `indeterminate` flag
|
|
5693
|
+
* instead.
|
|
5694
|
+
*
|
|
5695
|
+
* Return contract (note the DIFFERENT judge vs. `isAnyDescendantSessionAlive`):
|
|
5696
|
+
* - `true` → some descendant session is `busy`/`retry` (genuinely ongoing).
|
|
5697
|
+
* - `false` → enumeration succeeded, EVERY candidate's MEMBERSHIP was
|
|
5698
|
+
* confirmed either way (`resolveSessionMembership` never
|
|
5699
|
+
* returned `null`), and every CONFIRMED descendant's status read
|
|
5700
|
+
* succeeded and is not ongoing (includes "no descendant session
|
|
5701
|
+
* exists at all" — e.g. a plain, non-`task` tool call).
|
|
5702
|
+
* - `null` → INDETERMINATE: `listSessions` failed, OR at least one
|
|
5703
|
+
* candidate's MEMBERSHIP could not be confirmed
|
|
5704
|
+
* (`resolveSessionMembership` returned `null` — a fetch failure
|
|
5705
|
+
* or pathological chain partway through the parent walk), OR at
|
|
5706
|
+
* least one CONFIRMED descendant's `isSessionOngoing` read
|
|
5707
|
+
* failed — and no OTHER candidate was already confirmed `true`.
|
|
5708
|
+
* The caller MUST NOT treat `null` the same as `false` here
|
|
5709
|
+
* (unlike the recovery cross-check's contract) — see
|
|
5710
|
+
* `isB2AbandonmentConfirmed`.
|
|
5711
|
+
*/
|
|
5712
|
+
async isAnyDescendantSessionOngoing(rootSessionId) {
|
|
5713
|
+
const sessions = await listSessions(this.port);
|
|
5714
|
+
if (!sessions) {
|
|
5715
|
+
this.log({
|
|
5716
|
+
level: "warn",
|
|
5717
|
+
message: `Could not enumerate sessions to check descendant liveness for root ${rootSessionId} (listSessions failed) \u2014 treating descendant liveness as indeterminate`
|
|
5718
|
+
});
|
|
5719
|
+
return null;
|
|
5720
|
+
}
|
|
5721
|
+
let indeterminate = false;
|
|
5722
|
+
for (const candidate of sessions) {
|
|
5723
|
+
if (!candidate?.id || candidate.id === rootSessionId) continue;
|
|
5724
|
+
const membership = await this.resolveSessionMembership(candidate.id, rootSessionId);
|
|
5725
|
+
if (membership === null) {
|
|
5726
|
+
indeterminate = true;
|
|
5727
|
+
continue;
|
|
5728
|
+
}
|
|
5729
|
+
if (membership === false) continue;
|
|
5730
|
+
const ongoing = await isSessionOngoing(this.port, candidate.id);
|
|
5731
|
+
if (ongoing === true) return true;
|
|
5732
|
+
if (ongoing === null) indeterminate = true;
|
|
5733
|
+
}
|
|
5734
|
+
return indeterminate ? null : false;
|
|
5735
|
+
}
|
|
3721
5736
|
/**
|
|
3722
5737
|
* Cheap decision-telemetry label for a running row's LAST correlated reply
|
|
3723
5738
|
* (WI-2 Task 2.2): which running SHAPE it is, for the status-gated recovery log.
|
|
@@ -3790,7 +5805,7 @@ var ChannelDriver = class {
|
|
|
3790
5805
|
// Evident API calls (combinedAuth thread routes)
|
|
3791
5806
|
async getPendingConversations() {
|
|
3792
5807
|
const res = await this.fetchImpl(
|
|
3793
|
-
`${this.apiUrl}/
|
|
5808
|
+
`${this.apiUrl}/runners/${this.agentId}/conversations/pending`,
|
|
3794
5809
|
{
|
|
3795
5810
|
headers: { Authorization: this.getAuthHeader() }
|
|
3796
5811
|
}
|
|
@@ -3808,7 +5823,7 @@ var ChannelDriver = class {
|
|
|
3808
5823
|
}
|
|
3809
5824
|
async getPendingMessages(conversationId) {
|
|
3810
5825
|
const res = await this.fetchImpl(
|
|
3811
|
-
`${this.apiUrl}/
|
|
5826
|
+
`${this.apiUrl}/runners/${this.agentId}/threads/${conversationId}/messages?status=pending`,
|
|
3812
5827
|
{ headers: { Authorization: this.getAuthHeader() } }
|
|
3813
5828
|
);
|
|
3814
5829
|
this.assertAuth(res, "fetching pending messages");
|
|
@@ -3832,7 +5847,7 @@ var ChannelDriver = class {
|
|
|
3832
5847
|
*/
|
|
3833
5848
|
async getProcessingMessages() {
|
|
3834
5849
|
const res = await this.fetchImpl(
|
|
3835
|
-
`${this.apiUrl}/
|
|
5850
|
+
`${this.apiUrl}/runners/${this.agentId}/conversations/processing`,
|
|
3836
5851
|
{ headers: { Authorization: this.getAuthHeader() } }
|
|
3837
5852
|
);
|
|
3838
5853
|
this.assertAuth(res, "fetching processing messages");
|
|
@@ -3846,6 +5861,32 @@ var ChannelDriver = class {
|
|
|
3846
5861
|
}
|
|
3847
5862
|
return messages;
|
|
3848
5863
|
}
|
|
5864
|
+
/**
|
|
5865
|
+
* The `opencode_session_id` fragment of a status PATCH body — `{}` when this
|
|
5866
|
+
* conversation has ABANDONED that session (#553). The field is optional
|
|
5867
|
+
* server-side and an absent one leaves the persisted binding untouched, so
|
|
5868
|
+
* omitting it is how a routine status write stops resurrecting it.
|
|
5869
|
+
*
|
|
5870
|
+
* ONLY for writes whose sole cost is a lost deep link. The `processing` notice
|
|
5871
|
+
* degrades to no "View in Evident" link (the reaction swap still fires) and the
|
|
5872
|
+
* turn-failure notice is built from the PATCH's own `error` text with a link off
|
|
5873
|
+
* the persisted row — neither loses content the user came for. `markDone`
|
|
5874
|
+
* deliberately does NOT use this helper: the server fetches the reply text
|
|
5875
|
+
* THROUGH the session id it is given, so suppressing there would replace the
|
|
5876
|
+
* agent's answer with a bare "✅ Done!" (the #183/#187 failure). The
|
|
5877
|
+
* `ensureSession` guard, not this suppression, is what makes the self-heal
|
|
5878
|
+
* stick.
|
|
5879
|
+
*/
|
|
5880
|
+
sessionIdBody(sessionId, conversationId, messageId, status2) {
|
|
5881
|
+
if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
|
|
5882
|
+
this.log({
|
|
5883
|
+
level: "debug",
|
|
5884
|
+
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 ${conversationId.slice(0, 8)}`,
|
|
5885
|
+
conversation_id: conversationId,
|
|
5886
|
+
message_id: messageId
|
|
5887
|
+
});
|
|
5888
|
+
return {};
|
|
5889
|
+
}
|
|
3849
5890
|
/**
|
|
3850
5891
|
* EXISTING combinedAuth route — now fired by the watcher on queued→running
|
|
3851
5892
|
* (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
|
|
@@ -3869,13 +5910,13 @@ var ChannelDriver = class {
|
|
|
3869
5910
|
*/
|
|
3870
5911
|
async markProcessing(conversationId, messageId, sessionId, opencodeMessageId, title) {
|
|
3871
5912
|
const res = await this.fetchImpl(
|
|
3872
|
-
`${this.apiUrl}/
|
|
5913
|
+
`${this.apiUrl}/runners/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
|
|
3873
5914
|
{
|
|
3874
5915
|
method: "PATCH",
|
|
3875
5916
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
3876
5917
|
body: JSON.stringify({
|
|
3877
5918
|
status: "processing",
|
|
3878
|
-
|
|
5919
|
+
...this.sessionIdBody(sessionId, conversationId, messageId, "processing"),
|
|
3879
5920
|
...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
|
|
3880
5921
|
...title ? { title } : {}
|
|
3881
5922
|
})
|
|
@@ -3918,12 +5959,17 @@ var ChannelDriver = class {
|
|
|
3918
5959
|
*/
|
|
3919
5960
|
async markDone(conversationId, messageId, sessionId, opencodeMessageId, title, usage) {
|
|
3920
5961
|
const res = await this.fetchImpl(
|
|
3921
|
-
`${this.apiUrl}/
|
|
5962
|
+
`${this.apiUrl}/runners/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
|
|
3922
5963
|
{
|
|
3923
5964
|
method: "PATCH",
|
|
3924
5965
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
3925
5966
|
body: JSON.stringify({
|
|
3926
5967
|
status: "done",
|
|
5968
|
+
// ALWAYS sent, even for a session this conversation has abandoned
|
|
5969
|
+
// (#553): the server reads the reply text back out of THIS session id
|
|
5970
|
+
// to deliver it. Omitting it would leave the user with "✅ Done!"
|
|
5971
|
+
// instead of the answer — a worse regression than the resurrection it
|
|
5972
|
+
// would prevent, which `ensureSession`'s guard handles anyway.
|
|
3927
5973
|
opencode_session_id: sessionId,
|
|
3928
5974
|
...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
|
|
3929
5975
|
...title ? { title } : {},
|
|
@@ -3940,20 +5986,35 @@ var ChannelDriver = class {
|
|
|
3940
5986
|
}
|
|
3941
5987
|
/**
|
|
3942
5988
|
* Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
|
|
3943
|
-
* when provided (issue #182)
|
|
3944
|
-
* `
|
|
3945
|
-
*
|
|
3946
|
-
*
|
|
5989
|
+
* when provided (issue #182). Three states for `sessionId`:
|
|
5990
|
+
* - omitted (`undefined`) → don't send the field, leave the persisted
|
|
5991
|
+
* session untouched (unused today; kept for API symmetry).
|
|
5992
|
+
* - a real id (`string`) → send it, update the persisted session (the
|
|
5993
|
+
* turn-failure call sites: an errored OpenCode turn).
|
|
5994
|
+
* - explicit `null` → send it, CLEAR the persisted session (issue
|
|
5995
|
+
* #485's dispatch-handoff-failure call site: the session id still
|
|
5996
|
+
* exists but is wedged, so the next attempt must get a fresh one
|
|
5997
|
+
* instead of reusing it — see WI-1's server-side null-clearing PATCH).
|
|
3947
5998
|
*/
|
|
3948
|
-
async markFailed(conversationId, messageId, sessionId, error2, usage) {
|
|
5999
|
+
async markFailed(conversationId, messageId, sessionId, error2, usage, failure) {
|
|
3949
6000
|
const body = { status: "failed" };
|
|
3950
|
-
if (sessionId
|
|
6001
|
+
if (sessionId === null) {
|
|
6002
|
+
body.opencode_session_id = null;
|
|
6003
|
+
} else if (sessionId !== void 0) {
|
|
6004
|
+
Object.assign(body, this.sessionIdBody(sessionId, conversationId, messageId, "failed"));
|
|
6005
|
+
}
|
|
3951
6006
|
if (error2 !== void 0) body.error = error2;
|
|
3952
6007
|
if (usage) Object.assign(body, usage);
|
|
6008
|
+
if (failure) {
|
|
6009
|
+
body.failure_kind = failure.kind;
|
|
6010
|
+
body.failure_provider_id = failure.providerId;
|
|
6011
|
+
body.failure_model_id = failure.modelId;
|
|
6012
|
+
body.failure_reason = failure.reason;
|
|
6013
|
+
}
|
|
3953
6014
|
await this.callWithRetry(
|
|
3954
6015
|
"marking message as failed",
|
|
3955
6016
|
() => this.fetchImpl(
|
|
3956
|
-
`${this.apiUrl}/
|
|
6017
|
+
`${this.apiUrl}/runners/${this.agentId}/threads/${conversationId}/messages/${messageId}`,
|
|
3957
6018
|
{
|
|
3958
6019
|
method: "PATCH",
|
|
3959
6020
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
@@ -3962,6 +6023,29 @@ var ChannelDriver = class {
|
|
|
3962
6023
|
)
|
|
3963
6024
|
);
|
|
3964
6025
|
}
|
|
6026
|
+
/**
|
|
6027
|
+
* Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
|
|
6028
|
+
*
|
|
6029
|
+
* `messageFailure` alone (structured OpenCode error → `model_auth`) covers
|
|
6030
|
+
* most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
|
|
6031
|
+
* to the P1-2b zero-provider check — one extra loopback call to
|
|
6032
|
+
* `hasAnyConfiguredProvider`, only reached when the structured classifier
|
|
6033
|
+
* couldn't place it. Fails open (never throws): a fallback probe failure
|
|
6034
|
+
* (`null`/indeterminate) leaves the classification `null`, which produces
|
|
6035
|
+
* today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
|
|
6036
|
+
*/
|
|
6037
|
+
async classifyModelAuthFailure(messages, userMessageId) {
|
|
6038
|
+
const classified = messageFailure(messages, userMessageId);
|
|
6039
|
+
if (classified != null) return classified;
|
|
6040
|
+
const reply = findLastAssistantReplyFor(messages, userMessageId);
|
|
6041
|
+
const hasProvider = await hasAnyConfiguredProvider(this.port);
|
|
6042
|
+
return applyZeroProviderFallback(
|
|
6043
|
+
classified,
|
|
6044
|
+
hasProvider,
|
|
6045
|
+
reply?.info?.providerID ?? null,
|
|
6046
|
+
reply?.info?.modelID ?? null
|
|
6047
|
+
);
|
|
6048
|
+
}
|
|
3965
6049
|
/**
|
|
3966
6050
|
* Best-effort, SINGLE-ATTEMPT runner→server telemetry ping
|
|
3967
6051
|
* (queued-followup-redrive). `POST .../messages/:id/signal {signal, ...extra}`
|
|
@@ -3980,7 +6064,7 @@ var ChannelDriver = class {
|
|
|
3980
6064
|
async postSignal(conversationId, messageId, signal, extra) {
|
|
3981
6065
|
try {
|
|
3982
6066
|
const res = await this.fetchImpl(
|
|
3983
|
-
`${this.apiUrl}/
|
|
6067
|
+
`${this.apiUrl}/runners/${this.agentId}/threads/${conversationId}/messages/${messageId}/signal`,
|
|
3984
6068
|
{
|
|
3985
6069
|
method: "POST",
|
|
3986
6070
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
@@ -4009,7 +6093,7 @@ var ChannelDriver = class {
|
|
|
4009
6093
|
}
|
|
4010
6094
|
async persistSession(conversationId, sessionId) {
|
|
4011
6095
|
const res = await this.fetchImpl(
|
|
4012
|
-
`${this.apiUrl}/
|
|
6096
|
+
`${this.apiUrl}/runners/${this.agentId}/threads/${conversationId}`,
|
|
4013
6097
|
{
|
|
4014
6098
|
method: "PATCH",
|
|
4015
6099
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
@@ -4035,7 +6119,7 @@ var ChannelDriver = class {
|
|
|
4035
6119
|
await this.callWithRetry(
|
|
4036
6120
|
"reporting interactive event",
|
|
4037
6121
|
() => this.fetchImpl(
|
|
4038
|
-
`${this.apiUrl}/
|
|
6122
|
+
`${this.apiUrl}/runners/${this.agentId}/threads/${conversationId}/interactive-event`,
|
|
4039
6123
|
{
|
|
4040
6124
|
method: "POST",
|
|
4041
6125
|
headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
|
|
@@ -4114,10 +6198,16 @@ var ChannelDriver = class {
|
|
|
4114
6198
|
import chalk5 from "chalk";
|
|
4115
6199
|
import ora2 from "ora";
|
|
4116
6200
|
import { select as select2 } from "@inquirer/prompts";
|
|
6201
|
+
var INTERACTIVE_START_TIMEOUT_MS = 3e4;
|
|
4117
6202
|
async function ensureOpenCodeRunning(ctx) {
|
|
4118
6203
|
const healthCheck = await checkOpenCodeHealth(ctx.port);
|
|
4119
6204
|
if (healthCheck.healthy) {
|
|
4120
|
-
return {
|
|
6205
|
+
return {
|
|
6206
|
+
port: ctx.port,
|
|
6207
|
+
process: null,
|
|
6208
|
+
version: healthCheck.version ?? null,
|
|
6209
|
+
notReadyReason: null
|
|
6210
|
+
};
|
|
4121
6211
|
}
|
|
4122
6212
|
const runningInstances = await findHealthyOpenCodeInstances();
|
|
4123
6213
|
if (runningInstances.length > 0) {
|
|
@@ -4138,7 +6228,7 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4138
6228
|
console.log(chalk5.yellow("Tip: Run with the correct port:"));
|
|
4139
6229
|
console.log(
|
|
4140
6230
|
chalk5.dim(
|
|
4141
|
-
` ${getCliName()} run --
|
|
6231
|
+
` ${getCliName()} run --runner ${ctx.agentId} --port ${runningInstances[0].port}`
|
|
4142
6232
|
)
|
|
4143
6233
|
);
|
|
4144
6234
|
}
|
|
@@ -4158,14 +6248,22 @@ async function ensureOpenCodeRunning(ctx) {
|
|
|
4158
6248
|
if (!ctx.interactive) {
|
|
4159
6249
|
ctx.log(`OpenCode is not running on port ${ctx.port}. Starting it automatically...`);
|
|
4160
6250
|
const proc = await startOpenCode(ctx.port);
|
|
4161
|
-
const health = await waitForOpenCodeHealth(ctx.port,
|
|
6251
|
+
const health = await waitForOpenCodeHealth(ctx.port, ctx.startTimeoutMs);
|
|
4162
6252
|
if (!health.healthy) {
|
|
4163
|
-
|
|
4164
|
-
|
|
4165
|
-
|
|
6253
|
+
return {
|
|
6254
|
+
port: ctx.port,
|
|
6255
|
+
process: proc,
|
|
6256
|
+
version: null,
|
|
6257
|
+
notReadyReason: `it did not become healthy within ${Math.round(ctx.startTimeoutMs / 1e3)}s`
|
|
6258
|
+
};
|
|
4166
6259
|
}
|
|
4167
6260
|
ctx.log(`OpenCode started on port ${ctx.port}${health.version ? ` (v${health.version})` : ""}`);
|
|
4168
|
-
return {
|
|
6261
|
+
return {
|
|
6262
|
+
port: ctx.port,
|
|
6263
|
+
process: proc,
|
|
6264
|
+
version: health.version ?? null,
|
|
6265
|
+
notReadyReason: null
|
|
6266
|
+
};
|
|
4169
6267
|
}
|
|
4170
6268
|
let port = ctx.port;
|
|
4171
6269
|
if (isPortInUse(port)) {
|
|
@@ -4218,122 +6316,15 @@ Port ${port} is already in use.`));
|
|
|
4218
6316
|
if (action === "start") {
|
|
4219
6317
|
const spinner = ora2("Starting OpenCode...").start();
|
|
4220
6318
|
const proc = await startOpenCode(port);
|
|
4221
|
-
const health = await waitForOpenCodeHealth(port,
|
|
6319
|
+
const health = await waitForOpenCodeHealth(port, INTERACTIVE_START_TIMEOUT_MS);
|
|
4222
6320
|
if (!health.healthy) {
|
|
4223
6321
|
spinner.fail("Failed to start OpenCode");
|
|
4224
6322
|
throw new Error("OpenCode failed to start");
|
|
4225
6323
|
}
|
|
4226
6324
|
spinner.stop();
|
|
4227
|
-
return { port, process: proc, version: health.version ?? null };
|
|
4228
|
-
}
|
|
4229
|
-
return { port, process: null, version: null };
|
|
4230
|
-
}
|
|
4231
|
-
|
|
4232
|
-
// src/commands/agent-lookup.ts
|
|
4233
|
-
async function readErrorMessage(response) {
|
|
4234
|
-
const text = await response.text().catch(() => "");
|
|
4235
|
-
if (!text) return response.statusText || void 0;
|
|
4236
|
-
try {
|
|
4237
|
-
const data = JSON.parse(text);
|
|
4238
|
-
const message = data.message ?? data.error;
|
|
4239
|
-
if (typeof message === "string" && message.trim()) {
|
|
4240
|
-
return message;
|
|
4241
|
-
}
|
|
4242
|
-
} catch {
|
|
4243
|
-
}
|
|
4244
|
-
return text.trim() || response.statusText || void 0;
|
|
4245
|
-
}
|
|
4246
|
-
function authFailureHint(apiUrl, serverMessage) {
|
|
4247
|
-
const reason = serverMessage ? `: ${serverMessage}` : "";
|
|
4248
|
-
return `Authentication failed${reason}. Your credentials were rejected by ${apiUrl}. This usually means you logged in against a different environment, or your session expired \u2014 log in again pointing at this endpoint and retry.`;
|
|
4249
|
-
}
|
|
4250
|
-
async function resolveAgentIdFromKey(authHeader) {
|
|
4251
|
-
const apiUrl = getApiUrlConfig();
|
|
4252
|
-
try {
|
|
4253
|
-
const response = await fetch(`${apiUrl}/me`, {
|
|
4254
|
-
headers: { Authorization: authHeader }
|
|
4255
|
-
});
|
|
4256
|
-
if (response.status === 401) {
|
|
4257
|
-
const serverMessage = await readErrorMessage(response);
|
|
4258
|
-
return { error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
4259
|
-
}
|
|
4260
|
-
if (!response.ok) {
|
|
4261
|
-
const serverMessage = await readErrorMessage(response);
|
|
4262
|
-
return {
|
|
4263
|
-
error: `Failed to resolve runner from key (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
4264
|
-
};
|
|
4265
|
-
}
|
|
4266
|
-
const data = await response.json();
|
|
4267
|
-
if (data.auth_type === "agent_key" && data.agent_id) {
|
|
4268
|
-
return { agent_id: data.agent_id };
|
|
4269
|
-
}
|
|
4270
|
-
return {
|
|
4271
|
-
error: "Cannot resolve runner ID: auth type is not agent_key. Please provide --agent explicitly."
|
|
4272
|
-
};
|
|
4273
|
-
} catch (error2) {
|
|
4274
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
4275
|
-
return { error: `Failed to resolve runner from key: ${message}` };
|
|
4276
|
-
}
|
|
4277
|
-
}
|
|
4278
|
-
async function notifyAgentDisconnected(agentId, authHeader) {
|
|
4279
|
-
const apiUrl = getApiUrlConfig();
|
|
4280
|
-
try {
|
|
4281
|
-
const response = await fetch(`${apiUrl}/agents/${agentId}/disconnect`, {
|
|
4282
|
-
method: "POST",
|
|
4283
|
-
headers: { Authorization: authHeader }
|
|
4284
|
-
});
|
|
4285
|
-
if (!response.ok) {
|
|
4286
|
-
const serverMessage = await readErrorMessage(response);
|
|
4287
|
-
return {
|
|
4288
|
-
ok: false,
|
|
4289
|
-
error: `HTTP ${response.status}${serverMessage ? `: ${serverMessage}` : ""}`
|
|
4290
|
-
};
|
|
4291
|
-
}
|
|
4292
|
-
return { ok: true };
|
|
4293
|
-
} catch (error2) {
|
|
4294
|
-
return { ok: false, error: error2 instanceof Error ? error2.message : String(error2) };
|
|
4295
|
-
}
|
|
4296
|
-
}
|
|
4297
|
-
async function getAgentInfo(agentId, authHeader) {
|
|
4298
|
-
const apiUrl = getApiUrlConfig();
|
|
4299
|
-
try {
|
|
4300
|
-
const response = await fetch(`${apiUrl}/agents/${agentId}`, {
|
|
4301
|
-
headers: { Authorization: authHeader }
|
|
4302
|
-
});
|
|
4303
|
-
if (response.status === 401) {
|
|
4304
|
-
const serverMessage = await readErrorMessage(response);
|
|
4305
|
-
return { valid: false, error: authFailureHint(apiUrl, serverMessage), authFailed: true };
|
|
4306
|
-
}
|
|
4307
|
-
if (response.status === 403) {
|
|
4308
|
-
const serverMessage = await readErrorMessage(response);
|
|
4309
|
-
return {
|
|
4310
|
-
valid: false,
|
|
4311
|
-
error: serverMessage ?? "You do not have access to this runner (it may belong to a different team or organization)."
|
|
4312
|
-
};
|
|
4313
|
-
}
|
|
4314
|
-
if (response.status === 404) {
|
|
4315
|
-
const serverMessage = await readErrorMessage(response);
|
|
4316
|
-
return { valid: false, error: serverMessage ?? `Runner ${agentId} not found` };
|
|
4317
|
-
}
|
|
4318
|
-
if (!response.ok) {
|
|
4319
|
-
const serverMessage = await readErrorMessage(response);
|
|
4320
|
-
return {
|
|
4321
|
-
valid: false,
|
|
4322
|
-
error: `API error (HTTP ${response.status})${serverMessage ? `: ${serverMessage}` : ""}`
|
|
4323
|
-
};
|
|
4324
|
-
}
|
|
4325
|
-
const agent = await response.json();
|
|
4326
|
-
if (agent.agent_type !== "local") {
|
|
4327
|
-
return {
|
|
4328
|
-
valid: false,
|
|
4329
|
-
error: `Runner is type '${agent.agent_type}', must be 'local' for CLI connection`
|
|
4330
|
-
};
|
|
4331
|
-
}
|
|
4332
|
-
return { valid: true, agent };
|
|
4333
|
-
} catch (error2) {
|
|
4334
|
-
const message = error2 instanceof Error ? error2.message : "Unknown error";
|
|
4335
|
-
return { valid: false, error: `Failed to validate runner: ${message}` };
|
|
6325
|
+
return { port, process: proc, version: health.version ?? null, notReadyReason: null };
|
|
4336
6326
|
}
|
|
6327
|
+
return { port, process: null, version: null, notReadyReason: "you chose to continue without it" };
|
|
4337
6328
|
}
|
|
4338
6329
|
|
|
4339
6330
|
// src/commands/run.ts
|
|
@@ -4341,6 +6332,7 @@ var MAX_ACTIVITY_LOG_ENTRIES = 10;
|
|
|
4341
6332
|
var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
|
|
4342
6333
|
var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
|
|
4343
6334
|
var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
|
|
6335
|
+
var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
|
|
4344
6336
|
function resolveLogLevel(options) {
|
|
4345
6337
|
const accepted = Object.keys(LOG_LEVELS);
|
|
4346
6338
|
const validate = (value, source) => {
|
|
@@ -4364,6 +6356,63 @@ function resolveLogLevel(options) {
|
|
|
4364
6356
|
}
|
|
4365
6357
|
return "info";
|
|
4366
6358
|
}
|
|
6359
|
+
function resolveFileSyncDirectories(raw, homeDir) {
|
|
6360
|
+
const directories = [];
|
|
6361
|
+
for (const entry of raw ?? []) {
|
|
6362
|
+
const trimmed = entry.trim();
|
|
6363
|
+
if (trimmed === "") {
|
|
6364
|
+
throw new Error("--enable-file-sync-to requires a directory path (got an empty value)");
|
|
6365
|
+
}
|
|
6366
|
+
const expanded = trimmed === "~" ? homeDir : trimmed.startsWith("~/") ? join3(homeDir, trimmed.slice(2)) : trimmed;
|
|
6367
|
+
if (!isAbsolute2(expanded)) {
|
|
6368
|
+
throw new Error(`--enable-file-sync-to requires an absolute directory path; got "${entry}"`);
|
|
6369
|
+
}
|
|
6370
|
+
const normalized = resolvePath(expanded);
|
|
6371
|
+
if (parse(normalized).root === normalized) {
|
|
6372
|
+
throw new Error(
|
|
6373
|
+
`--enable-file-sync-to will not allow-list the filesystem root ("${entry}"); name the specific directory the credentials belong in (for example ~/.claude)`
|
|
6374
|
+
);
|
|
6375
|
+
}
|
|
6376
|
+
if (!directories.includes(normalized)) {
|
|
6377
|
+
directories.push(normalized);
|
|
6378
|
+
}
|
|
6379
|
+
}
|
|
6380
|
+
if (directories.length > MAX_FILE_SYNC_DIRECTORIES) {
|
|
6381
|
+
throw new Error(
|
|
6382
|
+
`--enable-file-sync-to accepts at most ${MAX_FILE_SYNC_DIRECTORIES} directories; got ${directories.length}`
|
|
6383
|
+
);
|
|
6384
|
+
}
|
|
6385
|
+
return directories;
|
|
6386
|
+
}
|
|
6387
|
+
var DEFAULT_OPENCODE_START_TIMEOUT_SECONDS = 180;
|
|
6388
|
+
var MAX_OPENCODE_START_TIMEOUT_SECONDS = 3600;
|
|
6389
|
+
var OPENCODE_START_TIMEOUT_ENV = "EVIDENT_OPENCODE_START_TIMEOUT";
|
|
6390
|
+
function resolveOpenCodeStartTimeoutMs(options, env = process.env) {
|
|
6391
|
+
const defaultMs = DEFAULT_OPENCODE_START_TIMEOUT_SECONDS * 1e3;
|
|
6392
|
+
let raw;
|
|
6393
|
+
let source;
|
|
6394
|
+
if (options.opencodeStartTimeout !== void 0) {
|
|
6395
|
+
raw = options.opencodeStartTimeout;
|
|
6396
|
+
source = "--opencode-start-timeout";
|
|
6397
|
+
} else if (env[OPENCODE_START_TIMEOUT_ENV] !== void 0 && env[OPENCODE_START_TIMEOUT_ENV] !== "") {
|
|
6398
|
+
raw = env[OPENCODE_START_TIMEOUT_ENV];
|
|
6399
|
+
source = OPENCODE_START_TIMEOUT_ENV;
|
|
6400
|
+
} else {
|
|
6401
|
+
return { timeoutMs: defaultMs, warnings: [] };
|
|
6402
|
+
}
|
|
6403
|
+
const trimmed = raw.trim();
|
|
6404
|
+
const seconds = Number(trimmed);
|
|
6405
|
+
const isPositiveInteger = /^\d+$/.test(trimmed) && Number.isInteger(seconds) && seconds > 0;
|
|
6406
|
+
if (!isPositiveInteger || seconds > MAX_OPENCODE_START_TIMEOUT_SECONDS) {
|
|
6407
|
+
return {
|
|
6408
|
+
timeoutMs: defaultMs,
|
|
6409
|
+
warnings: [
|
|
6410
|
+
`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`
|
|
6411
|
+
]
|
|
6412
|
+
};
|
|
6413
|
+
}
|
|
6414
|
+
return { timeoutMs: seconds * 1e3, warnings: [] };
|
|
6415
|
+
}
|
|
4367
6416
|
function meetsThreshold(state, level) {
|
|
4368
6417
|
return LOG_LEVELS[level] >= LOG_LEVELS[state.logLevel];
|
|
4369
6418
|
}
|
|
@@ -4385,6 +6434,10 @@ function log2(state, message, level = "info") {
|
|
|
4385
6434
|
function logActivity(state, entry) {
|
|
4386
6435
|
const level = entry.level ?? (entry.type === "error" ? "error" : "info");
|
|
4387
6436
|
if (!meetsThreshold(state, level)) return;
|
|
6437
|
+
forwardRunnerActivity(
|
|
6438
|
+
{ level, message: entry.message, error: entry.error },
|
|
6439
|
+
{ agentId: state.agentId, authHeader: state.authHeader }
|
|
6440
|
+
);
|
|
4388
6441
|
const fullEntry = {
|
|
4389
6442
|
...entry,
|
|
4390
6443
|
level,
|
|
@@ -4484,23 +6537,46 @@ async function handleAuthError(state, error2) {
|
|
|
4484
6537
|
}
|
|
4485
6538
|
async function driveChannels(state, driver) {
|
|
4486
6539
|
let idlePolls = 0;
|
|
6540
|
+
let idleMs = 0;
|
|
6541
|
+
let consecutiveDrainFailures = 0;
|
|
6542
|
+
let unreachableMs = 0;
|
|
4487
6543
|
let lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
6544
|
+
let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
4488
6545
|
while (state.running) {
|
|
6546
|
+
const cycleStartedAtMs = performance.now();
|
|
6547
|
+
let idleThisCycle = false;
|
|
6548
|
+
let unreachableThisCycle = false;
|
|
4489
6549
|
if (state.connection?.reconnecting && state.connection.reconnectPromise) {
|
|
4490
6550
|
logActivity(state, { type: "info", message: "Waiting for tunnel reconnection..." });
|
|
4491
6551
|
if (state.interactive) displayStatus(state);
|
|
4492
6552
|
await state.connection.reconnectPromise;
|
|
4493
6553
|
}
|
|
6554
|
+
const carriedOverFileSync = driver.fileSyncActivity().inFlight;
|
|
6555
|
+
void driver.syncPendingFiles().catch(
|
|
6556
|
+
(error2) => logActivity(state, {
|
|
6557
|
+
type: "error",
|
|
6558
|
+
error: `Runner file sync failed: ${error2 instanceof Error ? error2.message : String(error2)}`
|
|
6559
|
+
})
|
|
6560
|
+
);
|
|
4494
6561
|
try {
|
|
4495
6562
|
const processed = await driver.drainPending();
|
|
6563
|
+
consecutiveDrainFailures = 0;
|
|
6564
|
+
unreachableMs = 0;
|
|
4496
6565
|
state.messageCount += processed;
|
|
4497
6566
|
const proxiedActivity = state.lastProxiedActivityAt !== lastSeenProxiedActivityAt;
|
|
4498
6567
|
lastSeenProxiedActivityAt = state.lastProxiedActivityAt;
|
|
4499
|
-
|
|
6568
|
+
const appliedFiles = driver.fileSyncActivity().appliedFiles;
|
|
6569
|
+
const filesApplied = appliedFiles !== lastSeenAppliedFiles;
|
|
6570
|
+
const fileActivity = carriedOverFileSync || filesApplied;
|
|
6571
|
+
lastSeenAppliedFiles = appliedFiles;
|
|
6572
|
+
if (filesApplied) state.claudeUsageRearm?.();
|
|
6573
|
+
if (processed > 0 || driver.hasInFlightWatchers() || proxiedActivity || fileActivity) {
|
|
4500
6574
|
idlePolls = 0;
|
|
6575
|
+
idleMs = 0;
|
|
4501
6576
|
if (processed > 0 && state.interactive) displayStatus(state);
|
|
4502
6577
|
} else if (state.idleTimeout !== null) {
|
|
4503
6578
|
idlePolls++;
|
|
6579
|
+
idleThisCycle = true;
|
|
4504
6580
|
if (idlePolls === 1) {
|
|
4505
6581
|
logActivity(state, {
|
|
4506
6582
|
type: "info",
|
|
@@ -4524,21 +6600,44 @@ async function driveChannels(state, driver) {
|
|
|
4524
6600
|
const errorMessage = error2 instanceof Error ? error2.message : String(error2);
|
|
4525
6601
|
logActivity(state, { type: "error", error: `Channel processing error: ${errorMessage}` });
|
|
4526
6602
|
if (state.interactive) displayStatus(state);
|
|
4527
|
-
|
|
4528
|
-
|
|
4529
|
-
|
|
4530
|
-
|
|
4531
|
-
|
|
4532
|
-
|
|
4533
|
-
if (
|
|
4534
|
-
|
|
6603
|
+
if (driver.hasInFlightWatchers()) {
|
|
6604
|
+
consecutiveDrainFailures = 0;
|
|
6605
|
+
unreachableMs = 0;
|
|
6606
|
+
} else if (state.idleTimeout !== null) {
|
|
6607
|
+
consecutiveDrainFailures++;
|
|
6608
|
+
unreachableThisCycle = true;
|
|
6609
|
+
if (consecutiveDrainFailures === 1) {
|
|
6610
|
+
logActivity(state, {
|
|
6611
|
+
type: "info",
|
|
6612
|
+
message: `Cannot reach Evident, will exit if this persists past the idle timeout (timeout: ${state.idleTimeout}s)...`
|
|
6613
|
+
});
|
|
6614
|
+
if (state.interactive) displayStatus(state);
|
|
6615
|
+
}
|
|
4535
6616
|
}
|
|
4536
6617
|
}
|
|
6618
|
+
await new Promise((resolve3) => setTimeout(resolve3, CHANNEL_POLL_INTERVAL_MS));
|
|
6619
|
+
const cycleMs = performance.now() - cycleStartedAtMs;
|
|
6620
|
+
if (idleThisCycle) idleMs += cycleMs;
|
|
6621
|
+
if (unreachableThisCycle) unreachableMs += cycleMs;
|
|
6622
|
+
if (state.idleTimeout !== null && consecutiveDrainFailures >= 2 && unreachableMs > state.idleTimeout * 1e3) {
|
|
6623
|
+
logActivity(state, {
|
|
6624
|
+
type: "info",
|
|
6625
|
+
level: "warn",
|
|
6626
|
+
message: `Exiting: could not reach Evident for ${consecutiveDrainFailures} consecutive polls (${Math.round(unreachableMs / 1e3)}s)`
|
|
6627
|
+
});
|
|
6628
|
+
if (state.interactive) displayStatus(state);
|
|
6629
|
+
break;
|
|
6630
|
+
}
|
|
6631
|
+
if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
|
|
6632
|
+
logActivity(state, { type: "info", message: "Idle timeout reached" });
|
|
6633
|
+
if (state.interactive) displayStatus(state);
|
|
6634
|
+
break;
|
|
6635
|
+
}
|
|
4537
6636
|
}
|
|
4538
6637
|
}
|
|
4539
6638
|
var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
|
|
4540
|
-
async function runSweep(state, driver,
|
|
4541
|
-
const mode = `age=${
|
|
6639
|
+
async function runSweep(state, driver, config) {
|
|
6640
|
+
const mode = `age=${config.maxAgeMs ?? "\u2014"} count=${config.maxCount ?? "\u2014"}`;
|
|
4542
6641
|
try {
|
|
4543
6642
|
const sessions = await listSessions(state.port);
|
|
4544
6643
|
if (sessions === null) {
|
|
@@ -4551,8 +6650,8 @@ async function runSweep(state, driver, config2) {
|
|
|
4551
6650
|
const toDelete = selectSessionsToDelete(
|
|
4552
6651
|
sessions.map((s) => ({ id: s.id, lastActivityMs: sessionLastActivityMs(s) })),
|
|
4553
6652
|
{
|
|
4554
|
-
maxAgeMs:
|
|
4555
|
-
maxCount:
|
|
6653
|
+
maxAgeMs: config.maxAgeMs,
|
|
6654
|
+
maxCount: config.maxCount,
|
|
4556
6655
|
nowMs: Date.now(),
|
|
4557
6656
|
protectedIds: driver.protectedSessionIds()
|
|
4558
6657
|
}
|
|
@@ -4588,7 +6687,7 @@ async function runSweep(state, driver, config2) {
|
|
|
4588
6687
|
}
|
|
4589
6688
|
}
|
|
4590
6689
|
function scheduleSessionCleanup(state, driver, options) {
|
|
4591
|
-
const
|
|
6690
|
+
const config = resolveSessionCleanupConfig(
|
|
4592
6691
|
{
|
|
4593
6692
|
maxAge: options.sessionCleanupMaxAge,
|
|
4594
6693
|
maxCount: options.sessionCleanupMaxCount,
|
|
@@ -4596,21 +6695,129 @@ function scheduleSessionCleanup(state, driver, options) {
|
|
|
4596
6695
|
},
|
|
4597
6696
|
process.env
|
|
4598
6697
|
);
|
|
4599
|
-
for (const warning2 of
|
|
6698
|
+
for (const warning2 of config.warnings) {
|
|
4600
6699
|
logActivity(state, { type: "info", level: "warn", message: `Session cleanup: ${warning2}` });
|
|
4601
6700
|
}
|
|
4602
|
-
if (!
|
|
6701
|
+
if (!config.enabled) return;
|
|
4603
6702
|
logActivity(state, {
|
|
4604
6703
|
type: "info",
|
|
4605
|
-
message: `Session cleanup enabled (age=${
|
|
6704
|
+
message: `Session cleanup enabled (age=${config.maxAgeMs ?? "\u2014"}, count=${config.maxCount ?? "\u2014"}, interval=${config.intervalMs}ms)`
|
|
4606
6705
|
});
|
|
4607
|
-
const interval = setInterval(() => void runSweep(state, driver,
|
|
6706
|
+
const interval = setInterval(() => void runSweep(state, driver, config), config.intervalMs);
|
|
4608
6707
|
const firstSweep = setTimeout(
|
|
4609
|
-
() => void runSweep(state, driver,
|
|
6708
|
+
() => void runSweep(state, driver, config),
|
|
4610
6709
|
SESSION_CLEANUP_FIRST_SWEEP_MS
|
|
4611
6710
|
);
|
|
4612
6711
|
state.sessionCleanupTimers.push(interval, firstSweep);
|
|
4613
6712
|
}
|
|
6713
|
+
function claudeUsageFailureStreakSuffix(consecutiveFailures) {
|
|
6714
|
+
return consecutiveFailures > 1 ? ` (${consecutiveFailures} consecutive failures)` : "";
|
|
6715
|
+
}
|
|
6716
|
+
function scheduleClaudeUsageReporting(state, options) {
|
|
6717
|
+
const { mode, warnings } = resolveClaudeUsageReportingMode(
|
|
6718
|
+
options.claudeUsageReporting,
|
|
6719
|
+
process.env
|
|
6720
|
+
);
|
|
6721
|
+
for (const warning2 of warnings) {
|
|
6722
|
+
logActivity(state, {
|
|
6723
|
+
type: "info",
|
|
6724
|
+
level: "warn",
|
|
6725
|
+
message: `Claude usage reporting: ${warning2}`
|
|
6726
|
+
});
|
|
6727
|
+
}
|
|
6728
|
+
if (mode === "off") {
|
|
6729
|
+
logActivity(state, {
|
|
6730
|
+
type: "info",
|
|
6731
|
+
level: "debug",
|
|
6732
|
+
message: "Claude usage reporting is off (--claude-usage-reporting off)"
|
|
6733
|
+
});
|
|
6734
|
+
return null;
|
|
6735
|
+
}
|
|
6736
|
+
let consecutiveFailures = 0;
|
|
6737
|
+
let armed = false;
|
|
6738
|
+
let rearmRequested = false;
|
|
6739
|
+
const scheduleNextTick = () => {
|
|
6740
|
+
armed = true;
|
|
6741
|
+
rearmRequested = false;
|
|
6742
|
+
state.claudeUsageTimer = setTimeout(() => void tick(false), nextReportDelayMs());
|
|
6743
|
+
};
|
|
6744
|
+
const rearm = () => {
|
|
6745
|
+
if (armed) {
|
|
6746
|
+
rearmRequested = true;
|
|
6747
|
+
return;
|
|
6748
|
+
}
|
|
6749
|
+
rearmRequested = false;
|
|
6750
|
+
armed = true;
|
|
6751
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6752
|
+
};
|
|
6753
|
+
const tick = async (isProbe) => {
|
|
6754
|
+
try {
|
|
6755
|
+
const usage = await getClaudeUsage();
|
|
6756
|
+
const result = await reportClaudeUsage(state.agentId, state.authHeader, usage);
|
|
6757
|
+
if (result.ok) {
|
|
6758
|
+
if (consecutiveFailures > 0) {
|
|
6759
|
+
logActivity(state, {
|
|
6760
|
+
type: "info",
|
|
6761
|
+
level: "info",
|
|
6762
|
+
message: "Claude usage reporting recovered"
|
|
6763
|
+
});
|
|
6764
|
+
}
|
|
6765
|
+
consecutiveFailures = 0;
|
|
6766
|
+
logActivity(state, {
|
|
6767
|
+
type: "info",
|
|
6768
|
+
level: "debug",
|
|
6769
|
+
message: "Reported Claude usage to Evident"
|
|
6770
|
+
});
|
|
6771
|
+
} else {
|
|
6772
|
+
consecutiveFailures++;
|
|
6773
|
+
logActivity(state, {
|
|
6774
|
+
type: "info",
|
|
6775
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6776
|
+
message: `Failed to report Claude usage: ${result.error}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6777
|
+
});
|
|
6778
|
+
}
|
|
6779
|
+
scheduleNextTick();
|
|
6780
|
+
} catch (error2) {
|
|
6781
|
+
if (error2 instanceof ClaudeUsageError && isLocalCredentialProblem(error2)) {
|
|
6782
|
+
if (mode === "on") {
|
|
6783
|
+
logActivity(state, {
|
|
6784
|
+
type: "info",
|
|
6785
|
+
level: "warn",
|
|
6786
|
+
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"
|
|
6787
|
+
});
|
|
6788
|
+
scheduleNextTick();
|
|
6789
|
+
} else if (isProbe) {
|
|
6790
|
+
logActivity(state, {
|
|
6791
|
+
type: "info",
|
|
6792
|
+
level: "debug",
|
|
6793
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6794
|
+
});
|
|
6795
|
+
armed = false;
|
|
6796
|
+
if (rearmRequested) rearm();
|
|
6797
|
+
} else {
|
|
6798
|
+
logActivity(state, {
|
|
6799
|
+
type: "info",
|
|
6800
|
+
level: "debug",
|
|
6801
|
+
message: `Claude usage reporting: ${error2.message}`
|
|
6802
|
+
});
|
|
6803
|
+
scheduleNextTick();
|
|
6804
|
+
}
|
|
6805
|
+
} else {
|
|
6806
|
+
consecutiveFailures++;
|
|
6807
|
+
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
6808
|
+
logActivity(state, {
|
|
6809
|
+
type: "info",
|
|
6810
|
+
level: claudeUsageFailureLogLevel(consecutiveFailures),
|
|
6811
|
+
message: `Claude usage reporting failed: ${message}${claudeUsageFailureStreakSuffix(consecutiveFailures)}`
|
|
6812
|
+
});
|
|
6813
|
+
scheduleNextTick();
|
|
6814
|
+
}
|
|
6815
|
+
}
|
|
6816
|
+
};
|
|
6817
|
+
armed = true;
|
|
6818
|
+
state.claudeUsageTimer = setTimeout(() => void tick(true), FIRST_REPORT_DELAY_MS);
|
|
6819
|
+
return rearm;
|
|
6820
|
+
}
|
|
4614
6821
|
async function notifyOffline(state) {
|
|
4615
6822
|
if (!state.agentId || !state.authHeader) return;
|
|
4616
6823
|
if (!state.connected) {
|
|
@@ -4628,13 +6835,29 @@ async function notifyOffline(state) {
|
|
|
4628
6835
|
if (state.interactive) displayStatus(state);
|
|
4629
6836
|
}
|
|
4630
6837
|
}
|
|
6838
|
+
async function timeShutdownPhase(state, durations, name, run2) {
|
|
6839
|
+
const startedAt = Date.now();
|
|
6840
|
+
try {
|
|
6841
|
+
return await run2();
|
|
6842
|
+
} finally {
|
|
6843
|
+
const elapsedMs = Date.now() - startedAt;
|
|
6844
|
+
durations[name] = elapsedMs;
|
|
6845
|
+
log2(state, `Shutdown phase ${name}: ${elapsedMs}ms`);
|
|
6846
|
+
}
|
|
6847
|
+
}
|
|
4631
6848
|
async function cleanup(state, opts = {}) {
|
|
6849
|
+
const durations = {};
|
|
4632
6850
|
state.running = false;
|
|
4633
6851
|
for (const timer of state.sessionCleanupTimers) {
|
|
4634
6852
|
clearInterval(timer);
|
|
4635
6853
|
clearTimeout(timer);
|
|
4636
6854
|
}
|
|
4637
6855
|
state.sessionCleanupTimers = [];
|
|
6856
|
+
if (state.claudeUsageTimer) {
|
|
6857
|
+
clearTimeout(state.claudeUsageTimer);
|
|
6858
|
+
state.claudeUsageTimer = null;
|
|
6859
|
+
}
|
|
6860
|
+
state.claudeUsageRearm = null;
|
|
4638
6861
|
if (opts.graceful && state.channelDriver) {
|
|
4639
6862
|
state.channelDriver.stop();
|
|
4640
6863
|
log2(state, "Draining in-flight channel work before shutdown...");
|
|
@@ -4642,7 +6865,13 @@ async function cleanup(state, opts = {}) {
|
|
|
4642
6865
|
logActivity(state, { type: "info", message: "Draining in-flight work before shutdown..." });
|
|
4643
6866
|
displayStatus(state);
|
|
4644
6867
|
}
|
|
4645
|
-
const
|
|
6868
|
+
const driver = state.channelDriver;
|
|
6869
|
+
const settled = await timeShutdownPhase(
|
|
6870
|
+
state,
|
|
6871
|
+
durations,
|
|
6872
|
+
"drain",
|
|
6873
|
+
() => driver.waitForInFlight(SHUTDOWN_DRAIN_TIMEOUT_MS)
|
|
6874
|
+
);
|
|
4646
6875
|
if (!settled) {
|
|
4647
6876
|
logActivity(state, {
|
|
4648
6877
|
type: "info",
|
|
@@ -4651,13 +6880,15 @@ async function cleanup(state, opts = {}) {
|
|
|
4651
6880
|
if (state.interactive) displayStatus(state);
|
|
4652
6881
|
}
|
|
4653
6882
|
}
|
|
4654
|
-
await notifyOffline(state);
|
|
6883
|
+
await timeShutdownPhase(state, durations, "offline_notify", () => notifyOffline(state));
|
|
4655
6884
|
if (state.connection) {
|
|
4656
|
-
state.connection
|
|
6885
|
+
const connection = state.connection;
|
|
6886
|
+
await timeShutdownPhase(state, durations, "tunnel_close", () => connection.close());
|
|
4657
6887
|
state.connection = null;
|
|
4658
6888
|
}
|
|
4659
6889
|
if (state.opencodeProcess) {
|
|
4660
|
-
|
|
6890
|
+
const opencodeProcess = state.opencodeProcess;
|
|
6891
|
+
await timeShutdownPhase(state, durations, "opencode_stop", () => stopOpenCode(opencodeProcess));
|
|
4661
6892
|
if (state.interactive) {
|
|
4662
6893
|
logActivity(state, { type: "info", message: "Stopped OpenCode process" });
|
|
4663
6894
|
displayStatus(state);
|
|
@@ -4666,12 +6897,15 @@ async function cleanup(state, opts = {}) {
|
|
|
4666
6897
|
}
|
|
4667
6898
|
state.opencodeProcess = null;
|
|
4668
6899
|
}
|
|
6900
|
+
return durations;
|
|
4669
6901
|
}
|
|
4670
6902
|
async function run(options) {
|
|
4671
6903
|
const interactive = isInteractive(options.json);
|
|
4672
6904
|
let logLevel;
|
|
6905
|
+
let fileSyncDirectories;
|
|
4673
6906
|
try {
|
|
4674
6907
|
logLevel = resolveLogLevel(options);
|
|
6908
|
+
fileSyncDirectories = resolveFileSyncDirectories(options.enableFileSyncTo, homedir3());
|
|
4675
6909
|
} catch (error2) {
|
|
4676
6910
|
const message = error2 instanceof Error ? error2.message : String(error2);
|
|
4677
6911
|
if (options.json) {
|
|
@@ -4684,7 +6918,7 @@ async function run(options) {
|
|
|
4684
6918
|
return;
|
|
4685
6919
|
}
|
|
4686
6920
|
const state = {
|
|
4687
|
-
agentId: options.agent || "",
|
|
6921
|
+
agentId: options.runner || options.agent || "",
|
|
4688
6922
|
agentName: null,
|
|
4689
6923
|
port: options.port ?? 4096,
|
|
4690
6924
|
conversationFilter: options.conversation ?? null,
|
|
@@ -4704,8 +6938,29 @@ async function run(options) {
|
|
|
4704
6938
|
messageCount: 0,
|
|
4705
6939
|
lastProxiedActivityAt: null,
|
|
4706
6940
|
sessionCleanupTimers: [],
|
|
6941
|
+
claudeUsageTimer: null,
|
|
6942
|
+
claudeUsageRearm: null,
|
|
4707
6943
|
authHeader: ""
|
|
4708
6944
|
};
|
|
6945
|
+
setTelemetryAuthProvider(() => ({ authHeader: state.authHeader, agentId: state.agentId }));
|
|
6946
|
+
if (fileSyncDirectories.length > 0) {
|
|
6947
|
+
log2(state, `File sync enabled for: ${fileSyncDirectories.join(", ")}`);
|
|
6948
|
+
} else {
|
|
6949
|
+
log2(state, "File sync is disabled (no --enable-file-sync-to given)", "debug");
|
|
6950
|
+
}
|
|
6951
|
+
if (!options.runner && options.agent) {
|
|
6952
|
+
telemetry.info(
|
|
6953
|
+
EventTypes.DEPRECATED_AGENT_FLAG_USED,
|
|
6954
|
+
"Deprecated --agent flag used instead of --runner",
|
|
6955
|
+
{ command: "run" },
|
|
6956
|
+
state.agentId
|
|
6957
|
+
);
|
|
6958
|
+
const agentFlagNotice = "--agent is deprecated, use --runner instead; will be removed in a future release.";
|
|
6959
|
+
log2(state, agentFlagNotice, "warn");
|
|
6960
|
+
if (state.interactive && !state.json) {
|
|
6961
|
+
logActivity(state, { type: "info", level: "warn", message: agentFlagNotice });
|
|
6962
|
+
}
|
|
6963
|
+
}
|
|
4709
6964
|
if (state.idleTimeout === null && (process.env.GITHUB_ACTIONS || process.env.CI)) {
|
|
4710
6965
|
log2(
|
|
4711
6966
|
state,
|
|
@@ -4716,14 +6971,38 @@ async function run(options) {
|
|
|
4716
6971
|
const handleSignal = async () => {
|
|
4717
6972
|
if (state.shuttingDown) return;
|
|
4718
6973
|
state.shuttingDown = true;
|
|
6974
|
+
const shutdownStartedAt = Date.now();
|
|
4719
6975
|
if (state.interactive) {
|
|
4720
6976
|
logActivity(state, { type: "info", message: "Shutting down..." });
|
|
4721
6977
|
displayStatus(state);
|
|
4722
6978
|
} else {
|
|
4723
6979
|
log2(state, "Shutting down...");
|
|
4724
6980
|
}
|
|
4725
|
-
await cleanup(state, { graceful: true });
|
|
4726
|
-
|
|
6981
|
+
const durations = await cleanup(state, { graceful: true });
|
|
6982
|
+
const telemetryBudgetMs = Number(process.env.EVIDENT_TELEMETRY_SHUTDOWN_MS) || TELEMETRY_SHUTDOWN_TIMEOUT_MS;
|
|
6983
|
+
await timeShutdownPhase(state, durations, "telemetry_flush", async () => {
|
|
6984
|
+
let timer;
|
|
6985
|
+
const flushed = shutdownTelemetry().then(
|
|
6986
|
+
() => true,
|
|
6987
|
+
(error2) => {
|
|
6988
|
+
log2(
|
|
6989
|
+
state,
|
|
6990
|
+
`Telemetry flush failed during shutdown: ${error2 instanceof Error ? error2.message : String(error2)}`,
|
|
6991
|
+
"warn"
|
|
6992
|
+
);
|
|
6993
|
+
return true;
|
|
6994
|
+
}
|
|
6995
|
+
);
|
|
6996
|
+
const timedOut = new Promise((resolve3) => {
|
|
6997
|
+
timer = setTimeout(() => resolve3(false), telemetryBudgetMs);
|
|
6998
|
+
});
|
|
6999
|
+
if (!await Promise.race([flushed, timedOut])) {
|
|
7000
|
+
log2(state, `Telemetry flush exceeded ${telemetryBudgetMs}ms \u2014 exiting anyway`, "warn");
|
|
7001
|
+
}
|
|
7002
|
+
clearTimeout(timer);
|
|
7003
|
+
});
|
|
7004
|
+
const breakdown = Object.entries(durations).map(([phase, ms]) => `${phase}=${ms}ms`).join(" ");
|
|
7005
|
+
log2(state, `Shutdown complete in ${Date.now() - shutdownStartedAt}ms (${breakdown})`);
|
|
4727
7006
|
process.exit(0);
|
|
4728
7007
|
};
|
|
4729
7008
|
process.on("SIGINT", handleSignal);
|
|
@@ -4734,10 +7013,13 @@ async function run(options) {
|
|
|
4734
7013
|
if (!interactive) {
|
|
4735
7014
|
printError("Authentication required");
|
|
4736
7015
|
blank();
|
|
4737
|
-
console.log(
|
|
7016
|
+
console.log(
|
|
7017
|
+
chalk6.dim("Set EVIDENT_RUNNER_KEY (or EVIDENT_AGENT_KEY) environment variable for CI")
|
|
7018
|
+
);
|
|
4738
7019
|
console.log(chalk6.dim("Or run `evident login` for interactive authentication"));
|
|
4739
7020
|
blank();
|
|
4740
7021
|
process.exit(1);
|
|
7022
|
+
return;
|
|
4741
7023
|
}
|
|
4742
7024
|
blank();
|
|
4743
7025
|
console.log(chalk6.yellow("You are not logged in to Evident."));
|
|
@@ -4748,6 +7030,25 @@ async function run(options) {
|
|
|
4748
7030
|
);
|
|
4749
7031
|
}
|
|
4750
7032
|
state.authHeader = getAuthHeader(credentials2);
|
|
7033
|
+
if (credentials2.notice) {
|
|
7034
|
+
log2(state, credentials2.notice, "warn");
|
|
7035
|
+
if (state.interactive && !state.json) {
|
|
7036
|
+
logActivity(state, { type: "info", level: "warn", message: credentials2.notice });
|
|
7037
|
+
}
|
|
7038
|
+
}
|
|
7039
|
+
if (credentials2.keySource === "agent_key") {
|
|
7040
|
+
telemetry.info(
|
|
7041
|
+
EventTypes.DEPRECATED_AGENT_KEY_ENV_USED,
|
|
7042
|
+
"Deprecated EVIDENT_AGENT_KEY env var used instead of EVIDENT_RUNNER_KEY",
|
|
7043
|
+
{ command: "run" },
|
|
7044
|
+
state.agentId
|
|
7045
|
+
);
|
|
7046
|
+
const agentKeyNotice = "EVIDENT_AGENT_KEY is deprecated, use EVIDENT_RUNNER_KEY instead; will be removed in a future release.";
|
|
7047
|
+
log2(state, agentKeyNotice, "warn");
|
|
7048
|
+
if (state.interactive && !state.json) {
|
|
7049
|
+
logActivity(state, { type: "info", level: "warn", message: agentKeyNotice });
|
|
7050
|
+
}
|
|
7051
|
+
}
|
|
4751
7052
|
if (!state.agentId) {
|
|
4752
7053
|
if (credentials2.authType === "agent_key") {
|
|
4753
7054
|
const resolved = await resolveAgentIdFromKey(state.authHeader);
|
|
@@ -4763,13 +7064,21 @@ async function run(options) {
|
|
|
4763
7064
|
} else {
|
|
4764
7065
|
printError(resolved.error || "Failed to resolve runner ID from key");
|
|
4765
7066
|
process.exit(1);
|
|
7067
|
+
return;
|
|
4766
7068
|
}
|
|
4767
7069
|
} else {
|
|
4768
|
-
printError(
|
|
7070
|
+
printError(
|
|
7071
|
+
"--runner (or --agent) is required when not using EVIDENT_RUNNER_KEY or EVIDENT_AGENT_KEY"
|
|
7072
|
+
);
|
|
4769
7073
|
blank();
|
|
4770
|
-
console.log(
|
|
7074
|
+
console.log(
|
|
7075
|
+
chalk6.dim(
|
|
7076
|
+
"Either provide --runner/--agent <id> or set EVIDENT_RUNNER_KEY/EVIDENT_AGENT_KEY"
|
|
7077
|
+
)
|
|
7078
|
+
);
|
|
4771
7079
|
blank();
|
|
4772
7080
|
process.exit(1);
|
|
7081
|
+
return;
|
|
4773
7082
|
}
|
|
4774
7083
|
}
|
|
4775
7084
|
telemetry.info(
|
|
@@ -4810,25 +7119,67 @@ async function run(options) {
|
|
|
4810
7119
|
}
|
|
4811
7120
|
spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
|
|
4812
7121
|
state.agentName = validation.agent.name;
|
|
7122
|
+
const microvmId = process.env.MICROVM_ID?.trim();
|
|
7123
|
+
if (microvmId) {
|
|
7124
|
+
const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
|
|
7125
|
+
if (reported.ok) {
|
|
7126
|
+
log2(state, "Reported MicroVM identity so this runner can be resumed rather than restarted");
|
|
7127
|
+
} else {
|
|
7128
|
+
const message = `Could not report MicroVM identity (future wakes will cold-start): ${reported.error}`;
|
|
7129
|
+
log2(state, message, "warn");
|
|
7130
|
+
if (state.interactive && !state.json) {
|
|
7131
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
7132
|
+
}
|
|
7133
|
+
}
|
|
7134
|
+
} else {
|
|
7135
|
+
log2(state, "Not running in a MicroVM (MICROVM_ID unset) \u2014 nothing to report", "debug");
|
|
7136
|
+
}
|
|
7137
|
+
const { timeoutMs: opencodeStartTimeoutMs, warnings: opencodeStartTimeoutWarnings } = resolveOpenCodeStartTimeoutMs(options, process.env);
|
|
7138
|
+
for (const warning2 of opencodeStartTimeoutWarnings) {
|
|
7139
|
+
logActivity(state, { type: "info", level: "warn", message: warning2 });
|
|
7140
|
+
}
|
|
4813
7141
|
const ocSpinner = interactive && !state.json ? ora3("Checking OpenCode...").start() : null;
|
|
4814
7142
|
try {
|
|
4815
7143
|
const oc = await ensureOpenCodeRunning({
|
|
4816
7144
|
port: state.port,
|
|
4817
7145
|
interactive: state.interactive,
|
|
4818
7146
|
agentId: state.agentId,
|
|
4819
|
-
log: (message) => log2(state, message)
|
|
7147
|
+
log: (message) => log2(state, message),
|
|
7148
|
+
startTimeoutMs: opencodeStartTimeoutMs
|
|
4820
7149
|
});
|
|
4821
7150
|
state.port = oc.port;
|
|
4822
7151
|
state.opencodeProcess = oc.process;
|
|
4823
7152
|
state.opencodeVersion = oc.version;
|
|
4824
|
-
state.opencodeConnected = oc.
|
|
7153
|
+
state.opencodeConnected = oc.notReadyReason === null;
|
|
4825
7154
|
const version2 = state.opencodeVersion ? ` (v${state.opencodeVersion})` : "";
|
|
4826
7155
|
ocSpinner?.succeed(`OpenCode running on port ${state.port}${version2}`);
|
|
4827
|
-
|
|
4828
|
-
|
|
4829
|
-
|
|
4830
|
-
|
|
4831
|
-
|
|
7156
|
+
if (!state.interactive && oc.notReadyReason !== null) {
|
|
7157
|
+
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}).`;
|
|
7158
|
+
logActivity(state, { type: "info", level: "warn", message });
|
|
7159
|
+
} else {
|
|
7160
|
+
const versionWarning = buildOpenCodeVersionWarning(state.opencodeVersion);
|
|
7161
|
+
if (versionWarning) {
|
|
7162
|
+
log2(state, versionWarning, "warn");
|
|
7163
|
+
if (state.interactive && !state.json) {
|
|
7164
|
+
logActivity(state, { type: "info", level: "warn", message: versionWarning });
|
|
7165
|
+
}
|
|
7166
|
+
}
|
|
7167
|
+
const noProviderWarning = buildNoProviderWarning(
|
|
7168
|
+
await hasAnyConfiguredProvider(state.port)
|
|
7169
|
+
);
|
|
7170
|
+
if (noProviderWarning) {
|
|
7171
|
+
log2(state, noProviderWarning, "warn");
|
|
7172
|
+
if (state.interactive && !state.json) {
|
|
7173
|
+
logActivity(state, { type: "info", level: "warn", message: noProviderWarning });
|
|
7174
|
+
blank();
|
|
7175
|
+
console.log(chalk6.yellow("\u26A0 No OpenCode model provider is configured."));
|
|
7176
|
+
console.log(
|
|
7177
|
+
chalk6.dim(
|
|
7178
|
+
`Run ${chalk6.cyan("opencode auth login")} to set one up \u2014 messages will fail until then.`
|
|
7179
|
+
)
|
|
7180
|
+
);
|
|
7181
|
+
blank();
|
|
7182
|
+
}
|
|
4832
7183
|
}
|
|
4833
7184
|
}
|
|
4834
7185
|
} catch (error2) {
|
|
@@ -4843,6 +7194,10 @@ async function run(options) {
|
|
|
4843
7194
|
getAuthHeader: () => state.authHeader,
|
|
4844
7195
|
conversationFilter: state.conversationFilter,
|
|
4845
7196
|
stuckQueuedMs: CHANNEL_STUCK_QUEUED_MS,
|
|
7197
|
+
// #559: the `--enable-file-sync-to` allow-list. Empty ⇒ pending files are
|
|
7198
|
+
// REJECTED with `file_sync_disabled` on the ack, not silently ignored.
|
|
7199
|
+
fileSyncDirectories,
|
|
7200
|
+
homeDir: homedir3(),
|
|
4846
7201
|
log: (entry) => (
|
|
4847
7202
|
// Thread the driver's real level straight through so `debug`/`warn`
|
|
4848
7203
|
// survive the sink filter (they no longer collapse to info). `type`
|
|
@@ -4869,6 +7224,18 @@ async function run(options) {
|
|
|
4869
7224
|
type: "info",
|
|
4870
7225
|
message: `Tunnel ${isReconnect ? "reconnected" : "connected"} (runner: ${agentId})`
|
|
4871
7226
|
});
|
|
7227
|
+
if (options.tunnelReadyFile) {
|
|
7228
|
+
const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
|
|
7229
|
+
if (marker.ok) {
|
|
7230
|
+
log2(state, `Wrote tunnel readiness marker to ${options.tunnelReadyFile}`, "debug");
|
|
7231
|
+
} else {
|
|
7232
|
+
log2(
|
|
7233
|
+
state,
|
|
7234
|
+
`Failed to write tunnel readiness marker to ${options.tunnelReadyFile}: ${marker.error}`,
|
|
7235
|
+
"error"
|
|
7236
|
+
);
|
|
7237
|
+
}
|
|
7238
|
+
}
|
|
4872
7239
|
emitAgentConnected(state.agentId, {
|
|
4873
7240
|
port: state.port,
|
|
4874
7241
|
cli_version: getCliVersion(),
|
|
@@ -4924,6 +7291,12 @@ async function run(options) {
|
|
|
4924
7291
|
onDrainPing: () => {
|
|
4925
7292
|
if (!state.running) return;
|
|
4926
7293
|
logActivity(state, { type: "info", message: "Drain ping received \u2014 draining" });
|
|
7294
|
+
void channelDriver.syncPendingFiles().catch(
|
|
7295
|
+
(error2) => logActivity(state, {
|
|
7296
|
+
type: "error",
|
|
7297
|
+
error: `Runner file sync failed on ping: ${error2 instanceof Error ? error2.message : String(error2)}`
|
|
7298
|
+
})
|
|
7299
|
+
);
|
|
4927
7300
|
channelDriver.drainPending().then((processed) => {
|
|
4928
7301
|
if (processed > 0) {
|
|
4929
7302
|
state.messageCount += processed;
|
|
@@ -4953,6 +7326,7 @@ async function run(options) {
|
|
|
4953
7326
|
throw error2;
|
|
4954
7327
|
}
|
|
4955
7328
|
scheduleSessionCleanup(state, channelDriver, options);
|
|
7329
|
+
state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
|
|
4956
7330
|
if (!interactive || state.json) {
|
|
4957
7331
|
log2(state, "Driving channel messages...");
|
|
4958
7332
|
}
|
|
@@ -4982,7 +7356,7 @@ async function run(options) {
|
|
|
4982
7356
|
}
|
|
4983
7357
|
telemetry.error(EventTypes.CLI_ERROR, `Run command failed: ${message}`, {
|
|
4984
7358
|
command: "run",
|
|
4985
|
-
agentId: options.agent
|
|
7359
|
+
agentId: options.runner || options.agent
|
|
4986
7360
|
});
|
|
4987
7361
|
await shutdownTelemetry();
|
|
4988
7362
|
process.exit(1);
|
|
@@ -5007,10 +7381,18 @@ program.name("evident").description("Run OpenCode locally and connect it to Evid
|
|
|
5007
7381
|
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);
|
|
5008
7382
|
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 }));
|
|
5009
7383
|
program.command("whoami").description("Show the currently logged in user").action(whoami);
|
|
5010
|
-
program.command("
|
|
7384
|
+
program.command("status").description("Check whether the configured credentials can reach Evident").option("--json", "Output in JSON format").action((options) => status({ json: options.json }));
|
|
7385
|
+
program.command("claude-usage").description("[spike] Show Claude subscription usage (requires a local `claude login`)").action(claudeUsage);
|
|
7386
|
+
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(
|
|
7387
|
+
"-a, --agent [id]",
|
|
7388
|
+
"Deprecated alias for --runner (still supported; --runner wins if both are given)"
|
|
7389
|
+
).option("-p, --port <port>", "OpenCode port (default: 4096)", "4096").option(
|
|
5011
7390
|
"--log-level <level>",
|
|
5012
7391
|
"Log verbosity: debug | info | warn | error (default: info). Env: EVIDENT_LOG_LEVEL"
|
|
5013
|
-
).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(
|
|
7392
|
+
).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(
|
|
7393
|
+
"--opencode-start-timeout <seconds>",
|
|
7394
|
+
"Seconds to wait for OpenCode to become healthy when the runner starts it (default: 180). Env: EVIDENT_OPENCODE_START_TIMEOUT"
|
|
7395
|
+
).option("--json", "Output in JSON format").option(
|
|
5014
7396
|
"--session-cleanup-max-age <duration>",
|
|
5015
7397
|
"Delete OpenCode sessions idle longer than this (e.g. 7d, 24h). Enables cleanup. Env: EVIDENT_SESSION_CLEANUP_MAX_AGE"
|
|
5016
7398
|
).option(
|
|
@@ -5019,10 +7401,22 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5019
7401
|
).option(
|
|
5020
7402
|
"--session-cleanup-interval <duration>",
|
|
5021
7403
|
"How often the cleanup sweep runs (default: 1h). Env: EVIDENT_SESSION_CLEANUP_INTERVAL"
|
|
7404
|
+
).option(
|
|
7405
|
+
"--claude-usage-reporting <mode>",
|
|
7406
|
+
"Report Claude subscription usage to Evident: auto | on | off (default: auto). Env: EVIDENT_CLAUDE_USAGE_REPORTING"
|
|
7407
|
+
).option(
|
|
7408
|
+
"--enable-file-sync-to <dir>",
|
|
7409
|
+
"Allow Evident to write files into this directory (repeatable). Omit to disable file sync entirely.",
|
|
7410
|
+
(value, previous) => previous.concat([value]),
|
|
7411
|
+
[]
|
|
7412
|
+
).option(
|
|
7413
|
+
"--tunnel-ready-file <path>",
|
|
7414
|
+
"Path to write once the tunnel is connected (opt-in; unused unless an operator/image sets it)"
|
|
5022
7415
|
).action(
|
|
5023
7416
|
(options) => {
|
|
5024
7417
|
run({
|
|
5025
7418
|
agent: options.agent,
|
|
7419
|
+
runner: options.runner,
|
|
5026
7420
|
port: parseInt(options.port, 10),
|
|
5027
7421
|
// Raw string — validation/precedence is single-sourced in run.ts's
|
|
5028
7422
|
// resolveLogLevel (flag > -v > EVIDENT_LOG_LEVEL > info).
|
|
@@ -5030,11 +7424,21 @@ program.command("run").description("Connect to Evident and process messages").op
|
|
|
5030
7424
|
verbose: options.verbose,
|
|
5031
7425
|
conversation: options.conversation,
|
|
5032
7426
|
idleTimeout: options.idleTimeout ? parseInt(options.idleTimeout, 10) : void 0,
|
|
7427
|
+
// Raw string — validation/precedence is single-sourced in run.ts's
|
|
7428
|
+
// resolveOpenCodeStartTimeoutMs (flag > EVIDENT_OPENCODE_START_TIMEOUT > 180s default).
|
|
7429
|
+
opencodeStartTimeout: options.opencodeStartTimeout,
|
|
5033
7430
|
json: options.json,
|
|
5034
7431
|
// Raw strings — the resolver in run.ts single-sources parsing (M1).
|
|
5035
7432
|
sessionCleanupMaxAge: options.sessionCleanupMaxAge,
|
|
5036
7433
|
sessionCleanupMaxCount: options.sessionCleanupMaxCount,
|
|
5037
|
-
sessionCleanupInterval: options.sessionCleanupInterval
|
|
7434
|
+
sessionCleanupInterval: options.sessionCleanupInterval,
|
|
7435
|
+
// Raw string — the resolver in run.ts single-sources parsing
|
|
7436
|
+
// (resolveClaudeUsageReportingMode).
|
|
7437
|
+
claudeUsageReporting: options.claudeUsageReporting,
|
|
7438
|
+
// Raw values — expansion/validation is single-sourced in run.ts's
|
|
7439
|
+
// resolveFileSyncDirectories.
|
|
7440
|
+
enableFileSyncTo: options.enableFileSyncTo,
|
|
7441
|
+
tunnelReadyFile: options.tunnelReadyFile
|
|
5038
7442
|
});
|
|
5039
7443
|
}
|
|
5040
7444
|
);
|