praxis-agent 0.62.3 → 0.62.6

package/README.md

@@ -92,7 +92,7 @@ For Anthropic Messages:
 ```sh
 export PRAXIS_PROVIDER="anthropic"
 export PRAXIS_API_KEY="your-api-key"
-export PRAXIS_MODEL="claude-sonnet-4-20250514"
+export PRAXIS_MODEL="claude-sonnet-4-6"
 
 cd /path/to/project
 praxis
@@ -100,7 +100,7 @@ praxis
 
 Anthropic models use a 200,000-token context window by default, including
 unknown model IDs. Add the exact terminal `[1m]` suffix (for example,
-`claude-sonnet-4-20250514[1m]`) to request a 1,000,000-token context window;
+`claude-sonnet-4-6[1m]`) to request a 1,000,000-token context window;
 Praxis keeps that selected model public, removes the suffix on the wire, and
 adds the `context-1m-2025-08-07` Anthropic beta once. An explicit
 `PRAXIS_CONTEXT_WINDOW_TOKENS` value overrides either inferred window.

package/dist/application/background-agent-manager.js

@@ -77,6 +77,7 @@ const modelUsageCounterFields = [
     'outputTokens',
     'cacheReadInputTokens',
     'cacheCreationInputTokens',
+    'cacheCreationInputTokens1h',
     'webSearchRequests',
 ];
 const modelUsageMetadataFields = ['contextWindow', 'maxOutputTokens'];
@@ -90,6 +91,10 @@ function assertValidModelUsageEntry(model, usage) {
             throw new Error(`Model usage for "${model}" has an invalid ${field} counter`);
         }
     }
+    if ((usage.cacheCreationInputTokens1h ?? 0) >
+        (usage.cacheCreationInputTokens ?? 0)) {
+        throw new Error(`Model usage for "${model}" has an invalid cacheCreationInputTokens1h counter`);
+    }
     for (const field of modelUsageMetadataFields) {
         const value = usage[field];
         if (value !== undefined && (!Number.isSafeInteger(value) || value < 1)) {
@@ -120,14 +125,20 @@ function addUsageChecked(model, left, right) {
     const outputTokens = left.outputTokens + right.outputTokens;
     const cacheReadInputTokens = (left.cacheReadInputTokens ?? 0) + (right.cacheReadInputTokens ?? 0);
     const cacheCreationInputTokens = (left.cacheCreationInputTokens ?? 0) + (right.cacheCreationInputTokens ?? 0);
+    const cacheCreationInputTokens1h = (left.cacheCreationInputTokens1h ?? 0) +
+        (right.cacheCreationInputTokens1h ?? 0);
     const webSearchRequests = (left.webSearchRequests ?? 0) + (right.webSearchRequests ?? 0);
     if (!Number.isSafeInteger(inputTokens) ||
         !Number.isSafeInteger(outputTokens) ||
         !Number.isSafeInteger(cacheReadInputTokens) ||
         !Number.isSafeInteger(cacheCreationInputTokens) ||
+        !Number.isSafeInteger(cacheCreationInputTokens1h) ||
         !Number.isSafeInteger(webSearchRequests)) {
         throw new Error('Model usage total overflow');
     }
+    if (cacheCreationInputTokens1h > cacheCreationInputTokens) {
+        throw new Error(`Model usage${model === undefined ? '' : ` for "${model}"`} has an invalid cacheCreationInputTokens1h counter`);
+    }
     // Aggregates without a model stay counter-only; per-model rows merge their
     // capability metadata with conflict rejection.
     const metadata = model === undefined ? {} : mergeModelUsageMetadata(model, left, right);
@@ -136,6 +147,7 @@ function addUsageChecked(model, left, right) {
         outputTokens,
         ...(cacheReadInputTokens === 0 ? {} : { cacheReadInputTokens }),
         ...(cacheCreationInputTokens === 0 ? {} : { cacheCreationInputTokens }),
+        ...(cacheCreationInputTokens1h === 0 ? {} : { cacheCreationInputTokens1h }),
         ...(webSearchRequests === 0 ? {} : { webSearchRequests }),
         ...metadata,
     };
@@ -147,6 +159,10 @@ function assertValidResultUsage(usage) {
             throw new Error(`Model usage total has an invalid ${field} counter`);
         }
     }
+    if ((usage.cacheCreationInputTokens1h ?? 0) >
+        (usage.cacheCreationInputTokens ?? 0)) {
+        throw new Error('Model usage total has an invalid cacheCreationInputTokens1h counter');
+    }
 }
 function addApiDuration(value, total, field) {
     if (!Number.isFinite(value) || value < 0) {

package/dist/application/session-service.js

@@ -14,7 +14,7 @@ import { projectNativeSessionEntries } from './native-session-projection.js';
 import { classifyClaudeInterruption, } from '../native/interruption.js';
 import { findUnresolvedClaudeToolCalls, getClaudeContentBlocks, } from '../native/tool-links.js';
 import { createClaudeAgentSettingEntry, createClaudeHookAttachmentEntries, createClaudeRuleAttachmentEntry, translateProviderEvents, } from '../native/translation.js';
-import { AgentRunCancelledError, AgentRuntime, MALFORMED_TOOL_INPUT_MESSAGE, ModelProviderError, } from '../core/runtime.js';
+import { AgentBudgetExceededError, AgentRunCancelledError, AgentRuntime, MALFORMED_TOOL_INPUT_MESSAGE, ModelProviderError, } from '../core/runtime.js';
 import { resolveToolSchedulingPolicy } from '../core/tool-scheduling-policy.js';
 import { BackgroundTaskRuntime, } from './background-task-runtime.js';
 import { backgroundAgentNotificationMarkers, } from './background-agent-manager.js';
@@ -115,14 +115,30 @@ const emptyToolRegistry = {
     execute: async () => ({ content: '', isError: false }),
 };
 function mergeUsage(left, right) {
+    for (const usage of [left, right]) {
+        const oneHour = usage.cacheCreationInputTokens1h;
+        if (oneHour !== undefined &&
+            (!Number.isSafeInteger(oneHour) ||
+                oneHour < 0 ||
+                oneHour > (usage.cacheCreationInputTokens ?? 0))) {
+            throw new Error('Model usage has an invalid cacheCreationInputTokens1h counter');
+        }
+    }
     const cacheReadInputTokens = (left.cacheReadInputTokens ?? 0) + (right.cacheReadInputTokens ?? 0);
     const cacheCreationInputTokens = (left.cacheCreationInputTokens ?? 0) + (right.cacheCreationInputTokens ?? 0);
+    const cacheCreationInputTokens1h = (left.cacheCreationInputTokens1h ?? 0) +
+        (right.cacheCreationInputTokens1h ?? 0);
     const webSearchRequests = (left.webSearchRequests ?? 0) + (right.webSearchRequests ?? 0);
+    if (!Number.isSafeInteger(cacheCreationInputTokens1h) ||
+        cacheCreationInputTokens1h > cacheCreationInputTokens) {
+        throw new Error('Model usage has an invalid cacheCreationInputTokens1h counter');
+    }
     return {
         inputTokens: left.inputTokens + right.inputTokens,
         outputTokens: left.outputTokens + right.outputTokens,
         ...(cacheReadInputTokens === 0 ? {} : { cacheReadInputTokens }),
         ...(cacheCreationInputTokens === 0 ? {} : { cacheCreationInputTokens }),
+        ...(cacheCreationInputTokens1h === 0 ? {} : { cacheCreationInputTokens1h }),
         ...(webSearchRequests === 0 ? {} : { webSearchRequests }),
     };
 }
@@ -131,6 +147,7 @@ const sessionUsageCounterFields = [
     'outputTokens',
     'cacheReadInputTokens',
     'cacheCreationInputTokens',
+    'cacheCreationInputTokens1h',
     'webSearchRequests',
 ];
 const sessionUsageMetadataFields = ['contextWindow', 'maxOutputTokens'];
@@ -144,6 +161,10 @@ function assertValidSessionUsageEntry(model, usage) {
             throw new Error(`Model usage for "${model}" has an invalid ${field} counter`);
         }
     }
+    if ((usage.cacheCreationInputTokens1h ?? 0) >
+        (usage.cacheCreationInputTokens ?? 0)) {
+        throw new Error(`Model usage for "${model}" has an invalid cacheCreationInputTokens1h counter`);
+    }
     for (const field of sessionUsageMetadataFields) {
         const value = usage[field];
         if (value !== undefined && (!Number.isSafeInteger(value) || value < 1)) {
@@ -174,20 +195,27 @@ function addSessionUsageChecked(model, left, right) {
     const outputTokens = left.outputTokens + right.outputTokens;
     const cacheReadInputTokens = (left.cacheReadInputTokens ?? 0) + (right.cacheReadInputTokens ?? 0);
     const cacheCreationInputTokens = (left.cacheCreationInputTokens ?? 0) + (right.cacheCreationInputTokens ?? 0);
+    const cacheCreationInputTokens1h = (left.cacheCreationInputTokens1h ?? 0) +
+        (right.cacheCreationInputTokens1h ?? 0);
     const webSearchRequests = (left.webSearchRequests ?? 0) + (right.webSearchRequests ?? 0);
     if (!Number.isSafeInteger(inputTokens) ||
         !Number.isSafeInteger(outputTokens) ||
         !Number.isSafeInteger(cacheReadInputTokens) ||
         !Number.isSafeInteger(cacheCreationInputTokens) ||
+        !Number.isSafeInteger(cacheCreationInputTokens1h) ||
         !Number.isSafeInteger(webSearchRequests)) {
         throw new Error('Model usage total overflow');
     }
+    if (cacheCreationInputTokens1h > cacheCreationInputTokens) {
+        throw new Error(`Model usage for "${model}" has an invalid cacheCreationInputTokens1h counter`);
+    }
     const metadata = mergeSessionUsageMetadata(model, left, right);
     return {
         inputTokens,
         outputTokens,
         ...(cacheReadInputTokens === 0 ? {} : { cacheReadInputTokens }),
         ...(cacheCreationInputTokens === 0 ? {} : { cacheCreationInputTokens }),
+        ...(cacheCreationInputTokens1h === 0 ? {} : { cacheCreationInputTokens1h }),
         ...(webSearchRequests === 0 ? {} : { webSearchRequests }),
         ...metadata,
     };
@@ -228,6 +256,7 @@ function hasNonZeroUsage(usage) {
         usage.outputTokens > 0 ||
         (usage.cacheReadInputTokens ?? 0) > 0 ||
         (usage.cacheCreationInputTokens ?? 0) > 0 ||
+        (usage.cacheCreationInputTokens1h ?? 0) > 0 ||
         (usage.webSearchRequests ?? 0) > 0);
 }
 function requireUsageCounter(value, field) {
@@ -244,6 +273,12 @@ function requireManualCompactUsage(usage) {
     if (usage.cacheCreationInputTokens !== undefined) {
         requireUsageCounter(usage.cacheCreationInputTokens, 'usage.cacheCreationInputTokens');
     }
+    if (usage.cacheCreationInputTokens1h !== undefined) {
+        requireUsageCounter(usage.cacheCreationInputTokens1h, 'usage.cacheCreationInputTokens1h');
+        if (usage.cacheCreationInputTokens1h > (usage.cacheCreationInputTokens ?? 0)) {
+            throw new TypeError('usage.cacheCreationInputTokens1h must not exceed usage.cacheCreationInputTokens');
+        }
+    }
     if (usage.webSearchRequests !== undefined) {
         requireUsageCounter(usage.webSearchRequests, 'usage.webSearchRequests');
     }
@@ -3935,9 +3970,23 @@ export class ClaudeSessionService {
                         permissionUpdates: this.sessionPermissionUpdates.get(sessionId) ?? [],
                         onPermissionUpdates: (updates) => this.applyPermissionUpdates(sessionId, updates),
                     };
-                    const attemptMainTurn = () => signal
-                        ? runtime.run({ ...runtimeRequest, signal })
-                        : runtime.run(runtimeRequest);
+                    const attemptMainTurn = async () => {
+                        try {
+                            return signal
+                                ? await runtime.run({ ...runtimeRequest, signal })
+                                : await runtime.run(runtimeRequest);
+                        }
+                        catch (error) {
+                            if (error instanceof AgentBudgetExceededError) {
+                                turnAccounting.complete({
+                                    kind: 'runtime',
+                                    recovery: recoveryResults,
+                                    result: error.result,
+                                });
+                            }
+                            throw error;
+                        }
+                    };
                     const surfaceExhaustedRecovery = (error) => {
                         this.options.eventSink?.({
                             type: 'failed',
@@ -3975,6 +4024,9 @@ export class ClaudeSessionService {
                                     retryError instanceof AgentRunCancelledError) {
                                     throw new AgentRunCancelledError();
                                 }
+                                if (retryError instanceof AgentBudgetExceededError) {
+                                    throw retryError;
+                                }
                                 surfaceExhaustedRecovery(error);
                                 throw error;
                             }

package/dist/application/subagent-service.js

@@ -33,6 +33,7 @@ function mergeSubagentUsage(left, right) {
             'outputTokens',
             'cacheReadInputTokens',
             'cacheCreationInputTokens',
+            'cacheCreationInputTokens1h',
             'webSearchRequests',
         ]) {
             const value = usage[field];
@@ -46,6 +47,10 @@ function mergeSubagentUsage(left, right) {
                 throw new Error(`Subagent usage has an invalid ${field}`);
             }
         }
+        if ((usage.cacheCreationInputTokens1h ?? 0) >
+            (usage.cacheCreationInputTokens ?? 0)) {
+            throw new Error('Subagent usage has an invalid cacheCreationInputTokens1h counter');
+        }
     }
     const counters = {
         inputTokens: left.inputTokens + right.inputTokens,
@@ -53,11 +58,16 @@ function mergeSubagentUsage(left, right) {
         cacheReadInputTokens: (left.cacheReadInputTokens ?? 0) + (right.cacheReadInputTokens ?? 0),
         cacheCreationInputTokens: (left.cacheCreationInputTokens ?? 0) +
             (right.cacheCreationInputTokens ?? 0),
+        cacheCreationInputTokens1h: (left.cacheCreationInputTokens1h ?? 0) +
+            (right.cacheCreationInputTokens1h ?? 0),
         webSearchRequests: (left.webSearchRequests ?? 0) + (right.webSearchRequests ?? 0),
     };
     if (Object.values(counters).some((value) => !Number.isSafeInteger(value))) {
         throw new Error('Subagent usage total overflow');
     }
+    if (counters.cacheCreationInputTokens1h > counters.cacheCreationInputTokens) {
+        throw new Error('Subagent usage has an invalid cacheCreationInputTokens1h counter');
+    }
     const metadata = (field) => {
         const leftValue = left[field];
         const rightValue = right[field];
@@ -79,6 +89,11 @@ function mergeSubagentUsage(left, right) {
         ...(counters.cacheCreationInputTokens === 0
             ? {}
             : { cacheCreationInputTokens: counters.cacheCreationInputTokens }),
+        ...(counters.cacheCreationInputTokens1h === 0
+            ? {}
+            : {
+                cacheCreationInputTokens1h: counters.cacheCreationInputTokens1h,
+            }),
         ...(counters.webSearchRequests === 0
             ? {}
             : { webSearchRequests: counters.webSearchRequests }),

package/dist/application/turn-accounting.js

@@ -5,6 +5,7 @@ const counterFields = [
     'outputTokens',
     'cacheReadInputTokens',
     'cacheCreationInputTokens',
+    'cacheCreationInputTokens1h',
     'webSearchRequests',
 ];
 const metadataFields = ['contextWindow', 'maxOutputTokens'];
@@ -25,19 +26,26 @@ function mergeUsage(left, right) {
     }
     const cacheReadInputTokens = (left.cacheReadInputTokens ?? 0) + (right.cacheReadInputTokens ?? 0);
     const cacheCreationInputTokens = (left.cacheCreationInputTokens ?? 0) + (right.cacheCreationInputTokens ?? 0);
+    const cacheCreationInputTokens1h = (left.cacheCreationInputTokens1h ?? 0) +
+        (right.cacheCreationInputTokens1h ?? 0);
     const webSearchRequests = (left.webSearchRequests ?? 0) + (right.webSearchRequests ?? 0);
     if (!Number.isSafeInteger(left.inputTokens + right.inputTokens) ||
         !Number.isSafeInteger(left.outputTokens + right.outputTokens) ||
         !Number.isSafeInteger(cacheReadInputTokens) ||
         !Number.isSafeInteger(cacheCreationInputTokens) ||
+        !Number.isSafeInteger(cacheCreationInputTokens1h) ||
         !Number.isSafeInteger(webSearchRequests)) {
         throw new Error('Model usage total overflow');
     }
+    if (cacheCreationInputTokens1h > cacheCreationInputTokens) {
+        throw new Error('Model usage has an invalid cacheCreationInputTokens1h counter');
+    }
     return {
         inputTokens: left.inputTokens + right.inputTokens,
         outputTokens: left.outputTokens + right.outputTokens,
         ...(cacheReadInputTokens === 0 ? {} : { cacheReadInputTokens }),
         ...(cacheCreationInputTokens === 0 ? {} : { cacheCreationInputTokens }),
+        ...(cacheCreationInputTokens1h === 0 ? {} : { cacheCreationInputTokens1h }),
         ...(webSearchRequests === 0 ? {} : { webSearchRequests }),
     };
 }
@@ -51,6 +59,10 @@ function assertValidSessionUsageEntry(model, usage) {
             throw new Error(`Model usage for "${model}" has an invalid ${field} counter`);
         }
     }
+    if ((usage.cacheCreationInputTokens1h ?? 0) >
+        (usage.cacheCreationInputTokens ?? 0)) {
+        throw new Error(`Model usage for "${model}" has an invalid cacheCreationInputTokens1h counter`);
+    }
     for (const field of metadataFields) {
         const value = usage[field];
         if (value !== undefined && (!Number.isSafeInteger(value) || value < 1)) {
@@ -81,20 +93,27 @@ function addSessionUsageChecked(model, left, right) {
     const outputTokens = left.outputTokens + right.outputTokens;
     const cacheReadInputTokens = (left.cacheReadInputTokens ?? 0) + (right.cacheReadInputTokens ?? 0);
     const cacheCreationInputTokens = (left.cacheCreationInputTokens ?? 0) + (right.cacheCreationInputTokens ?? 0);
+    const cacheCreationInputTokens1h = (left.cacheCreationInputTokens1h ?? 0) +
+        (right.cacheCreationInputTokens1h ?? 0);
     const webSearchRequests = (left.webSearchRequests ?? 0) + (right.webSearchRequests ?? 0);
     if (!Number.isSafeInteger(inputTokens) ||
         !Number.isSafeInteger(outputTokens) ||
         !Number.isSafeInteger(cacheReadInputTokens) ||
         !Number.isSafeInteger(cacheCreationInputTokens) ||
+        !Number.isSafeInteger(cacheCreationInputTokens1h) ||
         !Number.isSafeInteger(webSearchRequests)) {
         throw new Error('Model usage total overflow');
     }
+    if (cacheCreationInputTokens1h > cacheCreationInputTokens) {
+        throw new Error(`Model usage for "${model}" has an invalid cacheCreationInputTokens1h counter`);
+    }
     const metadata = mergeSessionUsageMetadata(model, left, right);
     return {
         inputTokens,
         outputTokens,
         ...(cacheReadInputTokens === 0 ? {} : { cacheReadInputTokens }),
         ...(cacheCreationInputTokens === 0 ? {} : { cacheCreationInputTokens }),
+        ...(cacheCreationInputTokens1h === 0 ? {} : { cacheCreationInputTokens1h }),
         ...(webSearchRequests === 0 ? {} : { webSearchRequests }),
         ...metadata,
     };
@@ -119,6 +138,7 @@ function hasNonZeroUsage(usage) {
         usage.outputTokens > 0 ||
         (usage.cacheReadInputTokens ?? 0) > 0 ||
         (usage.cacheCreationInputTokens ?? 0) > 0 ||
+        (usage.cacheCreationInputTokens1h ?? 0) > 0 ||
         (usage.webSearchRequests ?? 0) > 0);
 }
 function validateMetricUsage(usage) {
@@ -134,6 +154,10 @@ function validateMetricUsage(usage) {
             throw new TypeError(`usage.${field} must be a nonnegative safe integer`);
         }
     }
+    if ((usage.cacheCreationInputTokens1h ?? 0) >
+        (usage.cacheCreationInputTokens ?? 0)) {
+        throw new TypeError('usage.cacheCreationInputTokens1h must not exceed usage.cacheCreationInputTokens');
+    }
 }
 function requireMetric(value, field) {
     if (typeof value !== 'number' || !Number.isFinite(value) || value < 0) {

package/dist/application/workflow-manager.js

@@ -85,6 +85,7 @@ const workflowUsageCounterFields = [
     'outputTokens',
     'cacheReadInputTokens',
     'cacheCreationInputTokens',
+    'cacheCreationInputTokens1h',
     'webSearchRequests',
 ];
 const workflowUsageMetadataFields = [
@@ -98,6 +99,10 @@ function assertValidWorkflowUsage(usage) {
             throw new Error(`Workflow agent usage has an invalid ${field} counter`);
         }
     }
+    if ((usage.cacheCreationInputTokens1h ?? 0) >
+        (usage.cacheCreationInputTokens ?? 0)) {
+        throw new Error('Workflow agent usage has an invalid cacheCreationInputTokens1h counter');
+    }
 }
 function addWorkflowApiDuration(value, total, field) {
     if (!Number.isFinite(value) || value < 0) {
@@ -115,14 +120,20 @@ function addWorkflowUsageChecked(model, left, right) {
     const outputTokens = left.outputTokens + right.outputTokens;
     const cacheReadInputTokens = (left.cacheReadInputTokens ?? 0) + (right.cacheReadInputTokens ?? 0);
     const cacheCreationInputTokens = (left.cacheCreationInputTokens ?? 0) + (right.cacheCreationInputTokens ?? 0);
+    const cacheCreationInputTokens1h = (left.cacheCreationInputTokens1h ?? 0) +
+        (right.cacheCreationInputTokens1h ?? 0);
     const webSearchRequests = (left.webSearchRequests ?? 0) + (right.webSearchRequests ?? 0);
     if (!Number.isSafeInteger(inputTokens) ||
         !Number.isSafeInteger(outputTokens) ||
         !Number.isSafeInteger(cacheReadInputTokens) ||
         !Number.isSafeInteger(cacheCreationInputTokens) ||
+        !Number.isSafeInteger(cacheCreationInputTokens1h) ||
         !Number.isSafeInteger(webSearchRequests)) {
         throw new Error('Workflow model usage total overflow');
     }
+    if (cacheCreationInputTokens1h > cacheCreationInputTokens) {
+        throw new Error(`Workflow model usage${model === undefined ? '' : ` for "${model}"`} has an invalid cacheCreationInputTokens1h counter`);
+    }
     // Aggregates without a model stay counter-only; per-model rows merge their
     // capability metadata with conflict rejection.
     const metadata = model === undefined ? {} : mergeWorkflowUsageMetadata(model, left, right);
@@ -131,6 +142,7 @@ function addWorkflowUsageChecked(model, left, right) {
         outputTokens,
         ...(cacheReadInputTokens === 0 ? {} : { cacheReadInputTokens }),
         ...(cacheCreationInputTokens === 0 ? {} : { cacheCreationInputTokens }),
+        ...(cacheCreationInputTokens1h === 0 ? {} : { cacheCreationInputTokens1h }),
         ...(webSearchRequests === 0 ? {} : { webSearchRequests }),
         ...metadata,
     };

package/dist/cli/protocol.js

@@ -1,4 +1,5 @@
 import { randomUUID } from 'node:crypto';
+import { cacheCreationInputTokenSplit } from '../core/usage.js';
 export function projectProtocolTimings(startedAt, requestAt, outputAt) {
     return {
         ...(requestAt === undefined
@@ -13,6 +14,7 @@ export function projectProtocolTimings(startedAt, requestAt, outputAt) {
     };
 }
 function protocolUsage(usage) {
+    const cacheCreation = cacheCreationInputTokenSplit(usage);
     return {
         input_tokens: usage.inputTokens,
         cache_creation_input_tokens: usage.cacheCreationInputTokens ?? 0,
@@ -24,8 +26,8 @@ function protocolUsage(usage) {
         },
         service_tier: 'standard',
         cache_creation: {
-            ephemeral_1h_input_tokens: 0,
-            ephemeral_5m_input_tokens: 0,
+            ephemeral_1h_input_tokens: cacheCreation.oneHour,
+            ephemeral_5m_input_tokens: cacheCreation.fiveMinute,
         },
         inference_geo: '',
         iterations: [],

package/dist/core/runtime.d.ts

@@ -74,6 +74,8 @@ export interface ModelUsage {
     outputTokens: number;
     cacheReadInputTokens?: number;
     cacheCreationInputTokens?: number;
+    /** One-hour cache writes included in cacheCreationInputTokens. */
+    cacheCreationInputTokens1h?: number;
     webSearchRequests?: number;
     /** Positive safe integer when the completing model's context window is known. */
     contextWindow?: number;
@@ -563,6 +565,11 @@ export declare class AgentRunCancelledError extends Error {
     readonly name = "AgentRunCancelledError";
     constructor();
 }
+export declare class AgentBudgetExceededError extends Error {
+    readonly name = "AgentBudgetExceededError";
+    readonly result: AgentRunResult;
+    constructor(message: string, result: AgentRunResult);
+}
 export type RuntimeEventSink = (event: RuntimeEvent) => void;
 export type ProviderErrorKind = 'authentication_failed' | 'billing_error' | 'rate_limit' | 'invalid_request' | 'server_error' | 'timeout' | 'overloaded' | 'api_error' | 'prompt_too_long' | 'transport_error' | 'cancelled' | 'unknown' | 'max_output_tokens';
 export declare function modelProviderErrorKind(error: ModelProviderError): ProviderErrorKind;

package/dist/core/runtime.js

@@ -55,6 +55,14 @@ export class AgentRunCancelledError extends Error {
         super('Agent run cancelled');
     }
 }
+export class AgentBudgetExceededError extends Error {
+    name = 'AgentBudgetExceededError';
+    result;
+    constructor(message, result) {
+        super(message);
+        this.result = result;
+    }
+}
 const emptyUsage = () => ({ inputTokens: 0, outputTokens: 0 });
 export function modelProviderErrorKind(error) {
     if (error.kind !== undefined)
@@ -76,14 +84,30 @@ export function modelProviderErrorKind(error) {
 const unsupportedImageResult = 'Provider does not support image tool results';
 const unsupportedDocumentResult = 'Provider does not support document tool results';
 function addUsage(left, right) {
+    for (const usage of [left, right]) {
+        const oneHour = usage.cacheCreationInputTokens1h;
+        if (oneHour !== undefined &&
+            (!Number.isSafeInteger(oneHour) ||
+                oneHour < 0 ||
+                oneHour > (usage.cacheCreationInputTokens ?? 0))) {
+            throw new Error('Model usage has an invalid cacheCreationInputTokens1h counter');
+        }
+    }
     const cacheReadInputTokens = (left.cacheReadInputTokens ?? 0) + (right.cacheReadInputTokens ?? 0);
     const cacheCreationInputTokens = (left.cacheCreationInputTokens ?? 0) + (right.cacheCreationInputTokens ?? 0);
+    const cacheCreationInputTokens1h = (left.cacheCreationInputTokens1h ?? 0) +
+        (right.cacheCreationInputTokens1h ?? 0);
     const webSearchRequests = (left.webSearchRequests ?? 0) + (right.webSearchRequests ?? 0);
+    if (!Number.isSafeInteger(cacheCreationInputTokens1h) ||
+        cacheCreationInputTokens1h > cacheCreationInputTokens) {
+        throw new Error('Model usage has an invalid cacheCreationInputTokens1h counter');
+    }
     return {
         inputTokens: left.inputTokens + right.inputTokens,
         outputTokens: left.outputTokens + right.outputTokens,
         ...(cacheReadInputTokens === 0 ? {} : { cacheReadInputTokens }),
         ...(cacheCreationInputTokens === 0 ? {} : { cacheCreationInputTokens }),
+        ...(cacheCreationInputTokens1h === 0 ? {} : { cacheCreationInputTokens1h }),
         ...(webSearchRequests === 0 ? {} : { webSearchRequests }),
     };
 }
@@ -126,6 +150,7 @@ const modelUsageCounterFields = [
     'outputTokens',
     'cacheReadInputTokens',
     'cacheCreationInputTokens',
+    'cacheCreationInputTokens1h',
     'webSearchRequests',
 ];
 const modelUsageMetadataFields = ['contextWindow', 'maxOutputTokens'];
@@ -134,6 +159,7 @@ function hasNonZeroModelUsage(usage) {
         usage.outputTokens > 0 ||
         (usage.cacheReadInputTokens ?? 0) > 0 ||
         (usage.cacheCreationInputTokens ?? 0) > 0 ||
+        (usage.cacheCreationInputTokens1h ?? 0) > 0 ||
         (usage.webSearchRequests ?? 0) > 0);
 }
 function assertValidModelUsageEntry(model, usage) {
@@ -146,6 +172,10 @@ function assertValidModelUsageEntry(model, usage) {
             throw new Error(`Model usage for "${model}" has an invalid ${field} counter`);
         }
     }
+    if ((usage.cacheCreationInputTokens1h ?? 0) >
+        (usage.cacheCreationInputTokens ?? 0)) {
+        throw new Error(`Model usage for "${model}" has an invalid cacheCreationInputTokens1h counter`);
+    }
     for (const field of modelUsageMetadataFields) {
         const value = usage[field];
         if (value !== undefined && (!Number.isSafeInteger(value) || value < 1)) {
@@ -176,20 +206,27 @@ function addUsageChecked(model, left, right) {
     const outputTokens = left.outputTokens + right.outputTokens;
     const cacheReadInputTokens = (left.cacheReadInputTokens ?? 0) + (right.cacheReadInputTokens ?? 0);
     const cacheCreationInputTokens = (left.cacheCreationInputTokens ?? 0) + (right.cacheCreationInputTokens ?? 0);
+    const cacheCreationInputTokens1h = (left.cacheCreationInputTokens1h ?? 0) +
+        (right.cacheCreationInputTokens1h ?? 0);
     const webSearchRequests = (left.webSearchRequests ?? 0) + (right.webSearchRequests ?? 0);
     if (!Number.isSafeInteger(inputTokens) ||
         !Number.isSafeInteger(outputTokens) ||
         !Number.isSafeInteger(cacheReadInputTokens) ||
         !Number.isSafeInteger(cacheCreationInputTokens) ||
+        !Number.isSafeInteger(cacheCreationInputTokens1h) ||
         !Number.isSafeInteger(webSearchRequests)) {
         throw new Error('Model usage total overflow');
     }
+    if (cacheCreationInputTokens1h > cacheCreationInputTokens) {
+        throw new Error(`Model usage for "${model}" has an invalid cacheCreationInputTokens1h counter`);
+    }
     const metadata = mergeModelUsageMetadata(model, left, right);
     return {
         inputTokens,
         outputTokens,
         ...(cacheReadInputTokens === 0 ? {} : { cacheReadInputTokens }),
         ...(cacheCreationInputTokens === 0 ? {} : { cacheCreationInputTokens }),
+        ...(cacheCreationInputTokens1h === 0 ? {} : { cacheCreationInputTokens1h }),
         ...(webSearchRequests === 0 ? {} : { webSearchRequests }),
         ...metadata,
     };
@@ -337,6 +374,30 @@ export class AgentRuntime {
         let linesRemoved = 0;
         let activeAttemptHasPresentation = false;
         let activeAttemptDiscarded = false;
+        const materializeResult = (text) => {
+            const modelUsage = modelUsageByModel.size === 0
+                ? undefined
+                : Object.fromEntries(modelUsageByModel);
+            return {
+                text,
+                usage,
+                ...(modelUsage === undefined ? {} : { modelUsage }),
+                ...(sawExternallyMeteredSummary
+                    ? {
+                        unrecordedModelUsage: Object.fromEntries(unrecordedModelUsageByModel),
+                        unrecordedDurationApiMs,
+                        unrecordedDurationApiWithoutRetriesMs,
+                    }
+                    : {}),
+                ...(durationApiMs === 0 ? {} : { durationApiMs }),
+                ...(durationApiMs === 0 && durationApiWithoutRetriesMs === 0
+                    ? {}
+                    : { durationApiWithoutRetriesMs }),
+                ...(durationToolMs === 0 ? {} : { durationToolMs }),
+                ...(linesAdded === 0 ? {} : { linesAdded }),
+                ...(linesRemoved === 0 ? {} : { linesRemoved }),
+            };
+        };
         const maxModelTurns = request.maxModelTurns ?? this.options.maxModelTurns;
         if (maxModelTurns !== undefined &&
             (!Number.isSafeInteger(maxModelTurns) || maxModelTurns <= 0)) {
@@ -399,7 +460,7 @@ export class AgentRuntime {
                 if (this.options.maxBudgetUsd !== undefined &&
                     spent !== undefined &&
                     spent >= this.options.maxBudgetUsd) {
-                    throw new Error(`Maximum budget of $${this.options.maxBudgetUsd.toFixed(6)} exceeded`);
+                    throw new AgentBudgetExceededError(`Maximum budget of $${this.options.maxBudgetUsd.toFixed(6)} exceeded`, materializeResult(''));
                 }
                 modelTurns += 1;
                 this.emit({ type: 'state', state: 'awaiting-model' });
@@ -698,28 +759,7 @@ export class AgentRuntime {
                         continue;
                     }
                     this.emit({ type: 'state', state: 'completed' });
-                    const modelUsage = modelUsageByModel.size === 0
-                        ? undefined
-                        : Object.fromEntries(modelUsageByModel);
-                    return {
-                        text,
-                        usage,
-                        ...(modelUsage === undefined ? {} : { modelUsage }),
-                        ...(sawExternallyMeteredSummary
-                            ? {
-                                unrecordedModelUsage: Object.fromEntries(unrecordedModelUsageByModel),
-                                unrecordedDurationApiMs,
-                                unrecordedDurationApiWithoutRetriesMs,
-                            }
-                            : {}),
-                        ...(durationApiMs === 0 ? {} : { durationApiMs }),
-                        ...(durationApiMs === 0 && durationApiWithoutRetriesMs === 0
-                            ? {}
-                            : { durationApiWithoutRetriesMs }),
-                        ...(durationToolMs === 0 ? {} : { durationToolMs }),
-                        ...(linesAdded === 0 ? {} : { linesAdded }),
-                        ...(linesRemoved === 0 ? {} : { linesRemoved }),
-                    };
+                    return materializeResult(text);
                 }
                 const scheduledToolResults = await toolScheduler.settle();
                 const followUpUserMessages = [];

package/dist/core/usage.d.ts

@@ -4,6 +4,9 @@ export interface ModelPricing {
     outputPerMillionUsd: number;
     cacheReadInputPerMillionUsd?: number;
     cacheCreationInputPerMillionUsd?: number;
+    /** Falls back to cacheCreationInputPerMillionUsd when omitted. */
+    cacheCreationInputPerMillionUsd1h?: number;
+    webSearchPerRequestUsd?: number;
 }
 export type ModelPricingTable = Readonly<Record<string, ModelPricing>>;
 export type ModelPricingSource = 'builtin' | 'environment' | 'unknown';
@@ -18,9 +21,15 @@ export declare class ModelPricingRegistry {
     private readonly table;
     private readonly environmentModels;
     constructor(overrides?: ModelPricingTable);
+    private resolveKey;
     resolve(model: string): ModelPricing | undefined;
     diagnose(model: string): ModelPricingDiagnosis;
     static fromEnvironment(value: string | undefined): ModelPricingRegistry;
 }
+export declare function cacheCreationInputTokenSplit(usage: ModelUsage): {
+    total: number;
+    fiveMinute: number;
+    oneHour: number;
+};
 export declare function usageCostUsd(usage: ModelUsage, pricing: ModelPricing): number;
 //# sourceMappingURL=usage.d.ts.map

package/dist/core/usage.js

@@ -1,28 +1,67 @@
+function anthropicPricing(inputPerMillionUsd, outputPerMillionUsd, cacheReadInputPerMillionUsd, cacheCreationInputPerMillionUsd) {
+    return {
+        inputPerMillionUsd,
+        outputPerMillionUsd,
+        cacheReadInputPerMillionUsd,
+        cacheCreationInputPerMillionUsd,
+        cacheCreationInputPerMillionUsd1h: inputPerMillionUsd * 2,
+        webSearchPerRequestUsd: 0.01,
+    };
+}
 const BUILTIN_PRICING = {
     'claude-3-5-sonnet-20241022': {
         inputPerMillionUsd: 3,
         outputPerMillionUsd: 15,
         cacheReadInputPerMillionUsd: 0.3,
         cacheCreationInputPerMillionUsd: 3.75,
+        cacheCreationInputPerMillionUsd1h: 6,
+        webSearchPerRequestUsd: 0.01,
     },
     'claude-3-7-sonnet-20250219': {
         inputPerMillionUsd: 3,
         outputPerMillionUsd: 15,
         cacheReadInputPerMillionUsd: 0.3,
         cacheCreationInputPerMillionUsd: 3.75,
+        cacheCreationInputPerMillionUsd1h: 6,
+        webSearchPerRequestUsd: 0.01,
     },
     'claude-sonnet-4-20250514': {
         inputPerMillionUsd: 3,
         outputPerMillionUsd: 15,
         cacheReadInputPerMillionUsd: 0.3,
         cacheCreationInputPerMillionUsd: 3.75,
+        cacheCreationInputPerMillionUsd1h: 6,
+        webSearchPerRequestUsd: 0.01,
     },
     'claude-opus-4-20250514': {
         inputPerMillionUsd: 15,
         outputPerMillionUsd: 75,
         cacheReadInputPerMillionUsd: 1.5,
         cacheCreationInputPerMillionUsd: 18.75,
+        cacheCreationInputPerMillionUsd1h: 30,
+        webSearchPerRequestUsd: 0.01,
     },
+    'claude-fable-5-1': anthropicPricing(10, 50, 0.25, 12.5),
+    'claude-opus-5': anthropicPricing(5, 25, 0.5, 6.25),
+    'claude-opus-4-8': anthropicPricing(5, 25, 0.5, 6.25),
+    'claude-opus-4-7': anthropicPricing(5, 25, 0.5, 6.25),
+    'claude-opus-4-6': anthropicPricing(5, 25, 0.5, 6.25),
+    'claude-opus-4-5': anthropicPricing(5, 25, 0.5, 6.25),
+    'claude-opus-4-5-20251101': anthropicPricing(5, 25, 0.5, 6.25),
+    'claude-sonnet-5': anthropicPricing(2, 10, 0.2, 2.5),
+    'claude-sonnet-4-6': anthropicPricing(3, 15, 0.3, 3.75),
+    'claude-sonnet-4-5': anthropicPricing(3, 15, 0.3, 3.75),
+    'claude-sonnet-4-5-20250929': anthropicPricing(3, 15, 0.3, 3.75),
+    'claude-haiku-4-5': anthropicPricing(1, 5, 0.1, 1.25),
+    'claude-haiku-4-5-20251001': anthropicPricing(1, 5, 0.1, 1.25),
+    'claude-opus-4': anthropicPricing(15, 75, 1.5, 18.75),
+    'claude-opus-4-1': anthropicPricing(15, 75, 1.5, 18.75),
+    'claude-opus-4-1-20250805': anthropicPricing(15, 75, 1.5, 18.75),
+    'claude-sonnet-4': anthropicPricing(3, 15, 0.3, 3.75),
+    'claude-3-5-sonnet-20240620': anthropicPricing(3, 15, 0.3, 3.75),
+    'claude-3-5-sonnet-latest': anthropicPricing(3, 15, 0.3, 3.75),
+    'claude-3-5-haiku-20241022': anthropicPricing(0.8, 4, 0.08, 1),
+    'claude-3-5-haiku-latest': anthropicPricing(0.8, 4, 0.08, 1),
     'gpt-4o': {
         inputPerMillionUsd: 5,
         outputPerMillionUsd: 15,
@@ -47,12 +86,20 @@ function parsePricing(value, label) {
     }
     const cacheRead = record.cacheReadInputPerMillionUsd;
     const cacheCreation = record.cacheCreationInputPerMillionUsd;
+    const cacheCreation1h = record.cacheCreationInputPerMillionUsd1h;
+    const webSearch = record.webSearchPerRequestUsd;
     if (cacheRead !== undefined && !validRate(cacheRead)) {
         throw new Error(`${label}.cacheReadInputPerMillionUsd must be non-negative`);
     }
     if (cacheCreation !== undefined && !validRate(cacheCreation)) {
         throw new Error(`${label}.cacheCreationInputPerMillionUsd must be non-negative`);
     }
+    if (cacheCreation1h !== undefined && !validRate(cacheCreation1h)) {
+        throw new Error(`${label}.cacheCreationInputPerMillionUsd1h must be non-negative`);
+    }
+    if (webSearch !== undefined && !validRate(webSearch)) {
+        throw new Error(`${label}.webSearchPerRequestUsd must be non-negative`);
+    }
     return {
         inputPerMillionUsd: input,
         outputPerMillionUsd: output,
@@ -62,6 +109,10 @@ function parsePricing(value, label) {
         ...(cacheCreation === undefined
             ? {}
             : { cacheCreationInputPerMillionUsd: cacheCreation }),
+        ...(cacheCreation1h === undefined
+            ? {}
+            : { cacheCreationInputPerMillionUsd1h: cacheCreation1h }),
+        ...(webSearch === undefined ? {} : { webSearchPerRequestUsd: webSearch }),
     };
 }
 export class ModelPricingRegistry {
@@ -71,12 +122,36 @@ export class ModelPricingRegistry {
         this.table = { ...BUILTIN_PRICING, ...overrides };
         this.environmentModels = new Set(Object.keys(overrides));
     }
+    resolveKey(model) {
+        if (Object.prototype.hasOwnProperty.call(this.table, model)) {
+            return model;
+        }
+        const suffix = '[1m]';
+        if (!model.endsWith(suffix))
+            return undefined;
+        const base = model.slice(0, -suffix.length);
+        if (base.length === 0 || base.endsWith(suffix))
+            return undefined;
+        return Object.prototype.hasOwnProperty.call(this.table, base)
+            ? base
+            : undefined;
+    }
     resolve(model) {
-        return this.table[model];
+        const key = this.resolveKey(model);
+        return key === undefined ? undefined : this.table[key];
     }
     diagnose(model) {
-        const pricing = this.resolve(model);
-        if (!pricing) {
+        const key = this.resolveKey(model);
+        if (key === undefined) {
+            return {
+                model,
+                source: 'unknown',
+                policy: 'fail-closed',
+                budgetBehavior: 'reject-before-provider',
+            };
+        }
+        const pricing = this.table[key];
+        if (pricing === undefined) {
             return {
                 model,
                 source: 'unknown',
@@ -86,7 +161,7 @@ export class ModelPricingRegistry {
         }
         return {
             model,
-            source: this.environmentModels.has(model) ? 'environment' : 'builtin',
+            source: this.environmentModels.has(key) ? 'environment' : 'builtin',
             pricing,
             policy: 'fail-closed',
             budgetBehavior: 'enforce',
@@ -115,19 +190,36 @@ export class ModelPricingRegistry {
         return new ModelPricingRegistry(overrides);
     }
 }
+export function cacheCreationInputTokenSplit(usage) {
+    const total = usage.cacheCreationInputTokens ?? 0;
+    const oneHour = usage.cacheCreationInputTokens1h ?? 0;
+    if (!Number.isSafeInteger(total) ||
+        total < 0 ||
+        !Number.isSafeInteger(oneHour) ||
+        oneHour < 0 ||
+        oneHour > total) {
+        throw new Error('cacheCreationInputTokens1h must be a nonnegative safe integer no greater than cacheCreationInputTokens');
+    }
+    return { total, fiveMinute: total - oneHour, oneHour };
+}
 function regularInputTokens(usage) {
-    return Math.max(0, usage.inputTokens -
-        (usage.cacheReadInputTokens ?? 0) -
-        (usage.cacheCreationInputTokens ?? 0));
+    const { total } = cacheCreationInputTokenSplit(usage);
+    return Math.max(0, usage.inputTokens - (usage.cacheReadInputTokens ?? 0) - total);
 }
 export function usageCostUsd(usage, pricing) {
+    const cacheCreation = cacheCreationInputTokenSplit(usage);
     const regularInput = regularInputTokens(usage);
     const input = regularInput * pricing.inputPerMillionUsd;
     const output = usage.outputTokens * pricing.outputPerMillionUsd;
     const cacheRead = (usage.cacheReadInputTokens ?? 0) *
         (pricing.cacheReadInputPerMillionUsd ?? pricing.inputPerMillionUsd);
-    const cacheCreation = (usage.cacheCreationInputTokens ?? 0) *
-        (pricing.cacheCreationInputPerMillionUsd ?? pricing.inputPerMillionUsd);
-    return (input + output + cacheRead + cacheCreation) / 1_000_000;
+    const cacheCreationCost = cacheCreation.fiveMinute *
+        (pricing.cacheCreationInputPerMillionUsd ?? pricing.inputPerMillionUsd) +
+        cacheCreation.oneHour *
+            (pricing.cacheCreationInputPerMillionUsd1h ??
+                pricing.cacheCreationInputPerMillionUsd ??
+                pricing.inputPerMillionUsd);
+    return ((input + output + cacheRead + cacheCreationCost) / 1_000_000 +
+        (usage.webSearchRequests ?? 0) * (pricing.webSearchPerRequestUsd ?? 0));
 }
 //# sourceMappingURL=usage.js.map

package/dist/persistence/subagent-lifecycle-store.d.ts

@@ -8,6 +8,7 @@ export interface PersistedSubagentRunResult {
         outputTokens: number;
         cacheReadInputTokens?: number;
         cacheCreationInputTokens?: number;
+        cacheCreationInputTokens1h?: number;
         webSearchRequests?: number;
         contextWindow?: number;
         maxOutputTokens?: number;

package/dist/persistence/subagent-lifecycle-store.js

@@ -38,10 +38,14 @@ function isUsage(value) {
     for (const field of [
         'cacheReadInputTokens',
         'cacheCreationInputTokens',
+        'cacheCreationInputTokens1h',
         'webSearchRequests',
     ])
         if (usage[field] !== undefined && !isNonnegativeInteger(usage[field]))
             return false;
+    if ((usage.cacheCreationInputTokens1h ?? 0) >
+        (usage.cacheCreationInputTokens ?? 0))
+        return false;
     for (const field of ['contextWindow', 'maxOutputTokens'])
         if (usage[field] !== undefined &&
             (!Number.isSafeInteger(usage[field]) || Number(usage[field]) < 1))

package/dist/providers/anthropic-compatible.js

@@ -3,7 +3,7 @@ import { transportFailureKind } from './provider-errors.js';
 import { reportProviderTransportActivity } from './provider-transport-activity.js';
 import { markNonStreamingFallbackEligible } from './non-streaming-fallback-provider.js';
 import { createAnthropicPromptCachePolicyResolver, } from './anthropic-prompt-cache.js';
-import { resolveAnthropicModelSpec } from './anthropic-model-spec.js';
+import { resolveAnthropicEffort, resolveAnthropicModelSpec, } from './anthropic-model-spec.js';
 function isRecord(value) {
     return typeof value === 'object' && value !== null && !Array.isArray(value);
 }
@@ -16,6 +16,28 @@ function readNonNegativeTokenCount(usage, field, required) {
     }
     return value;
 }
+function readCacheCreationInputTokens1h(usage) {
+    const detail = usage.cache_creation;
+    if (detail === undefined)
+        return undefined;
+    if (!isRecord(detail)) {
+        throw new ModelProviderError('Provider returned an invalid cache_creation usage object', { retryable: false });
+    }
+    const total = readNonNegativeTokenCount(usage, 'cache_creation_input_tokens', true);
+    if (total === undefined) {
+        throw new ModelProviderError('Provider returned cache_creation usage without cache_creation_input_tokens', { retryable: false });
+    }
+    const fiveMinute = readNonNegativeTokenCount(detail, 'ephemeral_5m_input_tokens', true);
+    const oneHour = readNonNegativeTokenCount(detail, 'ephemeral_1h_input_tokens', true);
+    if (fiveMinute === undefined || oneHour === undefined) {
+        throw new ModelProviderError('Provider returned an incomplete cache_creation usage object', { retryable: false });
+    }
+    const sum = fiveMinute + oneHour;
+    if (!Number.isSafeInteger(sum) || sum !== total) {
+        throw new ModelProviderError('Provider returned cache_creation usage whose TTL counters do not sum to cache_creation_input_tokens', { retryable: false });
+    }
+    return oneHour;
+}
 function webSearchLinks(value) {
     if (!Array.isArray(value))
         return [];
@@ -207,6 +229,10 @@ function parseSseEvent(data, state, maxToolArgumentsBytes, maxToolCallsPerRespon
                 typeof usage.cache_creation_input_tokens === 'number'
                     ? usage.cache_creation_input_tokens
                     : 0;
+            const cacheCreationInputTokens1h = readCacheCreationInputTokens1h(usage);
+            if (cacheCreationInputTokens1h !== undefined) {
+                state.cacheCreationInputTokens1h = cacheCreationInputTokens1h;
+            }
             state.cacheReadInputTokens =
                 typeof usage.cache_read_input_tokens === 'number'
                     ? usage.cache_read_input_tokens
@@ -466,6 +492,10 @@ function parseSseEvent(data, state, maxToolArgumentsBytes, maxToolCallsPerRespon
                     ...(state.cacheCreationInputTokens === 0
                         ? {}
                         : { cacheCreationInputTokens: state.cacheCreationInputTokens }),
+                    ...(state.cacheCreationInputTokens1h === undefined ||
+                        state.cacheCreationInputTokens1h === 0
+                        ? {}
+                        : { cacheCreationInputTokens1h: state.cacheCreationInputTokens1h }),
                     ...(state.webSearchRequests === 0
                         ? {}
                         : { webSearchRequests: state.webSearchRequests }),
@@ -666,12 +696,7 @@ export class AnthropicCompatibleProvider {
         this.providerBetas = Object.freeze([...modelSpec.betas]);
         this.streaming = options.streaming ?? true;
         this.fetchImplementation = options.fetchImplementation ?? fetch;
-        this.maxOutputTokens = positiveInteger(options.maxOutputTokens ??
-            (this.wireModel.includes('claude-opus-4-6')
-                ? 64_000
-                : this.wireModel.startsWith('claude-')
-                    ? 32_000
-                    : 8192), 'Max output tokens');
+        this.maxOutputTokens = positiveInteger(options.maxOutputTokens ?? modelSpec.defaultMaxOutputTokens, 'Max output tokens');
         this.capabilities = {
             streaming: this.streaming,
             usage: true,
@@ -739,6 +764,7 @@ export class AnthropicCompatibleProvider {
         ].filter((beta, index, all) => all.indexOf(beta) === index);
         const serialized = serializeMessages(request.messages, request.stableSystemMessageCount, this.promptCaching);
         const requestTools = request.tools;
+        const effort = resolveAnthropicEffort(this.wireModel, request.effort);
         const requestInit = {
             method: 'POST',
             headers: {
@@ -754,9 +780,7 @@ export class AnthropicCompatibleProvider {
                 messages: serialized.messages,
                 stream: this.streaming,
                 ...(thinkingPayload ? { thinking: thinkingPayload } : {}),
-                ...(request.effort
-                    ? { output_config: { effort: request.effort } }
-                    : {}),
+                ...(effort ? { output_config: { effort } } : {}),
                 ...(serialized.system.length ? { system: serialized.system } : {}),
                 ...(request.webSearch
                     ? {
@@ -926,6 +950,8 @@ export class AnthropicCompatibleProvider {
                 const outputTokens = readNonNegativeTokenCount(payload.usage, 'output_tokens', true);
                 const cacheReadInputTokens = readNonNegativeTokenCount(payload.usage, 'cache_read_input_tokens', false);
                 const cacheCreationInputTokens = readNonNegativeTokenCount(payload.usage, 'cache_creation_input_tokens', false);
+                const cacheCreationInputTokens1h = readCacheCreationInputTokens1h(payload.usage);
+                const normalizedCacheCreationInputTokens = cacheCreationInputTokens ?? 0;
                 const usage = {
                     input_tokens: inputTokens,
                     output_tokens: outputTokens,
@@ -935,6 +961,15 @@ export class AnthropicCompatibleProvider {
                     ...(cacheCreationInputTokens === undefined
                         ? {}
                         : { cache_creation_input_tokens: cacheCreationInputTokens }),
+                    ...(cacheCreationInputTokens1h === undefined
+                        ? {}
+                        : {
+                            cache_creation: {
+                                ephemeral_5m_input_tokens: normalizedCacheCreationInputTokens -
+                                    cacheCreationInputTokens1h,
+                                ephemeral_1h_input_tokens: cacheCreationInputTokens1h,
+                            },
+                        }),
                     ...(payload.usage.server_tool_use === undefined
                         ? {}
                         : { server_tool_use: payload.usage.server_tool_use }),

package/dist/providers/anthropic-model-spec.d.ts

@@ -1,8 +1,10 @@
 export declare const ANTHROPIC_LONG_CONTEXT_BETA = "context-1m-2025-08-07";
+export declare function resolveAnthropicEffort(wireModel: string, effort: string | undefined): string | undefined;
 export interface ResolvedAnthropicModelSpec {
     readonly model: string;
     readonly wireModel: string;
     readonly contextWindowTokens: number;
+    readonly defaultMaxOutputTokens: number;
     readonly betas: readonly string[];
     readonly supportsAdaptiveThinking: boolean;
 }

package/dist/providers/anthropic-model-spec.js

@@ -1,4 +1,21 @@
 export const ANTHROPIC_LONG_CONTEXT_BETA = 'context-1m-2025-08-07';
+function defaultMaxOutputTokens(wireModel) {
+    if (/(?:^|[^a-z0-9])claude-(?:opus-4-[678]|opus-5|sonnet-5)(?=$|[-@.:/_])/.test(wireModel)) {
+        return 64_000;
+    }
+    return wireModel.startsWith('claude-') ? 32_000 : 8192;
+}
+export function resolveAnthropicEffort(wireModel, effort) {
+    if (effort === undefined)
+        return undefined;
+    if (wireModel === 'claude-sonnet-4-6' || wireModel === 'claude-opus-4-6') {
+        return effort === 'xhigh' ? 'high' : effort;
+    }
+    if (/^claude-opus-4-5(?:-|$)/.test(wireModel)) {
+        return effort === 'xhigh' || effort === 'max' ? 'high' : effort;
+    }
+    return effort;
+}
 export function resolveAnthropicModelSpec(model, explicitContextWindowTokens) {
     const longContext = model.endsWith('[1m]');
     const wireModel = longContext ? model.slice(0, -'[1m]'.length) : model;
@@ -8,6 +25,7 @@ export function resolveAnthropicModelSpec(model, explicitContextWindowTokens) {
     return Object.freeze({
         model,
         wireModel,
+        defaultMaxOutputTokens: defaultMaxOutputTokens(wireModel),
         contextWindowTokens: explicitContextWindowTokens ?? (longContext ? 1_000_000 : 200_000),
         betas: Object.freeze(longContext ? [ANTHROPIC_LONG_CONTEXT_BETA] : []),
         supportsAdaptiveThinking: wireModel === 'claude-sonnet-4-6' || wireModel === 'claude-opus-4-6',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "praxis-agent",
-  "version": "0.62.3",
+  "version": "0.62.6",
   "description": "Local-first, single-user general agent for the command line.",
   "license": "MIT",
   "author": "wuqisen",