@stage-labs/metro 0.1.0-beta.321 → 0.1.0-beta.322

package/runtime/node_modules/@metro-labs/{daemon/src/gateway → inference/src}/usage.ts

@@ -1,3 +1,4 @@
+import { currentAttempt } from './attempts.js';
 type Key = string;
 
 export interface UsageWindow {
@@ -92,28 +93,49 @@ export function tallyTokens(key: Key, counted: Counted, now = new Date()): void
   });
 }
 
-const TOKEN_FIELD = /"(input_tokens|output_tokens|cache_read_input_tokens|cache_creation_input_tokens)"\s*:\s*(\d+)/g;
+const TOKEN_FIELD = /"(input_tokens|output_tokens|cache_read_input_tokens|cache_creation_input_tokens|ephemeral_5m_input_tokens|ephemeral_1h_input_tokens|cost)"\s*:\s*(\d+(?:\.\d+)?)/g;
+
+const FIELDS: Record<string, keyof Seen> = {
+  input_tokens: 'input',
+  output_tokens: 'output',
+  cache_read_input_tokens: 'read',
+  cache_creation_input_tokens: 'creation',
+  ephemeral_5m_input_tokens: 'write5m',
+  ephemeral_1h_input_tokens: 'write1h',
+  cost: 'cost',
+};
+
+interface Seen {
+  input: number;
+  output: number;
+  read: number;
+  creation: number;
+  write5m: number;
+  write1h: number;
+  cost: number;
+}
 
 export class UsageScanner {
   private carry = '';
-  private readonly max = { input: 0, output: 0, read: 0, creation: 0 };
+  private readonly max: Seen = { input: 0, output: 0, read: 0, creation: 0, write5m: 0, write1h: 0, cost: 0 };
   private touched = false;
+  private costed = false;
 
   constructor(private readonly key: Key) {}
 
   private scan(text: string): void {
     for (const hit of text.matchAll(TOKEN_FIELD)) {
+      const field = FIELDS[hit[1] ?? ''];
       const value = Number(hit[2]);
-      if (!Number.isFinite(value)) continue;
-      this.touched = true;
-      if (hit[1] === 'input_tokens') this.max.input = Math.max(this.max.input, value);
-      else if (hit[1] === 'output_tokens') this.max.output = Math.max(this.max.output, value);
-      else if (hit[1] === 'cache_read_input_tokens') this.max.read = Math.max(this.max.read, value);
-      else this.max.creation = Math.max(this.max.creation, value);
+      if (field === undefined || !Number.isFinite(value)) continue;
+      if (field === 'cost') this.costed = true;
+      else this.touched = true;
+      this.max[field] = Math.max(this.max[field], value);
     }
   }
 
   feed(text: string): void {
+    currentAttempt()?.firstByte();
     const joined = this.carry + text;
     const cut = joined.lastIndexOf('\n');
     if (cut === -1) {
@@ -128,7 +150,10 @@ export class UsageScanner {
     if (this.carry !== '') this.scan(this.carry);
     this.carry = '';
     if (!this.touched) return;
-    tallyTokens(this.key, { input: this.max.input, output: this.max.output, cached: this.max.read + this.max.creation }, now);
+    const { input, output, read, creation, write5m, write1h, cost } = this.max;
+    const split = write5m + write1h;
+    currentAttempt()?.counted({ input, output, cacheRead: read, cacheWrite5m: split > 0 ? write5m : creation, cacheWrite1h: write1h }, this.costed ? cost : null);
+    tallyTokens(this.key, { input, output, cached: read + creation }, now);
     this.touched = false;
   }
 }

package/runtime/runtime.json

@@ -1,3 +1,3 @@
 {
-  "version": "0.1.0-beta.321"
+  "version": "0.1.0-beta.322"
 }

package/runtime/sdk-runner/node_modules/@metro-labs/core/src/runner-activity.ts

@@ -8,17 +8,17 @@ export type RunnerTaskState = (typeof TASK_STATES)[number];
 export type RunnerEventKind = (typeof EVENTS)[number];
 
 const FAILURE = {
-  authentication: 'The provider refused authentication. Check the Model page.',
-  billing: 'The provider refused billing. Check the Model page.',
-  rate_limit: 'The provider rate limit was reached. Retry later or check the Model page.',
-  provider_unavailable: 'The provider is unavailable. Retry later or check the Model page.',
-  invalid_request: 'The provider rejected the request. Check Conversations and the Model page.',
+  authentication: 'The provider refused authentication. Check the Gateway page.',
+  billing: 'The provider refused billing. Check the Gateway page.',
+  rate_limit: 'The provider rate limit was reached. Retry later or check the Gateway page.',
+  provider_unavailable: 'The provider is unavailable. Retry later or check the Gateway page.',
+  invalid_request: 'The provider rejected the request. Check Conversations and the Gateway page.',
   output_limit: 'The model reached its output token limit. Check Conversations.',
-  provider_error: 'SDK turn failed (provider_error). Check Conversations and the Model page.',
-  error_during_execution: 'SDK turn failed (error_during_execution). Check Conversations and the Model page.',
-  error_max_turns: 'SDK turn failed (error_max_turns). Check Conversations and the Model page.',
-  error_max_budget_usd: 'SDK turn failed (error_max_budget_usd). Check Conversations and the Model page.',
-  error_max_structured_output_retries: 'SDK turn failed (error_max_structured_output_retries). Check Conversations and the Model page.',
+  provider_error: 'SDK turn failed (provider_error). Check Conversations and the Gateway page.',
+  error_during_execution: 'SDK turn failed (error_during_execution). Check Conversations and the Gateway page.',
+  error_max_turns: 'SDK turn failed (error_max_turns). Check Conversations and the Gateway page.',
+  error_max_budget_usd: 'SDK turn failed (error_max_budget_usd). Check Conversations and the Gateway page.',
+  error_max_structured_output_retries: 'SDK turn failed (error_max_structured_output_retries). Check Conversations and the Gateway page.',
   read_token_limit: 'Read exceeded its token limit. Read a smaller range with offset and limit, or search the file.',
   tool_error: 'A tool failed. See its result in Conversations.',
   task_error: 'A background task failed. See Conversations and Terminal.',
@@ -27,12 +27,12 @@ const FAILURE = {
   saved_state: 'The saved Agent SDK state cannot be read. Restore it before starting.',
   executable: 'The SDK executable could not start. Prepare the SDK runtime again.',
   module: 'The SDK runtime could not load a required module. Prepare the SDK runtime again.',
-  configuration: 'The SDK launch configuration is missing or invalid. Check Harness and the Model page before starting.',
+  configuration: 'The SDK launch configuration is missing or invalid. Check Harness and the Gateway page before starting.',
   process_exit: 'The SDK process exited unexpectedly. Check its reported exit code or signal before retrying.',
   connection: 'The SDK connection failed. Check the server and provider connection.',
   session_ended: 'The Agent SDK session ended unexpectedly.',
   interrupted: 'The SDK stopped before finishing a chat request. Check Conversations before retrying; actions may already have run.',
-  session_error: 'The Agent SDK session failed. Check Terminal and the Model page.',
+  session_error: 'The Agent SDK session failed. Check Terminal and the Gateway page.',
 } as const;
 export type RunnerFailureCode = keyof typeof FAILURE;
 

package/runtime/sdk-runner/runner.json

@@ -1,5 +1,5 @@
 {
-  "version": "0.1.0-beta.321",
+  "version": "0.1.0-beta.322",
   "dependencies": {
     "@anthropic-ai/claude-agent-sdk": "0.3.287",
     "@modelcontextprotocol/sdk": "^1.29.0",

package/runtime/sdk-runner/src/app.ts

@@ -40,7 +40,7 @@ async function connected(cfg: RunnerConfig, hooks: AppHooks, release: () => Prom
     callState: (state) => { runner.calls.reconcile(state.route); },
     toolsChanged: () => { tools?.changed(); },
     model: (model) => {
-      log.info({ model }, 'sdk-runner: the Model page picked a model');
+      log.info({ model }, 'sdk-runner: the Gateway page picked a model');
       runner.switchModel(model).catch((err: unknown) => {
         log.warn({ err: errMsg(err) }, 'sdk-runner: could not switch the model');
       });

package/runtime/sdk-runner/src/runner.ts

@@ -243,7 +243,7 @@ export class Runner {
         debugFailure('model switch', err);
         const failure = new SessionError(this.diagnostics.failure(err));
         this.parts.activity?.fail(failure);
-        log.warn({ reason: failure.failure.code, diagnostic: failure.failure.diagnostic, summary: failureSummary(failure) }, 'sdk-runner: the session could not take the model the Model page picked');
+        log.warn({ reason: failure.failure.code, diagnostic: failure.failure.diagnostic, summary: failureSummary(failure) }, 'sdk-runner: the session could not take the model the Gateway page picked');
       });
   }
 

package/runtime/node_modules/@metro-labs/daemon/src/gateway/served.ts

@@ -1,18 +0,0 @@
-export interface Served {
-  connection: string;
-  provider: string;
-  model: string;
-  at: string;
-}
-
-let last: Served | null = null;
-
-export const noteServed = (served: Served): void => {
-  last = served;
-};
-
-export const lastServed = (): Served | null => last;
-
-export const forgetServed = (): void => {
-  last = null;
-};