AI Prompt Studio

Type a task in plain English and run it through the Claude CLI or GitHub Copilot CLI, with optional skills attached as hints.

packages/ai-prompt

AI Prompt is a one-box front end for agent CLIs. You describe a task (“List all Docker containers and their port mappings”), the app wraps it in a prompt that asks for a JSON summary, runs the agent in a subprocess, and displays the steps and result.

Requirements

You need at least one agent CLI installed and signed in:

  • claude (Claude Code). The app runs claude --print --output-format json and sends the prompt on standard input.
  • gh with the Copilot extension. The app runs gh copilot -- -p <prompt> --allow-all-tools --silent.

The agent is chosen in the prompt box (<robos-ai-textarea>); the app falls back to Copilot if the box reports none. If the agent is not signed in, the box offers a login that opens a terminal with claude /login or gh auth login. A run is stopped after five minutes. The run itself always uses the agent CLI, because the agent runs commands. The ? button in the prompt box is different: it answers a quick question without running anything, through your active agent CLI or the optional API mode (AI agents and API mode).

The agent is asked to run commands with its own shell tool on your machine, and the Copilot call uses --allow-all-tools. Treat a prompt like a command you type yourself.

What you do in it

  1. Optionally pick skills in the Skills sidebar (filter box at the top). Built-in skills come from packages/skills-manager/skills-data.js; custom ones from ~/.config/robos/skills.json. A selected skill’s name and command are added to the prompt as hints, and $UPPER_CASE placeholders in a command become parameter fields.
  2. Type the task under What do you want to do?. Use @ to mention a file; the typeahead lists the directory or searches the text indexes in ~/.config/robos/search-index/.
  3. Click Run with AI, or press Ctrl+Enter (Cmd+Enter on macOS).
  4. Read the Summary, the list of steps (action, command, output) and the Result. New Prompt resets the form.

Skills Manager opens the skills editor (/usr/local/share/robos/skills-manager). History shows past runs; Clear All deletes them.

Data and security

  • History: ~/.config/robos/ai-prompt-history.json, trimmed to the last 50 runs.
  • Before a prompt leaves the app it passes through PromptSecurityGuard from robos-lib. Depending on the policy, the prompt is blocked or sensitive values are redacted.
  • If the agent answers in plain text instead of JSON, the app shows that text as the result with no steps.

Source: packages/ai-prompt.

More in AI & Agent Workflows

Agent Chat A chat window with session history, a model and mode selector and tool cards; replies are currently scripted samples. Agent Sidebar A prototype 320-pixel sidebar that mocks up an agent's plan, tool feed and approve/refine buttons. Agents Manager Browse and continue sessions of Claude Code, GitHub Copilot and Codex, and configure each provider's login, models and MCP servers. Context Manager Register GitHub repositories and local folders as context sources, check which agent instruction files they contain, and build one combined context text. Desktop Agents Viewer A prototype viewer that mocks up watching several agents on their own virtual desktops. MCP Manager Lists the MCP servers registered on this machine, lets you call their tools from a form, and records which agents should have which servers. RobOS Voice Speech-to-text dictation and spoken replies for the desktop, with a local HTTP API on port 19188. Skills Manager Catalog of command snippets and prompt templates for RobOS apps, with custom entries and installable packs from GitHub. Task Implementer Pick a GitHub or Jira task, choose a developer persona, and run the Claude Code CLI on it with streamed output. Task Planner Describe a feature or pick a template, have GitHub Copilot draft epics and tasks, edit them, and create them as issues on your task server.