---

name: open-interface description: "Use Open Interface for LLM-driven desktop automation runtime, configuration, packaging, and safe repository maintenance workflows." disable-model-invocation: true metadata: disco-role: operating license: GPL 3.0

Open Interface repo skill

Use this repo skill when a task is about Open Interface, the Python desktop app that sends screenshots and user objectives to an LLM backend, receives JSON instructions, and executes keyboard/mouse/wait steps to operate a computer.

This skill is self-contained guidance for future Open Interface checkouts. It bundles safe diagnostics and distilled runtime/build knowledge; it does not bundle the application source code, provider credentials, desktop permissions, or release signing credentials.

Quick routes

Task signal	Read next
Run the app, configure API keys, choose GPT/Gemini/custom models, inspect the request loop, debug malformed steps/done JSON, map OpenAI computer-use actions, or troubleshoot screenshots/pyautogui/display permissions	sub-skills/desktop-runtime/SKILL.md
Package or release the app, inspect PyInstaller options, hidden imports, resource inclusion, version bumps, Linux/Windows onefile behavior, macOS signing/notarization, or stale build artifacts	sub-skills/packaging/SKILL.md
The symptom could be either runtime or packaging	references/troubleshooting.md
Check whether this skill matches the current repository version	references/repo-provenance.md

Safe first checks

These bundled helpers are designed for inspection. They do not launch the GUI, use provider API keys, move the mouse/keyboard, capture screenshots, run PyInstaller, sign, notarize, or create release artifacts.

python scripts/check_runtime_contract.py --pretty
python sub-skills/desktop-runtime/scripts/inspect_action_map.py --pretty
python sub-skills/packaging/scripts/build_preflight.py --repo-root <target-checkout> --json

Use a target checkout only for source/build preflight facts. Do not point these helpers at private files containing credentials.

Runtime prerequisites summary

For normal app use in a target Open Interface checkout:

• Python 3.12 or newer is the documented script baseline.
• Runtime dependencies come from the project's dependency pins and include OpenAI, Google GenAI, Pillow, PyAutoGUI, ttkbootstrap, and platform-specific GUI/audio/build dependencies.
• A real interactive desktop is required for screenshots and keyboard/mouse automation. Headless sessions are suitable only for bundled static helpers.
• Model provider settings need an OpenAI, Gemini, or compatible custom endpoint API key. Settings changes require app restart.
• macOS runtime use requires Accessibility and Screen Recording permissions.

See the desktop-runtime sub-skill for detailed provider, settings, and JSON contract guidance.

Packaging prerequisites summary

For packaging or release planning:

• Start with the bundled packaging preflight helper rather than running the source build helper directly.
• Full builds can install dependencies, run PyInstaller, create/overwrite artifacts, prompt interactively, and use platform credentials.
• macOS signing/notarization requires private operator state; keep identities, keychain profiles, Apple credentials, and notary logs out of skill files and public reports.

See the packaging sub-skill for exact PyInstaller option expectations and safe release boundaries.

Approval boundaries

Ask before performing any operation that:

• Launches the GUI, captures screenshots, or controls the real keyboard/mouse.
• Sends requests to OpenAI, Gemini, or a custom model provider.
• Installs or upgrades dependencies or system packages.
• Runs PyInstaller or the source build helper.
• Deletes, overwrites, signs, notarizes, or archives build artifacts.
• Reads, writes, or displays secrets such as API keys or signing credentials.

Refresh and import notes

• Read references/repo-provenance.md before relying on this skill for a different commit or changed checkout.
• Structured router metadata for managed imports is in references/repo-routing-metadata.json.
• This production run was configured not to import the skill automatically. Import later only after verification approval and the repo-skill locked import helper are used by the verification workflow.