Skip to main content
GitHub repository: github.com/vavo/runpod-skill Runpod Codex Skill is a skill set designed for AI coding agents like OpenAI Codex. It provides structured guidance and tooling so agents can work with Runpod infrastructure without relying on stale documentation or incomplete knowledge. The skill covers Serverless workers, Pods, Flash SDK, Public Endpoints, runpodctl, MCP, Python SDK, REST, GraphQL, storage, and debugging. It includes a docs helper script that searches and fetches current Runpod documentation directly.

​
What it covers

The skill provides comprehensive guidance across the Runpod platform:
  • Serverless workers, handlers, endpoint jobs, local testing, Docker packaging, cold starts, logs, model caching, vLLM, and worker fitness checks.
  • Pods, templates, exposed ports, SSH, persistent storage, network volumes, billing guardrails, and runpodctl workflows.
  • Flash SDK and CLI workflows including runpod_flash, Endpoint, local dev, deployment, app environments, custom containers, and storage.
  • Public Endpoints model APIs for hosted image, video, audio, and text models with model-page lookup.
  • Runpod MCP setup and operation notes for Codex-compatible infrastructure management.
  • Python SDK usage for endpoint requests, worker patterns, and API key handling.
  • REST OpenAPI and GraphQL routing including the GraphQL spec at https://graphql-spec.runpod.io.

​
Why use a Codex skill?

Most AI agent guidance only covers parts of Runpod: a handler example here, a CLI command there. This skill ties everything together so Codex knows which tool to use, where to look up current docs, and how to avoid stale assumptions when making infrastructure changes. It favors live documentation over hardcoded values that can drift out of date.

​
Installation

Copy the skill folder into your Codex skills directory: 
mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"
rsync -a runpod/ "${CODEX_HOME:-$HOME/.codex}/skills/runpod/"

​
Usage

Invoke the skill explicitly in your prompts: 
$runpod deploy and troubleshoot this Serverless worker
$runpod check this Dockerfile before I push it to Runpod
$runpod use Public Endpoints to call the current Flux Dev model

​
Docs helper script

The skill includes a helper script that searches and fetches current Runpod documentation: 
# Search for topics
python runpod/scripts/runpod_docs.py search serverless handler
python runpod/scripts/runpod_docs.py search public endpoints models

# Fetch specific pages
python runpod/scripts/runpod_docs.py page flash/create-endpoints

# Get OpenAPI spec
python runpod/scripts/runpod_docs.py openapi --output /tmp/runpod-openapi.json

​
Repository structure

runpod/
├── SKILL.md
├── agents/openai.yaml
├── references/
│   ├── doc-map.md
│   ├── flash.md
│   ├── graphql.md
│   ├── pods-cli-api.md
│   ├── public-endpoints.md
│   ├── runpodctl.md
│   └── serverless.md
└── scripts/
    └── runpod_docs.py