switchroom 0.14.82 → 0.14.83
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/dist/cli/switchroom.js +121 -6
- package/dist/cli/ui/index.html +67 -10
- package/package.json +1 -1
- package/telegram-plugin/dist/gateway/gateway.js +76 -4
- package/telegram-plugin/gateway/gateway.ts +40 -0
- package/telegram-plugin/quota-watch.ts +93 -0
- package/telegram-plugin/tests/quota-watch.test.ts +72 -0
package/dist/cli/switchroom.js
CHANGED
|
@@ -14863,6 +14863,39 @@ async function parseSseOrJson(resp) {
|
|
|
14863
14863
|
const payload = dataLine ? dataLine.slice("data: ".length) : text;
|
|
14864
14864
|
return JSON.parse(payload);
|
|
14865
14865
|
}
|
|
14866
|
+
async function fetchHindsightToolsList(apiUrl, opts) {
|
|
14867
|
+
const fetchImpl = opts?.fetchImpl ?? fetch;
|
|
14868
|
+
const timeoutMs = opts?.timeoutMs ?? 4000;
|
|
14869
|
+
const bankId = opts?.bankId ?? "__doctor_probe__";
|
|
14870
|
+
const controller = new AbortController;
|
|
14871
|
+
const timeout = setTimeout(() => controller.abort(), timeoutMs);
|
|
14872
|
+
try {
|
|
14873
|
+
const resp = await fetchImpl(`${apiUrl}`, {
|
|
14874
|
+
method: "POST",
|
|
14875
|
+
headers: {
|
|
14876
|
+
"Content-Type": "application/json",
|
|
14877
|
+
Accept: "application/json, text/event-stream",
|
|
14878
|
+
"X-Bank-Id": bankId
|
|
14879
|
+
},
|
|
14880
|
+
body: JSON.stringify({ jsonrpc: "2.0", id: 1, method: "tools/list" }),
|
|
14881
|
+
signal: controller.signal
|
|
14882
|
+
});
|
|
14883
|
+
clearTimeout(timeout);
|
|
14884
|
+
if (!resp.ok)
|
|
14885
|
+
return { ok: false, reason: `HTTP ${resp.status}` };
|
|
14886
|
+
const parsed = await parseSseOrJson(resp);
|
|
14887
|
+
const raw = parsed.result?.tools;
|
|
14888
|
+
if (!Array.isArray(raw))
|
|
14889
|
+
return { ok: false, reason: "no tools in tools/list response" };
|
|
14890
|
+
const tools = raw.filter((t) => typeof t?.name === "string").map((t) => ({ name: t.name, required: t.inputSchema?.required ?? [] }));
|
|
14891
|
+
return { ok: true, tools };
|
|
14892
|
+
} catch (err) {
|
|
14893
|
+
clearTimeout(timeout);
|
|
14894
|
+
if (err.name === "AbortError")
|
|
14895
|
+
return { ok: false, reason: "Timeout" };
|
|
14896
|
+
return { ok: false, reason: String(err.message ?? err) };
|
|
14897
|
+
}
|
|
14898
|
+
}
|
|
14866
14899
|
async function probeHindsight(apiUrl, opts) {
|
|
14867
14900
|
const fetchImpl = opts?.fetchImpl ?? fetch;
|
|
14868
14901
|
const timeoutMs = opts?.timeoutMs ?? 3000;
|
|
@@ -14991,8 +15024,7 @@ async function ensureUserProfileMentalModel(apiUrl, bankId, opts) {
|
|
|
14991
15024
|
name: "create_mental_model",
|
|
14992
15025
|
arguments: {
|
|
14993
15026
|
name: "user-profile",
|
|
14994
|
-
source_query: "What are the key facts, preferences, context, and communication style about the user I talk to? Summarize what matters for making the agent feel like it knows them."
|
|
14995
|
-
types: ["world", "experience"]
|
|
15027
|
+
source_query: "What are the key facts, preferences, context, and communication style about the user I talk to? Summarize what matters for making the agent feel like it knows them."
|
|
14996
15028
|
}
|
|
14997
15029
|
}
|
|
14998
15030
|
}),
|
|
@@ -15077,6 +15109,12 @@ async function createBank(apiUrl, bankId, opts) {
|
|
|
15077
15109
|
if (!toolResponse.ok) {
|
|
15078
15110
|
return { ok: false, reason: `Tool call HTTP ${toolResponse.status}` };
|
|
15079
15111
|
}
|
|
15112
|
+
try {
|
|
15113
|
+
const created = await parseSseOrJson(toolResponse);
|
|
15114
|
+
if (created.result?.isError === true) {
|
|
15115
|
+
return { ok: false, reason: created.result.content?.[0]?.text ?? "create_bank returned isError" };
|
|
15116
|
+
}
|
|
15117
|
+
} catch {}
|
|
15080
15118
|
return { ok: true };
|
|
15081
15119
|
} catch (err) {
|
|
15082
15120
|
if (err.name === "AbortError") {
|
|
@@ -15137,8 +15175,8 @@ async function updateBankMissions(apiUrl, bankId, missions, opts) {
|
|
|
15137
15175
|
name: "update_bank",
|
|
15138
15176
|
arguments: {
|
|
15139
15177
|
bank_id: bankId,
|
|
15140
|
-
mission: missions.bank_mission,
|
|
15141
|
-
retain_mission: missions.retain_mission
|
|
15178
|
+
...missions.bank_mission != null ? { mission: missions.bank_mission } : {},
|
|
15179
|
+
...missions.retain_mission != null ? { config_updates: { retain_mission: missions.retain_mission } } : {}
|
|
15142
15180
|
}
|
|
15143
15181
|
}
|
|
15144
15182
|
}),
|
|
@@ -15148,6 +15186,12 @@ async function updateBankMissions(apiUrl, bankId, missions, opts) {
|
|
|
15148
15186
|
if (!toolResponse.ok) {
|
|
15149
15187
|
return { ok: false, reason: `Tool call HTTP ${toolResponse.status}` };
|
|
15150
15188
|
}
|
|
15189
|
+
try {
|
|
15190
|
+
const updated = await parseSseOrJson(toolResponse);
|
|
15191
|
+
if (updated.result?.isError === true) {
|
|
15192
|
+
return { ok: false, reason: updated.result.content?.[0]?.text ?? "update_bank returned isError" };
|
|
15193
|
+
}
|
|
15194
|
+
} catch {}
|
|
15151
15195
|
return { ok: true };
|
|
15152
15196
|
} catch (err) {
|
|
15153
15197
|
if (err.name === "AbortError") {
|
|
@@ -28959,6 +29003,30 @@ var init_manifest = __esm(() => {
|
|
|
28959
29003
|
]);
|
|
28960
29004
|
});
|
|
28961
29005
|
|
|
29006
|
+
// src/memory/hindsight-tools.ts
|
|
29007
|
+
var EXPECTED_HINDSIGHT_TOOLS;
|
|
29008
|
+
var init_hindsight_tools = __esm(() => {
|
|
29009
|
+
EXPECTED_HINDSIGHT_TOOLS = {
|
|
29010
|
+
recall: { required: ["query"] },
|
|
29011
|
+
reflect: { required: ["query"] },
|
|
29012
|
+
retain: { required: ["content"] },
|
|
29013
|
+
sync_retain: { required: ["content"] },
|
|
29014
|
+
delete_document: { required: ["document_id"] },
|
|
29015
|
+
create_directive: { required: ["content", "name"] },
|
|
29016
|
+
list_directives: { required: [] },
|
|
29017
|
+
delete_directive: { required: ["directive_id"] },
|
|
29018
|
+
create_bank: { required: ["bank_id"] },
|
|
29019
|
+
update_bank: { required: [] },
|
|
29020
|
+
list_banks: { required: [] },
|
|
29021
|
+
create_mental_model: { required: ["name", "source_query"] },
|
|
29022
|
+
list_mental_models: { required: [] },
|
|
29023
|
+
update_mental_model: { required: ["mental_model_id"] },
|
|
29024
|
+
refresh_mental_model: { required: ["mental_model_id"] },
|
|
29025
|
+
list_memories: { required: [] },
|
|
29026
|
+
get_memory: { required: ["memory_id"] }
|
|
29027
|
+
};
|
|
29028
|
+
});
|
|
29029
|
+
|
|
28962
29030
|
// src/cli/doctor-memory.ts
|
|
28963
29031
|
import { execFileSync as execFileSync17 } from "node:child_process";
|
|
28964
29032
|
function classifyShmSize(bytes) {
|
|
@@ -29030,8 +29098,51 @@ function checkHindsightContainerHealth(opts) {
|
|
|
29030
29098
|
} catch {}
|
|
29031
29099
|
return results;
|
|
29032
29100
|
}
|
|
29101
|
+
function classifyToolContract(advertised) {
|
|
29102
|
+
const byName = new Map(advertised.map((t) => [t.name, t]));
|
|
29103
|
+
const results = [];
|
|
29104
|
+
for (const [tool, spec] of Object.entries(EXPECTED_HINDSIGHT_TOOLS)) {
|
|
29105
|
+
const real = byName.get(tool);
|
|
29106
|
+
if (real === undefined) {
|
|
29107
|
+
results.push({
|
|
29108
|
+
name: `hindsight contract: ${tool}`,
|
|
29109
|
+
status: "fail",
|
|
29110
|
+
detail: `switchroom calls \`${tool}\` but the server no longer advertises it ` + `(renamed/removed upstream) \u2014 every callsite silently no-ops`,
|
|
29111
|
+
fix: "Upstream hindsight changed its MCP tool contract. Update the callsite " + "+ EXPECTED_HINDSIGHT_TOOLS (src/memory/hindsight-tools.ts) to the new " + "name, refresh tests/fixtures/hindsight-tools-list.snapshot.json, or pin " + "the prior hindsight image."
|
|
29112
|
+
});
|
|
29113
|
+
continue;
|
|
29114
|
+
}
|
|
29115
|
+
const missing = spec.required.filter((arg) => !real.required.includes(arg));
|
|
29116
|
+
const added = real.required.filter((arg) => !spec.required.includes(arg));
|
|
29117
|
+
if (added.length > 0) {
|
|
29118
|
+
results.push({
|
|
29119
|
+
name: `hindsight contract: ${tool}`,
|
|
29120
|
+
status: "fail",
|
|
29121
|
+
detail: `server now requires [${added.join(", ")}] on \`${tool}\` which ` + `switchroom does not track \u2014 calls may silently no-op`,
|
|
29122
|
+
fix: "Reconcile EXPECTED_HINDSIGHT_TOOLS + the callsite args with the new " + "server schema, then refresh the snapshot fixture."
|
|
29123
|
+
});
|
|
29124
|
+
} else if (missing.length > 0) {
|
|
29125
|
+
results.push({
|
|
29126
|
+
name: `hindsight contract: ${tool}`,
|
|
29127
|
+
status: "warn",
|
|
29128
|
+
detail: `switchroom treats [${missing.join(", ")}] as required on \`${tool}\` ` + `but the server no longer does (loosened upstream) \u2014 harmless, but the ` + `fixture is stale`,
|
|
29129
|
+
fix: "Refresh EXPECTED_HINDSIGHT_TOOLS + the snapshot fixture."
|
|
29130
|
+
});
|
|
29131
|
+
}
|
|
29132
|
+
}
|
|
29133
|
+
if (results.length === 0) {
|
|
29134
|
+
const used = Object.keys(EXPECTED_HINDSIGHT_TOOLS).length;
|
|
29135
|
+
results.push({
|
|
29136
|
+
name: "hindsight contract",
|
|
29137
|
+
status: "ok",
|
|
29138
|
+
detail: `${used} used tools present, required args satisfied (${advertised.length} advertised)`
|
|
29139
|
+
});
|
|
29140
|
+
}
|
|
29141
|
+
return results;
|
|
29142
|
+
}
|
|
29033
29143
|
var MIN_HINDSIGHT_SHM_BYTES;
|
|
29034
29144
|
var init_doctor_memory = __esm(() => {
|
|
29145
|
+
init_hindsight_tools();
|
|
29035
29146
|
MIN_HINDSIGHT_SHM_BYTES = 1024 * 1024 * 1024;
|
|
29036
29147
|
});
|
|
29037
29148
|
|
|
@@ -32043,6 +32154,10 @@ async function checkHindsight(config) {
|
|
|
32043
32154
|
status: "ok",
|
|
32044
32155
|
detail: `${probe2.serverName} ${probe2.serverVersion} at ${host}:${port}`
|
|
32045
32156
|
});
|
|
32157
|
+
const toolsList = await fetchHindsightToolsList(url);
|
|
32158
|
+
if (toolsList.ok) {
|
|
32159
|
+
results.push(...classifyToolContract(toolsList.tools));
|
|
32160
|
+
}
|
|
32046
32161
|
results.push(checkHindsightConsumer(config));
|
|
32047
32162
|
results.push(...checkHindsightContainerHealth());
|
|
32048
32163
|
for (const [agentName, agentConfig] of Object.entries(config.agents)) {
|
|
@@ -49700,8 +49815,8 @@ var {
|
|
|
49700
49815
|
} = import__.default;
|
|
49701
49816
|
|
|
49702
49817
|
// src/build-info.ts
|
|
49703
|
-
var VERSION = "0.14.
|
|
49704
|
-
var COMMIT_SHA = "
|
|
49818
|
+
var VERSION = "0.14.83";
|
|
49819
|
+
var COMMIT_SHA = "057ab099";
|
|
49705
49820
|
|
|
49706
49821
|
// src/cli/agent.ts
|
|
49707
49822
|
init_source();
|
package/dist/cli/ui/index.html
CHANGED
|
@@ -559,16 +559,42 @@
|
|
|
559
559
|
}
|
|
560
560
|
}
|
|
561
561
|
|
|
562
|
+
// Guards a second click from starting a second device-code flow (each
|
|
563
|
+
// start makes Microsoft send a sign-in email → the "2 emails" bug).
|
|
564
|
+
let msConnecting = false;
|
|
565
|
+
|
|
566
|
+
// The set of Microsoft account emails currently known to the broker.
|
|
567
|
+
// Used as a resilience baseline: the connect status lives only in the web
|
|
568
|
+
// process's memory, so if that process restarts mid-connect the status
|
|
569
|
+
// reads 'unknown' even when the token WAS stored. Diffing this list tells
|
|
570
|
+
// us the account really connected regardless of the lost status.
|
|
571
|
+
async function fetchMicrosoftAccountEmails() {
|
|
572
|
+
try {
|
|
573
|
+
const r = await fetch(`${API}/api/microsoft-accounts`, { headers: authHeaders() });
|
|
574
|
+
if (!r.ok) return new Set();
|
|
575
|
+
const list = await r.json();
|
|
576
|
+
return new Set((list || []).filter(a => a.brokerKnown).map(a => String(a.account).toLowerCase()));
|
|
577
|
+
} catch { return new Set(); }
|
|
578
|
+
}
|
|
579
|
+
|
|
562
580
|
// Start an in-browser Microsoft connect: show the device code + link,
|
|
563
581
|
// then poll until the operator completes sign-in on Microsoft's site.
|
|
564
582
|
async function connectMicrosoft() {
|
|
583
|
+
if (msConnecting) return; // double-submit guard
|
|
584
|
+
msConnecting = true;
|
|
585
|
+
const btn = document.getElementById('ms-connect-btn');
|
|
586
|
+
if (btn) btn.disabled = true;
|
|
587
|
+
const done = () => { msConnecting = false; const b = document.getElementById('ms-connect-btn'); if (b) b.disabled = false; };
|
|
565
588
|
const card = document.getElementById('ms-connect-card');
|
|
566
589
|
const show = (html) => { if (card) card.innerHTML = html; };
|
|
567
590
|
show('<div class="loading" style="padding:.8rem">Starting…</div>');
|
|
591
|
+
// Snapshot already-connected accounts BEFORE starting, for the
|
|
592
|
+
// restart-resilient terminal check below.
|
|
593
|
+
const before = await fetchMicrosoftAccountEmails();
|
|
568
594
|
try {
|
|
569
595
|
const res = await fetch(`${API}/api/connections/microsoft/connect`, { method: 'POST', headers: authHeaders() });
|
|
570
596
|
const data = await res.json();
|
|
571
|
-
if (!res.ok || !data.ok) { show(''); showError(data.error || `HTTP ${res.status}`); return; }
|
|
597
|
+
if (!res.ok || !data.ok) { show(''); showError(data.error || `HTTP ${res.status}`); done(); return; }
|
|
572
598
|
const url = data.verificationUri, code = data.userCode;
|
|
573
599
|
show(`<div class="account-card" style="border-color:var(--accent)">
|
|
574
600
|
<div class="account-card-header"><div class="account-label">Connect a Microsoft account</div></div>
|
|
@@ -580,27 +606,58 @@
|
|
|
580
606
|
<div id="ms-connect-status" style="color:var(--text-dim);margin-top:.3rem">Waiting for sign-in… (this card expires in ~15 min)</div>
|
|
581
607
|
</div>`);
|
|
582
608
|
const statusEl = () => document.getElementById('ms-connect-status');
|
|
583
|
-
const
|
|
609
|
+
const deadline = Date.now() + ((data.expiresInSec || 900) * 1000 + 30000);
|
|
610
|
+
const showConnected = (label) => {
|
|
611
|
+
show(`<div class="loading" style="padding:.8rem;color:var(--green)">✓ Connected ${escapeHtml(label)}. Use the access toggles below to grant an agent.</div>`);
|
|
612
|
+
fetchConnections();
|
|
613
|
+
};
|
|
614
|
+
// On any non-'connected' terminal state ('failed' or 'unknown'),
|
|
615
|
+
// re-check the broker's actual account list before declaring failure:
|
|
616
|
+
// a new account appearing means the connect really succeeded (e.g. the
|
|
617
|
+
// status was lost to a web restart). Only error if nothing new landed.
|
|
618
|
+
// Limitations (acceptable — the in-memory 'connected' state is the
|
|
619
|
+
// primary signal; this is only the lost-status fallback): a concurrent
|
|
620
|
+
// connect of a DIFFERENT account in another tab/CLI could be mistaken
|
|
621
|
+
// for this one; and re-connecting an ALREADY-known account (token
|
|
622
|
+
// refresh) shows no new account so falls through to the error.
|
|
623
|
+
const settleNonConnected = async (reason) => {
|
|
624
|
+
const after = await fetchMicrosoftAccountEmails();
|
|
625
|
+
const fresh = [...after].find(a => !before.has(a));
|
|
626
|
+
if (fresh) { showConnected(fresh); } else { show(''); showError(reason || 'connect failed'); }
|
|
627
|
+
done();
|
|
628
|
+
};
|
|
584
629
|
const poll = async () => {
|
|
585
|
-
|
|
586
|
-
|
|
630
|
+
let s;
|
|
631
|
+
try {
|
|
632
|
+
const sres = await fetch(`${API}/api/connections/microsoft/connect/${encodeURIComponent(data.requestId)}`, { headers: authHeaders() });
|
|
633
|
+
s = sres.ok ? await sres.json() : { state: 'failed', reason: `HTTP ${sres.status}` };
|
|
634
|
+
} catch {
|
|
635
|
+
// Transient fetch failure — most likely the web process restarting
|
|
636
|
+
// mid-connect (the exact case this flow must survive). Don't die
|
|
637
|
+
// (that would strand msConnecting=true and lock the button): keep
|
|
638
|
+
// polling until the device-code deadline, then settle via the
|
|
639
|
+
// broker re-check (which recovers a token stored before the restart).
|
|
640
|
+
if (Date.now() > deadline) { await settleNonConnected('connection lost'); return; }
|
|
641
|
+
setTimeout(poll, 3000);
|
|
642
|
+
return;
|
|
643
|
+
}
|
|
587
644
|
if (s.state === 'pending') {
|
|
588
|
-
if (Date.now()
|
|
645
|
+
if (Date.now() > deadline) { const e = statusEl(); if (e) e.textContent = 'Expired — click Connect to try again.'; done(); return; }
|
|
589
646
|
setTimeout(poll, 3000);
|
|
590
647
|
return;
|
|
591
648
|
}
|
|
592
649
|
if (s.state === 'connected') {
|
|
593
|
-
|
|
594
|
-
|
|
650
|
+
showConnected(`${s.account} (${s.accountType})`);
|
|
651
|
+
done();
|
|
595
652
|
} else {
|
|
596
|
-
|
|
597
|
-
showError(s.reason || 'connect failed');
|
|
653
|
+
await settleNonConnected(s.reason);
|
|
598
654
|
}
|
|
599
655
|
};
|
|
600
656
|
setTimeout(poll, 3000);
|
|
601
657
|
} catch (err) {
|
|
602
658
|
show('');
|
|
603
659
|
showError(err.message);
|
|
660
|
+
done();
|
|
604
661
|
}
|
|
605
662
|
}
|
|
606
663
|
|
|
@@ -1165,7 +1222,7 @@
|
|
|
1165
1222
|
<div style="margin-bottom:1.5rem">
|
|
1166
1223
|
<h3 style="margin:0 0 .6rem;font-size:.95rem;color:var(--text-dim);text-transform:uppercase;letter-spacing:.04em">
|
|
1167
1224
|
Microsoft 365
|
|
1168
|
-
<button onclick="connectMicrosoft()" class="usage-pill primary" style="margin-left:.6rem;cursor:pointer;border:none;text-transform:none;font-weight:600">+ Connect a Microsoft account</button>
|
|
1225
|
+
<button id="ms-connect-btn" onclick="connectMicrosoft()" class="usage-pill primary" style="margin-left:.6rem;cursor:pointer;border:none;text-transform:none;font-weight:600">+ Connect a Microsoft account</button>
|
|
1169
1226
|
</h3>
|
|
1170
1227
|
<div id="ms-connect-card"></div>
|
|
1171
1228
|
${msCards
|
package/package.json
CHANGED
|
@@ -52661,6 +52661,53 @@ function evaluateQuotaWatchAccount(args) {
|
|
|
52661
52661
|
}
|
|
52662
52662
|
return { kind: "skip", accountLabel: label, reason: "no-matching-transition" };
|
|
52663
52663
|
}
|
|
52664
|
+
var FLEET_ALL_EXHAUSTED_KEY = "__fleet_all_exhausted__";
|
|
52665
|
+
function evaluateFleetAllExhausted(args) {
|
|
52666
|
+
const { accounts, prev, now } = args;
|
|
52667
|
+
const allExhausted = accounts.length > 0 && accounts.every((a) => a.exhausted);
|
|
52668
|
+
const wasAlerting = prev.lastNotifiedHealth === "throttling";
|
|
52669
|
+
if (allExhausted && !wasAlerting) {
|
|
52670
|
+
return {
|
|
52671
|
+
kind: "notify",
|
|
52672
|
+
message: buildAllExhaustedMessage(accounts, now),
|
|
52673
|
+
newState: { lastNotifiedHealth: "throttling", lastNotifiedAt: now },
|
|
52674
|
+
transition: "entered"
|
|
52675
|
+
};
|
|
52676
|
+
}
|
|
52677
|
+
if (!allExhausted && wasAlerting) {
|
|
52678
|
+
return {
|
|
52679
|
+
kind: "notify",
|
|
52680
|
+
message: buildFleetRecoveredMessage(accounts),
|
|
52681
|
+
newState: { lastNotifiedHealth: "healthy", lastNotifiedAt: now },
|
|
52682
|
+
transition: "recovered"
|
|
52683
|
+
};
|
|
52684
|
+
}
|
|
52685
|
+
return { kind: "skip", reason: allExhausted ? "still-all-exhausted" : "not-all-exhausted" };
|
|
52686
|
+
}
|
|
52687
|
+
function buildAllExhaustedMessage(accounts, now) {
|
|
52688
|
+
const resets = accounts.map((a) => a.exhausted_until).filter((x) => typeof x === "number" && x > now);
|
|
52689
|
+
const earliest = resets.length > 0 ? Math.min(...resets) : null;
|
|
52690
|
+
const resetLine = earliest ? `Earliest reset: ${formatRelative(new Date(earliest), new Date(now))}.` : `Reset time unknown (no window data).`;
|
|
52691
|
+
return [
|
|
52692
|
+
`\uD83D\uDD34 <b>All accounts exhausted</b>`,
|
|
52693
|
+
``,
|
|
52694
|
+
`Every Anthropic account (${accounts.length}) is quota-walled \u2014 there is no healthy account to fail over to.`,
|
|
52695
|
+
resetLine,
|
|
52696
|
+
``,
|
|
52697
|
+
`<i>This is self-healing: agents resume and deferred scheduled jobs run automatically once a window resets. Nothing is lost. Add headroom with <code>/auth add</code> if this recurs.</i>`
|
|
52698
|
+
].join(`
|
|
52699
|
+
`);
|
|
52700
|
+
}
|
|
52701
|
+
function buildFleetRecoveredMessage(accounts) {
|
|
52702
|
+
const healthy = accounts.filter((a) => !a.exhausted).map((a) => a.label);
|
|
52703
|
+
const which = healthy.length > 0 ? ` (<code>${escapeHtml10(healthy[0])}</code>)` : "";
|
|
52704
|
+
return [
|
|
52705
|
+
`\uD83D\uDFE2 <b>Fleet recovered</b> \u2014 at least one account is healthy again${which}.`,
|
|
52706
|
+
``,
|
|
52707
|
+
`<i>Agents are back; any deferred scheduled jobs will run on their next occurrence.</i>`
|
|
52708
|
+
].join(`
|
|
52709
|
+
`);
|
|
52710
|
+
}
|
|
52664
52711
|
function buildThrottlingMessage(agentName3, snap) {
|
|
52665
52712
|
const q = snap.quota;
|
|
52666
52713
|
const fiveStr = fmtPct(q.fiveHourUtilizationPct);
|
|
@@ -52810,11 +52857,11 @@ function sweepStaleTurnActiveMarker(stateDir, opts) {
|
|
|
52810
52857
|
}
|
|
52811
52858
|
|
|
52812
52859
|
// ../src/build-info.ts
|
|
52813
|
-
var VERSION = "0.14.
|
|
52814
|
-
var COMMIT_SHA = "
|
|
52815
|
-
var COMMIT_DATE = "2026-06-07T12:
|
|
52860
|
+
var VERSION = "0.14.83";
|
|
52861
|
+
var COMMIT_SHA = "057ab099";
|
|
52862
|
+
var COMMIT_DATE = "2026-06-07T12:57:08+10:00";
|
|
52816
52863
|
var LATEST_PR = null;
|
|
52817
|
-
var COMMITS_AHEAD_OF_TAG =
|
|
52864
|
+
var COMMITS_AHEAD_OF_TAG = 4;
|
|
52818
52865
|
|
|
52819
52866
|
// gateway/boot-version.ts
|
|
52820
52867
|
function formatRelativeAgo(iso) {
|
|
@@ -61017,6 +61064,31 @@ async function runQuotaWatch() {
|
|
|
61017
61064
|
let watchState = loadQuotaWatchState(stateDir);
|
|
61018
61065
|
const now = Date.now();
|
|
61019
61066
|
const access = loadAccess();
|
|
61067
|
+
{
|
|
61068
|
+
const fleetPrev = watchState[FLEET_ALL_EXHAUSTED_KEY] ?? emptyAccountState();
|
|
61069
|
+
const fleetDecision = evaluateFleetAllExhausted({
|
|
61070
|
+
accounts: listStateData.accounts,
|
|
61071
|
+
prev: fleetPrev,
|
|
61072
|
+
now
|
|
61073
|
+
});
|
|
61074
|
+
if (fleetDecision.kind === "notify") {
|
|
61075
|
+
for (const chat_id of access.allowFrom) {
|
|
61076
|
+
await swallowingApiCall(() => bot.api.sendMessage(chat_id, fleetDecision.message, {
|
|
61077
|
+
parse_mode: "HTML",
|
|
61078
|
+
link_preview_options: { is_disabled: true }
|
|
61079
|
+
}), { chat_id, verb: "quota-watch.fleet-all-exhausted" });
|
|
61080
|
+
}
|
|
61081
|
+
watchState = patchQuotaWatchState(watchState, FLEET_ALL_EXHAUSTED_KEY, fleetDecision.newState);
|
|
61082
|
+
try {
|
|
61083
|
+
saveQuotaWatchState(stateDir, watchState);
|
|
61084
|
+
} catch (err) {
|
|
61085
|
+
process.stderr.write(`telegram gateway: quota-watch: fleet-state save failed: ${err}
|
|
61086
|
+
`);
|
|
61087
|
+
}
|
|
61088
|
+
process.stderr.write(`telegram gateway: quota-watch: fleet all-exhausted ${fleetDecision.transition}
|
|
61089
|
+
`);
|
|
61090
|
+
}
|
|
61091
|
+
}
|
|
61020
61092
|
const pendingTransitions = [];
|
|
61021
61093
|
const labelToSnapIndex = new Map(snapshots.map((s, i) => [s.label, i]));
|
|
61022
61094
|
for (const snap of snapshots) {
|
|
@@ -412,6 +412,8 @@ import {
|
|
|
412
412
|
} from '../credits-watch.js'
|
|
413
413
|
import {
|
|
414
414
|
evaluateQuotaWatchAccount,
|
|
415
|
+
evaluateFleetAllExhausted,
|
|
416
|
+
FLEET_ALL_EXHAUSTED_KEY,
|
|
415
417
|
loadQuotaWatchState,
|
|
416
418
|
saveQuotaWatchState,
|
|
417
419
|
patchQuotaWatchState,
|
|
@@ -14805,6 +14807,44 @@ async function runQuotaWatch(): Promise<void> {
|
|
|
14805
14807
|
const now = Date.now()
|
|
14806
14808
|
const access = loadAccess()
|
|
14807
14809
|
|
|
14810
|
+
// Fleet-wide all-exhausted check FIRST — must run before the per-account
|
|
14811
|
+
// early-return below. When every account is exhausted, the per-account loop
|
|
14812
|
+
// produces only 'blocked' skips → pendingTransitions empty → early return;
|
|
14813
|
+
// so this fleet-level alert (the one the trigger-based all-blocked card
|
|
14814
|
+
// misses during quiet periods / for the consumer+cron paths) would never
|
|
14815
|
+
// fire if placed after. Authoritative source: broker `exhausted` flags.
|
|
14816
|
+
{
|
|
14817
|
+
const fleetPrev = watchState[FLEET_ALL_EXHAUSTED_KEY] ?? emptyAccountState()
|
|
14818
|
+
const fleetDecision = evaluateFleetAllExhausted({
|
|
14819
|
+
accounts: listStateData.accounts,
|
|
14820
|
+
prev: fleetPrev,
|
|
14821
|
+
now,
|
|
14822
|
+
})
|
|
14823
|
+
if (fleetDecision.kind === 'notify') {
|
|
14824
|
+
for (const chat_id of access.allowFrom) {
|
|
14825
|
+
await swallowingApiCall(
|
|
14826
|
+
() =>
|
|
14827
|
+
bot.api.sendMessage(chat_id, fleetDecision.message, {
|
|
14828
|
+
parse_mode: 'HTML',
|
|
14829
|
+
link_preview_options: { is_disabled: true },
|
|
14830
|
+
}),
|
|
14831
|
+
{ chat_id, verb: 'quota-watch.fleet-all-exhausted' },
|
|
14832
|
+
)
|
|
14833
|
+
}
|
|
14834
|
+
// Persist immediately — the per-account early-return path below would
|
|
14835
|
+
// otherwise drop this flag change (edge-trigger would re-fire next poll).
|
|
14836
|
+
watchState = patchQuotaWatchState(watchState, FLEET_ALL_EXHAUSTED_KEY, fleetDecision.newState)
|
|
14837
|
+
try {
|
|
14838
|
+
saveQuotaWatchState(stateDir, watchState)
|
|
14839
|
+
} catch (err) {
|
|
14840
|
+
process.stderr.write(`telegram gateway: quota-watch: fleet-state save failed: ${err}\n`)
|
|
14841
|
+
}
|
|
14842
|
+
process.stderr.write(
|
|
14843
|
+
`telegram gateway: quota-watch: fleet all-exhausted ${fleetDecision.transition}\n`,
|
|
14844
|
+
)
|
|
14845
|
+
}
|
|
14846
|
+
}
|
|
14847
|
+
|
|
14808
14848
|
// First pass: evaluate all accounts against cached state. Collect
|
|
14809
14849
|
// labels that need a live probe (i.e. accounts with a detected transition
|
|
14810
14850
|
// that we're about to notify about). We probe those to get fresh
|
|
@@ -160,6 +160,99 @@ export function evaluateQuotaWatchAccount(args: {
|
|
|
160
160
|
return { kind: "skip", accountLabel: label, reason: "no-matching-transition" };
|
|
161
161
|
}
|
|
162
162
|
|
|
163
|
+
// ─── Fleet-level: all accounts exhausted ───────────────────────────────────────
|
|
164
|
+
|
|
165
|
+
/**
|
|
166
|
+
* Reserved key under which the fleet-wide "all accounts exhausted" alert state
|
|
167
|
+
* is stored in the same quota-watch.json map. Not a valid account label (emails
|
|
168
|
+
* can't contain this), so it never collides with a per-account entry, and the
|
|
169
|
+
* per-account loop (which iterates account snapshots, not state-map keys) never
|
|
170
|
+
* sees it. Encoded as a QuotaWatchAccountState so the existing load validator
|
|
171
|
+
* accepts it: lastNotifiedHealth "throttling" = currently alerting all-exhausted,
|
|
172
|
+
* "healthy"/null = not. Backward-compatible — old files simply lack the key.
|
|
173
|
+
*/
|
|
174
|
+
export const FLEET_ALL_EXHAUSTED_KEY = "__fleet_all_exhausted__";
|
|
175
|
+
|
|
176
|
+
export type FleetAllExhaustedDecision =
|
|
177
|
+
| { kind: "notify"; message: string; newState: QuotaWatchAccountState; transition: "entered" | "recovered" }
|
|
178
|
+
| { kind: "skip"; reason: string };
|
|
179
|
+
|
|
180
|
+
/**
|
|
181
|
+
* Fleet-wide all-exhausted alert (edge-triggered).
|
|
182
|
+
*
|
|
183
|
+
* Fires ONCE when every account enters the broker's exhausted state (no healthy
|
|
184
|
+
* account to fail over to — agents go quiet, crons defer, consumers/hindsight
|
|
185
|
+
* silently serve an exhausted account), and ONCE on recovery. This catches the
|
|
186
|
+
* cases the trigger-based interactive all-blocked card misses: a quiet period
|
|
187
|
+
* (no agent happens to 429 into the wall) and the consumer/cron paths.
|
|
188
|
+
*
|
|
189
|
+
* Authoritative source: the broker's per-account `exhausted` flag (set by
|
|
190
|
+
* mark-exhausted via failover + the consumer sensor), NOT probe-derived health
|
|
191
|
+
* — so there is no probe-failure false-alarm. Requires at least one account;
|
|
192
|
+
* an empty fleet never alerts.
|
|
193
|
+
*/
|
|
194
|
+
export function evaluateFleetAllExhausted(args: {
|
|
195
|
+
accounts: Array<{ label: string; exhausted: boolean; exhausted_until?: number }>;
|
|
196
|
+
prev: QuotaWatchAccountState;
|
|
197
|
+
now: number;
|
|
198
|
+
}): FleetAllExhaustedDecision {
|
|
199
|
+
const { accounts, prev, now } = args;
|
|
200
|
+
const allExhausted = accounts.length > 0 && accounts.every((a) => a.exhausted);
|
|
201
|
+
// "throttling" doubles as the "currently alerting all-exhausted" marker.
|
|
202
|
+
const wasAlerting = prev.lastNotifiedHealth === "throttling";
|
|
203
|
+
|
|
204
|
+
if (allExhausted && !wasAlerting) {
|
|
205
|
+
return {
|
|
206
|
+
kind: "notify",
|
|
207
|
+
message: buildAllExhaustedMessage(accounts, now),
|
|
208
|
+
newState: { lastNotifiedHealth: "throttling", lastNotifiedAt: now },
|
|
209
|
+
transition: "entered",
|
|
210
|
+
};
|
|
211
|
+
}
|
|
212
|
+
if (!allExhausted && wasAlerting) {
|
|
213
|
+
return {
|
|
214
|
+
kind: "notify",
|
|
215
|
+
message: buildFleetRecoveredMessage(accounts),
|
|
216
|
+
newState: { lastNotifiedHealth: "healthy", lastNotifiedAt: now },
|
|
217
|
+
transition: "recovered",
|
|
218
|
+
};
|
|
219
|
+
}
|
|
220
|
+
return { kind: "skip", reason: allExhausted ? "still-all-exhausted" : "not-all-exhausted" };
|
|
221
|
+
}
|
|
222
|
+
|
|
223
|
+
function buildAllExhaustedMessage(
|
|
224
|
+
accounts: Array<{ label: string; exhausted_until?: number }>,
|
|
225
|
+
now: number,
|
|
226
|
+
): string {
|
|
227
|
+
const resets = accounts
|
|
228
|
+
.map((a) => a.exhausted_until)
|
|
229
|
+
.filter((x): x is number => typeof x === "number" && x > now);
|
|
230
|
+
const earliest = resets.length > 0 ? Math.min(...resets) : null;
|
|
231
|
+
const resetLine = earliest
|
|
232
|
+
? `Earliest reset: ${formatRelative(new Date(earliest), new Date(now))}.`
|
|
233
|
+
: `Reset time unknown (no window data).`;
|
|
234
|
+
return [
|
|
235
|
+
`🔴 <b>All accounts exhausted</b>`,
|
|
236
|
+
``,
|
|
237
|
+
`Every Anthropic account (${accounts.length}) is quota-walled — there is no healthy account to fail over to.`,
|
|
238
|
+
resetLine,
|
|
239
|
+
``,
|
|
240
|
+
`<i>This is self-healing: agents resume and deferred scheduled jobs run automatically once a window resets. Nothing is lost. Add headroom with <code>/auth add</code> if this recurs.</i>`,
|
|
241
|
+
].join("\n");
|
|
242
|
+
}
|
|
243
|
+
|
|
244
|
+
function buildFleetRecoveredMessage(
|
|
245
|
+
accounts: Array<{ label: string; exhausted: boolean }>,
|
|
246
|
+
): string {
|
|
247
|
+
const healthy = accounts.filter((a) => !a.exhausted).map((a) => a.label);
|
|
248
|
+
const which = healthy.length > 0 ? ` (<code>${escapeHtml(healthy[0]!)}</code>)` : "";
|
|
249
|
+
return [
|
|
250
|
+
`🟢 <b>Fleet recovered</b> — at least one account is healthy again${which}.`,
|
|
251
|
+
``,
|
|
252
|
+
`<i>Agents are back; any deferred scheduled jobs will run on their next occurrence.</i>`,
|
|
253
|
+
].join("\n");
|
|
254
|
+
}
|
|
255
|
+
|
|
163
256
|
// ─── Message builders ─────────────────────────────────────────────────────────
|
|
164
257
|
|
|
165
258
|
function buildThrottlingMessage(agentName: string, snap: AccountSnapshot): string {
|
|
@@ -12,6 +12,7 @@ import { tmpdir } from "os";
|
|
|
12
12
|
import { join } from "path";
|
|
13
13
|
import {
|
|
14
14
|
evaluateQuotaWatchAccount,
|
|
15
|
+
evaluateFleetAllExhausted,
|
|
15
16
|
loadQuotaWatchState,
|
|
16
17
|
saveQuotaWatchState,
|
|
17
18
|
patchQuotaWatchState,
|
|
@@ -364,3 +365,74 @@ describe("patchQuotaWatchState", () => {
|
|
|
364
365
|
expect(current["bob@example.com"]).toBeUndefined();
|
|
365
366
|
});
|
|
366
367
|
});
|
|
368
|
+
|
|
369
|
+
describe("evaluateFleetAllExhausted", () => {
|
|
370
|
+
const notAlerting = { lastNotifiedHealth: null, lastNotifiedAt: 0 };
|
|
371
|
+
const alerting = { lastNotifiedHealth: "throttling" as const, lastNotifiedAt: 1000 };
|
|
372
|
+
|
|
373
|
+
it("notifies (entered) when every account is exhausted and we weren't alerting", () => {
|
|
374
|
+
const d = evaluateFleetAllExhausted({
|
|
375
|
+
accounts: [
|
|
376
|
+
{ label: "a", exhausted: true, exhausted_until: 5_000 },
|
|
377
|
+
{ label: "b", exhausted: true, exhausted_until: 9_000 },
|
|
378
|
+
],
|
|
379
|
+
prev: notAlerting,
|
|
380
|
+
now: 1_000,
|
|
381
|
+
});
|
|
382
|
+
expect(d.kind).toBe("notify");
|
|
383
|
+
if (d.kind === "notify") {
|
|
384
|
+
expect(d.transition).toBe("entered");
|
|
385
|
+
expect(d.newState.lastNotifiedHealth).toBe("throttling");
|
|
386
|
+
expect(d.message).toContain("All accounts exhausted");
|
|
387
|
+
// earliest reset is the 5_000 one
|
|
388
|
+
expect(d.message).toContain("Earliest reset");
|
|
389
|
+
}
|
|
390
|
+
});
|
|
391
|
+
|
|
392
|
+
it("skips (still) when all exhausted and already alerting — no re-spam", () => {
|
|
393
|
+
const d = evaluateFleetAllExhausted({
|
|
394
|
+
accounts: [{ label: "a", exhausted: true }, { label: "b", exhausted: true }],
|
|
395
|
+
prev: alerting,
|
|
396
|
+
now: 2_000,
|
|
397
|
+
});
|
|
398
|
+
expect(d.kind).toBe("skip");
|
|
399
|
+
});
|
|
400
|
+
|
|
401
|
+
it("notifies (recovered) when one account frees after we were alerting", () => {
|
|
402
|
+
const d = evaluateFleetAllExhausted({
|
|
403
|
+
accounts: [{ label: "a", exhausted: false }, { label: "b", exhausted: true }],
|
|
404
|
+
prev: alerting,
|
|
405
|
+
now: 3_000,
|
|
406
|
+
});
|
|
407
|
+
expect(d.kind).toBe("notify");
|
|
408
|
+
if (d.kind === "notify") {
|
|
409
|
+
expect(d.transition).toBe("recovered");
|
|
410
|
+
expect(d.newState.lastNotifiedHealth).toBe("healthy");
|
|
411
|
+
expect(d.message).toContain("Fleet recovered");
|
|
412
|
+
expect(d.message).toContain("a"); // names the healthy account
|
|
413
|
+
}
|
|
414
|
+
});
|
|
415
|
+
|
|
416
|
+
it("skips (not-all) when some account is healthy and we weren't alerting", () => {
|
|
417
|
+
const d = evaluateFleetAllExhausted({
|
|
418
|
+
accounts: [{ label: "a", exhausted: false }, { label: "b", exhausted: true }],
|
|
419
|
+
prev: notAlerting,
|
|
420
|
+
now: 4_000,
|
|
421
|
+
});
|
|
422
|
+
expect(d.kind).toBe("skip");
|
|
423
|
+
});
|
|
424
|
+
|
|
425
|
+
it("never alerts on an empty fleet", () => {
|
|
426
|
+
expect(evaluateFleetAllExhausted({ accounts: [], prev: notAlerting, now: 1 }).kind).toBe("skip");
|
|
427
|
+
});
|
|
428
|
+
|
|
429
|
+
it("shows reset-unknown when no exhausted_until is present", () => {
|
|
430
|
+
const d = evaluateFleetAllExhausted({
|
|
431
|
+
accounts: [{ label: "a", exhausted: true }],
|
|
432
|
+
prev: notAlerting,
|
|
433
|
+
now: 1_000,
|
|
434
|
+
});
|
|
435
|
+
expect(d.kind).toBe("notify");
|
|
436
|
+
if (d.kind === "notify") expect(d.message).toContain("Reset time unknown");
|
|
437
|
+
});
|
|
438
|
+
});
|