loop-constructor

What problem does it solve?

Medium and large AI-coding tasks often fail because they lack structured verification, clear role separation, and machine-checkable completion criteria. Without these, agents loop endlessly, produce slop, or require constant human intervention. This skill solves that by providing a rigorous selection procedure and deterministic linter to design closed-loop workflows that run autonomously.

Core Features & Use Cases

• D0-D6 Selection Procedure: A rigorous decision tree that derives loop shape from runnable checks, not vibes—covering decomposition, pattern selection, autonomy level, parallelism, guards, and iteration profile.
• Staged Loop Design: Decomposes tasks into gated sub-loops with machine-verifiable definitions of done, contract negotiation, and adversarial evaluator roles to prevent sycophancy.
• Deterministic Validation: A linter that rejects hollow checks, missing contracts, or cyclic dependencies, and a renderer that produces a runbook only for passing designs.
• Use Case: A developer needs to implement a medium-complexity feature with rate limiting and load verification. They use this skill to design a three-stage loop with a negotiated contract, separate evaluator context, and restart logic for broken builds.

Quick Start

Use the loop-constructor skill to design a gated sub-loop runbook for your medium or large autonomous coding task, emitted as a .loop/ runbook with machine-verifiable checks and an adversarial evaluator role.