自动运行并调试 Python、C 和 x86_64 汇编代码，错误时自动修复并返回可执行版本。

This skill does what it says (runs and auto-fixes code), but be cautious: it executes arbitrary user-supplied programs (compiled C/assembly and Python) on the host. Before installing, consider: 1) Only run this in an isolated, sandboxed environment (container/VM) because binaries could read files, access network, or perform malicious actions. 2) Require the registry metadata to list required system binaries (python3, gcc, nasm, ld) so you can ensure they exist in the sandbox. 3) Ask the publisher to document any resource limits (timeouts, memory), the exact behavior of clawhub.llm.generate, and whether the platform enforces sandboxing. 4) If you cannot guarantee isolation, do not install or invoke this skill with untrusted code. The mismatch between declared dependencies and actual runtime requirements is the main incoherence here.

Type: OpenClaw Skill Name: code-run-auto-fix Version: 1.0.0 The skill provides tools (run_python, run_c, run_assembly in skill.py) designed to execute arbitrary code and automatically 'fix' it using an LLM. While this behavior is consistent with the stated purpose in SKILL.md, the capability to compile and run C and Assembly code, combined with an automated execution loop based on LLM-generated output, creates a high-risk environment for arbitrary code execution. No evidence of intentional malice, such as data exfiltration or hardcoded backdoors, was found.