Ucm

Provides API marketplace access for AI agents. Discovers and calls external capabilities including web search, image generation, code execution, text-to-spee...

This skill appears internally consistent with an API marketplace aggregator, but consider the following before installing: (1) network access is required and calls to the marketplace will transmit request data and parameters off‑device — avoid sending sensitive secrets or private data to the service. (2) The registration script saves your API key to ~/.config/ucm/credentials.json and prints it to stdout; treat the key like any API secret and secure that file. (3) Calls to UCM are paid per endpoint — review pricing and quotas to avoid unexpected charges. (4) Verify the service (https://ucm.ai and registry.ucm.ai) and its privacy/policy terms if you plan to send user data. (5) If you need stronger local security, avoid using the registration helper and instead manage the key manually in a secure keystore.

Type: OpenClaw Skill Name: ucm Version: 1.1.1 The skill bundle is classified as suspicious due to its explicit provision of high-risk capabilities, notably `ucm/code-sandbox` for executing arbitrary code (Python/JS/Bash/R/Java) in a sandboxed environment, `ucm/web-scrape` for extracting webpage content, and `ucm/email` for sending emails. While these services are documented as part of the skill's purpose as an API marketplace, they introduce significant vulnerability surfaces. Misuse by an AI agent (e.g., via prompt injection) or vulnerabilities in the underlying sandbox could lead to arbitrary code execution, sensitive data exfiltration, or unauthorized communication. The `scripts/register.sh` includes input sanitization for JSON payloads, mitigating direct shell injection for those specific inputs, and the `SKILL.md` does not contain hidden malicious prompt injection directives. However, the inherent power and potential for misuse of the exposed services warrant a 'suspicious' classification rather than 'benign'.