feat: ensure that compaction message is tracked as agent initiated, consistent with github copilot extension and opencode

Author: caozhiyuan

README.md

@@ -198,7 +198,6 @@ The following command line options are available for the `start` command:
       "gpt-5-mini": "low"
     },
     "useFunctionApplyPatch": true,
-    "compactUseSmallModel": true,
     "useMessagesApi": true
   }
   ```
@@ -207,7 +206,6 @@ The following command line options are available for the `start` command:
 - **smallModel:** Fallback model used for tool-less warmup messages (e.g., Claude Code probe requests) to avoid spending premium requests; defaults to `gpt-5-mini`.
 - **modelReasoningEfforts:** Per-model `reasoning.effort` sent to the Copilot Responses API. Allowed values are `none`, `minimal`, `low`, `medium`, `high`, and `xhigh`. If a model isn’t listed, `high` is used by default.
 - **useFunctionApplyPatch:** When `true`, the server will convert any custom tool named `apply_patch` in Responses payloads into an OpenAI-style function tool (`type: "function"`) with a parameter schema so assistants can call it using function-calling semantics to edit files. Set to `false` to leave tools unchanged. Defaults to `true`.
-- **compactUseSmallModel:** When `true`, detected "compact" requests (e.g., from Claude Code or Opencode compact mode) will automatically use the configured `smallModel` to avoid consuming premium model usage for short/background tasks. Defaults to `true`.
 - **useMessagesApi:** When `true`, Claude-family models that support Copilot's native `/v1/messages` endpoint will use the Messages API; otherwise they fall back to `/chat/completions`. Set to `false` to disable Messages API routing and always use `/chat/completions`. Defaults to `true`.
 
 Edit this file to customize prompts or swap in your own fast model. Restart the server (or rerun the command) after changes so the cached config is refreshed.

src/lib/api-config.ts

@@ -75,6 +75,15 @@ export const getOauthAppConfig = (): OauthAppConfig => {
   }
 }
 
+export const prepareForCompact = (
+  headers: Record<string, string>,
+  isCompact?: boolean,
+) => {
+  if (isCompact) {
+    headers["x-initiator"] = "agent"
+  }
+}
+
 export const prepareInteractionHeaders = (
   sessionId: string | undefined,
   isSubagent: boolean,

src/lib/config.ts

@@ -62,7 +62,6 @@ const defaultConfig: AppConfig = {
     "gpt-5.3-codex": "xhigh",
   },
   useFunctionApplyPatch: true,
-  compactUseSmallModel: true,
   useMessagesApi: true,
 }
 
@@ -203,11 +202,6 @@ export function getReasoningEffortForModel(
   return config.modelReasoningEfforts?.[model] ?? "high"
 }
 
-export function shouldCompactUseSmallModel(): boolean {
-  const config = getConfig()
-  return config.compactUseSmallModel ?? true
-}
-
 export function isMessagesApiEnabled(): boolean {
   const config = getConfig()
   return config.useMessagesApi ?? true

src/routes/messages/handler.ts

@@ -7,7 +7,6 @@ import type { Model } from "~/services/copilot/get-models"
 import { awaitApproval } from "~/lib/approval"
 import {
   getSmallModel,
-  shouldCompactUseSmallModel,
   getReasoningEffortForModel,
   isMessagesApiEnabled,
 } from "~/lib/config"
@@ -90,9 +89,6 @@ export async function handleCompletion(c: Context) {
 
   if (isCompact) {
     logger.debug("Is compact request:", isCompact)
-    if (shouldCompactUseSmallModel()) {
-      anthropicPayload.model = getSmallModel()
-    }
   } else {
     // Merge tool_result and text blocks into tool_result to avoid consuming premium requests
     // (caused by skill invocations, edit hooks, plan or to do reminders)
@@ -119,6 +115,7 @@ export async function handleCompletion(c: Context) {
       selectedModel,
       requestId,
       sessionId,
+      isCompact,
     })
   }
 
@@ -128,13 +125,15 @@ export async function handleCompletion(c: Context) {
       selectedModel,
       requestId,
       sessionId,
+      isCompact,
     })
   }
 
   return await handleWithChatCompletions(c, anthropicPayload, {
     subagentMarker,
     requestId,
     sessionId,
+    isCompact,
   })
 }
 
@@ -148,9 +147,10 @@ const handleWithChatCompletions = async (
     subagentMarker?: SubagentMarker | null
     requestId: string
     sessionId?: string
+    isCompact?: boolean
   },
 ) => {
-  const { subagentMarker, requestId, sessionId } = options
+  const { subagentMarker, requestId, sessionId, isCompact } = options
   const openAIPayload = translateToOpenAI(anthropicPayload)
   logger.debug(
     "Translated OpenAI request payload:",
@@ -161,6 +161,7 @@ const handleWithChatCompletions = async (
     subagentMarker,
     requestId,
     sessionId,
+    isCompact,
   })
 
   if (isNonStreaming(response)) {
@@ -217,9 +218,11 @@ const handleWithResponsesApi = async (
     selectedModel?: Model
     requestId: string
     sessionId?: string
+    isCompact?: boolean
   },
 ) => {
-  const { subagentMarker, selectedModel, requestId, sessionId } = options
+  const { subagentMarker, selectedModel, requestId, sessionId, isCompact } =
+    options
 
   const responsesPayload =
     translateAnthropicMessagesToResponsesPayload(anthropicPayload)
@@ -243,6 +246,7 @@ const handleWithResponsesApi = async (
     subagentMarker,
     requestId,
     sessionId,
+    isCompact,
   })
 
   if (responsesPayload.stream && isAsyncIterable(response)) {
@@ -321,6 +325,7 @@ const handleWithMessagesApi = async (
     selectedModel?: Model
     requestId: string
     sessionId?: string
+    isCompact?: boolean
   },
 ) => {
   const {
@@ -329,6 +334,7 @@ const handleWithMessagesApi = async (
     selectedModel,
     requestId,
     sessionId,
+    isCompact,
   } = options
   // Pre-request processing: filter thinking blocks for Claude models so only
   // valid thinking blocks are sent to the Copilot Messages API.
@@ -366,6 +372,7 @@ const handleWithMessagesApi = async (
     subagentMarker,
     requestId,
     sessionId,
+    isCompact,
   })
 
   if (isAsyncIterable(response)) {

src/services/copilot/create-chat-completions.ts

@@ -6,6 +6,7 @@ import type { SubagentMarker } from "~/routes/messages/subagent-marker"
 import {
   copilotBaseUrl,
   copilotHeaders,
+  prepareForCompact,
   prepareInteractionHeaders,
 } from "~/lib/api-config"
 import { HTTPError } from "~/lib/error"
@@ -17,6 +18,7 @@ export const createChatCompletions = async (
     subagentMarker?: SubagentMarker | null
     requestId: string
     sessionId?: string
+    isCompact?: boolean
   },
 ) => {
   if (!state.copilotToken) throw new Error("Copilot token not found")
@@ -50,6 +52,8 @@ export const createChatCompletions = async (
     headers,
   )
 
+  prepareForCompact(headers, options.isCompact)
+
   const response = await fetch(`${copilotBaseUrl(state)}/chat/completions`, {
     method: "POST",
     headers,

src/services/copilot/create-messages.ts

@@ -10,6 +10,7 @@ import type { SubagentMarker } from "~/routes/messages/subagent-marker"
 import {
   copilotBaseUrl,
   copilotHeaders,
+  prepareForCompact,
   prepareInteractionHeaders,
 } from "~/lib/api-config"
 import { HTTPError } from "~/lib/error"
@@ -64,6 +65,7 @@ export const createMessages = async (
     subagentMarker?: SubagentMarker | null
     requestId: string
     sessionId?: string
+    isCompact?: boolean
   },
 ): Promise<CreateMessagesReturn> => {
   if (!state.copilotToken) throw new Error("Copilot token not found")
@@ -94,6 +96,8 @@ export const createMessages = async (
     headers,
   )
 
+  prepareForCompact(headers, options.isCompact)
+
   // align with vscode copilot extension anthropic-beta
   const anthropicBeta = buildAnthropicBetaHeader(
     anthropicBetaHeader,

src/services/copilot/create-responses.ts

@@ -6,6 +6,7 @@ import type { SubagentMarker } from "~/routes/messages/subagent-marker"
 import {
   copilotBaseUrl,
   copilotHeaders,
+  prepareForCompact,
   prepareInteractionHeaders,
 } from "~/lib/api-config"
 import { HTTPError } from "~/lib/error"
@@ -357,6 +358,7 @@ interface ResponsesRequestOptions {
   subagentMarker?: SubagentMarker | null
   requestId: string
   sessionId?: string
+  isCompact?: boolean
 }
 
 export const createResponses = async (
@@ -367,6 +369,7 @@ export const createResponses = async (
     subagentMarker,
     requestId,
     sessionId,
+    isCompact,
   }: ResponsesRequestOptions,
 ): Promise<CreateResponsesReturn> => {
   if (!state.copilotToken) throw new Error("Copilot token not found")
@@ -378,6 +381,8 @@ export const createResponses = async (
 
   prepareInteractionHeaders(sessionId, Boolean(subagentMarker), headers)
 
+  prepareForCompact(headers, isCompact)
+
   // service_tier is not supported by github copilot
   payload.service_tier = null