Feature: Automated Error Recovery and Self-Healing

[golemcore1]

Description

If a tool fails (e.g., 403 error or timeout), the bot logs the error but may get stuck in a loop trying the same approach.

Proposed Solution

• Introduce a 'Reflection Step' in the Auto Mode tick cycle.
• If a task fails twice, trigger a specialized model tier to analyze logs and propose an alternative strategy.

Benefits

• Increased robustness of autonomous operations.
• Reduced manual intervention for common technical failures.

[alexk-dev]

Closing as implemented. All key requirements addressed:

• Reflection step: AutoModeScheduler.buildReflectionMessage() triggers dedicated reflection runs on consecutive failures
• Failure detection: Configurable threshold with fingerprinting by command/error bucket, persisted in goals.json
• Self-healing: ToolFailureRecoveryService classifies shell failures and retries with recovery hints (max 2 per unique fingerprint)
• Alternative strategies: Reflection prompts include failure context; results applied via AutoModeService.applyReflectionResult()