pi-smart-router 0.6.0 → 0.6.1

package/.pi/extensions/smart-router/commands.ts

@@ -105,6 +105,38 @@ export function getSmartRouterArgumentCompletions(prefix: string): CompletionIte
   return filtered.length > 0 ? filtered : null;
 }
 
+/** Throw AbortError when ESC / ctx.signal has cancelled the command. */
+export function throwIfCommandAborted(signal: AbortSignal | undefined): void {
+  if (!signal?.aborted) {
+    return;
+  }
+  const reason = signal.reason;
+  if (reason instanceof Error) {
+    throw reason;
+  }
+  const error = new Error(typeof reason === 'string' && reason.length > 0 ? reason : 'Aborted');
+  error.name = 'AbortError';
+  throw error;
+}
+
+function isAbortError(error: unknown): boolean {
+  return error instanceof Error && error.name === 'AbortError';
+}
+
+/**
+ * Wrap fetch so LiteLLM pricing requests honor ctx.signal without changing
+ * refreshPricingCatalog's signature (out of File Scope for SP-172).
+ */
+export function createSignalAwareFetch(signal: AbortSignal | undefined): typeof fetch | undefined {
+  if (!signal) {
+    return undefined;
+  }
+  return ((input: Parameters<typeof fetch>[0], init?: RequestInit) => {
+    throwIfCommandAborted(signal);
+    return fetch(input, { ...init, signal });
+  }) as typeof fetch;
+}
+
 export function registerSmartRouterCommand(
   pi: ExtensionAPI,
   runtime: SmartRouterRuntime,
@@ -117,6 +149,7 @@ export function registerSmartRouterCommand(
       try {
         bindSharedModelRegistry(runtime, ctx.modelRegistry);
         const parsed = parseSmartRouterArgs(args);
+        const signal = ctx.signal;
 
         if (parsed.command === 'status') {
           ctx.ui.notify(formatStatusMessage(runtime, runtime.lastDecision), 'info');
@@ -124,13 +157,21 @@ export function registerSmartRouterCommand(
         }
 
         if (parsed.command === 'history') {
+          throwIfCommandAborted(signal);
           const rows = await runtime.store.listTelemetry({ limit: parsed.limit });
+          throwIfCommandAborted(signal);
           ctx.ui.notify(formatHistoryMessage(rows), 'info');
           return;
         }
 
         if (parsed.command === 'pricing') {
-          const { modelCount, lastUpdated } = await refreshPricingCatalog(runtime);
+          throwIfCommandAborted(signal);
+          const { modelCount, lastUpdated } = await refreshPricingCatalog(
+            runtime,
+            createSignalAwareFetch(signal),
+          );
+          // Abort before fleet mutation so cancel does not leave a half-rebuilt fleet.
+          throwIfCommandAborted(signal);
           await rebuildFleet(runtime, pi, ctx.cwd);
           ctx.ui.notify(
             `Pricing refreshed: ${modelCount} models loaded (last_updated: ${lastUpdated}). Fleet rebuilt (${runtime.streamDeps.fleet.length} models).`,
@@ -140,6 +181,7 @@ export function registerSmartRouterCommand(
         }
 
         if (parsed.command === 'export' && parsed.subcommand === 'dataset') {
+          throwIfCommandAborted(signal);
           const result = await exportDatasetToFile(runtime.store, ctx.cwd, parsed.limit);
           if (!result) {
             ctx.ui.notify('No routing dataset records to export.', 'info');
@@ -153,6 +195,7 @@ export function registerSmartRouterCommand(
         }
 
         if (parsed.command === 'export' && parsed.subcommand === 'telemetry-contrib') {
+          throwIfCommandAborted(signal);
           const result = await exportTelemetryContrib({
             store: runtime.store,
             cwd: ctx.cwd,
@@ -226,6 +269,7 @@ export function registerSmartRouterCommand(
           return;
         }
 
+        throwIfCommandAborted(signal);
         runtime.fleetMode = parsed.mode;
         await rebuildFleet(runtime, pi, ctx.cwd);
         pi.appendEntry(FLEET_MODE_ENTRY_TYPE, { mode: parsed.mode });
@@ -234,6 +278,10 @@ export function registerSmartRouterCommand(
           'info',
         );
       } catch (error) {
+        if (isAbortError(error) || ctx.signal?.aborted) {
+          ctx.ui.notify('Cancelled.', 'info');
+          return;
+        }
         const message = error instanceof Error ? error.message : String(error);
         ctx.ui.notify(message, 'error');
       }

package/.pi/extensions/smart-router/delegate-stream.ts

@@ -2,6 +2,7 @@ import {
   type Api,
   type AssistantMessage,
   type AssistantMessageEvent,
+  type AssistantMessageEventStream,
   type Context,
   type Model,
   type SimpleStreamOptions,
@@ -11,13 +12,31 @@ import {
 import { parseAssistantMessageError } from '../../../src/infrastructure/delegation/provider-error.js';
 import {
   buildDelegationContext,
+  forwardDelegatedEvent,
   modelToExecutionModel,
+  pushFailoverNotice,
   resolveDelegationOptions,
   type DelegatedStreamResult,
   type DelegationHeadroomContext,
+  type FailoverNoticeInfo,
+  type FlushDelegatedEventsOptions,
 } from './delegation-runtime.js';
 import type { StreamDelegationDeps } from './types.js';
+import { throwIfAborted } from './utils.js';
 
+function isTerminalEvent(
+  event: AssistantMessageEvent,
+): event is Extract<AssistantMessageEvent, { type: 'done' | 'error' }> {
+  return event.type === 'done' || event.type === 'error';
+}
+
+/**
+ * Buffer the full inner stream (no outer push).
+ *
+ * Used by the planning-delegate ephemeral sub-call: only the final observation
+ * text is injected into primary context — intermediate tokens must not reach the
+ * user-facing outer stream (SP-170: planning stays buffered by design).
+ */
 export async function collectDelegatedStream(
   targetModel: Model<Api>,
   context: Context,
@@ -25,9 +44,7 @@ export async function collectDelegatedStream(
   options: SimpleStreamOptions | undefined,
   headroomContext?: DelegationHeadroomContext,
 ): Promise<DelegatedStreamResult> {
-  if (options?.signal?.aborted) {
-    throw new Error('Request was aborted');
-  }
+  throwIfAborted(options);
 
   const delegationOptions = await resolveDelegationOptions(
     deps.modelRegistry,
@@ -41,9 +58,7 @@ export async function collectDelegatedStream(
   let finalMessage: AssistantMessage | undefined;
 
   for await (const event of inner) {
-    if (options?.signal?.aborted) {
-      throw new Error('Request was aborted');
-    }
+    throwIfAborted(options);
     events.push(event);
 
     if (event.type === 'done') {
@@ -60,31 +75,145 @@ export async function collectDelegatedStream(
   return { finalMessage, failed, events };
 }
 
-export async function delegateWithOutcome(
+export interface PipeDelegatedStreamOptions extends FlushDelegatedEventsOptions {
+  readonly outer: AssistantMessageEventStream;
+  /**
+   * When set, push a synthetic failover `text_delta` immediately after the first
+   * live `start` event (before further retry tokens).
+   */
+  readonly failoverNotice?: FailoverNoticeInfo;
+  /**
+   * When true (default), forward non-terminal events live and hold `done`/`error`
+   * until {@link commitPipedTerminal}. Set false to buffer only (no outer push).
+   */
+  readonly live?: boolean;
+}
+
+export interface PipedDelegatedStreamResult extends DelegatedStreamResult {
+  /** Terminal event held back so callers can decide failover before commit. */
+  readonly heldTerminal: AssistantMessageEvent | undefined;
+  readonly flushOptions: FlushDelegatedEventsOptions;
+  readonly outer: AssistantMessageEventStream | undefined;
+}
+
+/**
+ * Live-pipe provider events to `outer` as they arrive (SP-170).
+ *
+ * Non-terminal events (`start`, `text_delta`, …) are forwarded immediately so the
+ * UI is not frozen. Terminal `done`/`error` are held until the caller commits or
+ * discards them (failover discards without ending the outer stream).
+ */
+export async function pipeDelegatedStream(
   targetModel: Model<Api>,
   context: Context,
   deps: StreamDelegationDeps,
   options: SimpleStreamOptions | undefined,
-  sessionId: string | undefined,
-  headroomContext?: DelegationHeadroomContext,
-): Promise<DelegatedStreamResult> {
-  const delegationContext = buildDelegationContext(
-    context,
-    targetModel,
-    deps,
-    sessionId,
-  );
+  headroomContext: DelegationHeadroomContext | undefined,
+  pipe: PipeDelegatedStreamOptions,
+): Promise<PipedDelegatedStreamResult> {
+  throwIfAborted(options);
 
-  const result = await collectDelegatedStream(
+  const delegationOptions = await resolveDelegationOptions(
+    deps.modelRegistry,
     targetModel,
-    delegationContext,
-    deps,
     options,
     headroomContext,
   );
+  const delegateStream = deps.delegateStream ?? defaultDelegateStream;
+  const inner = delegateStream(targetModel, context, delegationOptions);
+  const events: AssistantMessageEvent[] = [];
+  let finalMessage: AssistantMessage | undefined;
+  let heldTerminal: AssistantMessageEvent | undefined;
+  let noticePushed = false;
+  const live = pipe.live !== false;
+  const flushOptions: FlushDelegatedEventsOptions = {
+    ...(pipe.sanitizeErrors !== undefined
+      ? { sanitizeErrors: pipe.sanitizeErrors }
+      : {}),
+    ...(pipe.contextWindow !== undefined
+      ? { contextWindow: pipe.contextWindow }
+      : {}),
+  };
+
+  for await (const event of inner) {
+    throwIfAborted(options);
+    events.push(event);
+
+    if (event.type === 'done') {
+      finalMessage = event.message;
+      heldTerminal = event;
+      continue;
+    }
+    if (event.type === 'error') {
+      finalMessage = event.error;
+      heldTerminal = event;
+      continue;
+    }
+
+    if (!live) {
+      continue;
+    }
 
+    forwardDelegatedEvent(pipe.outer, event, flushOptions);
+
+    if (
+      !noticePushed &&
+      pipe.failoverNotice &&
+      event.type === 'start'
+    ) {
+      pushFailoverNotice(pipe.outer, pipe.failoverNotice, event.partial);
+      noticePushed = true;
+    }
+  }
+
+  // Error-only streams never emit `start` — still surface the notice before commit
+  // when the caller is about to show a successful retry that also lacked start
+  // (handled by commit path via leftover failoverNotice on next pipe call).
+
+  const failed =
+    finalMessage !== undefined &&
+    (finalMessage.stopReason === 'error' || finalMessage.stopReason === 'aborted');
+
+  return {
+    finalMessage,
+    failed,
+    events,
+    heldTerminal,
+    flushOptions,
+    outer: pipe.outer,
+  };
+}
+
+/** Forward a held terminal event and end the outer stream. */
+export function commitPipedTerminal(
+  result: PipedDelegatedStreamResult,
+  overrides?: FlushDelegatedEventsOptions,
+): void {
+  const outer = result.outer;
+  if (!outer) {
+    return;
+  }
+  const opts = { ...result.flushOptions, ...overrides };
+  if (result.heldTerminal) {
+    forwardDelegatedEvent(outer, result.heldTerminal, opts);
+  }
+  const endMessage =
+    result.heldTerminal && isTerminalEvent(result.heldTerminal)
+      ? result.heldTerminal.type === 'done'
+        ? result.heldTerminal.message
+        : result.heldTerminal.error
+      : result.finalMessage;
+  outer.end(endMessage);
+}
+
+function recordDelegateOutcome(
+  targetModel: Model<Api>,
+  deps: StreamDelegationDeps,
+  sessionId: string | undefined,
+  result: DelegatedStreamResult,
+): void {
   if (!result.finalMessage) {
-    return result;
+    return;
   }
 
   if (result.failed) {
@@ -100,6 +229,46 @@ export async function delegateWithOutcome(
       id: targetModel.id,
     });
   }
+}
+
+/**
+ * Delegate with outcome recording. When `pipe` is provided, live-forwards to outer
+ * (holding the terminal event). Otherwise collects into a buffer (planning / probes).
+ */
+export async function delegateWithOutcome(
+  targetModel: Model<Api>,
+  context: Context,
+  deps: StreamDelegationDeps,
+  options: SimpleStreamOptions | undefined,
+  sessionId: string | undefined,
+  headroomContext?: DelegationHeadroomContext,
+  pipe?: PipeDelegatedStreamOptions,
+): Promise<PipedDelegatedStreamResult | DelegatedStreamResult> {
+  const delegationContext = buildDelegationContext(
+    context,
+    targetModel,
+    deps,
+    sessionId,
+  );
+
+  const result = pipe
+    ? await pipeDelegatedStream(
+        targetModel,
+        delegationContext,
+        deps,
+        options,
+        headroomContext,
+        pipe,
+      )
+    : await collectDelegatedStream(
+        targetModel,
+        delegationContext,
+        deps,
+        options,
+        headroomContext,
+      );
+
+  recordDelegateOutcome(targetModel, deps, sessionId, result);
 
   return result;
 }

package/.pi/extensions/smart-router/delegation-runtime.ts

@@ -271,6 +271,85 @@ export interface FlushDelegatedEventsOptions {
   readonly contextWindow?: number;
 }
 
+export interface FailoverNoticeInfo {
+  readonly failedModelId: string;
+  readonly alternateModelId: string;
+  readonly errorObj?: ReturnType<typeof parseAssistantMessageError>;
+}
+
+/** Build the operator-visible failover notice string. */
+export function buildFailoverNoticeText(
+  failedModelId: string,
+  alternateModelId: string,
+  errorObj?: ReturnType<typeof parseAssistantMessageError>,
+): string {
+  const reason = errorObj?.message || errorObj?.code || 'Unavailable';
+  return (
+    `> ⚠️ **pi-smart-router failover:** \`${failedModelId}\` failed (${reason}). ` +
+    `Retrying with \`${alternateModelId}\`...`
+  );
+}
+
+/**
+ * Push a synthetic failover `text_delta` to the live outer stream (SP-170).
+ * Call after the retry stream's `start` (or with a start partial) — no buffered-array mutation.
+ */
+export function pushFailoverNotice(
+  outer: AssistantMessageEventStream,
+  notice: FailoverNoticeInfo,
+  startPartial: AssistantMessage,
+): void {
+  const text = buildFailoverNoticeText(
+    notice.failedModelId,
+    notice.alternateModelId,
+    notice.errorObj,
+  );
+  const delta = `${text}\n\n`;
+  const partial: AssistantMessage = {
+    ...startPartial,
+    content: [{ type: 'text', text: delta }, ...startPartial.content],
+  };
+  outer.push({
+    type: 'text_delta',
+    contentIndex: 0,
+    delta,
+    partial,
+  });
+}
+
+/**
+ * Sanitize a single delegated event for live forwarding (length-stop + optional error UX).
+ */
+export function sanitizeDelegatedEvent(
+  event: AssistantMessageEvent,
+  options?: FlushDelegatedEventsOptions,
+): AssistantMessageEvent {
+  const lengthStopHints =
+    options?.contextWindow !== undefined
+      ? { contextWindow: options.contextWindow }
+      : undefined;
+  const events = [event];
+  sanitizeLengthStopEvents(events, lengthStopHints);
+  if (options?.sanitizeErrors) {
+    sanitizeDelegatedErrorEvents(events, lengthStopHints);
+  }
+  return events[0]!;
+}
+
+/** Push one sanitized event to the outer live stream (does not end the stream). */
+export function forwardDelegatedEvent(
+  outer: AssistantMessageEventStream,
+  event: AssistantMessageEvent,
+  options?: FlushDelegatedEventsOptions,
+): void {
+  outer.push(sanitizeDelegatedEvent(event, options));
+}
+
+/**
+ * Flush a buffered event list to outer and end the stream.
+ * Prefer live `forwardDelegatedEvent` / pipe paths for primary delegation (SP-170).
+ * Still used for terminal error-only buffers and legacy collect paths.
+ */
 export function flushDelegatedEvents(
   outer: AssistantMessageEventStream,
   events: readonly AssistantMessageEvent[],
@@ -291,15 +370,17 @@ export function flushDelegatedEvents(
   outer.end();
 }
 
+/**
+ * @deprecated SP-170 — prefer `pushFailoverNotice` on the live outer stream.
+ * Mutates a buffered event array (legacy collect-then-flush path only).
+ */
 export function injectFailoverNotice(
   events: AssistantMessageEvent[],
   failedModelId: string,
   alternateModelId: string,
   errorObj?: ReturnType<typeof parseAssistantMessageError>,
 ): void {
-  const reason = errorObj?.message || errorObj?.code || 'Unavailable';
-  const notice = `> ⚠️ **pi-smart-router failover:** \`${failedModelId}\` failed (${reason}). Retrying with \`${alternateModelId}\`...`;
-
+  const notice = buildFailoverNoticeText(failedModelId, alternateModelId, errorObj);
   let noticeInjected = false;
 
   for (let i = 0; i < events.length; i++) {

package/.pi/extensions/smart-router/planning-delegate.ts

@@ -32,6 +32,7 @@ import {
 import { collectDelegatedStream } from './delegate-stream.js';
 import { findFleetProfile, resolveRegistryModel } from './delegation-runtime.js';
 import type { StreamDelegationDeps } from './types.js';
+import { throwIfAborted } from './utils.js';
 
 /** Prefix for injected planning observations visible to the primary model. */
 export const PLANNING_DELEGATE_OBSERVATION_PREFIX =
@@ -169,7 +170,12 @@ export function injectPlanningDelegateObservation(
   };
 }
 
-/** Default frontier sub-call via provider stream (ephemeral one-shot delegate). */
+/** Default frontier sub-call via provider stream (ephemeral one-shot delegate).
+ *
+ * SP-170: intentionally uses collectDelegatedStream (buffered), not live outer
+ * piping. Only the final observation text is injected into the primary context;
+ * intermediate frontier tokens must not reach the user-facing stream.
+ */
 export async function defaultSpawnPlanningDelegate(
   frontierModel: Model<Api>,
   compressedContext: Context,
@@ -225,6 +231,9 @@ export async function resolvePlanningDelegatePath(
   options: SimpleStreamOptions | undefined,
   deps: StreamDelegationDeps,
 ): Promise<PlanningDelegateResolution> {
+  // Abort before planning sub-call work (SP-171 phase boundary).
+  throwIfAborted(options);
+
   const observability = decision.features!.planning_delegate!;
   const delegateModelId = observability.delegate_model_id!;
   const primaryModelId = decision.selected_model_id;
@@ -252,6 +261,7 @@ export async function resolvePlanningDelegatePath(
     context,
     observability.compressed_context,
   );
+  throwIfAborted(options);
   const spawnFn = deps.spawnPlanningDelegate ?? defaultSpawnPlanningDelegate;
   const spawnResult = await spawnFn(frontierModel, compressedContext, options, deps);
 

package/.pi/extensions/smart-router/route-and-delegate.ts

@@ -25,11 +25,16 @@ import {
   parseAssistantMessageError,
 } from '../../../src/infrastructure/delegation/provider-error.js';
 import { shouldFailoverOnProviderError } from '../../../src/infrastructure/gateway/gateway-dispatch.js';
-import { delegateWithOutcome } from './delegate-stream.js';
 import {
+  commitPipedTerminal,
+  delegateWithOutcome,
+  type PipedDelegatedStreamResult,
+} from './delegate-stream.js';
+import {
+  createErrorMessage,
   findFleetProfile,
   flushDelegatedEvents,
-  injectFailoverNotice,
+  type FailoverNoticeInfo,
   resolveRegistryModel,
 } from './delegation-runtime.js';
 import { buildRoutingRequest } from './routing-context.js';
@@ -39,6 +44,13 @@ import {
   resolvePlanningDelegatePath,
 } from './planning-delegate.js';
 import type { StreamDelegationDeps } from './types.js';
+import { isAbortError, throwIfAborted } from './utils.js';
+
+function isPipedResult(
+  result: Awaited<ReturnType<typeof delegateWithOutcome>>,
+): result is PipedDelegatedStreamResult {
+  return 'heldTerminal' in result;
+}
 
 function isRoutingLogEnabled(): boolean {
   return process.env.SMART_ROUTER_LOG_ROUTING === '1';
@@ -222,6 +234,10 @@ function resolveHeadroomFallbackTarget(
 /**
  * Route a request and delegate to the selected provider with failover.
  * Kept as one module to preserve the atomic failover state machine (#33).
+ *
+ * Abort checks run at phase boundaries (entry, fleet refresh, dispatch,
+ * planning delegate, each failover iteration). HyDRA/ONNX embedding inference
+ * cannot cancel mid-run — fail-fast only before/after that stage (SP-171).
  */
 export async function routeAndDelegate(
   context: Context,
@@ -229,8 +245,12 @@ export async function routeAndDelegate(
   deps: StreamDelegationDeps,
   outer: AssistantMessageEventStream,
 ): Promise<void> {
+  // Phase boundary: abort before any long work (fleet refresh, HyDRA dispatch, delegate).
+  throwIfAborted(options);
+
   const sessionId = options?.sessionId;
   if (deps.ensureFleetFresh) {
+    throwIfAborted(options);
     await deps.ensureFleetFresh();
   }
   let decision: RoutingDecision;
@@ -264,8 +284,14 @@ export async function routeAndDelegate(
       );
     }
     capturePreRouteOutcomes(request, deps, priorSnapshot, hadPin);
+    // Phase boundary: abort before HyDRA/dispatch (mid-ONNX cancel unsupported).
+    throwIfAborted(options);
     decision = await deps.router.dispatch.dispatch(request, { effectiveFleet });
   } catch (error) {
+    // Do not treat abort as a routing failure — never failover on cancel.
+    if (isAbortError(error, options)) {
+      throw error;
+    }
     const fallbackModel = resolveFallbackModel(deps, effectiveFleet);
     if (!fallbackModel) {
       throw error;
@@ -280,10 +306,19 @@ export async function routeAndDelegate(
       deps,
       options,
       sessionId,
+      undefined,
+      {
+        outer,
+        contextWindow: fallbackModel.contextWindow,
+      },
     );
-    flushDelegatedEvents(outer, fallbackResult.events, {
-      contextWindow: fallbackModel.contextWindow,
-    });
+    if (isPipedResult(fallbackResult)) {
+      commitPipedTerminal(fallbackResult);
+    } else {
+      flushDelegatedEvents(outer, fallbackResult.events, {
+        contextWindow: fallbackModel.contextWindow,
+      });
+    }
     return;
   }
 
@@ -294,6 +329,8 @@ export async function routeAndDelegate(
   let delegationContext: Context = context;
 
   if (isPlanningDelegateActive(decision)) {
+    // Phase boundary: abort before planning-delegate sub-call.
+    throwIfAborted(options);
     const planningResolution = await resolvePlanningDelegatePath(
       context,
       decision,
@@ -335,13 +372,11 @@ export async function routeAndDelegate(
   const headroomExcludedModelIds: string[] = [];
   const estimatedInputTokens =
     request.estimated_input_tokens ?? request.prompt_text.length;
-  let pendingFailoverInfo: {
-    failedModelId: string;
-    alternateModelId: string;
-    errorObj?: ReturnType<typeof parseAssistantMessageError>;
-  } | undefined;
+  let pendingFailoverInfo: FailoverNoticeInfo | undefined;
 
   while (true) {
+    // Phase boundary: abort before each failover / delegation attempt.
+    throwIfAborted(options);
     try {
       const targetProfile =
         findFleetProfile(effectiveFleet, targetModel.id) ??
@@ -392,6 +427,9 @@ export async function routeAndDelegate(
         ? { profile: targetProfile, estimatedInputTokens }
         : undefined;
 
+      const failoverNotice = pendingFailoverInfo;
+      pendingFailoverInfo = undefined;
+
       const result = await delegateWithOutcome(
         targetModel,
         delegationContext,
@@ -399,16 +437,19 @@ export async function routeAndDelegate(
         options,
         sessionId,
         headroomContext,
+        {
+          outer,
+          ...(failoverNotice !== undefined ? { failoverNotice } : {}),
+          contextWindow: targetModel.contextWindow,
+        },
       );
 
-      if (pendingFailoverInfo) {
-        injectFailoverNotice(
-          result.events,
-          pendingFailoverInfo.failedModelId,
-          pendingFailoverInfo.alternateModelId,
-          pendingFailoverInfo.errorObj,
-        );
-        pendingFailoverInfo = undefined;
+      if (!isPipedResult(result)) {
+        flushDelegatedEvents(outer, result.events, {
+          sanitizeErrors: result.failed,
+          contextWindow: targetModel.contextWindow,
+        });
+        return;
       }
 
       if (result.finalMessage && isZeroOutputLengthStop(result.finalMessage)) {
@@ -453,10 +494,7 @@ export async function routeAndDelegate(
         result.finalMessage &&
         isGeminiThoughtSignatureAssistantError(result.finalMessage)
       ) {
-        flushDelegatedEvents(outer, result.events, {
-          sanitizeErrors: true,
-          contextWindow: targetModel.contextWindow,
-        });
+        commitPipedTerminal(result, { sanitizeErrors: true });
         return;
       }
 
@@ -475,19 +513,13 @@ export async function routeAndDelegate(
             effectiveFleet,
           );
           if (!failover) {
-            flushDelegatedEvents(outer, result.events, {
-          sanitizeErrors: true,
-          contextWindow: targetModel.contextWindow,
-        });
+            commitPipedTerminal(result, { sanitizeErrors: true });
             return;
           }
 
           const alternateModel = resolveTargetModel(deps, failover);
           if (!alternateModel || alternateModel.id === targetModel.id) {
-            flushDelegatedEvents(outer, result.events, {
-          sanitizeErrors: true,
-          contextWindow: targetModel.contextWindow,
-        });
+            commitPipedTerminal(result, { sanitizeErrors: true });
             return;
           }
 
@@ -495,6 +527,7 @@ export async function routeAndDelegate(
             '[smart-router] infra error, failing over to alternate model',
             alternateModel.id,
           );
+          // Discard held terminal from the failed attempt — do not forward to outer.
           pendingFailoverInfo = {
             failedModelId: targetModel.id,
             alternateModelId: alternateModel.id,
@@ -506,12 +539,19 @@ export async function routeAndDelegate(
         }
       }
 
-      flushDelegatedEvents(outer, result.events, {
+      commitPipedTerminal(result, {
         sanitizeErrors: result.failed,
         contextWindow: targetModel.contextWindow,
       });
       return;
     } catch (error) {
+      if (isAbortError(error, options)) {
+        const abortMessage = createErrorMessage(targetModel, options, error);
+        outer.push({ type: 'error', reason: 'aborted', error: abortMessage });
+        outer.end(abortMessage);
+        return;
+      }
+
       deps.router.dispatch.recordOutcome(targetModel.id, { code: 'STREAM_DELEGATION_ERROR' });
 
       if (!failedModelIds.includes(targetModel.id)) {
@@ -558,24 +598,28 @@ export async function routeAndDelegate(
         errorObj: { message: error instanceof Error ? error.message : String(error) },
       };
 
+      const failoverNotice = pendingFailoverInfo;
+      pendingFailoverInfo = undefined;
       const fallbackResult = await delegateWithOutcome(
         fallbackModel,
         context,
         deps,
         options,
         sessionId,
+        undefined,
+        {
+          outer,
+          ...(failoverNotice !== undefined ? { failoverNotice } : {}),
+          contextWindow: fallbackModel.contextWindow,
+        },
       );
-      if (pendingFailoverInfo) {
-        injectFailoverNotice(
-          fallbackResult.events,
-          pendingFailoverInfo.failedModelId,
-          pendingFailoverInfo.alternateModelId,
-          pendingFailoverInfo.errorObj,
-        );
+      if (isPipedResult(fallbackResult)) {
+        commitPipedTerminal(fallbackResult);
+      } else {
+        flushDelegatedEvents(outer, fallbackResult.events, {
+          contextWindow: fallbackModel.contextWindow,
+        });
       }
-      flushDelegatedEvents(outer, fallbackResult.events, {
-      contextWindow: fallbackModel.contextWindow,
-    });
       return;
     }
   }

package/.pi/extensions/smart-router/utils.ts

@@ -56,3 +56,32 @@ export function resolveRateLimiter(store: StorePort): RateLimitPort | undefined
   }
   return createSqliteRateLimiter(store);
 }
+
+/** True when the request was aborted via AbortSignal or an abort-shaped error. */
+export function isAbortError(
+  error: unknown,
+  options?: { signal?: AbortSignal },
+): boolean {
+  if (options?.signal?.aborted) {
+    return true;
+  }
+  if (error instanceof DOMException && error.name === 'AbortError') {
+    return true;
+  }
+  if (error instanceof Error) {
+    if (error.name === 'AbortError') {
+      return true;
+    }
+    if (error.message === 'Request was aborted') {
+      return true;
+    }
+  }
+  return false;
+}
+
+/** Throw if `options.signal` is already aborted. */
+export function throwIfAborted(options?: { signal?: AbortSignal }): void {
+  if (options?.signal?.aborted) {
+    throw new Error('Request was aborted');
+  }
+}

package/README.md

@@ -383,6 +383,8 @@ When a **planning** turn would route primary inference to frontier while a warm
 2. **Pi extension** (`.pi/extensions/smart-router`) runs an ephemeral frontier sub-call with compressed context (tool execution history excluded by default), injects the result as an observation user message, then delegates **primary** streaming to the pinned economical model.
 3. **Fallback** — when delegate is disabled, spawn fails, or the delegate model is missing from the registry, the extension falls back to a **direct frontier** route with a documented `fallback_reason` in explain/telemetry.
 
+**Stream piping (SP-170):** Primary delegated inference **live-forwards** provider events to pi (`start` / `text_delta` / … as they arrive). The planning-delegate sub-call stays **buffered** — only the final observation text is injected into primary context; frontier tokens from the ephemeral sub-call are discarded and never reach the user-facing stream. On infra failover, a synthetic `text_delta` notice is pushed after the retry stream's `start` (no mutation of a buffered event array).
+
 | Knob | Env var | Default | Effect |
 |------|---------|---------|--------|
 | Delegate enabled | `SMART_ROUTER_PLANNING_DELEGATE_ENABLED` | `true` | When `false`, SAAR buffer allows direct frontier planning (`planning_direct_frontier` + `planning_delegate_disabled`) |
@@ -580,6 +582,8 @@ Exit code `2` signals regression above threshold — operators can wire this int
 
 The embedding matcher uses `@huggingface/transformers` with ONNX models (384-dim embeddings). Artifacts are downloaded at runtime and cached under `.pi-smart-router/models/` (configurable via `hydra.artifact_cache_path`). This directory is gitignored.
 
+**Abort / cancel limitation (SP-171):** `AbortSignal` is checked at phase boundaries before fleet refresh, HyDRA/dispatch, planning delegate, and each failover iteration. Mid-ONNX embedding inference cannot be cancelled — abort is fail-fast only before or after that stage, not during an in-flight ONNX run.
+
 | Encoder | Model | Context | Default |
 |---------|-------|---------|---------|
 | `minilm` | `Xenova/all-MiniLM-L6-v2` | 512 tokens | yes |
@@ -729,7 +733,8 @@ Contributors must run `npm run build` before publishing or consuming the library
 | Script | Purpose |
 |--------|---------|
 | `npm run build` | Compile library to `dist/` (`tsc --project tsconfig.build.json`) |
-| `npm run release:check` | Pre-release gate: `verify:ci` + consumer pack (`npm install --omit=dev` on tarball) |
+| `npm run release:check` | Pre-release gate: `verify:ci` + consumer pack + Tier 0 functional smoke |
+| `npm run release:functional-smoke` | Tier 0 functional smoke: calibration verify (`--skip-embed`), benchmark profiles, release gate assertions |
 | `npm run release:consumer-pack` | Pack tarball and verify production dependencies resolve (catches missing runtime deps) |
 | `npm run verify:ci` | Full CI parity: build, typecheck, lint, test, coverage |
 | `npm run typecheck` | TypeScript strict mode check (`tsc --noEmit`) |
@@ -794,10 +799,29 @@ npm run routing:verify-benchmark-profiles
 
 Tag-triggered publish via GitHub Actions (requires `NPMSECRET` repository secret). pi.dev gallery listing syncs automatically from npm (`pi-package` keyword); no separate submit step.
 
-1. `npm run release:check` (includes consumer pack verify)
+**Tier 0 functional smoke** (`release:functional-smoke`) runs before tag publish and chains:
+
+1. `routing:verify-calibration --skip-embed` — artifact shape + triage benchmark gates (no ONNX embedding)
+2. `routing:verify-benchmark-profiles` — checked-in capability profiles match fixture ingest
+3. `assert-release-gates --fixtures tests/eval/fixtures --baseline-version 0.6.0` — eval harness aggregate metrics vs `config/release-gates.json` and semver baseline regression vs `tests/eval/baselines/v0.6.0.json`
+
+`release:check` runs the full pre-release path: `verify:ci`, consumer pack verify, then Tier 0 functional smoke.
+
+**Baseline re-capture (post-tag):** after shipping a new semver (e.g. v0.7.0), freeze harness metrics for the next regression reference:
+
+```bash
+# Capture aggregate metrics from current fixtures (writes tests/eval/baselines/v0.7.0.json)
+npm run routing:capture-baseline -- --version 0.7.0
+
+# Point release gates at the new reference (config/release-gates.json + release:functional-smoke --baseline-version)
+```
+
+Commit the new baseline JSON and update `baseline_regression.reference_version` in `config/release-gates.json` plus the `--baseline-version` flag in `release:functional-smoke`. Re-run `npm run release:check` before tagging the next release.
+
+1. `npm run release:check` (CI parity + consumer pack + Tier 0 functional smoke)
 2. `npm version 0.1.1` (creates commit + `v0.1.1` tag)
 3. `git push && git push --tags`
-4. Actions → **Release** runs pack smoke, consumer pack verify, `npm publish`, and creates a GitHub Release
+4. Actions → **Release** runs pack smoke, consumer pack verify, Tier 0 functional smoke, `npm publish`, and creates a GitHub Release
 
 Re-publish a failed release: Actions → Release → Run workflow with existing tag (e.g. `v0.1.1`).
 

package/config/release-gates.json

@@ -0,0 +1,16 @@
+{
+  "version": 1,
+  "absolute_gates": {
+    "mean_capability_adequacy_rate_min": 0.85,
+    "mean_quality_retention_min": 0.7,
+    "mean_over_routing_rate_max": 0.15,
+    "mean_pin_preserved_rate_min": 0.6
+  },
+  "baseline_regression": {
+    "reference_version": "0.6.0",
+    "max_quality_retention_drop": 0.05,
+    "max_capability_adequacy_rate_drop": 0.05,
+    "max_pin_preserved_rate_drop": 0.05,
+    "max_over_routing_rate_increase": 0.05
+  }
+}

package/dist/infrastructure/pricing/litellm-fetch.d.ts

@@ -16,6 +16,8 @@ export interface LitellmFetchResult {
 export interface LitellmFetchDeps {
     readonly fetchFn?: typeof fetch;
     readonly pricingUrl?: string;
+    /** When set, aborts the LiteLLM pricing HTTP request (ESC / ctx.signal). */
+    readonly signal?: AbortSignal;
 }
 export declare class LitellmFetchError extends Error {
     readonly name = "LitellmFetchError";

package/dist/infrastructure/pricing/litellm-fetch.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"litellm-fetch.d.ts","sourceRoot":"","sources":["../../../src/infrastructure/pricing/litellm-fetch.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,6BAA6B,CAAC;AAE7E,eAAO,MAAM,2BAA2B,gGACuD,CAAC;AAIhG,MAAM,WAAW,kBAAkB;IACjC,QAAQ,CAAC,iBAAiB,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC;IAC7D,QAAQ,CAAC,wBAAwB,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC,CAAC;IACzE,+EAA+E;IAC/E,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;CAC9B;AAED,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,OAAO,CAAC,EAAE,OAAO,KAAK,CAAC;IAChC,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;CAC9B;AAED,qBAAa,iBAAkB,SAAQ,KAAK;IAC1C,SAAkB,IAAI,uBAAuB;CAC9C;AAED;;GAEG;AACH,wBAAgB,oBAAoB,CAAC,GAAG,GAAE,MAAM,CAAC,UAAwB,GAAG,MAAM,CAGjF;AAED;;;GAGG;AACH,wBAAgB,sBAAsB,CACpC,iBAAiB,EAAE,MAAM,EACzB,kBAAkB,EAAE,MAAM,GACzB,MAAM,CAIR;AA4CD;;;GAGG;AACH,wBAAgB,uBAAuB,CAAC,GAAG,EAAE,OAAO,GAAG,kBAAkB,CA8DxE;AAED,MAAM,WAAW,yBAAyB;IACxC,QAAQ,CAAC,OAAO,EAAE,YAAY,CAAC;IAC/B,sEAAsE;IACtE,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;CAC9B;AAED;;GAEG;AACH,wBAAsB,wBAAwB,CAC5C,IAAI,GAAE,gBAAqB,GAC1B,OAAO,CAAC,yBAAyB,CAAC,CAyCpC"}
+{"version":3,"file":"litellm-fetch.d.ts","sourceRoot":"","sources":["../../../src/infrastructure/pricing/litellm-fetch.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,6BAA6B,CAAC;AAE7E,eAAO,MAAM,2BAA2B,gGACuD,CAAC;AAIhG,MAAM,WAAW,kBAAkB;IACjC,QAAQ,CAAC,iBAAiB,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC;IAC7D,QAAQ,CAAC,wBAAwB,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC,CAAC;IACzE,+EAA+E;IAC/E,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;CAC9B;AAED,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,OAAO,CAAC,EAAE,OAAO,KAAK,CAAC;IAChC,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAC7B,4EAA4E;IAC5E,QAAQ,CAAC,MAAM,CAAC,EAAE,WAAW,CAAC;CAC/B;AAkBD,qBAAa,iBAAkB,SAAQ,KAAK;IAC1C,SAAkB,IAAI,uBAAuB;CAC9C;AAED;;GAEG;AACH,wBAAgB,oBAAoB,CAAC,GAAG,GAAE,MAAM,CAAC,UAAwB,GAAG,MAAM,CAGjF;AAED;;;GAGG;AACH,wBAAgB,sBAAsB,CACpC,iBAAiB,EAAE,MAAM,EACzB,kBAAkB,EAAE,MAAM,GACzB,MAAM,CAIR;AA4CD;;;GAGG;AACH,wBAAgB,uBAAuB,CAAC,GAAG,EAAE,OAAO,GAAG,kBAAkB,CA8DxE;AAED,MAAM,WAAW,yBAAyB;IACxC,QAAQ,CAAC,OAAO,EAAE,YAAY,CAAC;IAC/B,sEAAsE;IACtE,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;CAC9B;AAED;;GAEG;AACH,wBAAsB,wBAAwB,CAC5C,IAAI,GAAE,gBAAqB,GAC1B,OAAO,CAAC,yBAAyB,CAAC,CA4DpC"}

package/dist/infrastructure/pricing/litellm-fetch.js

@@ -7,6 +7,20 @@
  */
 export const DEFAULT_LITELLM_PRICING_URL = 'https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json';
 const CHAT_MODES = new Set(['chat', 'completion']);
+function throwIfAborted(signal) {
+    if (signal?.aborted) {
+        const reason = signal.reason;
+        if (reason instanceof Error) {
+            throw reason;
+        }
+        const error = new Error(typeof reason === 'string' && reason.length > 0 ? reason : 'Aborted');
+        error.name = 'AbortError';
+        throw error;
+    }
+}
+function isAbortError(error) {
+    return error instanceof Error && error.name === 'AbortError';
+}
 export class LitellmFetchError extends Error {
     name = 'LitellmFetchError';
 }
@@ -112,14 +126,23 @@ export function normalizeLitellmPricing(raw) {
 export async function fetchLitellmPriceCatalog(deps = {}) {
     const fetchFn = deps.fetchFn ?? fetch;
     const url = deps.pricingUrl ?? getLitellmPricingUrl();
+    const { signal } = deps;
+    throwIfAborted(signal);
     let response;
     try {
-        response = await fetchFn(url);
+        response = signal ? await fetchFn(url, { signal }) : await fetchFn(url);
     }
     catch (error) {
+        if (isAbortError(error) || signal?.aborted) {
+            throwIfAborted(signal);
+            if (isAbortError(error)) {
+                throw error;
+            }
+        }
         const detail = error instanceof Error ? error.message : String(error);
         throw new LitellmFetchError(`Failed to fetch LiteLLM pricing from ${url}: ${detail}`);
     }
+    throwIfAborted(signal);
     if (!response.ok) {
         throw new LitellmFetchError(`LiteLLM pricing fetch failed (${response.status} ${response.statusText}) from ${url}`);
     }
@@ -127,9 +150,16 @@ export async function fetchLitellmPriceCatalog(deps = {}) {
     try {
         raw = await response.json();
     }
-    catch {
+    catch (error) {
+        if (isAbortError(error) || signal?.aborted) {
+            throwIfAborted(signal);
+            if (isAbortError(error)) {
+                throw error;
+            }
+        }
         throw new LitellmFetchError(`LiteLLM pricing response from ${url} is not valid JSON.`);
     }
+    throwIfAborted(signal);
     const normalized = normalizeLitellmPricing(raw);
     return {
         catalog: {

package/dist/infrastructure/pricing/litellm-fetch.js.map

@@ -1 +1 @@
-{"version":3,"file":"litellm-fetch.js","sourceRoot":"","sources":["../../../src/infrastructure/pricing/litellm-fetch.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAIH,MAAM,CAAC,MAAM,2BAA2B,GACtC,6FAA6F,CAAC;AAEhG,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC,CAAC;AAcnD,MAAM,OAAO,iBAAkB,SAAQ,KAAK;IACxB,IAAI,GAAG,mBAAmB,CAAC;CAC9C;AAED;;GAEG;AACH,MAAM,UAAU,oBAAoB,CAAC,MAAyB,OAAO,CAAC,GAAG;IACvE,MAAM,UAAU,GAAG,GAAG,CAAC,mBAAmB,EAAE,IAAI,EAAE,CAAC;IACnD,OAAO,UAAU,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,2BAA2B,CAAC;AACxF,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,sBAAsB,CACpC,iBAAyB,EACzB,kBAA0B;IAE1B,MAAM,UAAU,GAAG,iBAAiB,GAAG,SAAS,CAAC;IACjD,MAAM,WAAW,GAAG,kBAAkB,GAAG,SAAS,CAAC;IACnD,OAAO,CAAC,UAAU,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;AACxC,CAAC;AAED,SAAS,QAAQ,CAAC,KAAc;IAC9B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;AAC9E,CAAC;AAED,SAAS,eAAe,CAAC,KAA8B,EAAE,KAAa;IACpE,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;IAC3B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AACjF,CAAC;AAED,SAAS,eAAe,CAAC,KAA8B,EAAE,KAAa;IACpE,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;IAC3B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;AACzF,CAAC;AAED,SAAS,oBAAoB,CAAC,KAA8B,EAAE,KAAa;IACzE,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;IAC3B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AAC/F,CAAC;AAED,SAAS,gBAAgB,CAAC,KAA8B;IACtD,MAAM,cAAc,GAAG,oBAAoB,CAAC,KAAK,EAAE,kBAAkB,CAAC,CAAC;IACvE,MAAM,eAAe,GACnB,oBAAoB,CAAC,KAAK,EAAE,mBAAmB,CAAC,IAAI,oBAAoB,CAAC,KAAK,EAAE,YAAY,CAAC,CAAC;IAEhG,IAAI,cAAc,KAAK,SAAS,IAAI,eAAe,KAAK,SAAS,EAAE,CAAC;QAClE,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,OAAO;QACL,GAAG,CAAC,cAAc,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7E,GAAG,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACjF,CAAC;AACJ,CAAC;AAED,SAAS,gBAAgB,CACvB,wBAAqD,EACrD,GAAW,EACX,MAAmB;IAEnB,wBAAwB,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC;AACzC,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,uBAAuB,CAAC,GAAY;IAClD,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;QACnB,MAAM,IAAI,iBAAiB,CACzB,2HAA2H,CAC5H,CAAC;IACJ,CAAC;IAED,MAAM,iBAAiB,GAA2B,EAAE,CAAC;IACrD,MAAM,wBAAwB,GAAgC,EAAE,CAAC;IACjE,IAAI,UAAU,GAAG,CAAC,CAAC;IAEnB,KAAK,MAAM,CAAC,QAAQ,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;QACpD,IAAI,QAAQ,KAAK,aAAa,EAAE,CAAC;YAC/B,SAAS;QACX,CAAC;QAED,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;YACrB,MAAM,IAAI,iBAAiB,CACzB,0BAA0B,QAAQ,iCAAiC,OAAO,KAAK,GAAG,CACnF,CAAC;QACJ,CAAC;QAED,MAAM,IAAI,GAAG,eAAe,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;QAC5C,IAAI,IAAI,KAAK,SAAS,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;YAChD,SAAS;QACX,CAAC;QAED,MAAM,SAAS,GAAG,eAAe,CAAC,KAAK,EAAE,sBAAsB,CAAC,CAAC;QACjE,MAAM,UAAU,GAAG,eAAe,CAAC,KAAK,EAAE,uBAAuB,CAAC,CAAC;QACnE,IAAI,SAAS,KAAK,SAAS,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;YACxD,SAAS;QACX,CAAC;QAED,MAAM,SAAS,GAAG,sBAAsB,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC;QAChE,iBAAiB,CAAC,QAAQ,CAAC,GAAG,SAAS,CAAC;QACxC,UAAU,IAAI,CAAC,CAAC;QAEhB,MAAM,MAAM,GAAG,gBAAgB,CAAC,KAAK,CAAC,CAAC;QACvC,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;YACzB,gBAAgB,CAAC,wBAAwB,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;QAC/D,CAAC;QAED,MAAM,QAAQ,GAAG,eAAe,CAAC,KAAK,EAAE,kBAAkB,CAAC,CAAC;QAC5D,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC3B,iBAAiB,CAAC,GAAG,QAAQ,IAAI,QAAQ,EAAE,CAAC,GAAG,SAAS,CAAC;YACzD,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;gBACzB,gBAAgB,CAAC,wBAAwB,EAAE,GAAG,QAAQ,IAAI,QAAQ,EAAE,EAAE,MAAM,CAAC,CAAC;YAChF,CAAC;QACH,CAAC;IACH,CAAC;IAED,IAAI,MAAM,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAChD,MAAM,IAAI,iBAAiB,CACzB,kFAAkF,CACnF,CAAC;IACJ,CAAC;IAED,OAAO;QACL,iBAAiB;QACjB,wBAAwB;QACxB,WAAW,EAAE,UAAU;KACxB,CAAC;AACJ,CAAC;AAQD;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,wBAAwB,CAC5C,OAAyB,EAAE;IAE3B,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,KAAK,CAAC;IACtC,MAAM,GAAG,GAAG,IAAI,CAAC,UAAU,IAAI,oBAAoB,EAAE,CAAC;IAEtD,IAAI,QAAkB,CAAC;IACvB,IAAI,CAAC;QACH,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC;IAChC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,MAAM,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACtE,MAAM,IAAI,iBAAiB,CACzB,wCAAwC,GAAG,KAAK,MAAM,EAAE,CACzD,CAAC;IACJ,CAAC;IAED,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,MAAM,IAAI,iBAAiB,CACzB,iCAAiC,QAAQ,CAAC,MAAM,IAAI,QAAQ,CAAC,UAAU,UAAU,GAAG,EAAE,CACvF,CAAC;IACJ,CAAC;IAED,IAAI,GAAY,CAAC;IACjB,IAAI,CAAC;QACH,GAAG,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;IAC9B,CAAC;IAAC,MAAM,CAAC;QACP,MAAM,IAAI,iBAAiB,CACzB,iCAAiC,GAAG,qBAAqB,CAC1D,CAAC;IACJ,CAAC;IAED,MAAM,UAAU,GAAG,uBAAuB,CAAC,GAAG,CAAC,CAAC;IAEhD,OAAO;QACL,OAAO,EAAE;YACP,iBAAiB,EAAE,UAAU,CAAC,iBAAiB;YAC/C,wBAAwB,EAAE,UAAU,CAAC,wBAAwB;YAC7D,cAAc,EAAE,EAAE;YAClB,YAAY,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACtC,MAAM,EAAE,UAAU;SACnB;QACD,WAAW,EAAE,UAAU,CAAC,WAAW;KACpC,CAAC;AACJ,CAAC"}
+{"version":3,"file":"litellm-fetch.js","sourceRoot":"","sources":["../../../src/infrastructure/pricing/litellm-fetch.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAIH,MAAM,CAAC,MAAM,2BAA2B,GACtC,6FAA6F,CAAC;AAEhG,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC,CAAC;AAgBnD,SAAS,cAAc,CAAC,MAA+B;IACrD,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;QACpB,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;QAC7B,IAAI,MAAM,YAAY,KAAK,EAAE,CAAC;YAC5B,MAAM,MAAM,CAAC;QACf,CAAC;QACD,MAAM,KAAK,GAAG,IAAI,KAAK,CAAC,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;QAC9F,KAAK,CAAC,IAAI,GAAG,YAAY,CAAC;QAC1B,MAAM,KAAK,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,YAAY,CAAC,KAAc;IAClC,OAAO,KAAK,YAAY,KAAK,IAAI,KAAK,CAAC,IAAI,KAAK,YAAY,CAAC;AAC/D,CAAC;AAED,MAAM,OAAO,iBAAkB,SAAQ,KAAK;IACxB,IAAI,GAAG,mBAAmB,CAAC;CAC9C;AAED;;GAEG;AACH,MAAM,UAAU,oBAAoB,CAAC,MAAyB,OAAO,CAAC,GAAG;IACvE,MAAM,UAAU,GAAG,GAAG,CAAC,mBAAmB,EAAE,IAAI,EAAE,CAAC;IACnD,OAAO,UAAU,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,2BAA2B,CAAC;AACxF,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,sBAAsB,CACpC,iBAAyB,EACzB,kBAA0B;IAE1B,MAAM,UAAU,GAAG,iBAAiB,GAAG,SAAS,CAAC;IACjD,MAAM,WAAW,GAAG,kBAAkB,GAAG,SAAS,CAAC;IACnD,OAAO,CAAC,UAAU,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;AACxC,CAAC;AAED,SAAS,QAAQ,CAAC,KAAc;IAC9B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;AAC9E,CAAC;AAED,SAAS,eAAe,CAAC,KAA8B,EAAE,KAAa;IACpE,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;IAC3B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AACjF,CAAC;AAED,SAAS,eAAe,CAAC,KAA8B,EAAE,KAAa;IACpE,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;IAC3B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;AACzF,CAAC;AAED,SAAS,oBAAoB,CAAC,KAA8B,EAAE,KAAa;IACzE,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;IAC3B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AAC/F,CAAC;AAED,SAAS,gBAAgB,CAAC,KAA8B;IACtD,MAAM,cAAc,GAAG,oBAAoB,CAAC,KAAK,EAAE,kBAAkB,CAAC,CAAC;IACvE,MAAM,eAAe,GACnB,oBAAoB,CAAC,KAAK,EAAE,mBAAmB,CAAC,IAAI,oBAAoB,CAAC,KAAK,EAAE,YAAY,CAAC,CAAC;IAEhG,IAAI,cAAc,KAAK,SAAS,IAAI,eAAe,KAAK,SAAS,EAAE,CAAC;QAClE,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,OAAO;QACL,GAAG,CAAC,cAAc,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7E,GAAG,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACjF,CAAC;AACJ,CAAC;AAED,SAAS,gBAAgB,CACvB,wBAAqD,EACrD,GAAW,EACX,MAAmB;IAEnB,wBAAwB,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC;AACzC,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,uBAAuB,CAAC,GAAY;IAClD,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;QACnB,MAAM,IAAI,iBAAiB,CACzB,2HAA2H,CAC5H,CAAC;IACJ,CAAC;IAED,MAAM,iBAAiB,GAA2B,EAAE,CAAC;IACrD,MAAM,wBAAwB,GAAgC,EAAE,CAAC;IACjE,IAAI,UAAU,GAAG,CAAC,CAAC;IAEnB,KAAK,MAAM,CAAC,QAAQ,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;QACpD,IAAI,QAAQ,KAAK,aAAa,EAAE,CAAC;YAC/B,SAAS;QACX,CAAC;QAED,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;YACrB,MAAM,IAAI,iBAAiB,CACzB,0BAA0B,QAAQ,iCAAiC,OAAO,KAAK,GAAG,CACnF,CAAC;QACJ,CAAC;QAED,MAAM,IAAI,GAAG,eAAe,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;QAC5C,IAAI,IAAI,KAAK,SAAS,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;YAChD,SAAS;QACX,CAAC;QAED,MAAM,SAAS,GAAG,eAAe,CAAC,KAAK,EAAE,sBAAsB,CAAC,CAAC;QACjE,MAAM,UAAU,GAAG,eAAe,CAAC,KAAK,EAAE,uBAAuB,CAAC,CAAC;QACnE,IAAI,SAAS,KAAK,SAAS,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;YACxD,SAAS;QACX,CAAC;QAED,MAAM,SAAS,GAAG,sBAAsB,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC;QAChE,iBAAiB,CAAC,QAAQ,CAAC,GAAG,SAAS,CAAC;QACxC,UAAU,IAAI,CAAC,CAAC;QAEhB,MAAM,MAAM,GAAG,gBAAgB,CAAC,KAAK,CAAC,CAAC;QACvC,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;YACzB,gBAAgB,CAAC,wBAAwB,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;QAC/D,CAAC;QAED,MAAM,QAAQ,GAAG,eAAe,CAAC,KAAK,EAAE,kBAAkB,CAAC,CAAC;QAC5D,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC3B,iBAAiB,CAAC,GAAG,QAAQ,IAAI,QAAQ,EAAE,CAAC,GAAG,SAAS,CAAC;YACzD,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;gBACzB,gBAAgB,CAAC,wBAAwB,EAAE,GAAG,QAAQ,IAAI,QAAQ,EAAE,EAAE,MAAM,CAAC,CAAC;YAChF,CAAC;QACH,CAAC;IACH,CAAC;IAED,IAAI,MAAM,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAChD,MAAM,IAAI,iBAAiB,CACzB,kFAAkF,CACnF,CAAC;IACJ,CAAC;IAED,OAAO;QACL,iBAAiB;QACjB,wBAAwB;QACxB,WAAW,EAAE,UAAU;KACxB,CAAC;AACJ,CAAC;AAQD;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,wBAAwB,CAC5C,OAAyB,EAAE;IAE3B,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,KAAK,CAAC;IACtC,MAAM,GAAG,GAAG,IAAI,CAAC,UAAU,IAAI,oBAAoB,EAAE,CAAC;IACtD,MAAM,EAAE,MAAM,EAAE,GAAG,IAAI,CAAC;IAExB,cAAc,CAAC,MAAM,CAAC,CAAC;IAEvB,IAAI,QAAkB,CAAC;IACvB,IAAI,CAAC;QACH,QAAQ,GAAG,MAAM,CAAC,CAAC,CAAC,MAAM,OAAO,CAAC,GAAG,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC;IAC1E,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,YAAY,CAAC,KAAK,CAAC,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YAC3C,cAAc,CAAC,MAAM,CAAC,CAAC;YACvB,IAAI,YAAY,CAAC,KAAK,CAAC,EAAE,CAAC;gBACxB,MAAM,KAAK,CAAC;YACd,CAAC;QACH,CAAC;QACD,MAAM,MAAM,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACtE,MAAM,IAAI,iBAAiB,CACzB,wCAAwC,GAAG,KAAK,MAAM,EAAE,CACzD,CAAC;IACJ,CAAC;IAED,cAAc,CAAC,MAAM,CAAC,CAAC;IAEvB,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,MAAM,IAAI,iBAAiB,CACzB,iCAAiC,QAAQ,CAAC,MAAM,IAAI,QAAQ,CAAC,UAAU,UAAU,GAAG,EAAE,CACvF,CAAC;IACJ,CAAC;IAED,IAAI,GAAY,CAAC;IACjB,IAAI,CAAC;QACH,GAAG,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;IAC9B,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,YAAY,CAAC,KAAK,CAAC,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YAC3C,cAAc,CAAC,MAAM,CAAC,CAAC;YACvB,IAAI,YAAY,CAAC,KAAK,CAAC,EAAE,CAAC;gBACxB,MAAM,KAAK,CAAC;YACd,CAAC;QACH,CAAC;QACD,MAAM,IAAI,iBAAiB,CACzB,iCAAiC,GAAG,qBAAqB,CAC1D,CAAC;IACJ,CAAC;IAED,cAAc,CAAC,MAAM,CAAC,CAAC;IAEvB,MAAM,UAAU,GAAG,uBAAuB,CAAC,GAAG,CAAC,CAAC;IAEhD,OAAO;QACL,OAAO,EAAE;YACP,iBAAiB,EAAE,UAAU,CAAC,iBAAiB;YAC/C,wBAAwB,EAAE,UAAU,CAAC,wBAAwB;YAC7D,cAAc,EAAE,EAAE;YAClB,YAAY,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACtC,MAAM,EAAE,UAAU;SACnB;QACD,WAAW,EAAE,UAAU,CAAC,WAAW;KACpC,CAAC;AACJ,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-smart-router",
-  "version": "0.6.0",
+  "version": "0.6.1",
   "description": "Auto-model router middleware for the pi.dev coding agent",
   "type": "module",
   "license": "MIT",
@@ -51,10 +51,12 @@
   "scripts": {
     "build": "tsc --project tsconfig.build.json",
     "prepublishOnly": "npm run release:check",
-    "release:check": "npm run verify:ci && npm run release:consumer-pack",
+    "release:check": "npm run verify:ci && npm run release:consumer-pack && npm run release:functional-smoke",
     "release:consumer-pack": "bash scripts/verify-consumer-pack.sh",
+    "release:functional-smoke": "npm run routing:verify-calibration -- --skip-embed && npm run routing:verify-benchmark-profiles && tsx scripts/eval/assert-release-gates.ts --fixtures tests/eval/fixtures --baseline-version 0.6.0",
     "typecheck": "tsc --noEmit",
     "test": "vitest run",
+    "test:release": "vitest run --testNamePattern '@release'",
     "coverage:check": "vitest run --coverage",
     "lint": "eslint . --ext .ts && node --input-type=module -e \"import { readFileSync } from 'node:fs'; import { parse } from 'yaml'; const doc = parse(readFileSync('config/models.yaml.example', 'utf8')); if (!Array.isArray(doc.models) || doc.models.length < 3) throw new Error('models.yaml.example must list at least one model per tier'); const tiers = new Set(doc.models.map((m) => m.tier)); for (const tier of ['zero-tier', 'economical-cloud', 'frontier-cloud']) { if (!tiers.has(tier)) throw new Error('missing tier: ' + tier); } console.log('models.yaml.example: valid');\"",
     "routing:bootstrap-centroids": "node --experimental-strip-types scripts/bootstrap-routing-centroids.ts",
@@ -62,10 +64,11 @@
     "routing:train-calibration": "node --experimental-strip-types scripts/train-routing-calibration.ts",
     "routing:ingest-benchmarks": "tsx scripts/ingest-benchmark-profiles.ts",
     "routing:verify-benchmark-profiles": "vitest run tests/unit/ingest-benchmark-profiles.test.ts -t \"checked-in artifact matches fixture ingest\"",
-    "routing:verify-calibration": "npm run build && node --experimental-strip-types scripts/verify-routing-calibration.ts",
+    "routing:verify-calibration": "bash -c 'if [[ \" $* \" == *\" --skip-embed \"* ]]; then vitest run tests/unit/train-routing-calibration.test.ts -t \"verifyRoutingCalibration passes|evaluates benchmark triage gates\"; else npm run build && tsx scripts/verify-routing-calibration.ts \"$@\"; fi' --",
     "routing:eval-replay": "tsx scripts/eval/counterfactual-replay.ts",
     "routing:eval-harness": "tsx scripts/eval/run-harness.ts",
     "routing:eval-harness:smoke": "tsx scripts/eval/run-harness.ts --summary-only",
+    "routing:capture-baseline": "tsx scripts/eval/capture-baseline.ts",
     "routing:test-projection": "npm run build && node --experimental-strip-types scripts/test-hydra-projection.ts",
     "benchmark:encoder": "tsx scripts/benchmark-encoder-latency.ts",
     "verify:ci": "npm run build && npm run typecheck && npm run lint && npm test && npm run coverage:check"

package/src/infrastructure/pricing/litellm-fetch.ts

@@ -23,6 +23,24 @@ export interface LitellmFetchResult {
 export interface LitellmFetchDeps {
   readonly fetchFn?: typeof fetch;
   readonly pricingUrl?: string;
+  /** When set, aborts the LiteLLM pricing HTTP request (ESC / ctx.signal). */
+  readonly signal?: AbortSignal;
+}
+
+function throwIfAborted(signal: AbortSignal | undefined): void {
+  if (signal?.aborted) {
+    const reason = signal.reason;
+    if (reason instanceof Error) {
+      throw reason;
+    }
+    const error = new Error(typeof reason === 'string' && reason.length > 0 ? reason : 'Aborted');
+    error.name = 'AbortError';
+    throw error;
+  }
+}
+
+function isAbortError(error: unknown): boolean {
+  return error instanceof Error && error.name === 'AbortError';
 }
 
 export class LitellmFetchError extends Error {
@@ -174,17 +192,28 @@ export async function fetchLitellmPriceCatalog(
 ): Promise<LitellmPriceCatalogResult> {
   const fetchFn = deps.fetchFn ?? fetch;
   const url = deps.pricingUrl ?? getLitellmPricingUrl();
+  const { signal } = deps;
+
+  throwIfAborted(signal);
 
   let response: Response;
   try {
-    response = await fetchFn(url);
+    response = signal ? await fetchFn(url, { signal }) : await fetchFn(url);
   } catch (error) {
+    if (isAbortError(error) || signal?.aborted) {
+      throwIfAborted(signal);
+      if (isAbortError(error)) {
+        throw error;
+      }
+    }
     const detail = error instanceof Error ? error.message : String(error);
     throw new LitellmFetchError(
       `Failed to fetch LiteLLM pricing from ${url}: ${detail}`,
     );
   }
 
+  throwIfAborted(signal);
+
   if (!response.ok) {
     throw new LitellmFetchError(
       `LiteLLM pricing fetch failed (${response.status} ${response.statusText}) from ${url}`,
@@ -194,12 +223,20 @@ export async function fetchLitellmPriceCatalog(
   let raw: unknown;
   try {
     raw = await response.json();
-  } catch {
+  } catch (error) {
+    if (isAbortError(error) || signal?.aborted) {
+      throwIfAborted(signal);
+      if (isAbortError(error)) {
+        throw error;
+      }
+    }
     throw new LitellmFetchError(
       `LiteLLM pricing response from ${url} is not valid JSON.`,
     );
   }
 
+  throwIfAborted(signal);
+
   const normalized = normalizeLitellmPricing(raw);
 
   return {