Long Task Harness for AI agents - feature-driven development with external memory
npx claudepluginhub mylukin/agent-foremanLong Task Harness for AI agents - task/feature-driven development with external memory
mylukin⚠️ DEPRECATED: This project is no longer maintained. Please migrate to ralph-dev - a simplified and improved version.
This project has been deprecated due to its complexity. A new, simplified version is now available:
👉 https://github.com/mylukin/ralph-dev
Please migrate to ralph-dev for continued support and updates.
Stop AI agents from half-building features. Ship complete code in one session.
AI coding agents face three common failure modes:
agent-foreman provides a structured harness that enables AI agents to:
/plugin install agent-foreman # 1. Install
/agent-foreman:init Build auth API # 2. Initialize
/agent-foreman:run # 3. Let AI work
# Quick install (binary)
curl -fsSL https://raw.githubusercontent.com/mylukin/agent-foreman/main/scripts/install.sh | bash
# Via npm
npm install -g agent-foreman
# Or use npx directly
npx agent-foreman --help
Manual download: GitHub Releases
/plugin marketplace add mylukin/agent-foreman
/plugin install agent-foreman
| Command | Description |
|---|---|
/agent-foreman:status | View project status and progress |
/agent-foreman:init | Initialize harness with project goal |
/agent-foreman:analyze | Analyze existing project structure |
/agent-foreman:spec | Transform requirements into tasks |
/agent-foreman:next | Get next priority task |
/agent-foreman:run | Auto-complete all pending tasks |
Transform requirements into tasks:
/agent-foreman:spec Build a user authentication system
Requirement → [PM→UX→Tech→QA] → Spec Files → BREAKDOWN Tasks → /run → Implementation
For standalone CLI usage without Claude Code:
| Command | Description |
|---|---|
init [goal] | Initialize or upgrade the harness |
next [feature_id] | Show next feature to work on |
status | Show current project status |
check [feature_id] | Verify code changes or task completion |
done <feature_id> | Verify, mark complete, and auto-commit |
fail <feature_id> | Mark a task as failed |
impact <feature_id> | Analyze impact of changes |
tdd [mode] | View or set TDD mode |
agents | Show available AI agents |
install | Install Claude Code plugin |
uninstall | Uninstall Claude Code plugin |
next → implement → check → done → repeat
| Step | Command | What Happens |
|---|---|---|
| 1 | next | Get task with acceptance criteria |
| 2 | implement | Write code to satisfy criteria |
| 3 | check | Verify implementation |
| 4 | done | Mark complete, auto-commit |
| File | Purpose |
|---|---|
ai/tasks/index.json | Task index with status summary |
ai/tasks/{module}/{id}.md | Individual task definitions |
ai/progress.log | Session handoff audit log |
ai/init.sh | Environment bootstrap script |
CLAUDE.md | AI agent instructions |
| Status | Meaning |
|---|---|
failing | Not yet implemented |
passing | Acceptance criteria met |
blocked | External dependency blocking |
needs_review | May be affected by changes |
failed | Verification failed |
deprecated | No longer needed |
AI agents need the same tooling that makes human teams effective:
Official prompts.chat marketplace - AI prompts, skills, and tools for Claude Code
Open Design — local-first design app exposed to coding agents over MCP. Install once with your agent's plugin command and projects/files/skills are reachable through stdio.
Behavioral guidelines to reduce common LLM coding mistakes, derived from Andrej Karpathy's observations