技能详情（站内镜像，无评论）
许可证：MIT-0
MIT-0 ·免费使用、修改和重新分发。无需归因。
版本：v1.0.0
统计：⭐ 0 · 30 · 0 current installs · 0 all-time installs
⭐ 0
安装量（当前） 0
🛡 VirusTotal ：可疑 · OpenClaw ：可疑
Package：alannjaf/karpathy-autoresearch
安全扫描（ClawHub）
- VirusTotal ：可疑
- OpenClaw ：可疑
OpenClaw 评估
The skill's files and instructions largely match its stated purpose (an autonomous mutate→evaluate loop), but there are several mismatches and risky behaviors the user should understand before running it (undeclared binary requirements, the loop can modify arbitrary files and run arbitrary eval commands, and monetization/contact instructions outside the codebase).
目的
The name/description (autoresearch loop) aligns with included code (loop.py, evaluate.py) and README. However the skill metadata declares no required binaries or credentials while the scripts implicitly require git and a working shell environment and realistically will need LLM/backtest tooling and possibly API keys to implement evals — a mild incoherence between declared requirements and what is needed in practice.
说明范围
SKILL.md precisely instructs the agent to locate a mutable file, create or use an eval function, initialize git, mutate the file, run evals, and commit/revert. That scope is consistent with the purpose. Caution: these instructions give the agent permission to modify and commit arbitrary files in the target workdir — if the 'mutable file' or working directory is pointed at sensitive configs or system files the loop could change them. The refere…
安装机制
No install spec (instruction-only) and included scripts are a reference implementation. This is low install risk — nothing is downloaded from arbitrary URLs. The files will exist on disk as part of the skill bundle, which is expected.
证书
The skill declares no required env vars, but realistic use (LLM-judge, backtest data, external scoring harnesses) will likely require API keys, data access, or other credentials that are not declared. The README also asks users to pay $99 USDT to a provided crypto address and DM a Telegram handle to unlock a 'Pro' tier — this is external monetization/contact and not a credential leak, but it is unrelated to skill functionality and could be a r…
持久
always:false (normal). The skill can be invoked autonomously (default). Combined with its ability to modify files and run arbitrary eval commands (shell subprocess with shell=True), autonomous runs could have a wide blast radius if the agent is allowed to operate on sensitive directories. The skill does not request persistent system-level privileges or modify other skills' configs.
安装（复制给龙虾 AI）
将下方整段复制到龙虾中文库对话中，由龙虾按 SKILL.md 完成安装。
请把本段交给龙虾中文库（龙虾 AI）执行：为本机安装 OpenClaw 技能「Karpathy Autoresearch」。简介：Autonomously optimize any OpenClaw skill by running it repeatedly, scoring outp…。
请 fetch 以下地址读取 SKILL.md 并按文档完成安装：https://raw.githubusercontent.com/openclaw/skills/refs/heads/main/skills/alannjaf/karpathy-autoresearch/SKILL.md
（来源：yingzhi8.cn 技能库）SKILL.md
暂无本地缓存内容，可在后台执行详情同步。