dsh-subscriptions 0.2.0 → 0.2.2

package/lib/providers/ollama.js

@@ -9,7 +9,7 @@ import { attributionHeaders, EMPTY_RESPONSE_CODE, errorChain, LlmAdapter, LlmErr
 import { DISCOVERY_TIMEOUT_MS, effortDisplayName, httpLlmError, idleWatchdog, isDiscoveryAborted, isMissingOrInvalidCredential, mapFetchFailure, ModelCatalogCache, readBoundedJson, withTimeout, } from './common.js';
 import { hostFetch } from '../http.js';
 import { resolveImages } from '../translate/resolved.js';
-import { streamChatCompletions, toChatMessages, toChatTools } from '../translate/chat-completions.js';
+import { streamChatCompletions, toChatMessages, toChatTools, } from '../translate/chat-completions.js';
 /** Route id owned by this adapter. Not a member of the OAuth `PROVIDER_IDS`. */
 export const OLLAMA_CLOUD_ROUTE = 'ollama-cloud';
 /** Default credential reference naming the Ollama Cloud API key. */
@@ -43,11 +43,15 @@ const OLLAMA_REASONING_FAMILIES = [
     // Recorded before the endpoint's level list was read: the thinking values
     // ride as `off`/`low`/`high`/`max` (`off` maps to the wire spelling
     // `none`), and the vendor publishes no `medium` level.
-    { prefix: 'deepseek-v4.1-flash', efforts: ['off', 'low', 'high', 'max'], defaultEffort: 'high' },
+    {
+        prefix: 'deepseek-v4.1-flash',
+        efforts: ['off', 'low', 'high', 'max'],
+        defaultEffort: 'high',
+    },
     { prefix: 'glm-5.3', efforts: ['low', 'high', 'max'], defaultEffort: 'max' },
 ];
 function reasoningFamily(id) {
-    return OLLAMA_REASONING_FAMILIES.find(family => id.startsWith(family.prefix));
+    return OLLAMA_REASONING_FAMILIES.find((family) => id.startsWith(family.prefix));
 }
 /**
  * The window the session plans against: the user's override, capped at what
@@ -93,16 +97,17 @@ export function normalizeOllamaModelId(id, baseURL) {
  * 403 may be a transient scope denial, so it falls back to the stored
  * catalog like any other discovery miss. */
 function isUnauthorized(error) {
-    return error instanceof LlmError && error.code === 'AUTH'
-        && error.failure?.status === 401;
+    return (error instanceof LlmError &&
+        error.code === 'AUTH' &&
+        error.failure?.status === 401);
 }
 /**
  * Find a discovered row for a catalog id, tolerating a listing endpoint that
  * still reports bare names (`glm-5.3` for catalogued `glm-5.3:cloud`).
  */
 function discoveredFor(catalogId, discovered) {
-    return discovered?.find(row => row.id === catalogId)
-        ?? discovered?.find(row => row.id === catalogId.replace(/:cloud$/, ''));
+    return (discovered?.find((row) => row.id === catalogId) ??
+        discovered?.find((row) => row.id === catalogId.replace(/:cloud$/, '')));
 }
 function reasoningFor(row, configuredDefault) {
     const family = reasoningFamily(row.id);
@@ -113,7 +118,10 @@ function reasoningFor(row, configuredDefault) {
         return undefined;
     // Every level here comes from the row or a family that named it; a
     // stored default the row does not list cannot ride the wire.
-    const efforts = row.efforts.map(effort => ({ id: ReasoningEffortId(effort), name: effortDisplayName(effort) }));
+    const efforts = row.efforts.map((effort) => ({
+        id: ReasoningEffortId(effort),
+        name: effortDisplayName(effort),
+    }));
     const storedDefault = row.efforts.includes(row.defaultEffort ?? '')
         ? row.defaultEffort
         : undefined;
@@ -124,10 +132,12 @@ function reasoningFor(row, configuredDefault) {
         familyDefault,
         storedDefault,
     ];
-    const defaultEffort = priority.find(entry => entry !== undefined && row.efforts.includes(entry));
+    const defaultEffort = priority.find((entry) => entry !== undefined && row.efforts.includes(entry));
     return {
         efforts,
-        ...defaultEffort === undefined ? {} : { defaultEffort: ReasoningEffortId(defaultEffort) },
+        ...(defaultEffort === undefined
+            ? {}
+            : { defaultEffort: ReasoningEffortId(defaultEffort) }),
     };
 }
 /**
@@ -143,14 +153,21 @@ function reasoningFor(row, configuredDefault) {
  * @returns the merged block, with a default that is always one of its levels.
  */
 function mergeDiscoveredReasoning(row, discovered, configuredDefault) {
-    const vendorLevels = discovered.efforts.map(entry => String(entry.id));
+    const vendorLevels = discovered.efforts.map((entry) => String(entry.id));
     // A published vendor set is closed: the stored row and the family table
     // must not widen it. When the endpoint names no levels, the stored row
     // supplies them and only a still-empty list falls to the family table.
     const efforts = vendorLevels.length > 0
         ? vendorLevels
-        : [...row.efforts, ...(row.efforts.length > 0 ? [] : reasoningFamily(row.id)?.efforts ?? [])];
-    const vendorDefault = discovered.defaultEffort === undefined ? undefined : String(discovered.defaultEffort);
+        : [
+            ...row.efforts,
+            ...(row.efforts.length > 0
+                ? []
+                : (reasoningFamily(row.id)?.efforts ?? [])),
+        ];
+    const vendorDefault = discovered.defaultEffort === undefined
+        ? undefined
+        : String(discovered.defaultEffort);
     // The user's explicit choice wins only within the levels the model still
     // offers: a stored default outside a retracted vendor set must not ride
     // the wire, so it falls through to the vendor or stored fallback.
@@ -158,28 +175,33 @@ function mergeDiscoveredReasoning(row, discovered, configuredDefault) {
         configuredDefault,
         vendorDefault,
         // The family default is the last resort, not a user preference.
-        ...configuredDefault !== undefined || vendorDefault !== undefined
+        ...(configuredDefault !== undefined || vendorDefault !== undefined
             ? [undefined, reasoningFamily(row.id)?.defaultEffort]
-            : [row.defaultEffort, reasoningFamily(row.id)?.defaultEffort],
+            : [row.defaultEffort, reasoningFamily(row.id)?.defaultEffort]),
     ];
-    const defaultEffort = priority.find(entry => entry !== undefined && efforts.includes(entry));
+    const defaultEffort = priority.find((entry) => entry !== undefined && efforts.includes(entry));
     return {
-        efforts: efforts.map(effort => ({ id: ReasoningEffortId(effort), name: effortDisplayName(effort) })),
-        ...defaultEffort === undefined ? {} : { defaultEffort: ReasoningEffortId(defaultEffort) },
+        efforts: efforts.map((effort) => ({
+            id: ReasoningEffortId(effort),
+            name: effortDisplayName(effort),
+        })),
+        ...(defaultEffort === undefined
+            ? {}
+            : { defaultEffort: ReasoningEffortId(defaultEffort) }),
     };
 }
 /** Strip image blocks for models that cannot accept vision input. Runs before
  * image resolution so a text-only model never requires the attachment store. */
 function withoutImages(messages) {
-    return messages.map(message => ({
+    return messages.map((message) => ({
         ...message,
-        content: message.content.filter(block => block.type !== 'image'),
+        content: message.content.filter((block) => block.type !== 'image'),
     }));
 }
 function discoveryHeaders(apiKey) {
     return {
-        'accept': 'application/json',
-        ...apiKey === undefined ? {} : { 'authorization': `Bearer ${apiKey}` },
+        accept: 'application/json',
+        ...(apiKey === undefined ? {} : { authorization: `Bearer ${apiKey}` }),
         ...attributionHeaders(),
     };
 }
@@ -205,8 +227,10 @@ function extractContextWindow(show) {
     const info = recordOf(show['model_info']);
     if (info !== undefined) {
         for (const [key, value] of Object.entries(info)) {
-            if (key.endsWith('context_length') && typeof value === 'number'
-                && Number.isSafeInteger(value) && value > 0)
+            if (key.endsWith('context_length') &&
+                typeof value === 'number' &&
+                Number.isSafeInteger(value) &&
+                value > 0)
                 return value;
         }
     }
@@ -228,7 +252,11 @@ function extractThinking(show) {
     const values = [];
     if (Array.isArray(rawValues)) {
         for (const entry of rawValues) {
-            const effort = entry === false ? 'off' : typeof entry === 'string' && entry.length > 0 ? entry : undefined;
+            const effort = entry === false
+                ? 'off'
+                : typeof entry === 'string' && entry.length > 0
+                    ? entry
+                    : undefined;
             if (effort !== undefined && !values.includes(effort))
                 values.push(effort);
         }
@@ -236,7 +264,9 @@ function extractThinking(show) {
     const rawDefault = thinking['default'];
     return {
         values,
-        ...typeof rawDefault === 'string' && rawDefault.length > 0 ? { default: rawDefault } : {},
+        ...(typeof rawDefault === 'string' && rawDefault.length > 0
+            ? { default: rawDefault }
+            : {}),
     };
 }
 function discoveredFromShow(id, rawId, show) {
@@ -249,8 +279,12 @@ function discoveredFromShow(id, rawId, show) {
     const vendor = thinking ? extractThinking(show) : undefined;
     // The vendor's level list is the truth about the live model; the family
     // table only answers a model whose thinking capability came without one.
-    const family = thinking && (vendor === undefined || vendor.values.length === 0) ? reasoningFamily(id) : undefined;
-    const efforts = vendor !== undefined && vendor.values.length > 0 ? vendor.values : family?.efforts ?? [];
+    const family = thinking && (vendor === undefined || vendor.values.length === 0)
+        ? reasoningFamily(id)
+        : undefined;
+    const efforts = vendor !== undefined && vendor.values.length > 0
+        ? vendor.values
+        : (family?.efforts ?? []);
     const defaultEffort = vendor?.default ?? family?.defaultEffort;
     const contextWindow = extractContextWindow(show);
     return {
@@ -258,14 +292,21 @@ function discoveredFromShow(id, rawId, show) {
         rawId,
         name: id,
         thinking,
-        ...contextWindow === undefined ? {} : { contextWindow },
+        ...(contextWindow === undefined ? {} : { contextWindow }),
         inputModalities: vision ? ['text', 'image'] : ['text'],
-        ...efforts.length === 0 ? {} : {
-            reasoning: {
-                efforts: efforts.map(effort => ({ id: ReasoningEffortId(effort), name: effortDisplayName(effort) })),
-                ...defaultEffort === undefined ? {} : { defaultEffort: ReasoningEffortId(defaultEffort) },
-            },
-        },
+        ...(efforts.length === 0
+            ? {}
+            : {
+                reasoning: {
+                    efforts: efforts.map((effort) => ({
+                        id: ReasoningEffortId(effort),
+                        name: effortDisplayName(effort),
+                    })),
+                    ...(defaultEffort === undefined
+                        ? {}
+                        : { defaultEffort: ReasoningEffortId(defaultEffort) }),
+                },
+            }),
     };
 }
 /**
@@ -284,7 +325,7 @@ export async function fetchOllamaModels(options) {
         try {
             tagsResponse = await fetchFn(`${base}/tags`, {
                 headers: discoveryHeaders(options.apiKey),
-                ...options.signal === undefined ? {} : { signal: options.signal },
+                ...(options.signal === undefined ? {} : { signal: options.signal }),
             });
             transportError = undefined;
             break;
@@ -307,7 +348,11 @@ export async function fetchOllamaModels(options) {
         const record = recordOf(entry);
         const raw = record === undefined
             ? undefined
-            : typeof record['model'] === 'string' ? record['model'] : typeof record['name'] === 'string' ? record['name'] : undefined;
+            : typeof record['model'] === 'string'
+                ? record['model']
+                : typeof record['name'] === 'string'
+                    ? record['name']
+                    : undefined;
         if (raw === undefined || raw.length === 0)
             continue;
         const wire = normalizeOllamaModelId(raw, options.baseURL);
@@ -330,16 +375,20 @@ export async function fetchOllamaModels(options) {
             try {
                 const showResponse = await fetchFn(`${base}/show`, {
                     method: 'POST',
-                    headers: { 'content-type': 'application/json', ...discoveryHeaders(options.apiKey) },
+                    headers: {
+                        'content-type': 'application/json',
+                        ...discoveryHeaders(options.apiKey),
+                    },
                     body: JSON.stringify({ model: raw }),
-                    ...options.signal === undefined ? {} : { signal: options.signal },
+                    ...(options.signal === undefined ? {} : { signal: options.signal }),
                 });
                 if (!showResponse.ok)
                     throw await httpLlmError(showResponse, 'ollama model details');
                 const show = recordOf(await readBoundedJson(showResponse, 'ollama model details', MAX_DISCOVERY_BYTES));
-                out[index] = show === undefined
-                    ? { id: wire, rawId: raw, name: wire, thinking: false }
-                    : discoveredFromShow(wire, raw, show);
+                out[index] =
+                    show === undefined
+                        ? { id: wire, rawId: raw, name: wire, thinking: false }
+                        : discoveredFromShow(wire, raw, show);
             }
             catch {
                 // One model's enrichment never sinks the listing.
@@ -354,17 +403,21 @@ export function ollamaChatRequestBody(options, messages, maxTokens) {
     return {
         model: options.model,
         messages,
-        ...options.tools !== undefined && options.tools.length > 0
+        ...(options.tools !== undefined && options.tools.length > 0
             ? { tools: toChatTools(options.tools), tool_choice: 'auto' }
-            : {},
+            : {}),
         // Ollama takes `max_tokens`. The catalog row supplies the default
         // (32,768, matching upstream); a row without one rides without it.
-        ...maxTokens !== undefined ? { max_tokens: maxTokens } : {},
+        ...(maxTokens !== undefined ? { max_tokens: maxTokens } : {}),
         // `off` rides as `none` (upstream parity); any other effort rides
         // verbatim. Only advertised efforts ever reach this branch.
-        ...options.reasoningEffort !== undefined
-            ? { reasoning_effort: String(options.reasoningEffort) === 'off' ? 'none' : String(options.reasoningEffort) }
-            : {},
+        ...(options.reasoningEffort !== undefined
+            ? {
+                reasoning_effort: String(options.reasoningEffort) === 'off'
+                    ? 'none'
+                    : String(options.reasoningEffort),
+            }
+            : {}),
         stream: true,
         stream_options: { include_usage: true },
     };
@@ -377,9 +430,11 @@ export class OllamaAdapter extends LlmAdapter {
         this.options = options;
     }
     /** Drop the discovery snapshot so the next picker pass re-reads the endpoint. */
-    invalidateModels() { this.catalogCache.invalidate(); }
+    invalidateModels() {
+        this.catalogCache.invalidate();
+    }
     rows() {
-        return this.options.catalog().map(entry => {
+        return this.options.catalog().map((entry) => {
             const family = reasoningFamily(entry.id);
             return {
                 id: entry.id,
@@ -390,7 +445,7 @@ export class OllamaAdapter extends LlmAdapter {
                 vision: entry.vision === true,
                 efforts: entry.efforts !== undefined && entry.efforts.length > 0
                     ? [...entry.efforts]
-                    : [...family?.efforts ?? []],
+                    : [...(family?.efforts ?? [])],
                 // The family default stays out of the stored row: the resolver's
                 // priority keeps it last, so a vendor change is not held back by
                 // the fallback baked in during construction.
@@ -416,7 +471,7 @@ export class OllamaAdapter extends LlmAdapter {
         if (this.options.discovery === false)
             return this.staticList(provider, catalog);
         try {
-            const discovered = await withTimeout(signal => this.catalogCache.get(() => fetchOllamaModels({
+            const discovered = await withTimeout((signal) => this.catalogCache.get(() => fetchOllamaModels({
                 baseURL: this.options.baseURL(),
                 apiKey,
                 fetchFn: this.options.fetchFn,
@@ -429,7 +484,7 @@ export class OllamaAdapter extends LlmAdapter {
             // Discovery enriches the stored allowlist but never extends it: the
             // user adopted an exact catalog, so unmatched discoveries stay out of
             // the picker.
-            return catalog.map(entry => {
+            return catalog.map((entry) => {
                 const match = discoveredFor(entry.id, discovered);
                 if (match === undefined) {
                     return {
@@ -446,7 +501,8 @@ export class OllamaAdapter extends LlmAdapter {
                     provider,
                     id: entry.id,
                     name: entry.name,
-                    inputModalities: match.inputModalities ?? (entry.vision ? ['text', 'image'] : ['text']),
+                    inputModalities: match.inputModalities ??
+                        (entry.vision ? ['text', 'image'] : ['text']),
                 };
             });
         }
@@ -462,7 +518,7 @@ export class OllamaAdapter extends LlmAdapter {
         }
     }
     staticList(provider, catalog) {
-        return catalog.map(entry => ({
+        return catalog.map((entry) => ({
             provider,
             id: entry.id,
             name: entry.name,
@@ -470,7 +526,7 @@ export class OllamaAdapter extends LlmAdapter {
         }));
     }
     async resolveModel(provider, model) {
-        const entry = this.rows().find(row => row.id === model);
+        const entry = this.rows().find((row) => row.id === model);
         // The pass-through native behavior is deliberately absent: an
         // uncatalogued id fails here instead of 404ing at the gateway.
         if (entry === undefined)
@@ -481,13 +537,14 @@ export class OllamaAdapter extends LlmAdapter {
         const discoveryKey = this.options.discovery === false ? undefined : await this.options.apiKey();
         const discovered = this.options.discovery === false
             ? undefined
-            : await withTimeout(signal => this.catalogCache.resolve(() => fetchOllamaModels({
+            : await withTimeout((signal) => this.catalogCache.resolve(() => fetchOllamaModels({
                 baseURL: this.options.baseURL(),
                 apiKey: discoveryKey,
                 fetchFn: this.options.fetchFn,
                 signal,
-            })), DISCOVERY_TIMEOUT_MS).then(models => discoveredFor(model, models ?? undefined), () => undefined);
-        const modalities = discovered?.inputModalities ?? (entry.vision ? ['text', 'image'] : ['text']);
+            })), DISCOVERY_TIMEOUT_MS).then((models) => discoveredFor(model, models ?? undefined), () => undefined);
+        const modalities = discovered?.inputModalities ??
+            (entry.vision ? ['text', 'image'] : ['text']);
         const configuredDefault = this.options.defaultEffortOf?.(model) ?? entry.defaultEffort;
         const reasoning = discovered?.reasoning === undefined
             ? reasoningFor(entry, configuredDefault)
@@ -497,9 +554,11 @@ export class OllamaAdapter extends LlmAdapter {
             id: model,
             name: entry.name,
             inputModalities: [...modalities],
-            context: { contextWindow: effectiveContextWindow(entry, discovered?.contextWindow) },
+            context: {
+                contextWindow: effectiveContextWindow(entry, discovered?.contextWindow),
+            },
             defaultMaxTokens: entry.maxTokens,
-            ...reasoning === undefined ? {} : { reasoning },
+            ...(reasoning === undefined ? {} : { reasoning }),
         };
     }
     async *stream(options) {
@@ -508,10 +567,12 @@ export class OllamaAdapter extends LlmAdapter {
             const apiKey = await this.options.apiKey();
             if (apiKey === undefined)
                 throw new LlmError('ollama-cloud API key is not configured', 'MISSING_CREDENTIAL');
-            const entry = this.rows().find(row => row.id === options.model);
+            const entry = this.rows().find((row) => row.id === options.model);
             if (entry === undefined)
                 throw new LlmError(`ollama-cloud has no configured model "${options.model}"`, 'UNKNOWN_MODEL');
-            const requestMessages = entry.vision ? options.messages : withoutImages(options.messages);
+            const requestMessages = entry.vision
+                ? options.messages
+                : withoutImages(options.messages);
             const resolved = await resolveImages(requestMessages, this.options.resolveAttachments?.(), watchdog.signal);
             const body = ollamaChatRequestBody(options, toChatMessages(resolved, options.system), options.maxTokens ?? entry.maxTokens);
             const chatBase = ollamaChatBaseURL(this.options.baseURL());
@@ -519,8 +580,8 @@ export class OllamaAdapter extends LlmAdapter {
             const response = await fetchFn(`${chatBase}/chat/completions`, {
                 method: 'POST',
                 headers: {
-                    'authorization': `Bearer ${apiKey}`,
-                    'accept': 'text/event-stream',
+                    authorization: `Bearer ${apiKey}`,
+                    accept: 'text/event-stream',
                     'content-type': 'application/json',
                     ...attributionHeaders(),
                 },
@@ -532,7 +593,9 @@ export class OllamaAdapter extends LlmAdapter {
             if (response.body === null) {
                 throw new LlmError('ollama-cloud API returned no response body', EMPTY_RESPONSE_CODE);
             }
-            const pulse = () => { watchdog.pulse(); };
+            const pulse = () => {
+                watchdog.pulse();
+            };
             yield* streamChatCompletions(response.body, pulse);
         }
         catch (error) {

package/lib/providers/pool-family.js

@@ -30,14 +30,22 @@ export function buildAccountPools(sources) {
                     entry = { members: [], info: model };
                     byModel.set(model.id, entry);
                 }
-                entry.members.push({ provider, account: catalog.account, model: model.id });
+                entry.members.push({
+                    provider,
+                    account: catalog.account,
+                    model: model.id,
+                });
             }
         }
         for (const [id, { members, info }] of byModel) {
             pools.set(poolKey(provider, id), {
                 members,
-                ...info.name === undefined || info.name === id ? {} : { name: info.name },
-                ...info.description === undefined ? {} : { description: info.description },
+                ...(info.name === undefined || info.name === id
+                    ? {}
+                    : { name: info.name }),
+                ...(info.description === undefined
+                    ? {}
+                    : { description: info.description }),
             });
         }
     }

package/lib/providers/pool-health.js

@@ -9,7 +9,7 @@
  * cooldown (transport blips say nothing about the account), or rethrow
  * (the request itself is at fault and another account would fail alike).
  */
-import { CONTEXT_WINDOW_EXCEEDED_CODE, LlmError, QUOTA_EXCEEDED_CODE } from '@deepseek-ai/dsh-llm';
+import { CONTEXT_WINDOW_EXCEEDED_CODE, LlmError, QUOTA_EXCEEDED_CODE, } from '@deepseek-ai/dsh-llm';
 /** Registry key for one pool member. */
 export function memberKey(provider, account, model) {
     return `${provider}/${account}/${model}`;
@@ -33,7 +33,10 @@ export const TRANSIENT_COOLDOWN_MS = 60_000;
  * wall means its siblings on the SAME account would too, so the cooldown
  * parks the account (other accounts of the provider are unaffected).
  */
-const MODEL_SCOPED_QUOTA_PROVIDERS = new Set(['claude', 'antigravity']);
+const MODEL_SCOPED_QUOTA_PROVIDERS = new Set([
+    'claude',
+    'antigravity',
+]);
 /** The `retry-after` an adapter propagated through `httpLlmError`, when any. */
 function retryAfterMs(error) {
     return error.failure.providerRetryAfterMs;
@@ -61,16 +64,28 @@ export function classifyPoolFailure(error, provider) {
                 action: 'switch',
                 cooldownMs: retryAfterMs(error) ?? DEFAULT_QUOTA_COOLDOWN_MS,
                 reason: error.code,
-                scope: MODEL_SCOPED_QUOTA_PROVIDERS.has(provider) ? 'member' : 'account',
+                scope: MODEL_SCOPED_QUOTA_PROVIDERS.has(provider)
+                    ? 'member'
+                    : 'account',
             };
         case 'AUTH':
         case 'INVALID_CREDENTIAL':
         case 'MISSING_CREDENTIAL':
-            return { action: 'switch', cooldownMs: AUTH_COOLDOWN_MS, reason: error.code, scope: 'account' };
+            return {
+                action: 'switch',
+                cooldownMs: AUTH_COOLDOWN_MS,
+                reason: error.code,
+                scope: 'account',
+            };
         case 'SERVER':
         case 'TIMEOUT':
         case 'EMPTY_RESPONSE':
-            return { action: 'switch', cooldownMs: TRANSIENT_COOLDOWN_MS, reason: error.code, scope: 'member' };
+            return {
+                action: 'switch',
+                cooldownMs: TRANSIENT_COOLDOWN_MS,
+                reason: error.code,
+                scope: 'member',
+            };
         case 'TRANSPORT':
             return { action: 'switch' };
         case 'HTTP_402':
@@ -78,7 +93,12 @@ export function classifyPoolFailure(error, provider) {
             // Plan/model availability is account-shaped: another account of the
             // same subscription may still serve. A 400 that is not a context-window
             // error stays HTTP_400 and throws (the request itself is at fault).
-            return { action: 'switch', cooldownMs: TRANSIENT_COOLDOWN_MS, reason: error.code, scope: 'member' };
+            return {
+                action: 'switch',
+                cooldownMs: TRANSIENT_COOLDOWN_MS,
+                reason: error.code,
+                scope: 'member',
+            };
         case CONTEXT_WINDOW_EXCEEDED_CODE:
         case 'ABORTED':
         default:
@@ -94,8 +114,8 @@ export class PoolHealthRegistry {
     records = new Map();
     /** Whether a member may serve: neither it nor its whole account is cooling. */
     isMemberAvailable(provider, account, model, now = Date.now()) {
-        return this.isAvailable(accountKey(provider, account), now)
-            && this.isAvailable(memberKey(provider, account, model), now);
+        return (this.isAvailable(accountKey(provider, account), now) &&
+            this.isAvailable(memberKey(provider, account, model), now));
     }
     /** Whether one registry key is clear right now. */
     isAvailable(key, now = Date.now()) {

package/lib/providers/pool-usage.js

@@ -102,7 +102,8 @@ export class PoolUsageTracker {
             return { supported: false };
         const key = `${provider}/${account}`;
         const entry = this.entries.get(key);
-        if (entry !== undefined && Date.now() - entry.at < (entry.cooldownMs ?? this.ttlMs)) {
+        if (entry !== undefined &&
+            Date.now() - entry.at < (entry.cooldownMs ?? this.ttlMs)) {
             if (entry.snapshot !== undefined) {
                 if (!force)
                     return entry.snapshot;
@@ -178,29 +179,33 @@ export class PoolUsageTracker {
             // fetch fails late and would otherwise cache a cooldown over a fresher
             // success that already landed.
             const sequence = ++this.nextSequence;
-            pending = fetcher().then((value) => {
+            pending = fetcher()
+                .then((value) => {
                 const at = Date.now();
                 const snapshot = { ...value, observedAt: at, stale: false };
                 // An invalidate() during the fetch means this result describes a world
                 // that has moved on; caching it would resurrect the old numbers.
-                if ((this.generation.get(key) ?? 0) === generation && (this.lastWritten.get(key) ?? 0) < sequence) {
+                if ((this.generation.get(key) ?? 0) === generation &&
+                    (this.lastWritten.get(key) ?? 0) < sequence) {
                     this.lastWritten.set(key, sequence);
                     this.entries.set(key, { snapshot, at });
                 }
                 return snapshot;
             }, (error) => {
-                if (!isMissingOrInvalidCredential(error) && (this.generation.get(key) ?? 0) === generation
-                    && (this.lastWritten.get(key) ?? 0) < sequence) {
+                if (!isMissingOrInvalidCredential(error) &&
+                    (this.generation.get(key) ?? 0) === generation &&
+                    (this.lastWritten.get(key) ?? 0) < sequence) {
                     this.lastWritten.set(key, sequence);
                     this.entries.set(key, {
                         error,
                         at: Date.now(),
                         cooldownMs: cooldownFor(error, this.ttlMs),
-                        ...lastSnapshot === undefined ? {} : { lastSnapshot },
+                        ...(lastSnapshot === undefined ? {} : { lastSnapshot }),
                     });
                 }
                 throw error;
-            }).finally(() => {
+            })
+                .finally(() => {
                 // Only clear our own slot: an invalidate may have replaced it.
                 if (this.inflight.get(key) === pending)
                     this.inflight.delete(key);
@@ -211,7 +216,7 @@ export class PoolUsageTracker {
     }
     /** Score one member against a snapshot's windows. */
     score(member, entry) {
-        const windows = (entry.snapshot.windows ?? []).filter(window => windowApplies(window, member.model));
+        const windows = (entry.snapshot.windows ?? []).filter((window) => windowApplies(window, member.model));
         let available = true;
         let urgency = 0;
         for (const window of windows) {
@@ -235,7 +240,9 @@ function degradedQuota(error) {
 }
 /** How long to hold a failure in the negative cache: the endpoint's own `retry-after`, or the default TTL. */
 function cooldownFor(error, defaultTtlMs) {
-    return error instanceof OAuthEndpointError && error.retryAfterMs !== undefined ? error.retryAfterMs : defaultTtlMs;
+    return error instanceof OAuthEndpointError && error.retryAfterMs !== undefined
+        ? error.retryAfterMs
+        : defaultTtlMs;
 }
 /**
  * Whether a window constrains this model: unscoped windows always do; a

package/lib/providers/pool.d.ts

@@ -105,6 +105,14 @@ export declare class PoolAdapter extends LlmAdapter {
     private select;
     /** Pin the serving member to the session (with bounded memory). */
     private remember;
+    /**
+     * Build the terminal exhaustion error and log it. Per-member failovers are
+     * already warned above; without this line dsh.log shows the failover chain
+     * but not the exhaustion that ends it. The message carries only the
+     * LlmError's own sanitized text and code — raw provider payloads never
+     * reach here (see the note in {@link exhausted}).
+     */
+    private exhaustedWithWarn;
     /** Report attempt failures or this pool's active cooldowns, never unrelated health records. */
     private exhausted;
 }