Model Setup

安全地管理 OpenClaw 模型配置。用于添加、测试和配置新模型到 models.json，包括 API key 验证、模型可访问性测试、工具调用功能检测、设置默认模型和配置到特定 agent。所有操作都会自动备份配置文件以确保安全。

This skill appears to implement the claimed model-management features, but exercise caution before installing/using it: - Required runtime: ensure Python 3 and curl are available (metadata omitted them). - Secret handling: the scripts will store API keys in models.json in cleartext and pass keys on the curl command line (visible to other local users). Avoid entering high-privilege or long-lived keys unless you accept that risk. Prefer short-lived keys or rotate keys after use. - Prefer to inspect and/or modify the scripts before use: consider changing test_model.py to use a Python HTTP client or pass Authorization via a safer mechanism (e.g., curl reading header from a file or using environment variables) to avoid command-line exposure. - Check file paths carefully: the tool will write to models.json and agent config.json paths you provide; back up those files and verify permissions (restrict to your user) before running. - Validate baseUrl: because test_model.py will POST to the provided baseUrl, only test against endpoints you trust to avoid sending credentials to a malicious server. If you want to proceed, review and harden the scripts (avoid CLI-exposed secrets, lock down models.json permissions, and confirm agent paths) or ask the author to address these issues.

Type: OpenClaw Skill Name: model-setup Version: 1.0.1 This skill bundle is classified as suspicious due to critical vulnerabilities that could lead to arbitrary command execution and arbitrary file writes. The `SKILL.md` instructs the AI agent to execute shell commands (`python3 scripts/test_model.py`, `python3 scripts/add_model.py`) using user-provided inputs (JSON strings, model IDs, agent paths). If the agent does not properly sanitize or quote these inputs when constructing the shell command, it creates a severe prompt/shell injection vulnerability, allowing arbitrary command execution. Additionally, `scripts/add_model.py` uses a user-provided `agent_path` to construct file paths (`Path(agent_path) / "agent" / "config.json"`), making it vulnerable to path traversal attacks that could allow writing to arbitrary `config.json` files outside the intended directories. The `scripts/test_model.py` also executes `curl` commands with user-provided API keys and base URLs, further expanding the attack surface.