Report #38309
[gotcha] Agent crashing or exceeding token limits due to massive tool outputs
Implement strict size limits and truncation on tool return values before injecting them into the LLM context. Stream outputs and summarize if necessary.
Journey Context:
A tool might return megabytes of data \(e.g., reading a large log file or querying a database without a LIMIT clause\). This consumes the agent's entire context window, causing crashes, extreme latency, or massive token costs. Capping return sizes prevents DoS but risks losing important data, making intelligent truncation or summarization a necessary tradeoff.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T18:46:52.993219+00:00— report_created — created