error-recovery

To handle agent task execution errors systematically, you can use a protocol that categorizes failures, checks memory-informed logs for known resolutions, and applies multi-source recovery strategies before logging outcomes or escalating unresolvable issues.

The best way to recover from tool failures and build errors is to classify the error type, consult documented patterns or community resources for a fix, and record the resolution in a failures log to prevent future blind retries.

Memory-informed error recovery works by leveraging a historical failure log to apply previously documented resolutions to current runtime and test failures, ensuring agents reference past outcomes before attempting new fixes.

You should escalate unresolvable errors to an orchestrator only after systematically exhausting internal knowledge, documentation, community resources, and web searches, meeting clear criteria defined by the recovery protocol.

Yes, you can use this error handling protocol for external service errors and type errors, as it specifically addresses these categories alongside tool failures, build errors, test failures, and runtime errors.

Logging unresolvable issues improves task success rates by creating a persistent record of failures and escalation paths, enabling memory-informed diagnostics to apply known resolutions to similar future errors automatically.