error-recovery-skill

Error classification distinguishes transient issues from permanent failures in agentic workflows, applying exponential backoff retries for temporary outages and graceful degradation to preserve partial results without stopping the process.

The best way to implement retry logic with exponential backoff is to use bounded retry attempts with increasing wait times, which safely handles rate limits and temporary outages while preventing infinite loops through permanent-error guardrails.

Graceful degradation works by switching to alternative approaches like backup services or cached data, returning partial results with structured error reporting when optional fields fail while still preserving required outputs for the agent loop.

You should use fallback patterns when error classification identifies a recoverable problem rather than a transient issue, allowing the workflow to switch to an alternative data source instead of attempting retries that will likely fail again.

Yes, you can use this error recovery approach for partial successes across service integrations, as it provides structured error and partial-result reporting to ensure required outputs are preserved while clearly detailing which optional fields failed.

Your workflow stops completely because it lacks a permanent-error guardrail and proper error classification; implementing contextual self-delegation prompts and structured reporting allows the system to escalate safely instead of crashing.