AgentLoop AutoHarness

What problem does it solve?

This Skill automates the orchestration of AI agent pipelines, enabling iterative improvements with human-in-the-loop feedback and speculative exploration.

Core Features & Use Cases

• Data-Driven Orchestration: Manages AI agent pipelines using Git-friendly YAML files, ensuring predictable and decoupled task executions.
• Human-in-the-Loop Checkpoints: Implements explicit pause gates for human review, allowing evaluation and approval/rejection via CLI or dashboard.
• Speculative Exploration: Spawns isolated Git branch worktrees to evaluate multiple fix approaches concurrently when human feedback is slow.
• Lifecycle Hooks: Automatically triggers post-completion actions, such as updating objective score charts, cloning pipelines for next iteration cycles, or executing custom cleanup scripts.
• Visual Dashboard: Serves a responsive, dark-themed HTML5 SPA displaying pipeline execution timelines, task decision logs, metric scores progression charts, and interactive DAGs.
• State-Based Ledger: Injects historical decisions and prerequisite context directly into task handoffs, optimizing LLM tokens.
• Doctor Self-Diagnostics: Includes a static validation engine that audits schemas, cross-references steps, verifies task-pipeline status alignments, and flags orphaned worktree branches.
• Use Case: Imagine you have a series of AI agent tasks to evaluate and improve. Use this Skill to manage the pipeline, collect human feedback, and iterate on the agent's behavior.

Quick Start

Use the AgentLoop AutoHarness to start a new iteration of the 'accuracy-iter-1' pipeline.

agentloop pipeline run accuracy-iter-1