⚠
Purpose & Capability
The skill claims to be an instruction-only virtual companion creator, but many included scripts require two external API keys (VIDU_KEY and TAVILY_API_KEY), expect an OpenClaw CLI and messaging integrations (Feishu/Telegram/etc.), read/write files under ~/.openclaw/workspace/skills/partner-creator, and can run a background push daemon. The registry metadata declared no required env vars or credentials, which conflicts with the SKILL.md and the scripts. Also a Feishu APP_SECRET and APP_ID are hardcoded in send-feishu-video.sh, which is not justified by the metadata and increases capability beyond the stated simple creation/chat functions.
⚠
Instruction Scope
SKILL.md instructs the agent to ask users for Vidu/Tavily API keys and to set them as session env vars, but the scripts read files and directories (assets/, references/, config/) under the user's home workspace, download images from third-party URLs, call external APIs (api.vidu.cn, api.tavily.com), and send media/messages to external platforms (via openclaw CLI or direct Feishu API). The instructions also describe downloading user-supplied photos and using a feishu_im_bot_image tool. There is contradictory guidance: SKILL.md forbids persisting API keys, yet push-daemon.sh can source a .env file in the skill directory and scripts reference persistent config/push-config.json — giving the agent/leverage to access or persist data beyond a single session.
ℹ
Install Mechanism
There is no formal install spec (instruction-only), which reduces installer-level risk, but the package includes a substantial set of executable scripts and NodeJS code that will run on the host when invoked. No external binary download/install occurs from unknown URLs, but the scripts themselves perform network operations (curl, node fetch) and will write to the user's workspace (assets/, config/, logs, PID files) when run. Because code is present, installing or running the skill results in code execution on-host — review scripts before use.
⚠
Credentials
Although registry metadata lists no required env vars, SKILL.md and the scripts clearly require at least two secrets: VIDU_KEY (vda_...) and TAVILY_API_KEY (tvly-...). The skill also expects TARGET_USER / chat IDs and may read a .env file under the skill directory. Critically, send-feishu-video.sh contains a hardcoded Feishu APP_ID and APP_SECRET; embedding these credentials in the skill is disproportionate and risky because it allows the script to obtain tokens and post messages as that application. The number and type of credentials used (user-supplied API keys plus embedded third-party app secret) are not justified by the metadata and should be questioned.
⚠
Persistence & Privilege
The skill does not force-install (always: false) but includes a push-daemon.sh and hourly-push.sh that implement a long-running background daemon writing PID, lock and log files under the skill directory and invoking hourly network and message-sending actions. This yields persistent behavior if the user starts the daemon. There is contradictory guidance about not persisting API keys while the daemon supports loading .env and config/push-config.json (which can contain target IDs). Because the skill can be made to run autonomously (user-started daemon + scripts that send messages), its persistence and ability to send data externally is a meaningful risk.