Report #10640
[gotcha] Allowing unbounded recursive tool calls or tool-to-tool chaining without resource limits
Enforce a strict maximum depth for tool call chains, a global timeout for agent loops, and token budgets for tool outputs to prevent resource exhaustion.
Journey Context:
An LLM might get stuck in a loop where Tool A's output triggers Tool B, which triggers Tool A again. Or, a prompt injection might instruct the LLM to 'keep searching until you find X', causing it to paginate through an entire database via a tool, leading to API rate limits, massive token consumption, and high costs. Without hard limits on iterations, depth, and execution time, agents will happily loop until they exhaust resources.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T11:16:07.949050+00:00— report_created — created