HotCodePush

Agent skills

The four HotCodePush skills a coding agent installs, the one prompt that sets a project up and what each skill runs for you.

The HotCodePush skills are four folders of instructions in the public hotcodepush-team/skills repository. A coding agent loads one when the task matches its description and runs the CLI's --json commands for you. It stops only where a person must approve something.

Set up from one prompt

Paste this prompt into the agent or assistant you use, Claude Code, Cursor, Claude Desktop, Gemini or any other:

Set up HotCodePush OTA updates in this project, following https://hotcodepush.com/docs/quickstart.md. If you can run commands, install the agent skills with `npx skills add hotcodepush-team/skills`, then run hotcodepush-init and hotcodepush-release. If you cannot run commands, read that page and give me the steps to do it myself.

An agent that runs shell commands in your project installs the skills, runs hotcodepush-init, then hotcodepush-release, and stops with the list of changed files for you to review and commit. It never commits, never prints a token and never creates a signing key unasked.

An assistant without a shell, a chat in the browser or on the desktop, cannot install or run anything. The prompt tells it to read the markdown twin of the quickstart instead and to give you its steps, so you run npx skills add hotcodepush-team/skills and the CLI's init and release create yourself.

The agent acts as the signed-in person and gets no right you lack. A person approves the sign-in once in the browser. In CI, HOTCODEPUSH_TOKEN stands in, as it does for the pipeline.

To install or update the skills by hand, run npx skills add hotcodepush-team/skills or npx skills update in the project. The skills CLI writes them into the agent's skills directory.

The skills

  • hotcodepush-init sets up over-the-air (OTA) updates in an existing project with init and handles the sign-in. It performs every manual step the CLI stops on and repeats doctor --json until it is clean. It is also the repair run when doctor reports a failed check.
  • hotcodepush-release builds where the framework needs it, previews the audience and publishes with release create --yes --json. In CI it labels the bundle from the commit, and it reports the release, the audience and the warnings.
  • hotcodepush-debug-no-update finds out why a device did not update. It walks the causes of Troubleshooting as a decision procedure over device probe --json and doctor --json, each with its reason code, the field that shows it and the fix.
  • hotcodepush-migrate moves an app to HotCodePush from Capawesome Cloud, Capgo, Codemagic Patch, CodePush, EAS Update, Hot Updater, Ionic Appflow, RevoPush or Stallion. It detects the source, freezes a branch for the old builds, removes the old client and hands off to hotcodepush-init and hotcodepush-release. It maps channels, the signing key and the binary ranges, and reports the differences the source's guide under Migrate to HotCodePush lists as a checklist.

Prompts

The agent picks the matching skill on its own:

Set up HotCodePush OTA updates in this project.
Release the current build to the staging channel.
My iPhone on 2.4.1 build 57 did not get the latest release. Why?
Migrate this app from CodePush to HotCodePush.

The skills cover the CLI, and the MCP server is the other half, with tools that manage apps, channels and releases from a chat.

Last updated on

On this page