agentic-loop-skill

Iterative planning and reflection for multi-step research is automated by running an Observe-Think-Act-Reflect-Decide loop. This agentic loop tracks state, validates intermediate results, and stops when explicit criteria are met or a maximum iteration count is reached.

An agentic loop is an autonomous problem-solving cycle that iteratively observes, thinks, acts, reflects, and decides. It applies self-delegation to carry forward compact context summaries, enabling multi-step computation and research workflows to adjust strategies based on intermediate results.

Iterative task automation handles error recovery by applying retry strategies for recoverable errors during the reflection phase. It prevents indefinite repetition by design through explicit stop conditions and iteration counters that halt the workflow when progress criteria are satisfied.

Yes, self-delegation supports state tracking by carrying forward a compact summary of context, progress, and errors across iterations. This ensures the workflow maintains continuity and applies explicit stop criteria without losing intermediate computation results.

Workflow control with an observe-think-act-reflect-decide cycle is needed when a single pass is unlikely to succeed. It suits complex tasks requiring repeated planning, execution, evaluation, and adjustment, such as multi-step research, synthesis, and computation workflows.

The best way to compare multiple items is using an agentic loop that iteratively gathers evidence, validates intermediate results, and synthesizes findings. The loop reflects on progress after each step and stops when the comparison is complete or after a set iteration limit.