diff --git a/packages/opencode/src/cli/cmd/github.ts b/packages/opencode/src/cli/cmd/github.ts
index 37aed2426d1..ceba8dab60a 100644
--- a/packages/opencode/src/cli/cmd/github.ts
+++ b/packages/opencode/src/cli/cmd/github.ts
@@ -147,6 +147,34 @@ export function parseGitHubRemote(url: string): { owner: string; repo: string }
   return { owner: match[1], repo: match[2] }
 }
 
+/**
+ * Extracts displayable text from assistant response parts.
+ * Handles text, reasoning, and tool-only responses gracefully.
+ */
+export function extractResponseText(parts: MessageV2.Part[]): string {
+  // Priority 1: Look for text parts
+  const textPart = parts.findLast((p) => p.type === "text")
+  if (textPart) return textPart.text
+
+  // Priority 2: Check reasoning parts (for thinking models)
+  const reasoningPart = parts.findLast((p) => p.type === "reasoning")
+  if (reasoningPart) return `[Reasoning] ${reasoningPart.text}`
+
+  // Priority 3: Tool-only responses - generate summary from completed tools
+  const toolParts = parts.filter(
+    (p): p is MessageV2.ToolPart & { state: MessageV2.ToolStateCompleted } =>
+      p.type === "tool" && p.state.status === "completed",
+  )
+  if (toolParts.length > 0) {
+    const summaries = toolParts.map((p) => `- ${p.tool}${p.state.title ? `: ${p.state.title}` : ""}`)
+    return `Completed ${toolParts.length} tool call(s):\n${summaries.join("\n")}`
+  }
+
+  // Improved error with debugging info
+  const partTypes = parts.map((p) => p.type).join(", ") || "none"
+  throw new Error(`Failed to parse response. Part types found: [${partTypes}]`)
+}
+
 export const GithubCommand = cmd({
   command: "github",
   describe: "manage GitHub agent",
@@ -856,10 +884,7 @@ export const GithubRunCommand = cmd({
           )
         }
 
-        const match = result.parts.findLast((p) => p.type === "text")
-        if (!match) throw new Error("Failed to parse the text response")
-
-        return match.text
+        return extractResponseText(result.parts)
       }
 
       async function getOidcToken() {
diff --git a/packages/opencode/test/cli/github-action.test.ts b/packages/opencode/test/cli/github-action.test.ts
new file mode 100644
index 00000000000..fa9a73e9e63
--- /dev/null
+++ b/packages/opencode/test/cli/github-action.test.ts
@@ -0,0 +1,138 @@
+import { test, expect, describe } from "bun:test"
+import { extractResponseText } from "../../src/cli/cmd/github"
+import type { MessageV2 } from "../../src/session/message-v2"
+
+// Helper to create minimal valid parts
+function createTextPart(text: string): MessageV2.Part {
+  return {
+    id: "1",
+    sessionID: "s",
+    messageID: "m",
+    type: "text" as const,
+    text,
+  }
+}
+
+function createReasoningPart(text: string): MessageV2.Part {
+  return {
+    id: "1",
+    sessionID: "s",
+    messageID: "m",
+    type: "reasoning" as const,
+    text,
+    time: { start: 0 },
+  }
+}
+
+function createToolPart(tool: string, title: string, status: "completed" | "running" = "completed"): MessageV2.Part {
+  if (status === "completed") {
+    return {
+      id: "1",
+      sessionID: "s",
+      messageID: "m",
+      type: "tool" as const,
+      callID: "c1",
+      tool,
+      state: {
+        status: "completed",
+        input: {},
+        output: "",
+        title,
+        metadata: {},
+        time: { start: 0, end: 1 },
+      },
+    }
+  }
+  return {
+    id: "1",
+    sessionID: "s",
+    messageID: "m",
+    type: "tool" as const,
+    callID: "c1",
+    tool,
+    state: {
+      status: "running",
+      input: {},
+      time: { start: 0 },
+    },
+  }
+}
+
+function createStepStartPart(): MessageV2.Part {
+  return {
+    id: "1",
+    sessionID: "s",
+    messageID: "m",
+    type: "step-start" as const,
+  }
+}
+
+describe("extractResponseText", () => {
+  test("returns text from text part", () => {
+    const parts = [createTextPart("Hello world")]
+    expect(extractResponseText(parts)).toBe("Hello world")
+  })
+
+  test("returns last text part when multiple exist", () => {
+    const parts = [createTextPart("First"), createTextPart("Last")]
+    expect(extractResponseText(parts)).toBe("Last")
+  })
+
+  test("returns text even when tool parts follow", () => {
+    const parts = [createTextPart("I'll help with that."), createToolPart("todowrite", "3 todos")]
+    expect(extractResponseText(parts)).toBe("I'll help with that.")
+  })
+
+  test("falls back to reasoning part with prefix", () => {
+    const parts = [createReasoningPart("Let me think about this...")]
+    expect(extractResponseText(parts)).toBe("[Reasoning] Let me think about this...")
+  })
+
+  test("handles tool-only response with summary (reproduces original bug)", () => {
+    // This is the exact scenario from the bug report - todowrite with no text
+    const parts = [createToolPart("todowrite", "8 todos")]
+    const result = extractResponseText(parts)
+    expect(result).toContain("Completed 1 tool call")
+    expect(result).toContain("todowrite: 8 todos")
+  })
+
+  test("handles multiple completed tools", () => {
+    const parts = [
+      createToolPart("read", "src/file.ts"),
+      createToolPart("edit", "src/file.ts"),
+      createToolPart("bash", "bun test"),
+    ]
+    const result = extractResponseText(parts)
+    expect(result).toContain("Completed 3 tool call")
+    expect(result).toContain("read: src/file.ts")
+    expect(result).toContain("edit: src/file.ts")
+    expect(result).toContain("bash: bun test")
+  })
+
+  test("handles tool with empty title", () => {
+    const parts = [createToolPart("todowrite", "")]
+    const result = extractResponseText(parts)
+    expect(result).toContain("Completed 1 tool call")
+    expect(result).toContain("- todowrite")
+    expect(result).not.toContain("todowrite:")
+  })
+
+  test("ignores running tool parts", () => {
+    const parts = [createToolPart("bash", "", "running")]
+    expect(() => extractResponseText(parts)).toThrow("Failed to parse response")
+  })
+
+  test("throws with part types on empty array", () => {
+    expect(() => extractResponseText([])).toThrow("Part types found: [none]")
+  })
+
+  test("throws with part types on unhandled parts", () => {
+    const parts = [createStepStartPart()]
+    expect(() => extractResponseText(parts)).toThrow("Part types found: [step-start]")
+  })
+
+  test("prefers text over reasoning when both present", () => {
+    const parts = [createReasoningPart("Internal thinking..."), createTextPart("Final answer")]
+    expect(extractResponseText(parts)).toBe("Final answer")
+  })
+})