跳转到正文
报告库
用途分类 / 其他用途

Caveman Optimize Skill 安全审计

作者说它能做什么(原文)

>

第三方安全检查结论

发现安全风险

已检查文件
1
发现的风险
3
会不会运行危险命令?检查是否下载程序后直接运行、让他人远程控制电脑,或藏起要运行的命令。发现 1 项风险
中风险

获批评估会执行未预先限定的夹具命令和仓库检查

原文依据:3 处
发现了什么

技能要求记录并运行“确切命令”,还会执行仓库的 focused code checks,但没有限定允许的命令、网络访问或进程权限。如果仓库或测试脚本不可信,这些命令可能执行任意项目代码。

为什么需要注意

恶意或被篡改的脚本可能读取当前进程可见的凭据、访问网络、修改工作区文件,或调用本机工具。操作员批准评估设计能降低风险,但不会自动证明其中命令安全。

这段代码的正常用途

这里的命令执行属于经操作者明确批准的评估流程:技能要求先提出包含确切夹具和命令的方案并获得批准,然后才编辑和运行配对评估。仓库检查确实可能执行项目代码,但原文没有要求执行隐藏、任意或未经批准的命令。用户仍应在批准时确认命令及其网络和进程权限。

这项判断针对展示的代码和适用条件,不表示风险已经实际发生。
SKILL.md:67来自说明文档打开原文件
Propose one minimal candidate change and a **paired eval** before editing. Theevaluation must run baseline and candidate on identical fixed inputs and record:- the task-outcome or quality check that must remain acceptable;- the same token, byte, or provider-counted cost measure for both arms;- the exact fixture, command, and environment used; and- any confounder that prevents a fair comparison.
查看另外 2 个位置
SKILL.md:82来自说明文档打开原文件
Keep the diff at the evidenced callsite and preserve existing safety controls.Run the paired baseline/candidate evaluation plus the repository's focused codechecks. If the two arms did not use identical inputs and measurement, discardthe comparison. If quality regresses or the resource result is inconclusive,
SKILL.md:75来自说明文档打开原文件
Ask for approval of the candidate and eval design. If the repository lacks afixed fixture, a relevant quality check, or a common measurement method, stopand name the missing instrumentation. Ordinary unit tests alone do not prove anoptimization.
会不会泄露文件和密钥?检查是否发送含密码或密钥的文件,以及代码里是否直接写了密钥。发现 1 项风险
中风险

要求记录“确切环境”可能把环境中的秘密带入报告

原文依据:2 处
发现了什么

配对评估要求记录使用的“exact fixture, command, and environment”,最终报告又包含命令和实际结果。文本没有说明环境记录必须只包含非敏感配置,也没有要求清除令牌、密钥或命令输出中的秘密。

为什么需要注意

如果代理将完整环境变量或含秘密的命令输出复制到报告,报告被分享、提交或上传时可能泄露 API 密钥、会话令牌、本机路径或内部服务信息。

这段证据能说明什么

“记录确切环境”含义不明确,可能只是记录运行条件,也可能被误解为完整导出环境变量;后者可能泄露令牌或密钥。原文没有明确要求读取或披露环境变量,最终报告模板也只要求列出检查命令和结果,因此现有证据不足以确认会收集秘密。用户可要求作者明确仅记录非敏感环境信息并对输出脱敏。

这项判断针对展示的代码和适用条件,不表示风险已经实际发生。
SKILL.md:70来自说明文档打开原文件
- the task-outcome or quality check that must remain acceptable;- the same token, byte, or provider-counted cost measure for both arms;- the exact fixture, command, and environment used; and- any confounder that prevents a fair comparison.
查看另外 1 个位置
SKILL.md:99来自说明文档打开原文件
Recorded profile: <server observation, verbatim>Candidate: <file:line and approved change>Paired eval: <identical input/fixture, baseline result, candidate result>Quality check: <actual result>Code checks: <commands and actual results>Accounting: report-only profile; $0 opportunity band; no inferred or verified savings
会不会删除文件或一直在后台运行?检查是否大范围删除文件、改写磁盘,或设置自动启动。发现 1 项风险
低风险

技能会进行持久代码修改,并可能自动撤销候选改动

原文依据:4 处
发现了什么

在两次明确批准后,技能会编辑调用点;若评估质量下降或资源结果不明确,它被指示撤销该候选编辑。文本未规定如何检测或保留编辑后同一区域出现的并发用户改动。

为什么需要注意

正常情况下这是预期且受控的优化流程;但若用户或其他工具同时修改相同代码,机械撤销可能覆盖或冲突于这些后续改动。

这段代码的正常用途

技能确实允许持久修改和在失败时撤销,但修改须经过两阶段明确选择与批准,并被限制在有证据的调用点;撤销也明确限定为“仅此候选编辑”。并发用户修改冲突是通用版本控制风险,所引原文没有指示覆盖或丢弃其他改动。用户批准前仍应确认工作区状态及撤销方式。

这项判断针对展示的代码和适用条件,不表示风险已经实际发生。
SKILL.md:75来自说明文档打开原文件
Ask for approval of the candidate and eval design. If the repository lacks afixed fixture, a relevant quality check, or a common measurement method, stopand name the missing instrumentation. Ordinary unit tests alone do not prove anoptimization.
查看另外 3 个位置
SKILL.md:82来自说明文档打开原文件
Keep the diff at the evidenced callsite and preserve existing safety controls.Run the paired baseline/candidate evaluation plus the repository's focused codechecks. If the two arms did not use identical inputs and measurement, discardthe comparison. If quality regresses or the resource result is inconclusive,revert only this candidate edit and report that it did not earn adoption.
SKILL.md:53来自说明文档打开原文件
Present the available supported observations without ranking them. Include theid, the exact title, the exact observation, and `last_seen_at`. Ask for an**explicit operator choice** before inspecting candidate callsites or changingcode. If no supported current observation exists, stop with no edit.
SKILL.md:80来自说明文档打开原文件
## 4. Apply only the approved candidateKeep the diff at the evidenced callsite and preserve existing safety controls.Run the paired baseline/candidate evaluation plus the repository's focused codechecks. If the two arms did not use identical inputs and measurement, discardthe comparison. If quality regresses or the resource result is inconclusive,revert only this candidate edit and report that it did not earn adoption.
会不会绕过安全保护?检查是否跳过网站安全验证、开放过多文件权限,或取消操作前的确认。未发现风险
会不会误导 AI 或隐藏内容?检查工作说明是否要求 AI 忽略你的指令、干扰检查结果,或夹带看不见的文字。未发现风险
会不会偷偷改推广链接或收款方?检查是否强制替换推广链接或收款对象,同时要求隐瞒更改。未发现风险

Skill 逻辑拆解

5 个说明模块

此技能先使用已登录的 Caveman CLI 读取报告,仅接受四种当前的 report-only profile,并明确排除生命周期数据、旧 ID 和历史提案。

查看原文
SKILL.md:18来自说明文档打开原文件
Require a logged-in Caveman CLI session and run:```bashcaveman opportunities list```Read only the `report_only_observations` array. Do not select from the lifecycle`data` array. Preserve each server-provided `title` and `observation` verbatim.Handle these exact repository-profile ids:
SKILL.md:40来自说明文档打开原文件
Never select or apply these retired ids:- `context-window-bloat`- `tool-catalog-utilization`- `verbose-tool-output`Treat any occurrence of a retired id in a stale proposal, local file, or oldresponse as historical context only. Never revive its money, recipe, orlifecycle claim. If the only actionable-looking item is `unlabeled-traffic`,

在查看候选调用点或修改代码前,技能要求操作员明确选择;在编辑前还需再次批准具体候选改动和配对评估设计。

查看原文
SKILL.md:53来自说明文档打开原文件
Present the available supported observations without ranking them. Include theid, the exact title, the exact observation, and `last_seen_at`. Ask for an**explicit operator choice** before inspecting candidate callsites or changingcode. If no supported current observation exists, stop with no edit.
SKILL.md:67来自说明文档打开原文件
Propose one minimal candidate change and a **paired eval** before editing. Theevaluation must run baseline and candidate on identical fixed inputs and record:- the task-outcome or quality check that must remain acceptable;- the same token, byte, or provider-counted cost measure for both arms;- the exact fixture, command, and environment used; and- any confounder that prevents a fair comparison.Ask for approval of the candidate and eval design. If the repository lacks afixed fixture, a relevant quality check, or a common measurement method, stop

获批后,技能会修改有证据支持的调用点、执行基线与候选评估及代码检查;若质量下降或结果不明确,会撤销该候选改动。

查看原文
SKILL.md:80来自说明文档打开原文件
## 4. Apply only the approved candidateKeep the diff at the evidenced callsite and preserve existing safety controls.Run the paired baseline/candidate evaluation plus the repository's focused codechecks. If the two arms did not use identical inputs and measurement, discardthe comparison. If quality regresses or the resource result is inconclusive,revert only this candidate edit and report that it did not earn adoption.

技能禁止创建实验、改变机会生命周期或启用优化器,并要求将结果描述为特定夹具上的观察,而非生产节省证明。

查看原文
SKILL.md:88来自说明文档打开原文件
Do not create a Caveman experiment or proposal, mark an opportunityimplemented, change its lifecycle, or switch on an optimizer. Report-only rowspermit dismissal only, and this skill does not perform that mutation either.
SKILL.md:107来自说明文档打开原文件
Never convert token or byte reduction into dollars without provider-complete,same-request accounting supplied by the product's verified methods. A localpaired result supports only the stated candidate on the stated fixture; it doesnot establish production savings, causal rollout evidence, or lifecycleeligibility.
从这里开始 · 工作说明SKILL.md
caveman-optimize
连线表示工作说明包含的模块,不是实际运行顺序。点击模块可查看原文。
文件与检查记录1 个文件

检查范围与遗漏

逐文件查看涉及的内容

下方列出本次涉及的原文范围;纳入检查不代表已查清所有问题。

  • SKILL.md已纳入全文

这份报告只针对上方版本。我们看了拿到的代码和说明文件,没有实际运行 Skill,也没有检查它另外安装的软件包。因此,这不是“保证安全”的承诺;换了版本或使用环境,结果也可能不同。

  • SKILL.md工作说明

代码和说明中提到的操作

运行命令
SKILL.md:20来自说明文档打开原文件
```bashcaveman opportunities list
读取了多少行
112
文件校验值(用于核对版本)
ca6eef52fb1a0f57a41e0d4c50cb3caa2ddd3e67aedbab9a8bad92dbb7554d04