LoopKit: A Finishing Harness for Claude Code

LoopKit is a sophisticated finishing harness designed to enhance the reliability and control of AI code generation, specifically for Claude Code. It focuses on externalizing critical decision-making processes from the AI model into robust file, script, and hook mechanisms, preventing drift and ensuring consistent execution. This approach does not aim to make the AI smarter but rather to make its outputs more dependable by managing decisions that are prone to error under pressure.

• State-File Memory: Manages memory through state files, with state/triage.md acting as a queue and specs/ defining the truth. Findings are persisted to disk rather than being lost in conversation history.

• Stage as a Lookup: The next stage of execution is determined by looking up the current state, ensuring a structured and predictable workflow.

• Blocking Hooks: Implements critical hooks that prevent dangerous operations such as arbitrary file deletion, history rewriting, or premature merging of pull requests. These hooks enforce governance rules and ensure code integrity.

• Stop Gate with Testing: A comprehensive stop gate mechanism runs prechecks, regression diffs, linting, type checking, and builds. It includes an acceptance review against specifications, with any failures blocking the process.

• Separated Powers: Differentiates between implementer and reviewer agents, with the reviewer adopting a default stance of skepticism until proven otherwise. Merge and approval actions are restricted to prevent self-approval or unauthorized changes.

• Escalation Door: Handles tasks requiring human intervention by routing them to an inbox (inbox/needs-human.md) with a structured format for review and decision-making.

• Memory Management: Utilizes adapters for graph, memory, and knowledge retrieval, allowing the system to recall past decisions and context without relying on traditional Retrieval-Augmented Generation (RAG) methods.

• Extensibility: Designed to be extended with project-specific skills, tools, and agents, allowing developers to integrate custom logic and workflows seamlessly.

• Robust Testing: Includes comprehensive testing for the plugin itself, covering hooks, scripts, and end-to-end scenarios in a scratch repository to ensure reliability and correctness.

LoopKit provides a framework for building more reliable and controllable AI-driven development workflows by enforcing strict processes and externalizing critical decision points.