self-check

Mid-execution drift is detected and corrected by performing self-observation checkpoints during long-running AI workflows to assess execution state, recognize patterns, and apply alignment adjustments before task completion or delegation.

A self-observation checkpoint protocol for multi-agent coordination is a formalized sequence that enters meta-mode, runs an alignment checklist, applies pattern recognition, and exits meta-mode with a decision to ensure tasks are only delegated when goals and scope are clearly aligned.

Mid-execution alignment checks should be triggered at specific progress milestones, such as 25% completion, to verify alignment and decide whether to continue the current execution path or adjust course in iterative AI workflows.

Workflow governance guardrails for delegating tasks are implemented by running a checkpoint sequence that verifies goals, scope, and progress are clearly aligned, ensuring tasks are only delegated when these criteria are met.

The self-check workflow requires no external dependencies, relying on a self-contained context-loaded Markdown body with YAML frontmatter and optional component directories for scripts, references, and assets.

Periodic checkpoints for drift detection are limited by their scheduling frequency, meaning drift occurring between checkpoint phases may not be immediately caught until the next scheduled alignment assessment.