hot.dev logo

Apache 2.0 on GitHub

Hot Language, runtime & platform Demos Runnable example projects JS / TS SDK @hot-dev/sdk for Node Python SDK hot-dev on PyPI Go SDK github.com/hot-dev/hot-go Rust SDK hot-dev on crates.io Java SDK dev.hot:hot-sdk on Maven Central VS Code Extension Hot language support Skills AI agent skills (SKILL.md)

AI Coding Assistants

Hot ships coding-agent context because conventional language assumptions do not apply to Hot syntax or execution semantics. Install the bundled instructions before asking an AI assistant to create or change Hot code.

This guide is about coding with AI assistants. To build an AI agent as part of your product, see Agents.

Install Project Context

From the root of a Hot project:

hot ai add

This installs:

  • AGENTS.md — concise repository instructions and critical Hot syntax rules
  • .skills/hot-language/ — detailed language guidance, references, and examples
  • .skills/hot-ai-agents/ — agent, model, tool, memory, and AI SDK integration guidance

Project-local context is the best default. It travels with the repository and lets the team review changes to the instructions alongside the code.

Install Skills Globally

To make the bundled Hot skills available across projects:

hot ai add --global

Global skills are installed under ~/.skills/. The exact skill set can vary by Hot release.

Inspect what is installed:

hot ai list

Refresh installed files after upgrading Hot:

hot ai update

hot ai add and hot ai update use snapshots bundled with the installed Hot release, so they work offline and stay version-aligned with the compiler.

Install the Latest Public Skills

The public skills are also published from hot-dev/hot-skills:

npx skills add hot-dev/hot-skills

Use the bundled hot ai add path when release alignment and offline installation matter. Use the public skills source when you intentionally want the newest published guidance.

Give the Assistant a Clear Task

Good coding-agent requests name the workflow behavior, existing namespace, and validation command:

Use the Hot language skill for this task.

Add an event handler in ::billing that listens for invoice:overdue,
posts a Slack notification, and retries transient failures three times.
Preserve the existing event payload shape. Run hot check and the relevant
Hot tests when finished.

For agent or model integrations, also ask the assistant to use the hot-ai-agents skill:

Use the hot-language and hot-ai-agents skills.

Add a support agent with project-specific memory and one permission-scoped
MCP tool. Follow the existing namespace and package patterns. Do not invent
package APIs; inspect the installed package docs first.

The explicit skill request is useful in tools that load skills on demand. AGENTS.md remains passive repository context for tools that support the standard.

Validate Generated Hot Code

AI-generated Hot code should be treated like any other code change:

hot check
hot test

When working inside this repository, use the project-specific commands in its AGENTS.md. Some Hot repositories run tests through Cargo:

cargo run test

Review generated code for the Hot-specific mistakes that general-purpose models make most often:

  • Assignment uses name value, not name = value.
  • Arithmetic and comparison use functions such as add and eq, not infix operators.
  • Conditional flows use if(...), cond, or match, not conventional if/else blocks.
  • Ordinary function bodies are serial; only explicit parallel flows run independent bindings concurrently.
  • Expected failures use Result.Err patterns; fail(...) is for broken invariants.
  • Event handlers with external side effects account for at-least-once delivery.
  • Namespaced package functions are inspected rather than guessed.

The bundled language skill contains the complete rules and examples.

LLM-Readable Documentation

Hot publishes two plain-text documentation resources:

Use the compact index for tools that can retrieve individual pages. Use the full document when a tool needs a single context artifact and its context window can hold the content.

The website documentation remains the source of truth for current public behavior. Bundled skills focus on the rules and workflows a coding agent needs to make correct changes.

Keep Secrets Out of Agent Context

Do not paste production API keys, model keys, customer data, .env contents, or private run payloads into an assistant unless the selected tool and environment are explicitly approved for that data.

Hot code reads secrets from context variables:

api-key ::hot::ctx/get("anthropic.api.key")

For local development, put the value in .env:

ANTHROPIC_API_KEY=your-provider-key

Then map it to a Hot context key in hot/ctx.hot:

::hot::run::ctx ns
::env ::hot::env

::hot::ctx/set({
  "anthropic.api.key": ::env/get("ANTHROPIC_API_KEY", "")
})

Deployed projects do not use hot/ctx.hot. Set the same context key in Context Variables in the Hot App for the target environment or project.

Use placeholders—not real values—in prompts and examples. See Context Variables and Context Requirements.

Troubleshooting

The assistant writes conventional syntax

Confirm AGENTS.md is visible from the working directory and explicitly ask the assistant to load the hot-language skill.

hot ai list

Instructions are from an older Hot release

Upgrade Hot, then refresh installed AI support:

hot update
hot ai update

A package API is invented

Direct the assistant to the installed Hot package docs and ask it to verify the exact namespace, function, type, and context requirements before editing code.

The skill is unavailable in a specific tool

Keep AGENTS.md in the repository and provide the relevant page from llms.txt or llms-full.txt as context. Skill discovery differs between coding tools, while plain repository instructions and Markdown remain widely usable.

Next Steps

Adopting Hot Demos