diff --git a/packages/opencode/src/session/prompt.ts b/packages/opencode/src/session/prompt.ts
index 2fc93c4825..936b0ff039 100644
--- a/packages/opencode/src/session/prompt.ts
+++ b/packages/opencode/src/session/prompt.ts
@@ -81,6 +81,12 @@ const STRUCTURED_OUTPUT_SYSTEM_PROMPT = `IMPORTANT: The user has requested struc
 const log = Log.create({ service: "session.prompt" })
 const elog = EffectLogger.create({ service: "session.prompt" })
 
+function isOrphanedInterruptedTool(part: MessageV2.ToolPart) {
+  // cleanup() marks abandoned tool_use blocks this way after retries/aborts.
+  // They are not pending work and must not trigger an assistant-prefill request.
+  return part.state.status === "error" && part.state.metadata?.interrupted === true
+}
+
 export interface Interface {
   readonly cancel: (sessionID: SessionID) => Effect.Effect<void>
   readonly prompt: (input: PromptInput) => Effect.Effect<MessageV2.WithParts, Image.Error>
@@ -1257,12 +1263,13 @@ export const layer = Layer.effect(
           const lastAssistantMsg = msgs.findLast(
             (msg) => msg.info.role === "assistant" && msg.info.id === lastAssistant?.id,
           )
-          // Some providers return "stop" even when the assistant message contains tool calls.
-          // Keep the loop running so tool results can be sent back to the model.
-          // Skip provider-executed tool parts — those were fully handled within the
-          // provider's stream (e.g. DWS Agent Platform) and don't need a re-loop.
+          // Some providers return "stop" even when the assistant message contains
+          // tool calls. Keep the loop running so tool results can be sent back to
+          // the model, but ignore cleanup-marked interrupted orphans.
           const hasToolCalls =
-            lastAssistantMsg?.parts.some((part) => part.type === "tool" && !part.metadata?.providerExecuted) ?? false
+            lastAssistantMsg?.parts.some(
+              (part) => part.type === "tool" && !part.metadata?.providerExecuted && !isOrphanedInterruptedTool(part),
+            ) ?? false
 
           if (
             lastAssistant?.finish &&
@@ -1270,6 +1277,16 @@ export const layer = Layer.effect(
             !hasToolCalls &&
             lastUser.id < lastAssistant.id
           ) {
+            const orphan = lastAssistantMsg?.parts.find(
+              (part): part is MessageV2.ToolPart => part.type === "tool" && isOrphanedInterruptedTool(part),
+            )
+            if (orphan) {
+              yield* slog.warn("loop exit with orphaned interrupted tool", {
+                messageID: lastAssistant.id,
+                tool: orphan.tool,
+                callID: orphan.callID,
+              })
+            }
             yield* slog.info("exiting loop")
             break
           }
diff --git a/packages/opencode/test/session/prompt.test.ts b/packages/opencode/test/session/prompt.test.ts
index ff9ded4d19..4c46474578 100644
--- a/packages/opencode/test/session/prompt.test.ts
+++ b/packages/opencode/test/session/prompt.test.ts
@@ -457,6 +457,35 @@ noLLMServer.instance(
   { config: cfg },
 )
 
+it.instance("loop exits without an LLM request for interrupted orphan tool calls", () =>
+  Effect.gen(function* () {
+    const { llm } = yield* useServerConfig(providerCfg)
+    const prompt = yield* SessionPrompt.Service
+    const sessions = yield* Session.Service
+    const chat = yield* sessions.create({ title: "Pinned" })
+    const seeded = yield* seed(chat.id, { finish: "stop" })
+    yield* sessions.updatePart({
+      id: PartID.ascending(),
+      messageID: seeded.assistant.id,
+      sessionID: chat.id,
+      type: "tool",
+      callID: "interrupted-call",
+      tool: "edit",
+      state: {
+        status: "error",
+        input: {},
+        error: "Tool execution aborted",
+        metadata: { interrupted: true },
+        time: { start: 1, end: 2 },
+      },
+    })
+
+    const result = yield* prompt.loop({ sessionID: chat.id })
+    expect(result.info.id).toBe(seeded.assistant.id)
+    expect(yield* llm.hits).toHaveLength(0)
+  }),
+)
+
 it.instance("loop calls LLM and returns assistant message", () =>
   Effect.gen(function* () {
     const { llm } = yield* useServerConfig(providerCfg)