What happens when a task fails repeatedly?

If a task fails at the retry threshold, the system triggers an Advisor Consult to decide whether to adjust the plan, attempt a local correction, or stop the loop for manual intervention.

Does it provide real-time updates while running in the background?

Yes, you can check the current progress using the status command, and the runner maintains detailed logs in the .claude/state/codex-loop/ directory.

What is the primary purpose of Harness Loop?

It automates a continuous loop of finding ready tasks, executing them, and reviewing the results in the background, allowing for autonomous progress on large projects.

How does it handle task dependencies?

It scans your Plans.md for tasks marked as cc:TODO or cc:WIP, verifies if their dependencies are met, and bundles them into 'ready batches' for the next execution cycle.

Can I run multiple loops simultaneously?

No, Harness Loop is designed to run one instance per project to ensure state consistency and prevent resource conflicts. It will show an 'already running' status if you try to start a second one.

Harness Loop Claude Code Skill | Autonomous Dev Workflow

Harness Loop is a specialized orchestrator for the Claude Code development harness that enables autonomous, multi-stage workflows. It moves beyond one-shot task execution by implementing a 'Plan→Work→Review' cycle that runs in the background. The skill identifies ready-to-execute tasks, manages dependencies, and utilizes the 'Breezing' executor for parallel task processing. With built-in state management, advisor consultations for high-risk decisions, and detailed logging, it allows developers to delegate complex, time-consuming backlogs to an autonomous agent while maintaining full control through status monitoring and stop commands.

주요 기능

01Autonomous background execution of 'Plan→Work→Review' cycles

02Intelligent dependency resolution for cc:TODO and cc:WIP tasks

03Parallel task processing via the Breezing batch executor

042,702 GitHub stars

05Advisor consultation system for high-risk tasks and failure handling

06Comprehensive state tracking and logging within the .claude/state directory

사용 사례

01Executing complex multi-step refactorings that require iterative feedback loops

02Processing large backlogs of development tasks overnight without human supervision

03Managing parallel feature implementation across multiple files using worker-based scaling

Tieline — product intent grounded in code

Harness Loop

주요 기능

사용 사례

Agent Skill Creator

Output Verifier

TaskFlow Inbox Triage

주요 기능

사용 사례