OpenClaw Security Audit — BLUF Report — 技能工具
v1.0.1Run a non-interactive OpenClaw security audit that produces a structured BLUF report with posture rating, ranked findings, and one-line fix commands.
0· 71·0 当前·0 累计
by @infectit007 安全扫描
OpenClaw
可疑
high confidenceThe skill's stated purpose (run and format a local OpenClaw audit) is plausible, but the instructions contain several clear contradictions and omissions (claims no network / read-only while showing Telegram network calls, file appends, cron scheduling, and an optional --fix mode) and the metadata does not declare the environment variables the runtime will actually use.
评估建议
This skill mostly does what it says (format OpenClaw audit output), but there are clear inconsistencies you should resolve before installing or automating it: 1) The SKILL.md asserts 'no network' and 'read-only' yet includes a Telegram POST and file-append examples—treat the Telegram path as a network behavior that requires BOT_TOKEN/CHAT_ID. 2) The metadata does not declare the TELEGRAM env vars referenced; ask the publisher to list them explicitly if you plan to use Telegram delivery. 3) The s...详细分析 ▾
⚠ 用途与能力
The skill claims to only run `openclaw security audit --deep` and be read-only, which matches its purpose. However, it also documents delivery via Telegram, appending to a memory file, and scheduling via `openclaw cron add` — behaviors outside a strict read-only formatter. The declared requirements list no env vars or config paths, yet the runtime instructions reference TELEGRAM_BOT_TOKEN and MASTER_TELEGRAM_ID.
⚠ 指令范围
SKILL.md contains contradictory statements: an explicit 'No network calls' / 'Read-only' claim, but later provides Python code that posts to Telegram (network) and shell examples that append to memory files (writes). It also instructs optional use of `--fix`, which would modify configuration if executed. The instructions therefore allow reading, writing, network I/O, and potentially config changes despite the trust claims.
✓ 安装机制
Instruction-only skill with no install spec or code files. That minimizes surface area and nothing is being downloaded or installed by the skill itself.
⚠ 凭证需求
Metadata declares no required environment variables, but instructions reference TELEGRAM_BOT_TOKEN and MASTER_TELEGRAM_ID for delivery. That omission is inconsistent: if the skill can deliver via Telegram it should declare those env vars. No unrelated third-party credentials are requested, but the mismatch between documentation and declared requirements is concerning.
⚠ 持久化与权限
always:false (good) and disable-model-invocation is default (normal), but the skill guides creating scheduled cron jobs (`openclaw cron add`) and appending to memory files—both are forms of persistence. The skill also documents an optional `--fix` mode that can change system state. The SKILL.md's claim of 'read-only' conflicts with these persistent/write-capable instructions.
安全有层次，运行前请审查代码。
运行时依赖
无特殊依赖
安装命令
点击复制官方npx clawhub@latest install eva-security-audit
镜像加速npx clawhub@latest install eva-security-audit --registry https://cn.longxiaskill.com 镜像可用
本土化适配说明
OpenClaw Security Audit — BLUF Report — 技能工具 安装说明： 安装命令：["openclaw skills install eva-security-audit","npx clawhub@latest install eva-security-audit","openclaw security audit --deep","openclaw security audit --deep > /tmp/audit-$(date +%Y%m%d-%H%M).txt 2>&1","openclaw cron add --name \"security-audit:daily\" --cron \"0 4 * * *\" --prompt \"Run the eva-security-audit skill and send the report to memory and Telegram.\"","openclaw cron list"]