diff --git a/src/llm/minimax.integration.test.ts b/src/llm/minimax.integration.test.ts
new file mode 100644
index 0000000..0bede30
--- /dev/null
+++ b/src/llm/minimax.integration.test.ts
@@ -0,0 +1,64 @@
+/**
+ * @fileoverview Integration tests for the MiniMax LLM adapter.
+ *
+ * These tests hit the live MiniMax API and are automatically skipped when
+ * MINIMAX_API_KEY is not set.  They do NOT mock the openai module so that
+ * the real HTTP client is exercised.
+ *
+ * Run with:
+ *   MINIMAX_API_KEY=<key> npm test
+ */
+
+import { describe, it, expect } from 'vitest'
+import { MiniMaxAdapter } from './minimax.js'
+
+const runIntegration = Boolean(process.env['MINIMAX_API_KEY'])
+
+function userMsg(text: string) {
+  return { role: 'user' as const, content: [{ type: 'text' as const, text }] }
+}
+
+describe.skipIf(!runIntegration)('MiniMaxAdapter (integration)', () => {
+  it('chat() returns a non-empty response from MiniMax-M2.7', async () => {
+    const adapter = new MiniMaxAdapter()
+    const response = await adapter.chat(
+      [userMsg('Reply with exactly the word "pong" and nothing else.')],
+      { model: 'MiniMax-M2.7', temperature: 0.01, maxTokens: 20 },
+    )
+
+    const text = response.content
+      .filter((b) => b.type === 'text')
+      .map((b) => (b as any).text)
+      .join('')
+
+    expect(text.toLowerCase()).toContain('pong')
+    expect(response.usage.input_tokens).toBeGreaterThan(0)
+  })
+
+  it('chat() works with MiniMax-M2.7-highspeed model', async () => {
+    const adapter = new MiniMaxAdapter()
+    const response = await adapter.chat(
+      [userMsg('Say only "ok".')],
+      { model: 'MiniMax-M2.7-highspeed', temperature: 0.01, maxTokens: 50 },
+    )
+    expect(response.content.length).toBeGreaterThan(0)
+    expect(['end_turn', 'max_tokens']).toContain(response.stop_reason)
+  })
+
+  it('stream() yields text events and a done event', { timeout: 15000 }, async () => {
+    const adapter = new MiniMaxAdapter()
+    const events: any[] = []
+    for await (const event of adapter.stream(
+      [userMsg('Count from 1 to 3, one number per line.')],
+      { model: 'MiniMax-M2.7', temperature: 0.5, maxTokens: 60 },
+    )) {
+      events.push(event)
+    }
+
+    const textEvents = events.filter((e) => e.type === 'text')
+    const doneEvent = events.find((e) => e.type === 'done')
+    expect(textEvents.length).toBeGreaterThan(0)
+    expect(doneEvent).toBeDefined()
+    expect(['end_turn', 'max_tokens']).toContain((doneEvent.data as any).stop_reason)
+  })
+})