btca-server 1.0.30 → 1.0.40

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "btca-server",
-	"version": "1.0.30",
+	"version": "1.0.40",
 	"description": "BTCA server for answering questions about your codebase using OpenCode AI",
 	"author": "Ben Davis",
 	"license": "MIT",
@@ -49,6 +49,7 @@
 		"prettier": "^3.7.4"
 	},
 	"dependencies": {
+		"@btca/shared": "workspace:*",
 		"@opencode-ai/sdk": "^1.0.208",
 		"hono": "^4.7.11",
 		"zod": "^3.25.76"

package/src/agent/service.ts

@@ -95,7 +95,7 @@ export namespace Agent {
 		const prompt = [
 			'You are an expert internal agent whose job is to answer questions about the collection.',
 			'You operate inside a collection directory.',
-			'Use the resources in this collection to answer the user\'s question.',
+			"Use the resources in this collection to answer the user's question.",
 			args.agentInstructions
 		].join('\n');
 

package/src/index.ts

@@ -62,7 +62,10 @@ const QuestionRequestSchema = z.object({
 	question: z
 		.string()
 		.min(1, 'Question cannot be empty')
-		.max(LIMITS.QUESTION_MAX, `Question too long (max ${LIMITS.QUESTION_MAX} chars)`),
+		.max(
+			LIMITS.QUESTION_MAX,
+			`Question too long (max ${LIMITS.QUESTION_MAX.toLocaleString()} chars). This includes conversation history - try starting a new thread or clearing the chat.`
+		),
 	resources: z
 		.array(ResourceNameField)
 		.max(
@@ -319,7 +322,11 @@ const createApp = (deps: {
 			} satisfies BtcaStreamMetaEvent;
 
 			Metrics.info('question.stream.start', { collectionKey });
-			const stream = StreamService.createSseStream({ meta, eventStream });
+			const stream = StreamService.createSseStream({
+				meta,
+				eventStream,
+				question: decoded.question
+			});
 
 			return new Response(stream, {
 				headers: {

package/src/stream/service.ts

@@ -1,6 +1,11 @@
 import type { OcEvent } from '../agent/types.ts';
 import { getErrorMessage, getErrorTag } from '../errors.ts';
 import { Metrics } from '../metrics/index.ts';
+import {
+	StreamingTagStripper,
+	extractCoreQuestion,
+	stripUserQuestionFromStart
+} from '@btca/shared';
 
 import type {
 	BtcaStreamDoneEvent,
@@ -41,6 +46,7 @@ export namespace StreamService {
 	export const createSseStream = (args: {
 		meta: BtcaStreamMetaEvent;
 		eventStream: AsyncIterable<OcEvent>;
+		question?: string; // Original question - used to filter echoed user message
 	}): ReadableStream<Uint8Array> => {
 		const encoder = new TextEncoder();
 
@@ -52,6 +58,15 @@ export namespace StreamService {
 		let textEvents = 0;
 		let reasoningEvents = 0;
 
+		// Create streaming tag stripper for filtering history markers
+		const tagStripper = new StreamingTagStripper();
+
+		// Extract the core question for stripping echoed user message from final response
+		const coreQuestion = extractCoreQuestion(args.question);
+
+		// Track total emitted text for accurate final text reconstruction
+		let emittedText = '';
+
 		const emit = (
 			controller: ReadableStreamDefaultController<Uint8Array>,
 			event: BtcaStreamEvent
@@ -73,17 +88,36 @@ export namespace StreamService {
 					try {
 						for await (const event of args.eventStream) {
 							if (event.type === 'message.part.updated') {
-								const part: any = (event.properties as any).part;
+								const props = event.properties as any;
+								const part: any = props?.part;
 								if (!part || typeof part !== 'object') continue;
 
+								// Skip user messages - only stream assistant responses
+								const messageRole = props?.message?.role ?? props?.role;
+								if (messageRole === 'user') {
+									continue;
+								}
+
 								if (part.type === 'text') {
 									const partId = String(part.id);
 									const nextText = String(part.text ?? '');
-									const delta = updateAccumulator(text, partId, nextText);
-									if (delta.length > 0) {
-										textEvents += 1;
-										const msg: BtcaStreamTextDeltaEvent = { type: 'text.delta', delta };
-										emit(controller, msg);
+
+									// Get the raw delta from accumulator
+									const rawDelta = updateAccumulator(text, partId, nextText);
+
+									if (rawDelta.length > 0) {
+										// Filter through the streaming tag stripper
+										const cleanDelta = tagStripper.process(rawDelta);
+
+										if (cleanDelta.length > 0) {
+											textEvents += 1;
+											emittedText += cleanDelta;
+											const msg: BtcaStreamTextDeltaEvent = {
+												type: 'text.delta',
+												delta: cleanDelta
+											};
+											emit(controller, msg);
+										}
 									}
 									continue;
 								}
@@ -120,18 +154,29 @@ export namespace StreamService {
 
 							if (event.type === 'session.idle') {
 								const tools = Array.from(toolsByCallId.values());
+
+								// Flush any remaining buffered content from the tag stripper
+								const flushed = tagStripper.flush();
+								if (flushed.length > 0) {
+									emittedText += flushed;
+								}
+
+								// Strip the echoed user question from the final text
+								let finalText = stripUserQuestionFromStart(emittedText, coreQuestion);
+
 								Metrics.info('stream.done', {
 									collectionKey: args.meta.collection.key,
-									textLength: text.combined.length,
+									textLength: finalText.length,
 									reasoningLength: reasoning.combined.length,
 									toolCount: tools.length,
 									toolUpdates,
 									textEvents,
 									reasoningEvents
 								});
+
 								const done: BtcaStreamDoneEvent = {
 									type: 'done',
-									text: text.combined,
+									text: finalText,
 									reasoning: reasoning.combined,
 									tools
 								};

package/src/validation/index.ts

@@ -47,8 +47,8 @@ export const LIMITS = {
 	NOTES_MAX: 500,
 	/** Maximum length for search paths */
 	SEARCH_PATH_MAX: 256,
-	/** Maximum length for questions */
-	QUESTION_MAX: 10_000,
+	/** Maximum length for questions (includes conversation history when formatted) */
+	QUESTION_MAX: 100_000,
 	/** Maximum number of resources per request */
 	MAX_RESOURCES_PER_REQUEST: 20
 } as const;