agentsop-bounded-loop

What problem does it solve?

It stops language-model-driven agent workflows from running forever when the loop logic lacks an explicit termination mechanism, causing runaway cost, rate-limits, and unusable failures.

Core Features & Use Cases

• Explicit termination contract: enforces a counter budget plus a checkable exit predicate that is outside the LM’s control.
• Progress witness + stagnation detection: requires each loop iteration to change state in a measurable way, and fast-fails when the same error or output repeats.
• Graceful escalation path: routes to an outer/human fallback with preserved state instead of crashing or endlessly retrying.

Use cases: bounding cyclic agent pipelines like tool-call→reflect→retry, plan→act→observe, optimizer/evaluator sweeps, code self-fix cycles, and multi-agent handoffs/supervisor patterns where delegation can create ping-pong loops.

Quick Start

Activate this skill when your agent hits a framework loop-depth/max-iteration error or shows repeated retries without improvement, and then refactor your workflow to add an iteration counter, progress witness, stagnation exit, and an escalation branch that preserves state.