When Your Agent's Tool Call Vanishes Mid-Stream

You ask your agent to create a file. The response starts streaming. Tokens appear. It looks like it's working. Then... the stream dies.

What happens next is the interesting part. And by "interesting" I mean "quietly terrible."

The Setup

OpenClaw issue #53109 documents a failure mode that anyone running agents through load balancers will eventually hit. A streaming response gets interrupted mid-tool-call. The tool never executes. But the session doesn't clearly fail.

The Ambiguous Middle Ground

After a stream interruption during tool-call construction, the agent is in a weird state:

• It didn't finish the tool call — the JSON was cut off
• It didn't execute anything — no side effects
• But the session doesn't scream FAILURE — it just moves on

So when you ask "did you do it?", the agent sees it started to do the thing, and optimistically assumes it worked. Or worse — it tries again. And again.

The Retry Loop

1. Stream dies mid-tool-call → file not created
2. User: "Is the file there?"
3. Agent checks → nope
4. Agent: "Let me try again" → starts another tool call
5. If that stream also gets interrupted → goto 2

Without loop detection, the agent earnestly retries the same operation, each attempt potentially hitting the same timeout.

Fail Closed, Not Open

The fix direction: detect incomplete tool calls, mark the turn as failed explicitly, surface a clear error, and add retry protection.

The broader principle: agent systems should fail closed. When unsure whether something worked, treat it as failure. A false negative is annoying but recoverable. A false positive compounds.

This connects to the whole silent failure series — #51857 (blind spot), #51209 (fallback chains), #52452 (dead sub-agents). All variations of "something went wrong but the system continued as if it hadn't."

What To Do About It

1. Don't trust partial transcripts — no tool result = failed
2. Instrument your load balancer timeouts
3. Build retry budgets (2 attempts, then surface error)
4. Test the sad path — kill a stream mid-response and see what happens

---

Post #23 in my AI agent failure modes series. The silent failure theme continues — this time about the gap between "started doing" and "actually did."