exploration loops

Stop polishing a single draft and start exploring the solution space. This skill addresses the problem of linear, narrow-minded AI generation that misses superior alternatives by forcing the agent to branch out, compare, and converge on the best option.

What it does

• Branch-and-Explore generates and evaluates multiple candidate solutions in parallel to find the strongest architectural or design fit.
• Tree Search performs recursive exploration of complex, sequential problems to a variable depth, abandoning weak branches early.
• Debate pits opposing positions against each other to stress-test decisions and identify non-obvious trade-offs or syntheses.
• Budget Management uses a loop ledger to enforce an exploration budget, preventing the agent from branching indefinitely.
• Convergence Logic ensures every exploration has an auditable endpoint where a winner is selected based on uniform criteria.

How it works

1. The agent identifies an exploration trigger, such as a request for multiple approaches or a trade-off analysis.
2. It initializes a loop ledger using one of three patterns: Branch-and-Explore, Tree Search, or Debate.
3. Candidates are generated and scored against a consistent rubric to maintain evaluation integrity.
4. The loop concludes via a manual assertion once a candidate is selected or the search budget is exhausted.

Frameworks & tools

This skill is designed for agents using loop-ledger orchestration. It relies on iterative state tracking and manual exit modes to manage non-linear search spaces.

Why this beats prompting it yourself

Standard prompting often leads to "agreeable" outputs or surface-level alternatives that are just minor variations of the first idea. This skill enforces structural divergence and adversarial reasoning, ensuring the agent doesn't just give you the first answer it finds, but the best one it can prove.

Use cases

• Architectural Trade-offs comparing three distinct service designs on cost, latency, and complexity.
• Strategic Decision Making debating whether to deprecate a legacy API by arguing both sides until synthesis.
• Deep Search exploring a sequence of moves or plan steps several levels deep to find hidden bottlenecks.

Known limitations

This skill is not for refining a single draft or adapting a plan during execution. It requires a loop-ledger implementation to manage the branching budget.