OpenClaw LLM Tools

Universal Tool Definition System for LLM function calling. Define tools once, use with any LLM provider (OpenAI, Anthropic, Gemini, etc.). JSON Schema valida...

This package implements a legitimate tool registry and format converters, but the README includes examples that are dangerous if used carelessly: avoid registering handlers that perform eval() or that read arbitrary filesystem paths unless you control and audit them. Before installing or exposing this skill to LLM-driven calls: 1) review any handler functions you register — never use eval on untrusted input; 2) avoid handlers that read sensitive files or run shell commands, or run them in a sandboxed environment; 3) prefer pure, constrained handlers that accept validated inputs; 4) be cautious when exposing tool execution to untrusted or third‑party models (an LLM could trigger calls); and 5) note the documentation/code mismatches (missing modules and the ephemeral decorator registry) and consider a code review or tests to ensure behavior matches your expectations.

Type: OpenClaw Skill Name: oc-llm-tools Version: 1.0.0 The skill bundle provides a framework for managing LLM tool definitions but includes high-risk examples and capabilities. Specifically, SKILL.md contains a documentation example using 'eval()' for a tool handler, which is a classic Remote Code Execution (RCE) vulnerability. Additionally, the skill's metadata defines broad 'read', 'write', and 'exec' tool permissions, which, while potentially useful for a tool management system, present a significant attack surface without explicit safety constraints.