When should I use the 'continue' command?

You should use it at the start of every new session after the initial project setup. Claude will read the progress logs to pick up exactly where the previous session left off.

Can I use this with existing projects?

Yes, you can initialize the harness in an existing repository to begin tracking future features and maintaining session context for ongoing development.

How does it prevent Claude from losing context?

The skill creates and updates structured artifacts in a .claude/ directory, specifically a features.json for task tracking and a progress.md for session history and handoff instructions.

Is this skill based on official Anthropic research?

Yes, it implements the architectural patterns described in Anthropic Engineering's research on effective harnesses for long-running agents.

What is a long-running harness for Claude?

It is a specialized workflow that allows Claude to handle large-scale development projects across multiple separate sessions by maintaining a persistent state and progress log.

Long-Running Project Harness

Name: Long-Running Project Harness
Author: dapi

bydapi

0•

生产力与工作流

Orchestrates multi-session development projects by maintaining structured progress tracking and context handoffs between Claude sessions.

This skill implements an advanced orchestration framework designed to solve the context-loss problem inherent in long-term AI development tasks. Based on Anthropic's research into effective agent harnesses, it provides a structured system for initializing projects with granular feature lists, tracking progress through dedicated markdown logs, and managing session handoffs. By breaking complex tasks into small, verifiable units and maintaining persistent state files across multiple interactions, it prevents common AI failures like 'one-shotting' or premature project completion, ensuring consistent incremental progress for large-scale software engineering.

主要功能

01Granular feature tracking supporting exhaustive lists of 30-200 individual tasks

02Automated project initialization with features.json and progress.md scaffolding

03Project status visualization to identify completed, in-progress, and remaining tasks

040 GitHub stars

05Multi-session context preservation using structured handoff logs and git history

06Environment setup automation via specialized init.sh scripts with smoke tests

使用场景

01Developing feature-heavy APIs or CLI tools through systematic, multi-session implementation cycles

02Managing large-scale refactoring projects where context must be preserved across incremental steps

03Building complex full-stack applications that require multiple days or weeks of development

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add dapi/claude-code-marketplace long-running-harness

For use in Claude.ai and ChatGPT

主要功能

01Granular feature tracking supporting exhaustive lists of 30-200 individual tasks

02Automated project initialization with features.json and progress.md scaffolding

03Project status visualization to identify completed, in-progress, and remaining tasks

040 GitHub stars

05Multi-session context preservation using structured handoff logs and git history

06Environment setup automation via specialized init.sh scripts with smoke tests

使用场景

01Developing feature-heavy APIs or CLI tools through systematic, multi-session implementation cycles

02Managing large-scale refactoring projects where context must be preserved across incremental steps

03Building complex full-stack applications that require multiple days or weeks of development