by 安全扫描
OpenClaw
可疑
medium confidenceThe skill mostly matches its stated KB/onboarding purpose, but there are small-but-important incoherences around declared network behavior and optional integrations (API keys / Telegram) and included install/test scripts — review these before installing.
评估建议
This skill appears to be a coherent knowledge-base/onboarding product, but review these before installing:
- Clarify network behavior: SKILL.md metadata says it does not make network requests, yet the skill advertises optional integrations (OpenAI/Anthropic, Notion, Confluence, Google Drive, Telegram) that necessarily require outbound calls. Assume that connecting any of those will send data off your host.
- Inspect scripts locally: open install.sh and test/smoke-test.sh before running bash ins...详细分析 ▾
ℹ 用途与能力
The name/description (knowledge base, onboarding, search, indexing) align with the files and config.yaml. Optional environment variables (OpenAI/Anthropic/Notion/Confluence/Google/Telegram) also make sense for the advertised integrations. The claim 'works without external SaaS' is defensible because config.yaml supports local storage, while external connectors are optional. Overall capability requests are proportionate to the described purpose, but the presence of many optional connector keys expands the attack surface if you enable them.
ℹ 指令范围
SKILL.md instructs the user to fill config.yaml, run install.sh and smoke-test, index documents, and optionally connect remote storage and LLM APIs. Those actions require reading and uploading company documents to embedding/LLM providers when connectors are enabled. SKILL.md metadata reports network_behavior.makes_requests: false and uses_agent_telegram: false which contradicts the described Telegram bot, external indexing, and API-key options — this mismatch is noteworthy and should be clarified before use.
✓ 安装机制
There is no platform install spec (instruction-only), which reduces automatic risk. Included scripts (build.sh shown) perform benign packaging and smoke-test activities; build.sh contains no network downloads and uses only local shell utilities. install.sh and test/smoke-test.sh exist but their contents are not shown here — they should be inspected locally before execution. No downloads from arbitrary URLs were observed in the included build.sh.
ℹ 凭证需求
No required env vars are declared, but SKILL.md lists several optional API keys (ANTHROPIC_API_KEY, OPENAI_API_KEY, NOTION_API_KEY, CONFLUENCE_API_TOKEN, GOOGLE_SERVICE_ACCOUNT_JSON, TELEGRAM_BOT_TOKEN). These map to supported integrations (embeddings/LLMs, remote KB sources, Telegram bot) and are therefore reasonable if you intend to enable those integrations. Because they are optional, the request for multiple credentials is not inherently disproportionate — but enabling them lets the skill access external services and your documents, so minimize privileges and scope when provisioning keys.
✓ 持久化与权限
always is false; the skill is user-invocable and may be invoked autonomously (default), which is normal. There is no indication the skill requests permanent platform-level presence or modifies other skills' configs. The package contains install/test scripts that run locally only if you execute them; nothing indicates forced persistence beyond normal skill behavior.
安全有层次，运行前请审查代码。
运行时依赖
无特殊依赖
版本
latestv3.5.12026/4/21
Initial marketplace release.
● 无害
安装命令
点击复制官方npx clawhub@latest install raai-knowledge-base-pro
镜像加速npx clawhub@latest install raai-knowledge-base-pro --registry https://cn.longxiaskill.com