fix context compressor summary prompt (#1041)

packages/server/src/lib/context-compressor/index.ts

@@ -15,6 +15,8 @@
 
 import { encodingForModel, getEncoding } from 'js-tiktoken'
 import { randomUUID } from 'crypto'
+import { mkdir, writeFile } from 'fs/promises'
+import { resolve } from 'path'
 import { logger } from '../../services/logger'
 import { AgentBridgeClient, type AgentBridgeRunResult } from '../../services/hermes/agent-bridge'
 import {
@@ -82,6 +84,25 @@ export interface SummarizerOptions {
   workerKey?: string
 }
 
+const SUMMARIZER_TRIGGER_MESSAGE = 'Generate the context checkpoint summary now.'
+const SUMMARIZER_DEBUG_DIR = 'logs/context-compressor'
+const SUMMARIZER_DEBUG_FILE = 'summarizer-debug.json'
+
+async function writeSummarizerDebugDump(payload: Record<string, unknown>): Promise<void> {
+  if (process.env.NODE_ENV !== 'development') return
+  try {
+    const debugDir = resolve(process.cwd(), SUMMARIZER_DEBUG_DIR)
+    await mkdir(debugDir, { recursive: true })
+    await writeFile(
+      resolve(debugDir, SUMMARIZER_DEBUG_FILE),
+      `${JSON.stringify(payload, null, 2)}\n`,
+      'utf8',
+    )
+  } catch (err) {
+    logger.warn(err, '[context-compressor] failed to write summarizer debug dump')
+  }
+}
+
 // ─── Token counting ─────────────────────────────────────
 
 let _encoder: ReturnType<typeof getEncoding> | null = null
@@ -444,24 +465,40 @@ export async function callSummarizer(
     ? { profile: summarizer }
     : summarizer || {}
   const profile = options.profile || 'default'
-  const convHistory: Array<{ role: string; content: string }> = [...history]
+  void history
+  const convHistory: Array<{ role: string; content: string }> = []
 
   if (previousSummary) {
     convHistory.unshift(
       { role: 'user', content: `[Previous summary]\n${previousSummary}` },
       { role: 'assistant', content: 'Understood, I will update the summary.' },
+      { role: 'user', content: prompt },
     )
+  } else {
+    convHistory.unshift({ role: 'user', content: prompt })
   }
 
   const bridge = new AgentBridgeClient({ timeoutMs: timeoutMs + 15_000 })
   const sessionId = `compress_${Date.now().toString(36)}_${randomUUID().replace(/-/g, '').slice(0, 12)}`
   const workerKey = options.workerKey || `${profile}:compression:${sessionId}`
+  const message = SUMMARIZER_TRIGGER_MESSAGE
+
+  await writeSummarizerDebugDump({
+    writtenAt: new Date().toISOString(),
+    sessionId,
+    workerKey,
+    profile,
+    model: options.model || null,
+    provider: options.provider || null,
+    message,
+    convHistory,
+  })
 
   try {
     const result = await bridge.request<AgentBridgeRunResult>({
       action: 'chat',
       session_id: sessionId,
-      message: prompt,
+      message,
       conversation_history: convHistory,
       profile,
       worker_key: workerKey,
@@ -622,10 +659,9 @@ export class ChatContextCompressor {
     try {
       const contentToSummarize = serializeForSummary(toCompress)
       const prompt = buildIncrementalPrompt(previousSummary, contentToSummarize, this.config.summaryBudget)
-      const history = buildConversationHistory(toCompress)
 
       const t0 = Date.now()
-      summary = await callSummarizer(upstream, apiKey, prompt, history, this.config.summarizationTimeoutMs, previousSummary, summarizer)
+      summary = await callSummarizer(upstream, apiKey, prompt, [], this.config.summarizationTimeoutMs, previousSummary, summarizer)
       logger.info('[context-compressor] incremental-llm done in %dms, %d chars', Date.now() - t0, summary.length)
     } catch (err: any) {
       logger.warn('[context-compressor] incremental-llm failed: %s — keeping new messages verbatim', err.message)
@@ -701,12 +737,11 @@ export class ChatContextCompressor {
 
     const contentToSummarize = serializeForSummary(toCompress)
     const prompt = buildFullPrompt(contentToSummarize, this.config.summaryBudget)
-    const history = buildConversationHistory(toCompress)
 
     let summary: string | null = null
     try {
       const t0 = Date.now()
-      summary = await callSummarizer(upstream, apiKey, prompt, history, this.config.summarizationTimeoutMs, undefined, summarizer)
+      summary = await callSummarizer(upstream, apiKey, prompt, [], this.config.summarizationTimeoutMs, undefined, summarizer)
       logger.info('[context-compressor] full-llm done in %dms, %d chars', Date.now() - t0, summary.length)
     } catch (err: any) {
       logger.warn('[context-compressor] full-llm failed: %s', err.message)

packages/server/src/services/hermes/run-chat/compression.ts

@@ -333,7 +333,6 @@ export async function buildCompressedHistory(
         history = await compressHistory(history, null, sessionId, upstream, apiKey, cState, totalTokens, emit, sessionMap, modelContext, compressionConfig.compressor, currentRunInputTokens)
       }
     }
-
     return history
   } catch (err) {
     if (isContextWindowTooSmallError(err)) throw err

tests/server/context-compressor.test.ts

@@ -178,8 +178,14 @@ describe('ChatContextCompressor', () => {
       action: 'chat',
       profile: 'default',
       worker_key: 'default:compression:s1',
+      message: 'Generate the context checkpoint summary now.',
       wait: true,
     }), expect.any(Object))
+    const request = bridgeRequestMock.mock.calls[0][0]
+    expect(request.conversation_history[0]).toEqual(expect.objectContaining({
+      role: 'user',
+      content: expect.stringContaining('TURNS TO SUMMARIZE:'),
+    }))
     const compressSessionId = bridgeRequestMock.mock.calls[0][0].session_id
     expect(String(compressSessionId)).toMatch(/^compress_/)
     expect(bridgeDestroyMock).toHaveBeenCalledWith(
@@ -471,6 +477,16 @@ describe('ChatContextCompressor', () => {
     const result = await compressor.compress(messages, 'http://upstream', undefined, 's1')
 
     expect(bridgeRequestMock).toHaveBeenCalledTimes(1)
+    const request = bridgeRequestMock.mock.calls[0][0]
+    expect(request.message).toBe('Generate the context checkpoint summary now.')
+    expect(request.conversation_history.slice(0, 3)).toEqual([
+      { role: 'user', content: '[Previous summary]\nprevious summary' },
+      { role: 'assistant', content: 'Understood, I will update the summary.' },
+      expect.objectContaining({
+        role: 'user',
+        content: expect.stringContaining('NEW TURNS TO INCORPORATE:'),
+      }),
+    ])
     expect(result.messages.map(m => m.content)).toEqual([
       'head 0',
       'head 1',