Does it require manual review for every task in the loop?

No, it is designed for autonomous operation, including automated lead reviews and git cherry-picking for approved changes, though users can recap progress at any time.

How does it handle task failures or stalls?

The skill includes plateau detection to identify when progress has stalled on a specific task and can escalate the issue to an advisor or pause for user intervention.

What is the primary purpose of the harness-loop skill?

It automates the execution of a sequence of development tasks by looping through a plan and using fresh contexts for each wake-up to ensure high-quality, autonomous output.

When should I avoid using harness-loop?

It should not be used for one-shot task execution, simple code reviews, or planning sessions that do not require an autonomous, repeating work cycle.

Can I control how frequently the agent wakes up to work?

Yes, it supports multiple pacing modes such as 'worker' (270s), 'ci' (270s), 'plateau' (1200s), and 'night' (3600s) to balance execution speed with resource management.

Harness Loop Claude Code Skill | Autonomous Dev Cycles

Harness Loop is a meta-skill for Claude Code designed to handle complex, multi-task development projects through an autonomous, long-running execution cycle. By leveraging dynamic mode and scheduled wake-ups, it ensures each task starts with a fresh context while maintaining progress through a centralized plan. It manages task prioritization, automated reviews, and plateau detection, making it ideal for background development, large-scale refactoring, or overnight batch processing where manual intervention should be minimized. It bridges the gap between manual coding and fully autonomous agent workflows.

Características Principales

01Automated task prioritization and roadmap execution from Plans.md.

022,702 GitHub stars

03Persistent session state management with fresh-context wake-ups via ScheduleWakeup.

04Customizable pacing modes optimized for CI, worker tasks, or overnight processing.

05Autonomous Plan-Work-Review execution cycles using Claude Code dynamic mode.

06Intelligent plateau detection and advisor escalation to prevent infinite loops.

Casos de Uso

01Managing a backlog of CI-style tasks that require autonomous self-correction and review loops.

02Executing large-scale codebase refactors across multiple modules autonomously.

03Running a series of non-blocking development tasks in the background during deep work or overnight.

Tieline — product intent grounded in code

Harness Loop

Características Principales

Casos de Uso

TaskFlow Inbox Triage

Document Production Pro

Agent Skill Creator

Características Principales

Casos de Uso