loop selector

Stop wasting tokens on loops that never converge or tasks that shouldn't be loops at all. This agent skill acts as a high-level router for iterative tasks, ensuring you select the right architectural pattern before writing a single line of logic. It evaluates tasks against a strict "loop-shaped" gate and then maps them to a specialized category—Quality, Memory, Planning, Exploration, or System Optimization.

What it does

• Pre-loop validation ensures a task has a clear progress signal, requires multiple passes, and possesses a natural stopping point before proceeding.
• Pattern routing maps ambiguous user requests like "optimize this" or "fix this" to one of twenty specific iterative patterns.
• Config generation produces immediate loop_start parameters including exit_mode, target_value, and max_iterations tailored to the specific pattern.
• Composite detection identifies when a task requires a sequence of different loop types rather than forcing a single, ill-fitting pattern.
• Budget enforcement recommends appropriate iteration limits to prevent infinite loops and manage compute costs.

How it works

1. Validation: Run the Step 0 check to confirm the task is actually loop-shaped based on signal availability and necessity.
2. Classification: Use the decision tree to match the task against five core categories: quality, memory, planning, exploration, or system optimization.
3. Pattern Selection: Identify the specific sub-pattern (e.g., Adversarial Critique vs. Judge Ensemble) within the chosen category.
4. Configuration: Generate the concrete loop_start JSON config with the correct exit modes and progress directions.

Frameworks & tools

This skill is designed for agents using iterative ledger patterns or loop-based architectures. It integrates with bantamkit for memory-related workflows and utilizes standard iteration controllers like critique-gate and prompt-evolve.

Why this beats prompting it yourself

Prompting an agent to "loop until finished" often leads to hallucinated stopping conditions or inefficient iteration. This skill provides a structured decision framework that enforces architectural best practices, ensuring you use the most efficient pattern for the specific problem domain.

Use cases

• Quality Control: Routing a content generation task to a Multi-Critic or Adversarial loop.
• Bug Fixing: Mapping a failing test suite to a Plan-Execute-Replan planning loop.
• Prompt Engineering: Configuring a System Optimization loop to tune templates against a metric.
• Design Research: Setting up an Exploration loop to branch and compare multiple UI candidates.

Known limitations

This skill is a router and does not contain the full implementation logic for every pattern. Users must refer to the specific category skills for deep wiring rationale and failure-mode notes.