Merge ff612a0d51 into 6af205a13a

src/agents/pi-embedded-runner/run.ts

@@ -367,6 +367,7 @@ export async function runEmbeddedPiAgent(
                 log.warn(
                   `context overflow detected; attempting auto-compaction for ${provider}/${modelId}`,
                 );
+                await params.onCompactionStart?.();
                 overflowCompactionAttempted = true;
                 const compactResult = await compactEmbeddedPiSessionDirect({
                   sessionId: params.sessionId,

src/agents/pi-embedded-runner/run/params.ts

@@ -86,6 +86,7 @@ export type RunEmbeddedPiAgentParams = {
   onBlockReplyFlush?: () => void | Promise<void>;
   blockReplyBreak?: "text_end" | "message_end";
   blockReplyChunking?: BlockReplyChunking;
+  onCompactionStart?: () => void | Promise<void>;
   onReasoningStream?: (payload: { text?: string; mediaUrls?: string[] }) => void | Promise<void>;
   onToolResult?: (payload: { text?: string; mediaUrls?: string[] }) => void | Promise<void>;
   onAgentEvent?: (evt: { stream: string; data: Record<string, unknown> }) => void;

src/auto-reply/reply/agent-runner-execution.ts

@@ -276,6 +276,13 @@ export async function runAgentTurnWithFallback(params: {
             abortSignal: params.opts?.abortSignal,
             blockReplyBreak: params.resolvedBlockStreamingBreak,
             blockReplyChunking: params.blockReplyChunking,
+            onCompactionStart: async () => {
+              if (params.opts?.onPartialReply) {
+                await params.opts.onPartialReply({
+                  text: "🧹 Compacting context, please wait...",
+                });
+              }
+            },
             onPartialReply: allowPartialStream
               ? async (payload) => {
                   const textForTyping = await handlePartialForTyping(payload);