diff --git a/packages/opencode/src/session/llm.ts b/packages/opencode/src/session/llm.ts
index c9a62c8645e0..6395d8cc4bac 100644
--- a/packages/opencode/src/session/llm.ts
+++ b/packages/opencode/src/session/llm.ts
@@ -99,28 +99,20 @@ export namespace LLM {
     // TODO: move this to a proper hook
     const isOpenaiOauth = provider.id === "openai" && auth?.type === "oauth"
 
-    const system: string[] = []
-    system.push(
-      [
-        // use agent prompt otherwise provider prompt
-        ...(input.agent.prompt ? [input.agent.prompt] : SystemPrompt.provider(input.model)),
-        // any custom prompt passed into this call
-        ...input.system,
-        // any custom prompt from last user message
-        ...(input.user.system ? [input.user.system] : []),
-      ]
-        .filter((x) => x)
-        .join("\n"),
-    )
+    const header = [...(input.agent.prompt ? [input.agent.prompt] : SystemPrompt.provider(input.model))]
+      .filter((x) => x)
+      .join("\n")
+    const tail = [...input.system, ...(input.user.system ? [input.user.system] : [])].filter((x) => x).join("\n")
+    const system = [header, tail].filter((x) => x)
 
-    const header = system[0]
     await Plugin.trigger(
       "experimental.chat.system.transform",
       { sessionID: input.sessionID, model: input.model },
       { system },
     )
-    // rejoin to maintain 2-part structure for caching if header unchanged
-    if (system.length > 2 && system[0] === header) {
+    // Keep a stable header separate from the mutable tail so provider prompt caching
+    // can reuse the static prefix even when turn-level instructions change.
+    if (header && system.length > 2 && system[0] === header) {
       const rest = system.slice(1)
       system.length = 0
       system.push(header, rest.join("\n"))
diff --git a/packages/opencode/test/session/llm.test.ts b/packages/opencode/test/session/llm.test.ts
index 1fa2e61eb241..209334e34e91 100644
--- a/packages/opencode/test/session/llm.test.ts
+++ b/packages/opencode/test/session/llm.test.ts
@@ -383,6 +383,93 @@ describe("session.llm.stream", () => {
     })
   })
 
+  test("keeps the system header separate from turn instructions", async () => {
+    const server = state.server
+    if (!server) {
+      throw new Error("Server not initialized")
+    }
+
+    const providerID = "vivgrid"
+    const modelID = "gemini-3.1-pro-preview"
+    const fixture = await loadFixture(providerID, modelID)
+    const model = fixture.model
+
+    const request = waitRequest(
+      "/chat/completions",
+      new Response(createChatStream("Hello"), {
+        status: 200,
+        headers: { "Content-Type": "text/event-stream" },
+      }),
+    )
+
+    await using tmp = await tmpdir({
+      init: async (dir) => {
+        await Bun.write(
+          path.join(dir, "opencode.json"),
+          JSON.stringify({
+            $schema: "https://opencode.ai/config.json",
+            enabled_providers: [providerID],
+            provider: {
+              [providerID]: {
+                options: {
+                  apiKey: "test-key",
+                  baseURL: `${server.url.origin}/v1`,
+                },
+              },
+            },
+          }),
+        )
+      },
+    })
+
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const resolved = await Provider.getModel(ProviderID.make(providerID), ModelID.make(model.id))
+        const sessionID = SessionID.make("session-test-system-split")
+        const agent = {
+          name: "test",
+          mode: "primary",
+          options: {},
+          prompt: "Header prompt",
+          permission: [{ permission: "*", pattern: "*", action: "allow" }],
+        } satisfies Agent.Info
+
+        const user = {
+          id: MessageID.make("user-system-split"),
+          sessionID,
+          role: "user",
+          time: { created: Date.now() },
+          agent: agent.name,
+          model: { providerID: ProviderID.make(providerID), modelID: resolved.id },
+          system: "Turn specific",
+        } satisfies MessageV2.User
+
+        const stream = await LLM.stream({
+          user,
+          sessionID,
+          model: resolved,
+          agent,
+          system: ["Context one", "Context two"],
+          abort: new AbortController().signal,
+          messages: [{ role: "user", content: "Hello" }],
+          tools: {},
+        })
+
+        for await (const _ of stream.fullStream) {
+        }
+
+        const body = (await request).body
+        const msgs = body.messages as Array<{ role: string; content: string }>
+        const system = msgs.filter((msg) => msg.role === "system")
+
+        expect(system).toHaveLength(2)
+        expect(system[0]?.content).toBe("Header prompt")
+        expect(system[1]?.content).toBe("Context one\nContext two\nTurn specific")
+      },
+    })
+  })
+
   test("raw stream abort signal cancels provider response body promptly", async () => {
     const server = state.server
     if (!server) throw new Error("Server not initialized")