Agent Benchmark

提供基于12项标准化任务的AI Agent能力评估，涵盖文件操作、数据处理、系统操作、健壮性和代码质量，自动评分生成报告。

What to check before installing or running this skill: - Do not run the skill with sensitive environment variables present. The runner forwards your process.env into any executed task code and tasks can read files and env vars. - The SKILL.md instructs running a PowerShell runner that is not present in the package; instead a Node-based runner (index.js) is included — confirm which runner you intend to use and why documentation and code differ. - Review index.js and any task files (tasks.json, default-tasks.json, extended tasks) before execution. The runner will write task code to disk and execute it; untrusted tasks could perform arbitrary I/O or network activity. - Note the runner writes reports to '../../memory/benchmark-results.md' (outside the skill folder). If you don't want persistent copies in agent memory, modify the code or run in an isolated workspace. - Ensure required interpreters (python/node/go) are either intentionally available or intentionally absent. If you don't want the skill to execute arbitrary interpreters, run it in a sandbox without those binaries. - Ask the publisher (author/owner) for clarification: why does documentation reference PowerShell runner while code is Node-based, and why are some files (benchmark-runner.ps1) referenced but missing? If you cannot verify the source or intent, avoid running it on sensitive machines.

Type: OpenClaw Skill Name: yuyonghao-agent-benchmark Version: 0.1.1 The skill bundle is a benchmarking tool for AI agents, providing a suite of tasks to evaluate capabilities in file operations, data processing, and system interaction. It includes a Node.js runner (index.js) for Python, Node.js, and Go tasks, and references a PowerShell-based runner for a larger set of system tasks. While the tool executes arbitrary code and accesses environment variables as part of its benchmarking process, these actions are consistent with its stated purpose, and no evidence of malicious intent, data exfiltration, or unauthorized persistence was found.