@openagents-org/agent-launcher 0.2.185 → 0.2.187

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.
@@ -76,7 +76,7 @@ const SECRET_ENV_VARS = [
76
76
  const ACT_ALLOW_TOOLS = ['shell', 'write'];
77
77
 
78
78
  // How long to wait for a turn before giving up (no output / hung CLI).
79
- const TURN_TIMEOUT_MS = 10 * 60 * 1000;
79
+ const TURN_IDLE_TIMEOUT_MS = 10 * 60 * 1000;
80
80
 
81
81
  class CopilotAdapter extends BaseAdapter {
82
82
  /**
@@ -565,10 +565,15 @@ class CopilotAdapter extends BaseAdapter {
565
565
  let pending = Promise.resolve();
566
566
  let settled = false;
567
567
 
568
- const timeout = setTimeout(() => {
569
- this._stopProcess(proc).catch(() => {});
570
- finish({ timedOut: true, finalText: accumulatedText() });
571
- }, TURN_TIMEOUT_MS);
568
+ let timeout = null;
569
+ const resetTimeout = () => {
570
+ if (timeout) clearTimeout(timeout);
571
+ timeout = setTimeout(() => {
572
+ this._stopProcess(proc).catch(() => {});
573
+ finish({ timedOut: true, finalText: accumulatedText() });
574
+ }, TURN_IDLE_TIMEOUT_MS);
575
+ };
576
+ resetTimeout();
572
577
 
573
578
  const finish = (summary) => {
574
579
  if (settled) return;
@@ -657,6 +662,9 @@ class CopilotAdapter extends BaseAdapter {
657
662
  }
658
663
  };
659
664
 
665
+ if (proc.stdout) proc.stdout.on('data', resetTimeout);
666
+ if (proc.stderr) proc.stderr.on('data', resetTimeout);
667
+
660
668
  proc.stdout.on('data', (chunk) => {
661
669
  for (const ev of parser.push(chunk)) {
662
670
  pending = pending.then(() => handle(ev)).catch(() => {});
@@ -25,6 +25,7 @@ const { spawn, bridgeSpawn } = require('../wsl');
25
25
 
26
26
  const BaseAdapter = require('./base');
27
27
  const { buildOpenclawSystemPrompt } = require('./workspace-prompt');
28
+ const { REASON } = require('./health-status');
28
29
  const { whichBinary, whereBinary } = require('../paths');
29
30
 
30
31
  const IS_WINDOWS = process.platform === 'win32';
@@ -123,6 +124,34 @@ class HermesAdapter extends BaseAdapter {
123
124
  return null;
124
125
  }
125
126
 
127
+ /**
128
+ * Preflight gate (run by the daemon before join). Hermes can do nothing
129
+ * without its CLI, so when none can be resolved we surface a precise
130
+ * 'runtime_missing' reason and skip the workspace join — instead of joining,
131
+ * reporting "online" and then failing every message. That contradiction is
132
+ * exactly what users hit: the workspace showed a green hermes and "all set"
133
+ * while the smoke test on the same screen said "Not installed", because
134
+ * `create_agent --install` treats a failed third-party install script as a
135
+ * warning and connects the agent anyway.
136
+ *
137
+ * Re-resolves rather than trusting the constructor's lookup: an install that
138
+ * landed after the daemon started must not need a restart to be seen. NOTE:
139
+ * 'runtime_missing' (binary gone/never landed at run time), NOT
140
+ * 'not_installed' — install detection lives in the installer.
141
+ */
142
+ preflight() {
143
+ if (!this._hermesBin) this._hermesBin = this._findHermesBinary();
144
+ if (!this._hermesBin) {
145
+ return {
146
+ ok: false,
147
+ reason: REASON.RUNTIME_MISSING,
148
+ message: `Hermes CLI not found — install with: ${HERMES_INSTALL_HINT}`,
149
+ };
150
+ }
151
+ this._log(`Hermes CLI resolved: ${this._hermesBin}`);
152
+ return { ok: true };
153
+ }
154
+
126
155
  _resolveProfile(explicit, agentName) {
127
156
  if (explicit && explicit !== '' && explicit !== 'auto') return explicit;
128
157
  // Match agent name to an existing ~/.hermes/profiles/<name> if present
@@ -371,6 +400,17 @@ class HermesAdapter extends BaseAdapter {
371
400
  const sender = msg.senderName || msg.senderType || 'user';
372
401
  this._log(`Processing workspace message from ${sender} in ${msgChannel}`);
373
402
 
403
+ // Re-resolve per message: the constructor's lookup is a snapshot, and an
404
+ // agent that was created before its CLI finished installing would
405
+ // otherwise stay broken until the daemon restarts.
406
+ if (!this._hermesBin) this._hermesBin = this._findHermesBinary();
407
+ if (!this._hermesBin) {
408
+ const message = `Hermes CLI not found — install with: ${HERMES_INSTALL_HINT}`;
409
+ this._reportStatus(REASON.RUNTIME_MISSING, message);
410
+ await this.sendError(msgChannel, message);
411
+ return;
412
+ }
413
+
374
414
  await this._autoTitleChannel(msgChannel, content);
375
415
  await this.sendStatus(msgChannel, 'thinking...');
376
416
 
@@ -158,7 +158,8 @@ class LlmDirectAdapter extends BaseAdapter {
158
158
  'Authorization': `Bearer ${this._apiKey}`,
159
159
  };
160
160
  const payload = JSON.stringify({
161
- model: this._model || 'gpt-4o',
161
+ // The model picked in the workspace, which modelLabel() reports, wins.
162
+ model: this.modelLabel() || 'gpt-4o',
162
163
  messages,
163
164
  stream: true,
164
165
  });
@@ -27,8 +27,10 @@ const { whichBinary, whereBinary, getEnhancedEnv } = require('../paths');
27
27
 
28
28
  const IS_WINDOWS = process.platform === 'win32';
29
29
 
30
- // Max wall-clock for a single `opencode run`. When it fires, Node kills the
31
- // child with SIGTERM, the 'close' handler sees signal='SIGTERM' / code=null.
30
+ // Max time an `opencode run` may stay completely silent. This is deliberately
31
+ // an IDLE timeout, not a wall-clock cap: long coding tasks routinely exceed
32
+ // five minutes while continuing to emit text/tool events. A fixed spawn
33
+ // timeout killed those healthy runs midway through their answer.
32
34
  const TIMEOUT_MS = 300000; // 5 minutes
33
35
 
34
36
  // Pinned OpenCode CLI version. Do NOT use @latest anywhere (registry.json /
@@ -158,6 +160,34 @@ class OpenCodeAdapter extends BaseAdapter {
158
160
  }
159
161
 
160
162
  /** Environment the CLI is spawned with. */
163
+
164
+ /** Test seam for the inactivity watchdog. */
165
+ _runTimeoutMs() {
166
+ return TIMEOUT_MS;
167
+ }
168
+
169
+ /**
170
+ * Arm a watchdog that expires only after a continuous period without CLI
171
+ * output. Every stdout/stderr chunk proves the process is still making
172
+ * progress and pushes the deadline out again.
173
+ */
174
+ _createIdleWatchdog(onTimeout) {
175
+ const timeoutMs = this._runTimeoutMs();
176
+ let timer = null;
177
+ let stopped = false;
178
+ const touch = () => {
179
+ if (stopped) return;
180
+ if (timer) clearTimeout(timer);
181
+ timer = setTimeout(onTimeout, timeoutMs);
182
+ };
183
+ const cancel = () => {
184
+ stopped = true;
185
+ if (timer) clearTimeout(timer);
186
+ timer = null;
187
+ };
188
+ touch();
189
+ return { touch, cancel, timeoutMs };
190
+ }
161
191
  _spawnEnv() {
162
192
  return { ...(this.agentEnv || process.env) };
163
193
  }
@@ -701,8 +731,8 @@ class OpenCodeAdapter extends BaseAdapter {
701
731
  * CRITICAL: opencode-ai does NOT read the OPENCODE_MODEL env var. When no
702
732
  * model is given on the command line AND none is set in opencode.json, the
703
733
  * non-interactive `run` command hangs forever waiting for interactive
704
- * provider/model selection — it emits zero output until the spawn timeout
705
- * kills it with SIGTERM (surfaced as the misleading "exited with code null").
734
+ * provider/model selection — it emits zero output until the inactivity
735
+ * watchdog kills it with SIGTERM.
706
736
  * Passing `--model` explicitly is what makes headless runs actually work.
707
737
  */
708
738
  _resolveModel() {
@@ -843,7 +873,6 @@ class OpenCodeAdapter extends BaseAdapter {
843
873
  stdio: ['pipe', 'pipe', 'pipe'],
844
874
  env: spawnEnv,
845
875
  cwd: runCwd,
846
- timeout: TIMEOUT_MS,
847
876
  detached: !IS_WINDOWS,
848
877
  // Never let a console window appear. Besides being ugly, an attached
849
878
  // console makes opencode start its interactive TUI and hang instead of
@@ -859,9 +888,16 @@ class OpenCodeAdapter extends BaseAdapter {
859
888
  let pendingEvents = Promise.resolve();
860
889
  const responseState = { finalText: '', allText: '', seenText: false };
861
890
  let settled = false;
891
+ let timedOut = false;
892
+ const watchdog = this._createIdleWatchdog(() => {
893
+ timedOut = true;
894
+ this._log(`opencode idle for ${this._runTimeoutMs() / 1000}s; terminating`);
895
+ this._stopProcess(proc).catch(() => {});
896
+ });
862
897
  const finish = (fn, arg) => {
863
898
  if (settled) return;
864
899
  settled = true;
900
+ watchdog.cancel();
865
901
  if (this._channelProcesses[msgChannel] === proc) {
866
902
  delete this._channelProcesses[msgChannel];
867
903
  }
@@ -870,6 +906,7 @@ class OpenCodeAdapter extends BaseAdapter {
870
906
 
871
907
  if (proc.stdout) {
872
908
  proc.stdout.on('data', (d) => {
909
+ watchdog.touch();
873
910
  const chunk = d.toString('utf-8');
874
911
  stdout += chunk;
875
912
  streamBuffer += chunk;
@@ -882,7 +919,10 @@ class OpenCodeAdapter extends BaseAdapter {
882
919
  }
883
920
  });
884
921
  }
885
- if (proc.stderr) proc.stderr.on('data', (d) => { stderr += d; });
922
+ if (proc.stderr) proc.stderr.on('data', (d) => {
923
+ watchdog.touch();
924
+ stderr += d;
925
+ });
886
926
 
887
927
  // Send the prompt via stdin, then close it so opencode knows the message
888
928
  // is complete and never blocks waiting for more input. Guard the write:
@@ -897,9 +937,8 @@ class OpenCodeAdapter extends BaseAdapter {
897
937
 
898
938
  proc.on('error', (err) => finish(reject, err));
899
939
  // 'close' (not 'exit') fires after stdout/stderr have fully drained, so we
900
- // never parse a truncated response. The handler receives (code, signal):
901
- // when the spawn timeout kills the child, code is null and signal is the
902
- // signal name — report that as a timeout rather than "exited with code null".
940
+ // never parse a truncated response. The explicit `timedOut` flag keeps a
941
+ // watchdog termination distinct from an external signal or crash.
903
942
  proc.on('close', async (code, signal) => {
904
943
  try { await pendingEvents; } catch {}
905
944
 
@@ -917,14 +956,18 @@ class OpenCodeAdapter extends BaseAdapter {
917
956
  // because stderr happened to be empty.
918
957
  const stdoutErr = OpenCodeAdapter._extractErrorFromStdout(stdout);
919
958
 
920
- if (signal) {
921
- this._log(`opencode killed by signal ${signal} after ${TIMEOUT_MS / 1000}s (output: ${!!stdout})`);
959
+ if (timedOut) {
960
+ this._log(`opencode timed out after ${watchdog.timeoutMs / 1000}s idle (output: ${!!stdout})`);
922
961
  const cls = this.constructor._classifyFailure({ code, signal, stdout, stderr, stdoutErr });
923
- // A signal almost always means our own timeout fired; only override to
924
- // a more specific provider error if stdout/stderr clearly show one.
925
962
  const category = (cls.category === 'unknown_error' || cls.category === 'process_crashed')
926
963
  ? 'timeout' : cls.category;
927
- return finish(reject, this._failure(category, cls.diagnostic || `signal=${signal}`, cls.detail));
964
+ return finish(reject, this._failure(category, cls.diagnostic || `idle=${watchdog.timeoutMs}ms`, cls.detail));
965
+ }
966
+
967
+ if (signal) {
968
+ this._log(`opencode killed by signal ${signal} (output: ${!!stdout})`);
969
+ const cls = this.constructor._classifyFailure({ code, signal, stdout, stderr, stdoutErr });
970
+ return finish(reject, this._failure(cls.category, cls.diagnostic || `signal=${signal}`, cls.detail));
928
971
  }
929
972
 
930
973
  if (code !== 0) {
@@ -98,7 +98,6 @@ const SERVER_READY_TIMEOUT_MS = 90000;
98
98
  const HEALTH_POLL_INTERVAL_MS = 400;
99
99
 
100
100
  // Max wall-clock for one turn, after which we interrupt and report it.
101
- const TURN_TIMEOUT_MS = 900000; // 15 minutes
102
101
 
103
102
  // Idle watchdog over the event stream, in the same shape the CLI adapters use.
104
103
  const WATCHDOG_INTERVAL_MS = 15000;
@@ -674,7 +673,6 @@ class OpenWorkerAdapter extends BaseAdapter {
674
673
  if (settled) return;
675
674
  settled = true;
676
675
  clearInterval(watchdog);
677
- clearTimeout(timeout);
678
676
  if (this._channelSockets[channel] === socket) delete this._channelSockets[channel];
679
677
  try { socket.close(); } catch {}
680
678
  const userStopped = this._stoppingChannels.has(channel);
@@ -821,12 +819,6 @@ class OpenWorkerAdapter extends BaseAdapter {
821
819
  }
822
820
  }, WATCHDOG_INTERVAL_MS);
823
821
 
824
- const timeout = setTimeout(() => {
825
- this._log(`Turn exceeded ${TURN_TIMEOUT_MS}ms — interrupting`);
826
- send({ type: 'interrupt' });
827
- if (!error) error = 'The turn ran past its time limit and was stopped.';
828
- finish({});
829
- }, TURN_TIMEOUT_MS);
830
822
  });
831
823
  }
832
824
 
@@ -88,19 +88,22 @@ function formatAttachmentsForPrompt(
88
88
  const url = att.url || `${base}/v1/files/${fileId}`;
89
89
  const curl = isWindows ? 'curl.exe' : 'curl';
90
90
  const tmpDir = isWindows ? '$env:TEMP' : '/tmp';
91
+ // A web upload's filename is its workspace path (uploaded_files/<stamp>_<name>)
92
+ // and curl -o does not create folders, so that download failed with exit 23.
93
+ const localName = filename.split(/[\\/]/).pop() || fileId || 'attachment';
91
94
  if (contentType.startsWith('image/')) {
92
95
  lines.push(
93
96
  `- Image: ${filename} (file_id: ${fileId}) — ` +
94
97
  `download with curl, then use your Read tool on the local file to view it:\n` +
95
- ` Step 1: ${curl} -s -H "X-Workspace-Token: ${tokenExpr}" "${url}" -o ${tmpDir}/${filename}\n` +
96
- ` Step 2: Use the Read tool on ${tmpDir}/${filename} to see the image`
98
+ ` Step 1: ${curl} -s -H "X-Workspace-Token: ${tokenExpr}" "${url}" -o ${tmpDir}/${localName}\n` +
99
+ ` Step 2: Use the Read tool on ${tmpDir}/${localName} to see the image`
97
100
  );
98
101
  } else {
99
102
  lines.push(
100
103
  `- File: ${filename} (file_id: ${fileId}, type: ${contentType}) — ` +
101
104
  `download with curl, then use your Read tool on the local file:\n` +
102
- ` Step 1: ${curl} -s -H "X-Workspace-Token: ${tokenExpr}" "${url}" -o ${tmpDir}/${filename}\n` +
103
- ` Step 2: Use the Read tool on ${tmpDir}/${filename} to read the file`
105
+ ` Step 1: ${curl} -s -H "X-Workspace-Token: ${tokenExpr}" "${url}" -o ${tmpDir}/${localName}\n` +
106
+ ` Step 2: Use the Read tool on ${tmpDir}/${localName} to read the file`
104
107
  );
105
108
  }
106
109
  } else {
package/src/daemon.js CHANGED
@@ -9,6 +9,7 @@ const { execSync, execFileSync } = require('child_process');
9
9
  const { spawn } = require('./wsl');
10
10
  const os = require('os');
11
11
  const { WorkspaceClient } = require('./workspace-client');
12
+ const { listEndpointModels } = require('./model-list');
12
13
  const { getEnhancedEnv, whichBinary, IS_WINDOWS, defaultAgentWorkdir } = require('./paths');
13
14
 
14
15
  /**
@@ -44,6 +45,23 @@ function filesystemRoots(platform = os.platform(), exists = fs.existsSync) {
44
45
  return roots;
45
46
  }
46
47
 
48
+ /**
49
+ * Hostname of a configured base URL, or null when there is none. The roster
50
+ * carries only this: a relay URL can hold a credential in its userinfo or
51
+ * path. A value written without a scheme (`localhost:4000`) still names a host.
52
+ */
53
+ function endpointHost(value) {
54
+ const raw = String(value || '').trim();
55
+ if (!raw) return null;
56
+ for (const candidate of [raw, `http://${raw}`]) {
57
+ try {
58
+ const host = new URL(candidate).hostname.toLowerCase();
59
+ if (host) return host;
60
+ } catch {}
61
+ }
62
+ return null;
63
+ }
64
+
47
65
  /**
48
66
  * Agent process lifecycle manager.
49
67
  *
@@ -326,7 +344,8 @@ class Daemon {
326
344
  // Model/key: per-agent env override, else the type-level saved env.
327
345
  // Both power the workspace agent cards; the key goes out MASKED only,
328
346
  // so the edit form can show a key is configured without the secret
329
- // ever leaving the device.
347
+ // ever leaving the device. The endpoint goes out as a hostname, so
348
+ // the workspace can tell a relay from the vendor its model list is for.
330
349
  let typeEnv = {};
331
350
  try { typeEnv = this.envManager.load(a.type) || {}; } catch {}
332
351
  const model = (a.env && a.env.LLM_MODEL) || typeEnv.LLM_MODEL || null;
@@ -338,6 +357,7 @@ class Daemon {
338
357
  model: model || null,
339
358
  workingDir: a.path || null,
340
359
  apiKeyMasked: apiKey ? maskApiKey(apiKey) : null,
360
+ baseUrlHost: endpointHost(this._configuredBaseUrl(a, typeEnv)),
341
361
  probe: this._probes[a.name] || null,
342
362
  });
343
363
  }
@@ -347,6 +367,52 @@ class Daemon {
347
367
  return roster;
348
368
  }
349
369
 
370
+ /**
371
+ * The base URL this agent's CLI is pointed at, found the way _buildAgentEnv
372
+ * builds its env: LLM_BASE_URL or a provider variable (OPENAI_BASE_URL,
373
+ * ANTHROPIC_BASE_URL, …) saved for it or mapped by resolve_env, and failing
374
+ * those, the daemon's own environment for the provider variable this type's
375
+ * LLM_BASE_URL maps to, which the spawned CLI inherits.
376
+ */
377
+ _configuredBaseUrl(a, typeEnv) {
378
+ const saved = { ...typeEnv, ...(a.env || {}) };
379
+ let env = saved;
380
+ try { env = { ...saved, ...this.envManager.resolve(a.type, saved, this.registry) }; } catch {}
381
+ const set = (v) => String(v || '').trim();
382
+ if (set(env.LLM_BASE_URL)) return env.LLM_BASE_URL;
383
+ const key = Object.keys(env).find((k) => /_BASE_URL$/.test(k) && set(env[k]));
384
+ if (key) return env[key];
385
+ let rules = [];
386
+ try { rules = (this.registry && this.registry.getResolveRules(a.type)) || []; } catch {}
387
+ const inherited = rules.find((r) => r && r.from === 'LLM_BASE_URL' && r.to && set(process.env[r.to]));
388
+ return inherited ? process.env[inherited.to] : null;
389
+ }
390
+
391
+ /**
392
+ * Where this agent's CLI sends requests and with which key, for listing that
393
+ * endpoint's models. The key is looked up the way the CLI would find it: the
394
+ * protocol's own variables for an Anthropic-protocol agent, else LLM_API_KEY
395
+ * or the variable resolve_env maps it to, inherited from the daemon or not.
396
+ */
397
+ _agentEndpoint(a) {
398
+ let typeEnv = {};
399
+ try { typeEnv = this.envManager.load(a.type) || {}; } catch {}
400
+ const env = this._buildAgentEnv(a);
401
+ let entry = null;
402
+ try { entry = this.registry.getEntry(a.type); } catch {}
403
+ const protocol = (entry && entry.protocol) || 'openai';
404
+ const rules = (entry && entry.resolve_env && entry.resolve_env.rules) || [];
405
+ const keyVars = protocol === 'anthropic'
406
+ ? ['ANTHROPIC_AUTH_TOKEN', 'ANTHROPIC_API_KEY', 'LLM_API_KEY']
407
+ : ['LLM_API_KEY', ...rules.filter((r) => r && r.from === 'LLM_API_KEY').map((r) => r.to), 'OPENAI_API_KEY'];
408
+ const keyVar = keyVars.find((k) => k && String(env[k] || '').trim());
409
+ return {
410
+ baseUrl: this._configuredBaseUrl(a, typeEnv),
411
+ apiKey: keyVar ? String(env[keyVar]).trim() : '',
412
+ protocol,
413
+ };
414
+ }
415
+
350
416
  /**
351
417
  * Filesystem hint for the working-directory picker: home, its immediate
352
418
  * (non-hidden) subfolders, and the device's filesystem roots. Roots let a
@@ -606,6 +672,20 @@ async _runNodeCommand(n, cmd) {
606
672
  this._nodeHeartbeat();
607
673
  ok = true;
608
674
  message = `Detected ${this._runtimes.length} runtime(s)`;
675
+ } else if (action === 'list_models') {
676
+ // The model picker asks what the agent's own endpoint serves. The
677
+ // request needs the key, which is why it runs here.
678
+ if (!name) throw new Error('Missing agent name');
679
+ const agent = this.config.getAgent(name);
680
+ if (!agent || !this._agentOnNodeWorkspace(agent, n)) {
681
+ throw new Error(`Agent '${name}' is not managed by this workspace`);
682
+ }
683
+ const endpoint = this._agentEndpoint(agent);
684
+ if (!endpoint.baseUrl) throw new Error(`Agent '${name}' uses its CLI's default endpoint`);
685
+ const listed = await listEndpointModels(endpoint);
686
+ ok = listed.models.length > 0;
687
+ message = ok ? `Listed ${listed.models.length} model(s)` : listed.error;
688
+ data = { models: listed.models };
609
689
  } else if (action === 'list_dir') {
610
690
  data = this._listDir(args.path);
611
691
  ok = true;