Part of Agent OS - Kernel-level governance for AI agents
Basic self-correction with retry for AI agents.
SCAK provides a lightweight self-correction loop that detects agent failures and retries with corrective feedback. When an agent produces an incorrect or incomplete result, SCAK catches the failure, applies a correction strategy, and retries the action.
pip install scakfrom scak import SelfCorrectingKernel
kernel = SelfCorrectingKernel()
result = await kernel.run(task="Summarize this document", max_retries=3)- Failure detection and classification
- Configurable retry strategies
- Memory of past corrections to avoid repeated mistakes
- Integration with Agent OS policy engine
See docs/ for guides and CONTRIBUTING.md for development setup.
MIT License - see LICENSE for details.