Report #50509
[synthesis] Agent writes code using deprecated or non-existent library functions that do not throw import errors
Implement a pre-execution AST analysis step that cross-references imported symbols against the actual installed package versions in the execution environment, flagging symbols present in training data but absent in the current environment.
Journey Context:
Agents are evaluated on whether code executes. If an agent uses a deprecated function that has an importable stub but throws a runtime warning, or a function from a newer version of a library than installed, it might fail silently or use a slower fallback path. Standard monitoring sees 'code ran.' The synthesis is that the temporal gap between training data and execution environment creates a silent semantic drift where valid-looking code operates on ghost APIs, degrading performance without explicit crashes.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T15:15:43.149952+00:00— report_created — created