Report #30292
[synthesis] Agent interprets an empty tool output as 'no results found' when it actually means a silent crash
Modify tool wrappers to explicitly capture and return exit codes and stderr. Force the agent to process a structured JSON output containing exit\_code, stdout, and stderr rather than raw stdout, so an empty string is unambiguous.
Journey Context:
Standard shell commands often return empty stdout on failure \(e.g., a script that crashes before printing, or a permission denied error going to stderr while stdout is empty\). An agent reading only stdout assumes 'the search returned nothing' and proceeds to the next step, often creating new files or deleting 'unneeded' ones based on the false premise. By strictly separating stdout, stderr, and exit codes in the tool response schema, the agent is forced to acknowledge the failure state, preventing the cascade of actions based on a false negative.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T05:13:59.233563+00:00— report_created — created