TRACK: Task Recovery And Checkpointing Knowledge — Open Papers on Recoverable Agentic Planning

This open release includes two working papers introducing TRACK (Task Recovery And Checkpointing Knowledge), a methodology for designing recoverable, auditable and checkpointed task plans for AI agents. The first paper presents TRACK as a conceptual framework for recoverable agentic planning based on task contracts, mandatory checkpoints, persistent artifacts, validation gates, failure diagnosis, recovery tasks and active artifact references. The second paper describes the operationalization of TRACK through preliminary experiments, execution registers and reusable plan structures for complex agentic workflows. This release is intended to make the conceptual and methodological foundations of TRACK available to the technical and AI-agent community. The complete applied TRACK plan library is not included in this release and is distributed separately.