Agent Skills › espressif/esp-claw › skills_lab_downloader

skills_lab_downloader

GitHub

从 ESP-Claw Hub 下载指定 Skill 或 App，验证元数据与硬件外设兼容性，执行文件安装并发布。

components/claw_capabilities/cap_skill_mgr/skills/skills_lab_downloader/SKILL.md espressif/esp-claw

触发场景

用户请求下载特定 ID 的 Skill 或 App 需要安装来自 skills-lab.esp-claw.com 的资源包

安装

npx skills add espressif/esp-claw --skill skills_lab_downloader -g -y

更多选项

非标准路径

npx skills add https://github.com/espressif/esp-claw/tree/master/components/claw_capabilities/cap_skill_mgr/skills/skills_lab_downloader -g -y

不安装直接使用

npx skills use espressif/esp-claw@skills_lab_downloader

指定 Agent (Claude Code)

npx skills add espressif/esp-claw --skill skills_lab_downloader -a claude-code -g -y

安装 repo 全部 skill

npx skills add espressif/esp-claw --all -g -y

预览 repo 内 skill

npx skills add espressif/esp-claw --list

SKILL.md

Frontmatter

{
    "name": "skills_lab_downloader",
    "metadata": {
        "cap_groups": [
            "cap_lua",
            "cap_http_request",
            "cap_boards",
            "cap_skill_manage",
            "cap_app_manage"
        ]
    },
    "description": "Download a Skill or App from the ESP-Claw hub by exact kind and id, check required peripherals, and publish it on the device."
}

Skills Lab Downloader

Download a Skill or App by its exact kind and id from skills-lab.esp-claw.com.

Rules

Require kind (skill or app) and the exact package id. Never infer kind from a title or default to Skill. If unknown, use skills_lab_search or ask the user.
Use {CUR_SKILL_DIR}/scripts/download_package.lua for both metadata and file downloads. Enable cap_http_request; if the host is blocked, tell the user to add *.esp-claw.com to the Web Console HTTP allowlist.
Read the returned metadata yourself; do not ask Lua to parse JSON. Verify kind and name match the requested kind and id. Stop on missing or mismatched metadata, HTTP errors, or missing files.
Before installation, compare metadata.peripherals with board_hardware_info. If required peripherals are missing, stop unless the user explicitly approves force installation.
Pass the complete metadata files array unchanged. It includes the manifest and nested resource paths; do not use extra_files or reconstruct a list from prose.
After all downloads succeed, call publish_app for an App or publish_skill for a Skill. Never publish after a download failure. Only report installation complete after publication succeeds; otherwise report that files were downloaded but publication failed.

Workflow

Fetch metadata using lua_run_script:

{
  "path": "{CUR_SKILL_DIR}/scripts/download_package.lua",
  "args": { "action": "fetch_metadata", "kind": "app", "id": "example-app" }
}

Check package identity and peripherals, then install using the exact files returned by the server. Set timeout_ms to 300000 for the complete download; the default 60-second script timeout may be too short for resource-heavy Apps. For example:

{
  "path": "{CUR_SKILL_DIR}/scripts/download_package.lua",
  "timeout_ms": 300000,
  "args": {
    "action": "install",
    "kind": "app",
    "id": "example-app",
    "files": ["launcher.json", "scripts/main.lua", "assets/audio/tone.pcm"]
  }
}

On success, call publish_app with {"app_id":"example-app"}. For a Skill, use kind: "skill" in both script calls and finish with publish_skill({"skill_id":"example-skill"}).

Storage and failures

Skills use /raw/<id>/ on the server and <DATA>/skills/<id>/ locally; SKILL.md is required.
Apps use /raw/apps/<id>/ on the server and <DATA>/apps/<id>/ locally; launcher.json is required.
Files stream directly to storage with a 20-second request timeout and a 10 MiB per-file limit. Nested directories are created automatically; paths must remain inside the package.
A failed or truncated download is removed. Previously downloaded files may remain; installation is not atomic and retrying overwrites files. Publication does not launch the App or activate the Skill.