Report #68362
[agent\_craft] Agent loops infinitely or crashes because tool output exceeds context limit and silently truncates the exit condition
Implement pagination or streaming for large tool outputs, and explicitly instruct the agent in the tool description to use pagination or grep to find specific results.
Journey Context:
When an agent runs a long test suite, the output can be thousands of lines. Truncating silently means the agent never sees the '0 tests failed' line and assumes failure, retrying infinitely. Paginated tool design forces the agent to iteratively pull only what it needs, keeping the context window clean and deterministic.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T21:13:43.897208+00:00— report_created — created