flash skill (Runpod): write and deploy code-first Runpod Serverless
Runpod's own skill that has a coding agent write, test and deploy Python on Runpod Serverless with the runpod-flash tool.
- Installs
- 14.4K
- on skills.sh
- Stars
- 47
- on GitHub
- Files
- 11
- besides SKILL.md
- Updated
- Oct 2026
- last repo commit
flash makes your coding agent write Python functions with the runpod-flash package and run them on remote Runpod GPUs or CPUs. It develops with flash dev, which hot-reloads your code and streams worker logs, then ships with flash deploy.
It needs the runpod-flash tool and a Runpod account. Code you run under flash dev executes on Runpod machines that bill your account, and flash deploy builds and uploads your code.
Good for
- Python developers who want serverless GPU code without writing a Dockerfile
- People building inference, model-serving or batch endpoints on Runpod
- Teams that want to call an existing endpoint or an external Docker image from Python
Install flash
Claude Code
- 1
Copy this prompt.
Install the flash skill in Claude Code. Run this command: /plugin marketplace add runpod/runpod-plugins-official /plugin install runpod@runpod /reload-plugins Then check that it loaded and tell me how to use it.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Pick one route, not both. Install the whole set, not only this skill: the runpod skills hand off to each other, and the worked golden-path examples live in the runpod router skill's folder. The plugin also registers the hosted Runpod MCP server, which stays inert until you sign in with /mcp.
From the project README. Source.
What's in the skill
When it loads
Write and deploy code-first Runpod Serverless with runpod-flash: @Endpoint/@remote, flash dev, resource configuration, and debugging. Use runpod-build-template for explicit Dockerfile/template project authoring, and runpodctl/runpod-mcp for infrastructure management.
Sections in SKILL.md
- Dev vs Deploy
- Autonomous Dev Loop
- Endpoint: Three Modes
- Mode 1: Your Code (Queue-Based Decorator)
- Mode 2: Your Code (Load-Balanced Routes)
- Mode 3: External Image (Client)
- How Mode Is Determined
- When to use image= (custom container) vs your own code
- Gotchas
- Resources
Key facts
- Author
- Runpod (vendor's own repo; the marketplace file and plugin manifest name Runpod as owner and author)
- Installs on skills.sh
- 14,355 as of 2026-11-02
- What is inside
- 12 files including SKILL.md: 3 reference guides, 8 eval cases and one Python eval fixture (evals/fixtures/dev-loop/main.py).
- Last change to this skill
- 2026-10-19
What it includes
Runs a local server whose functions execute on remote workers
Hot-reloads on save and streams worker logs. SKILL.md tells the agent to run it in the background, capture the log and drive it over HTTP.
Try: “Build a Whisper endpoint with flash and test it in dev.”
Runs code ·
Dev loop (flash dev)Builds an artifact and deploys a stable endpoint
Slow, so SKILL.md says to deploy only after the code works under flash dev.
Runs code ·
Deploy (flash deploy)One function is one endpoint with its own workers
@Endpoint with name, gpu, workers and dependencies. Imports must sit inside the function.
Runs code ·
Mode 1: queue-based decoratorSeveral HTTP routes share one worker pool
Routes like @api.post("/predict") share workers set as a range such as (1, 5).
Runs code ·
Mode 2: load-balanced routesDeploy a prebuilt image, or connect by ID
image= deploys a Docker image such as vLLM; id= connects to an existing endpoint without provisioning.
Runs code ·
Mode 3: external image or existing endpoint15 numbered traps with fixes
Covers function-body-only shipping, missing await, the 10 MB payload limit, runsync's 60 second timeout, a silent 401 when RUNPOD_API_KEY is wrong, and flash app delete for teardown.
Read ·
Gotchas listSetup and CLI, API and enums, patterns
Three guides: setup-and-cli.md, api.md and patterns.md.
Read ·
references/
From the official docs
Permissions
High risk
The agent runs your code on remote Runpod machines and deploys it as billable endpoints. Test in dev first, set worker limits, and delete endpoints you no longer need.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
Your code runs on billable remote workers
Even flash dev executes functions on remote GPUs or CPUs, and deployed endpoints keep their workers per your settings. Set worker limits and run flash app delete when done. github.com
Deploy uploads your project
flash deploy builds an artifact from your code and uploads it. Check what is in the project folder, and keep secrets out of it. github.com
Env values can end up in endpoint config
The external-image example passes env={"HF_TOKEN": "xxx"} to the endpoint. Use a scoped token and rotate it if you share the project. github.com
A wrong API key fails quietly
SKILL.md says flash dev can print its normal ready line even when provisioning failed with a 401. Check the provisioning log if endpoints do not appear. github.com
The plugin route installs all 16 skills and the Runpod MCP config
The marketplace file lists 16 skills for the runpod plugin. It also bundles .mcp.json, which registers the hosted Runpod MCP server at mcp.runpod.io, and a /runpod:migrate command. The git tree shows no hooks file in the plugin folder. github.com
Security advisories
No published security advisories as of Nov 2, 2026.
Technical details
| Talks to | Runpod APIs (api.runpod.io) and the workers it provisions. PyPI (the runpod-flash package). Hugging Face or other model hosts your code downloads from |
|---|---|
| Sign-in and keys | flash login stores a token, or RUNPOD_API_KEY from the environment is used. SKILL.md says a set RUNPOD_API_KEY overrides the flash login token, and a bad key fails quietly with a 401 in the provisioning log. |
| Scopes |
|
Questions
What does the flash skill do?
It has your coding agent write Python with the runpod-flash package, run it on remote Runpod GPUs or CPUs while you develop, and deploy it as a serverless endpoint.
How do I install flash?
Use the install box at the top of this page, and install the runpod-flash tool with uv. Then run flash login, or set RUNPOD_API_KEY, and use flash init to start a project.
Does it need an API key?
It needs a Runpod login. SKILL.md says to run flash login or export RUNPOD_API_KEY, and warns that a set RUNPOD_API_KEY overrides the flash login token.
Will it cost money?
Yes, Runpod bills the workers your code runs on, during flash dev and after flash deploy. The skill itself is free.
Is flash safe?
It runs code on remote machines under your Runpod account and can deploy and delete apps. Use worker limits, review what it deploys, and delete test endpoints.
flash or runpod-build-template?
SKILL.md says to use runpod-build-template when you explicitly want a Dockerfile or template project, and flash for the code-first workflow. For managing infrastructure it points to runpodctl and runpod-mcp.
Is this the official Runpod skill?
Yes. It is in Runpod's own runpod-plugins-official repo, and the plugin manifest names Runpod as author.
Does it work in the Claude app or Codex?
The README documents Claude Code and Codex plugins, plus the skills CLI for other agents. The skill installs a CLI and runs local Python, so there is no Claude app upload option here.
Information
updated Nov 2, 2026
- Repository
- runpod/runpod-plugins-official
- Folder
plugins/runpod/skills/flash
Sources
Docs, issue trackers and discussions this page is built from.
Show all 4Hide
- README.md · github.com
- marketplace.json · github.com
- SKILL.md · github.com
- README.md · github.com