install-opendirector — скилл для Claude Code: установка
Скилл install-opendirector для Claude Code ставит OpenDirector из исходников, проверяет запуск и выдаёт команды для старта. Он включается при первой установке или ремонте незавершённой.
Apache-2.0 ★ 413 обновлён 2026-11-17
npx skills add ThinkFlowLab/OpenDirector --skill install-opendirector -a claude-codeЧто делает install-opendirector
Скилл install-opendirector находит исходники OpenDirector, определяет ОС и архитектуру, проверяет права, менеджеры пакетов и системные зависимости. Он устанавливает недостающие Git, Node.js/npm, Python, FFmpeg и ffprobe, затем готовит окружение проекта.
Зависимости Python ставятся в .venv, после чего выполняются установка пакета и проверка через pip check. Сборочный backend запускает npm ci, сборку frontend и проверку встроенной версии OpenCode. Для исходной установки нужны Node.js >=24.22.1, Python >=3.11 и FFmpeg с точным энкодером libx264. После сборки скилл проверяет запуск и печатает команды для выбранной платформы.
Какие задачи решает
- установить OpenDirector из исходников
- найти недостающие системные зависимости
- создать или восстановить проектное окружение Python
- собрать frontend и приложение через установку пакета
- проверить запуск и получить команду для старта
Примеры запросов агенту
- Установи OpenDirector из исходников на текущей машине и проверь запуск.
- Восстанови незавершённую установку OpenDirector, установи недостающие зависимости и выдай команды для запуска.
- Запусти установку OpenDirector с проверкой
ffmpeg,ffprobe, Node.js и Python.
Как установить install-opendirector
1Через Skills CLI (Vercel)
npx skills add ThinkFlowLab/OpenDirector --skill install-opendirector -a claude-code2Вручную: положить файлы
~/.claude/skills/install-opendirector/SKILL.mdПапка скилла целиком; для одного проекта — .claude/skills/install-opendirector/
1Через Skills CLI (Vercel)
npx skills add ThinkFlowLab/OpenDirector --skill install-opendirector -a codex2Вручную: положить файлы
.agents/skills/install-opendirector/SKILL.mdДля всех проектов — ~/.agents/skills/
1Через Skills CLI (Vercel)
npx skills add ThinkFlowLab/OpenDirector --skill install-opendirector -a cursor2Вручную: положить файлы
.cursor/skills/install-opendirector/SKILL.mdCursor читает и .agents/skills/, и .claude/skills/
1GitHub CLI
gh skill install ThinkFlowLab/OpenDirector .agents/skills/install-opendirector2Вручную: положить файлы
.github/skills/install-opendirector/SKILL.md1Через Skills CLI (Vercel)
npx skills add ThinkFlowLab/OpenDirector --skill install-opendirector -a gemini-cli2Командой платформы
gemini skills install https://github.com/ThinkFlowLab/OpenDirector --path .agents/skills/install-opendirectorВручную: положить файлы
.devin/skills/install-opendirector/SKILL.mdСтарый путь .windsurf/skills/ тоже читается
Claude.ai
Упакуйте папку в zip и загрузите в Settings → Capabilities → Skills
По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов
Вопросы об install-opendirector
Как установить install-opendirector в Claude Code?
Для Claude Code выполните npx skills add ThinkFlowLab/OpenDirector --skill install-opendirector -a claude-code, затем откройте ~/.claude/skills/install-opendirector/SKILL.md.
Что делает install-opendirector?
Скилл устанавливает OpenDirector из исходников: проверяет платформу и зависимости, создаёт окружение Python, собирает приложение, проверяет запуск и выдаёт команды для старта. Он предназначен для первой установки и ремонта незавершённой установки, а не для обычного запуска.
Что нужно для работы install-opendirector?
Для исходной установки нужны Git, Node.js/npm, Python >=3.11, FFmpeg и ffprobe. Проверка FFmpeg требует точный энкодер libx264, а текущие требования исходников включают Node.js >=24.22.1.
Исходный текст
SKILL.md · Apache-2.0
--- name: install-opendirector description: Install OpenDirector from source on the current machine. Detect the platform, check and install missing prerequisites, prepare project dependencies, build the application, verify startup, and provide platform-specific launch commands. Use for first installation or repairing an incomplete installation; not for model deployment, application development, or routine startup alone. --- # Install OpenDirector Deliver a working local source installation and exact commands the user can run again. Reply in the user's language, supporting Chinese and English. A request to create or discuss this skill does not authorize installing the application. ## Locate the source and inspect the host 1. Honor the requested directory and revision. Otherwise reuse an OpenDirector checkout in the working directory or the repository containing this skill. Identify it by `package.json`, the root `pyproject.toml`, and `scripts/opendirector.py`, rather than its directory name. 2. If there is no checkout, use the source URL in the bundled platform guidance: `https://github.com/ThinkFlowLab/OpenDirector.git`. Clone into an available `OpenDirector` directory under the working directory. Never overwrite a nonempty directory. Record the resolved revision. ZIP sources are acceptable; their lack of Git metadata must not prevent installation. If Git is missing, inspect the host as below and bootstrap Git first, or use a source archive. Do not require Python just to discover the platform. 3. Inspect OS/version, architecture, shell, package managers, executable paths, and privileges using native commands before assuming Python or Node exists. Distinguish WSL from native Windows, and process architecture from host architecture when emulation is involved. Read only the relevant reference: [macOS](references/macos.md), [Linux/WSL](references/linux.md), or [Windows](references/windows.md). For another OS, report the unsupported installation path instead of applying a recipe for a different platform. 4. In the target source, read `README.md` or `README.zh-CN.md`, `.nvmrc`, `package.json` (including `engines`), the root `pyproject.toml`, `requirements.txt`, and the launcher. These files govern the selected revision; bundled commands are starting points, not permission to change its manifests. ## Check and install prerequisites Complete system prerequisite installation and user PATH setup before creating the project venv or installing application dependencies. Reuse compatible installed tools; install missing Git, Node.js/npm, Python, FFmpeg and ffprobe through the platform package manager or supported user-level installer. Python application packages still belong in the project venv. - Check actual paths, versions, and successful execution of Node.js, npm, Python, `ffmpeg`, and `ffprobe`; require FFmpeg's **exact** `libx264` encoder. Git is needed for cloning and for the current launcher in Git checkouts. Check Python's ability to create a venv with pip. An existing `.venv` needs its own interpreter check. Current source requirements are Node >=24.22.1 and Python >=3.11; reread the selected source instead of relying on these values. - Once Python 3.11+ is available, run this skill's standard-library-only helper with that interpreter, using absolute paths if necessary: ```sh python3 /path/to/install-opendirector/scripts/check_environment.py --root /path/to/OpenDirector ``` On Windows, use the selected `python.exe` with PowerShell's `&` operator. The helper prints JSON facts and stable status codes for the agent to explain in the user's language; exit 1 means action is needed, exit 2 means the source or its constraints could not be inspected. It does not install anything or start a service. It requires Python 3.11+ and supports simple `>=` runtime constraints; inspect newer constraint syntax manually if reported unsupported. Facts may include stale `.venv` or missing build files before installation. The helper's regression checks can be run with Python 3.11+ using `python3 -B /path/to/install-opendirector/scripts/test_check_environment.py`. - Reuse suitable tools; install only missing or unsuitable prerequisites using the relevant platform reference. Recheck after changes. Select an existing suitable executable before installing another version. Use a project venv for Python packages, never `sudo pip` or an upgrade of the system Python. - Make prerequisite tools discoverable on the user's PATH, not only through absolute paths or a temporary agent export. Inspect the user's shell and existing startup configuration, then add only the missing package-manager initialization or tool directories to the appropriate user configuration. Preserve existing entries and avoid duplicate additions. On Windows, use the installer-managed PATH or append missing directories to the user PATH without overwriting it. Carry these changes into the current process too. - Verify discovery and versions in a fresh instance of the user's normal shell without inheriting the agent's temporary PATH additions. Check `git`, `node`, `npm` (`npm.cmd` on Windows), the selected Python command, `ffmpeg`, and `ffprobe`; versioned Python commands such as `python3.12` are acceptable. If this fails, finish PATH setup before proceeding. Do not globally add the project's `.venv/bin` or `Scripts` directory to PATH. - Respect the execution environment's network and privilege approvals. Continue authorized ordinary installation without extra conversational confirmations. If an installer requires user interaction, report the precise pending action and continue independent checks. Do not bypass permissions, replace system executables, change global registries, or force dependency upgrades. ## Prepare dependencies and build Run from the source root, checking each command before proceeding. Use the selected compatible Python to create `.venv` only if a working suitable venv does not already exist. Preserve an incompatible or broken `.venv` by renaming it to an unused backup name before recreating it; do not delete its contents. POSIX (replace `python3` with the selected interpreter): ```sh python3 -m venv .venv .venv/bin/python -m pip install -v -e . .venv/bin/python -m pip check ``` Windows PowerShell (select the Python version before using `py -3`): ```powershell py -3 -m venv .venv & .\.venv\Scripts\python.exe -m pip install -v -e . & .\.venv\Scripts\python.exe -m pip check ``` The pip build backend runs npm ci, the frontend build, and the bundled OpenCode version check. Do not run these stages separately during installation. Wheel installs contain these resources and do not require Node.js; source installs do. If pip or npm downloads are slow or time out, retry the same installation with a user-specified or trusted alternative registry scoped to that process. Preserve global registry settings, lockfiles, and integrity checks. Skip the venv creation command when reusing one. Do not skip dependency installation solely because `node_modules` or `.venv` exists. `npm ci` requires the selected revision's matching lockfile; a mismatch is a concrete blocker, not a reason to regenerate it silently. Keep the source, user projects, model credentials, and configuration intact. Stop the failing stage on dependency resolution or build errors and report the actionable cause. Retry a transient network failure once with the same inputs; further attempts need new evidence or a changed network condition. ## Verify and hand off Follow [verification and delivery](references/verification.md), including the helper's `--installed` mode and a temporary launch through the real `serve` entrypoint. Report build success separately from startup and health results. Stop only the test processes you created, confirm port release, and leave the application stopped unless the user requested a running instance. Give the absolute source directory and copyable commands for the user's shell. The existing startup script is sufficient; do not create another launcher unless requested. POSIX uses `.venv/bin/python scripts/opendirector.py serve`; Windows uses `& .\.venv\Scripts\python.exe scripts/opendirector.py serve`. Explain the default URL `http://127.0.0.1:8000`, keeping the terminal open, `Ctrl+C` to stop, and `--port 8001` for a conflict. Include prerequisite PATH setup when needed. Installation does not deploy models or require GPU tooling. Say that AI features need model configuration in Settings / 设置 → Model access / 模型接入. Do not call paid model services, configure credentials, create user projects for a smoke test, enable a persistent service, or expose the server publicly as part of this workflow. Claim verification only for platforms and operations actually tested.
© ThinkFlowLab · Apache-2.0 · ThinkFlowLab/OpenDirector