diff --git a/packages/opencode/src/provider/transform.ts b/packages/opencode/src/provider/transform.ts
index 3fd0eaeb..4087dc7e 100644
--- a/packages/opencode/src/provider/transform.ts
+++ b/packages/opencode/src/provider/transform.ts
@@ -129,6 +129,59 @@ function normalizeMessages(
     }
   })
 
+  // Consolidate leading system messages and lower non-leading system messages to prevent
+  // Jinja chat template exceptions ("System message must be at the beginning.") on
+  // OpenAI-compatible, Ollama, vLLM, and local models.
+  const isAnthropic =
+    model.api.npm === "@ai-sdk/anthropic" || model.api.npm === "@ai-sdk/google-vertex/anthropic"
+
+  const leadingSystem: string[] = []
+  let inLeadingSystem = true
+  const processed: ModelMessage[] = []
+
+  for (const msg of msgs) {
+    if (msg.role === "system") {
+      const content = typeof msg.content === "string" ? msg.content : ""
+      if (inLeadingSystem) {
+        if (isAnthropic) {
+          processed.push(msg)
+        } else {
+          if (content.trim().length > 0) {
+            leadingSystem.push(content)
+          }
+        }
+      } else {
+        // Non-leading system message in conversation history:
+        // Lower to user-compatible <system-update> block to preserve chronological ordering
+        // without violating chat template constraints.
+        processed.push({
+          role: "user",
+          content: `<system-update>\n${content}\n</system-update>`,
+        })
+      }
+    } else {
+      if (inLeadingSystem) {
+        inLeadingSystem = false
+        if (!isAnthropic && leadingSystem.length > 0) {
+          processed.unshift({
+            role: "system",
+            content: leadingSystem.join("\n\n"),
+          })
+        }
+      }
+      processed.push(msg)
+    }
+  }
+
+  if (inLeadingSystem && !isAnthropic && leadingSystem.length > 0) {
+    processed.unshift({
+      role: "system",
+      content: leadingSystem.join("\n\n"),
+    })
+  }
+
+  msgs = processed
+
   // Anthropic rejects messages with empty content - filter out empty string messages
   // and remove empty text/reasoning parts from array content
   if (model.api.npm === "@ai-sdk/anthropic") {
diff --git a/packages/opencode/src/session/llm.ts b/packages/opencode/src/session/llm.ts
index 4ac098b2..e39bb90e 100644
--- a/packages/opencode/src/session/llm.ts
+++ b/packages/opencode/src/session/llm.ts
@@ -367,7 +367,7 @@ const live: Layer.Layer<
               {
                 specificationVersion: "v3" as const,
                 async transformParams(args) {
-                  if (args.type === "stream") {
+                  if (args.type === "stream" || args.type === "generate") {
                     // @ts-expect-error
                     args.params.prompt = ProviderTransform.message(
                       args.params.prompt,
diff --git a/packages/opencode/test/provider/transform.test.ts b/packages/opencode/test/provider/transform.test.ts
index 1d88b08c..2e85db35 100644
--- a/packages/opencode/test/provider/transform.test.ts
+++ b/packages/opencode/test/provider/transform.test.ts
@@ -4740,3 +4740,111 @@ describe("ProviderTransform.providerOptions - ai-gateway-provider", () => {
     expect(result).toEqual({ openaiCompatible: { reasoningEffort: "high" } })
   })
 })
+
+describe("ProviderTransform.message - system message consolidation and ordering", () => {
+  const openaiCompatibleModel = {
+    id: "ollama/qwen3.8-27b",
+    providerID: "ollama",
+    api: {
+      id: "qwen3.8-27b",
+      npm: "@ai-sdk/openai-compatible",
+    },
+    capabilities: { input: { text: true, image: false } },
+  } as any
+
+  const anthropicModel = {
+    id: "anthropic/claude-sonnet-4-20250514",
+    providerID: "anthropic",
+    api: {
+      id: "claude-sonnet-4-20250514",
+      npm: "@ai-sdk/anthropic",
+    },
+    capabilities: { input: { text: true, image: true } },
+  } as any
+
+  test("consolidates multiple leading system messages into a single system message for OpenAI-compatible providers", () => {
+    const msgs = [
+      { role: "system", content: "You are an AI pentesting agent." },
+      { role: "system", content: "<pentest-engagement>\nPhase: recon\n</pentest-engagement>" },
+      { role: "user", content: "Scan 10.10.10.5" },
+    ] as any[]
+
+    const result = ProviderTransform.message(msgs, openaiCompatibleModel, {})
+
+    expect(result).toHaveLength(2)
+    expect(result[0].role).toBe("system")
+    expect(result[0].content).toBe(
+      "You are an AI pentesting agent.\n\n<pentest-engagement>\nPhase: recon\n</pentest-engagement>",
+    )
+    expect(result[1].role).toBe("user")
+    expect(result[1].content).toBe("Scan 10.10.10.5")
+  })
+
+  test("lowers non-leading system messages into user-compatible <system-update> blocks", () => {
+    const msgs = [
+      { role: "system", content: "Base system prompt." },
+      { role: "user", content: "First question" },
+      { role: "assistant", content: "First answer" },
+      { role: "system", content: "Compaction summary or chronological update" },
+      { role: "user", content: "Second question" },
+    ] as any[]
+
+    const result = ProviderTransform.message(msgs, openaiCompatibleModel, {})
+
+    expect(result).toHaveLength(5)
+    expect(result[0].role).toBe("system")
+    expect(result[0].content).toBe("Base system prompt.")
+    expect(result[1].role).toBe("user")
+    expect(result[2].role).toBe("assistant")
+    expect(result[3].role).toBe("user")
+    expect(result[3].content).toBe(
+      "<system-update>\nCompaction summary or chronological update\n</system-update>",
+    )
+    expect(result[4].role).toBe("user")
+    expect(result[4].content).toBe("Second question")
+  })
+
+  test("preserves distinct leading system blocks for Anthropic prompt caching", () => {
+    const msgs = [
+      { role: "system", content: "Static system prompt" },
+      { role: "system", content: "Volatile engagement context" },
+      { role: "user", content: "Hello" },
+    ] as any[]
+
+    const result = ProviderTransform.message(msgs, anthropicModel, {})
+
+    expect(result).toHaveLength(3)
+    expect(result[0].role).toBe("system")
+    expect(result[0].content).toBe("Static system prompt")
+    expect(result[1].role).toBe("system")
+    expect(result[1].content).toBe("Volatile engagement context")
+    expect(result[2].role).toBe("user")
+  })
+
+  test("handles single leading system message without duplication", () => {
+    const msgs = [
+      { role: "system", content: "Sole system message" },
+      { role: "user", content: "Hello" },
+    ] as any[]
+
+    const result = ProviderTransform.message(msgs, openaiCompatibleModel, {})
+
+    expect(result).toHaveLength(2)
+    expect(result[0].role).toBe("system")
+    expect(result[0].content).toBe("Sole system message")
+    expect(result[1].role).toBe("user")
+  })
+
+  test("handles message sequence without any system message", () => {
+    const msgs = [
+      { role: "user", content: "Hello" },
+      { role: "assistant", content: "Hi" },
+    ] as any[]
+
+    const result = ProviderTransform.message(msgs, openaiCompatibleModel, {})
+
+    expect(result).toHaveLength(2)
+    expect(result[0].role).toBe("user")
+    expect(result[1].role).toBe("assistant")
+  })
+})