skills.rest
Pgooone
PgoooneCOMMUNITY
Explore all skills by this author →
GitHub

deploy-long-running-agent

Scaffolds autonomous multi-session AI coding workspaces from PRD documents.

1|Updated Apr 3, 2026
Search Tags:#code generation#claude code#task automation#prd#autonomous agent#workspace scaffolding#long-running agent
One-click install
npx skills add https://github.com/Pgooone/oh-pgone-claudecode --skill deploy-long-running-agent-pgooone
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: deploy-long-running-agent
Source: https://github.com/Pgooone/oh-pgone-claudecode/tree/main/.claude/skill/deploy-long-running-agent
Command: npx skills add https://github.com/Pgooone/oh-pgone-claudecode --skill deploy-long-running-agent-pgooone

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes assets (resource) components.

What problem does it solve? LLM context compaction loses project state across sessions, making long-running autonomous coding unreliable. This Skill converts a PRD into a complete agent workspace where state lives in files (CLAUDE.md, task.json, progress.txt, git history) instead of conversation history, so each fresh Claude Code session can resume work without prior context. ## Core Features & Use Cases - Create Mode: Reads a PRD, detects the tech stack (Python, Node.js, Go, Java, Rust), and generates a full workspace including architecture.md, task.json with 5-15 ordered tasks, CLAUDE.md workflow rules, init.sh, automation scripts, and stub entry points. - Update Mode: Detects an existing agent workspace and incrementally merges new PRD requirements, appending tasks while preserving completed work. - Dual-platform automation: Generates both bash and PowerShell runner scripts that loop Claude Code sessions, plus a next-task.txt prompt for semi-auto execution. - Use Case: You have a product requirements document for a price-monitoring tool. The Skill scaffolds the entire project, then offers Manual, Semi-Auto, or Full-Auto execution modes so the AI completes tasks one per session over hours. ## Quick Start Ask the AI to deploy your PRD.md file as an automatic coding agent workspace and choose a development mode when prompted.

Frequently Asked Questions about deploy-long-running-agent

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I turn a PRD into an autonomous AI coding project?▼

Provide the PRD file path and let the Skill analyze it, detect the tech stack, and generate a workspace containing CLAUDE.md workflow rules, a task.json task list, init.sh, and automation scripts. Each new Claude Code session then reads these files and completes one task.

What is the artifacts over context compaction approach for long-running agents?▼

Instead of relying on LLM context compression to remember state, project state is stored in files: CLAUDE.md for workflow rules, task.json as the task source of truth, progress.txt for session logs, and git commits for code history. New sessions start clean and rebuild context from these artifacts.

Can I update an existing agent workspace with new PRD requirements?▼

Yes. Update Mode detects an existing workspace by looking for CLAUDE.md, task.json, and init.sh, then merges new requirements by appending tasks with sequential IDs while preserving all tasks already marked passes: true.

Does the automation runner work on Windows PowerShell?▼

Yes. The Skill generates both run-automation.sh for Linux, macOS, and Git Bash, and run-automation.ps1 for Windows PowerShell. The PowerShell version handles temp-file prompts and separate stderr redirection.

What happens when the AI agent gets blocked on a task?▼

The generated CLAUDE.md defines blocking scenarios such as missing API keys or ambiguous requirements. The agent must not commit or mark the task passed; instead it logs the block in progress.txt and outputs a structured message requesting human help.

Which tech stacks does the workspace scaffolding support?▼

The Skill detects Python, Node.js/Next.js, Go, Java/Spring, and Rust from PRD keywords, then tailors the directory structure, init.sh dependency setup, and test commands such as pytest, npm test, or go test accordingly.

Related Skills

Explore More Software Engineering Skills→
chflechfle
3

sw-use-case-coder

Generate code scaffolding from tech-stack.yaml and use-cases.md definitions.

Community
Advanced
BaSui01BaSui01
3

record-session

Record development session summaries and git commit hashes to .trellis via Python scripts.

Community
Intermediate
vinwangvinwang

verification-loop

Automate end-to-end verification of Claude Code sessions across TypeScript, Python, and JavaScript projects.

Community
Advanced
multiplex-aimultiplex-ai
2

strategic-compact

Prompt users to compact memory at logical boundaries in Claude Code sessions.

Official
Basic