Tool Chain Saboteur

Break the guards around the tools.

Cumulus Operations' agents call tools: track shipments, read files, run jobs, post to status pages. The model never decides whether a call is allowed. The guards around each tool do, and each guard has a gap. Hijack a tool, smuggle an argument, chain harmless calls, then beat all three at once.

TOOL CHAIN SABOTEURCumulus Operations
Total 0
Tokens 0
Act score 1000
Penalty 1× per token
0%
Click objects to explore

Questions about Tool Chain Saboteur

What is Tool Chain Saboteur?

Tool Chain Saboteur is a free browser game from Oathera and the fourth chapter in the Prompt Heist series. You play a red team operator at a fictional cloud operations centre and attack the tools its agents can call, across four acts including a boss. After each act it explains what you exploited and what stops it.

What is tool-use hijacking?

Tool-use hijacking is getting an AI agent to call a tool it should not use for the person asking. It works when the agent holds more tools than its task needs and the only thing stopping a call is the model's own refusal, so a persuasive request can trigger it.

What is argument smuggling in an AI agent?

Argument smuggling is passing a tool arguments that pass a surface check but have a different effect. For example, a path that starts with an allowed folder can still climb out of it with ../ segments. The check looked at the text, but the tool acted on where it resolved.

How can safe tools be chained into a harmful result?

Each tool can be harmless on its own while the sequence is not. Reading an internal note, turning it into a public link and posting that link carries internal data to a public page, and no per-tool check sees the chain. Data flow across the whole chain has to be controlled.

How do you stop AI agent tool abuse?

Give an agent only the tools its task needs. Enforce who may call each tool outside the model against a verified identity. Validate arguments by their resolved effect, bind approvals to the exact action and a short expiry, and control where data from a chain may go.

Do I need to finish the earlier chapters first?

Yes. Tool Chain Saboteur unlocks when you have completed every act of Inbox Infiltrator, which unlocks after Data Vault Breach and Prompt Heist. Your progress is saved only in your browser's local storage.

Is it free, and does it collect what I type?

It is free. The game runs entirely in your browser and nothing you type leaves the page. Your best scores are saved only in your browser's local storage.

Keep reading: How AI agent tool calls get abused · Inbox Infiltrator, chapter 3 · Prompt Heist, chapter 1 · Enforce least privilege for an AI agent calling your systems