convez.ai
SkillDevOps & InfraestruturaAvançado
bowlandcup716

Grok GitHub Long-Running

bowlandcup716/grok-github-long-running

Execução resiliente de projetos Claude com persistência em GitHub

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill para gerenciar projetos de IA de longa duração com Claude/Grok, tratando automaticamente limites de tokens, rate limits, checkpoints e retomada após cooldown. Ideal para tarefas multi-hora ou multi-dia que seriam interrompidas.

Instalação

$npx convez-ai bowlandcup716/grok-github-long-running

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Refatoração de grandes codebases sem interrupção
  • Análise de dados prolongada com múltiplas chamadas à API
  • Automação de workflows complexos que excedem limites de tokens

Veredito

Skill robusto e bem estruturado para quem trabalha com projetos Claude longos. Implementação sólida com retry logic, checkpoints e persistência em GitHub, mas exige familiaridade com shell scripting e CLI para máximo aproveitamento.

README

A tradução para o português deste README ainda está na fila.

Grok GitHub Long-Running Skill

A robust skill for Grok AI that handles long-running AI projects (especially with Claude API) by automatically managing token limits, rate limits, retries, checkpoints, and GitHub persistence. Perfect for multi-hour or multi-day tasks that would otherwise get interrupted.

Features

See USE-CASES.md for detailed real-world use cases, benefits, and examples tailored to entrepreneurs (Root & Thread, RoofGuard, etc.).

  • Automatic Rate Limit Handling: Detects 429 errors, parses retry-after headers, and waits with exponential backoff.
  • Resumable Execution: Uses checkpoints (local JSON/files or GitHub) to resume exactly where it left off across sessions.
  • Smart Important Question Handling: When high-priority questions appear, it finishes all other pending subtasks first, then surfaces the important items at the end.
  • GitHub Persistence: Commits progress to a repo/branch for durability.
  • Loop Until Complete: Keeps running until the main task is done or manually stopped.
  • Proactive Self-Continuation: Near token limits (~97%), it announces it will auto-resume after cooldown and saves checkpoint for seamless continuation.
  • Claude Optimized: Built-in wrapper and best practices for Anthropic API.

One-Click Install (Recommended)

# 1. Clone and install in one go (replace YOURUSERNAME)
git clone https://github.com/YOURUSERNAME/grok-github-long-running.git
mkdir -p ~/.grok/skills/
cp -r grok-github-long-running/github-long-running ~/.grok/skills/
chmod +x ~/.grok/skills/github-long-running/scripts/claude-retry.sh
echo "✅ github-long-running skill installed! Restart Grok chat or say 'use github-long-running skill' to test."

Alternative (direct download): Download the latest release ZIP from Releases and extract to ~/.grok/skills/github-long-running/.

For Developers

Fork the repo and submit improvements!

Usage

In Grok Conversations

Trigger with natural language:

  • "Use github-long-running skill to build [big project] with Claude"
  • "Resume my long-running project with rate limit handling"
  • "Continue the Root & Thread AI design automation using github-long-running"

Grok will automatically:

  1. Break the task into checkpointable steps.
  2. Wrap Claude calls in retry logic.
  3. Save state regularly.
  4. Handle important questions intelligently.

Running the Script Directly

cd ~/.grok/skills/github-long-running/scripts
ANTHROPIC_API_KEY=your_key_here ./claude-retry.sh "Your prompt here"

Project Structure

github-long-running/
├── README.md                 ← This file
├── LICENSE                   ← MIT License
├── SKILL.md                  ← Grok's internal skill definition
├── .gitignore                ← Standard ignores
├── .github/workflows/test-skill.yml ← CI validation
├── scripts/
│   └── claude-retry.sh       ← Bash wrapper with retries
└── examples/
    └── template/             ← Ready-to-use project template
        ├── README.md
        └── task_breakdown.md
├── USE-CASES.md              ← Detailed use cases & benefits

Configuration

  • Set ANTHROPIC_API_KEY environment variable.
  • For GitHub persistence, provide a repo URL when starting the task.
  • Customize backoff in claude-retry.sh if needed.

Examples

Example 1: Large Codebase Refactor

"Use github-long-running skill to refactor my entire Next.js app with Claude, handling limits automatically."

Example 2: Content Generation Pipeline (Root & Thread)

"Generate 50 product descriptions + high-res images for Root & Thread Clothing Co. using github-long-running skill."

Example 3: AI Agent Development

"Build a complete RoofGuard OSHA compliance AI agent with github-long-running, including database schema, frontend, and deployment scripts. Resume if interrupted."

Example 4: Bulk Content + Automation

"Create 100 AI-optimized T-shirt designs and Shopify product uploads for Buffalo-themed merch using the long-running skill with checkpoints every 10 designs."

Advanced Example Project Template

See the examples/template/ folder for a ready-to-use structure including:

  • Project metadata guidelines
  • state.json example for checkpoints
  • Task breakdown best practices

Copy and customize this template for your own long-running projects.

Contributing

Improvements welcome! Fork and submit PRs. Priority areas:

  • Python retry class with better state management
  • Enhanced checkpoint JSON schemas with queuing for important questions
  • Support for other LLMs (OpenAI, Grok API, etc.)
  • Expanded GitHub Actions workflows
  • Documentation and bug fixes

License

MIT — feel free to use and modify.



Usage Story (from Max Baker, Buffalo NY AI entrepreneur):
"I was constantly hitting Claude Sonnet token limits while generating hundreds of high-res designs for my Root & Thread Clothing Co. POD store. This skill changed everything — it now auto-checkpoints at ~97% limit, announces the pause, waits for cooldown, and resumes seamlessly across chats. No more lost progress on long projects like bulk merch automation or RoofGuard AI agents. Saved me hours every week!"

Made for the Grok community by Max Baker (Buffalo, NY — Bowling Cub Digital / Root & Thread).

Ver o original no GitHub

Bom para

Relacionados