Is this based on official research?

Yes, this skill is based on Anthropic’s 'Effective harnesses for long-running agents' research published in late 2024.

How does it track progress?

It creates a .claude directory containing a features.json file for task status and a progress.md file that logs technical decisions and session history.

What happens during a 'continue' session?

Claude automatically performs a startup checklist, reading the progress log, checking the git status, and identifying the next priority feature to work on.

What problem does the Long-Running Project Harness solve?

It prevents 'context loss' where an AI agent forgets previous work or loses track of a project's overall goal between different chat sessions.

Can I use it with any programming language?

Yes, it is a language-agnostic orchestration tool that works with any stack, from React and Next.js to Python, Rails, or Go.

Long-Running Project Harness

Name: Long-Running Project Harness
Author: dapi

bydapi

•

生产力与工作流

Orchestrates complex, multi-session development projects by maintaining structured context and progress tracking between AI interactions.

The Long-Running Project Harness implements a sophisticated orchestration flow based on Anthropic’s research to solve context loss in AI-driven development. It breaks down ambitious projects into granular, verifiable features and maintains state through persistent artifacts like features.json and progress.md. This skill allows Claude to act as a consistent agent across dozens of sessions, ensuring that architectural decisions, completed tasks, and upcoming priorities are never forgotten, regardless of session timeouts or context window limits.

主要功能

01Persistent state management using structured JSON and Markdown progress logs.

02Automated project initialization with comprehensive feature roadmaps (up to 200 items).

03Session handoff protocols including automated startup checklists and git log analysis.

047 GitHub stars

05Environment preparation via init scripts to ensure immediate agent productivity.

06Granular feature verification tracking to prevent 'one-shotting' and premature completion.

使用场景

01Building full-stack applications from scratch over multiple days or weeks.

02Managing complex refactors that require incremental changes across many sessions.

03Resuming development on a project with a clear 'what's next' prompt after a break.

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

Download Skill