Skill Scan Wrapper#

当你要在安装一个本地 skill、归档或 release bundle 前做一次快速安全检查时，使用这个 skill。

⚠️ Security Notice#

This tool scans locally by default and requires user trust in the binary you run. Native LLM review sends a bounded text packet to a user-configured endpoint; external LLM review follows the external scanner's data policy. Always verify the checksum after downloading. For maximum security, build from source (recommended).

Binary Included#

PropertyValueLocationassets/bin/skillscanVersionv0.11.1PlatformLinux x64SHA-256bb3847f3afca734ef0cf6c061f553602b3df5dad23f726215d25812b74d01bf0

Verify locally before running:

bash

sha256sum assets/bin/skillscan
# Compare output with the SHA-256 value above

This bundled package includes a pre-compiled binary. You can still build from source if you prefer:

bash

git clone https://gitee.com/random_player/cmic-skill-scanner.git
cd cmic-skill-scanner && cargo build --release

前置条件#

• 默认 auto 模式会优先尝试本地可解析的 external scanner；没有可用 scanner 时回退到内置 native 引擎
• 不安装 external scanner 也可以使用：单二进制会回退到 native
• --upload-url 和 --use-llm 功能 默认禁用，仅在用户显式配置时启用

信任模型#

This is an open-source (MIT-0) package. The binary (bundled or downloaded) is a convenience only — it does not grant any additional trust.

Your options:

ApproachTrust RequirementVerificationBuild from sourceNone (you control everything)Manual code reviewBundled/downloaded binaryYou trust the release hostSHA-256 checksum

Default behavior and trust boundaries:

• CMIC does NOT upload reports unless you configure --upload-url
• CMIC does NOT configure an LLM endpoint unless you set --use-llm
• auto may execute a locally resolved external scanner; use --engine native to prevent that
• Does NOT access credentials or SSH configs as scan targets unless they are under the path you explicitly scan

工作流程#

1. 调用 skillscan：

bash

skillscan review /path/to/target --format markdown
skillscan review /path/to/skills --output-dir /tmp/skillscan-out

1. 阅读输出中的：输入类型、完整度、engine 执行状态、findings

网络上传功能 (默认禁用)#

⚠️ This feature is completely optional and disabled by default. It requires explicit user configuration via --upload-url.

This applies to report upload only. LLM review is a separate, explicit network feature; a resolved external scanner has its own dependency and network behavior.

What gets sent (only when you configure --upload-url):

• A structured JSON report containing detection findings
• An instance identifier you supply via --instance-id
• No skill source code, credentials, or system configuration is ever transmitted

Optional LLM Review#

⚠️ This feature is completely optional and disabled by default. It requires explicit --use-llm and works with native, external, and auto.

bash

skillscan review /path/to/target --engine native --use-llm   --llm-endpoint http://localhost:11434/v1   --llm-model your-model

With native, the endpoint receives static-finding summaries and up to 24 text files from the target package (up to 8 KiB each and 64 KiB total). Lines named api_key, token, password, secret, or authorization are redacted at a basic level. This is not a guarantee that all sensitive text is removed: use only a trusted endpoint and confirm the data-handling policy.

With a Cisco-compatible external engine, CMIC adds --use-llm and passes endpoint, model, and optional key to the child process through SKILL_SCANNER_LLM_* variables; the external scanner controls its own packet, redaction, and failure behavior. auto tries that external LLM first and falls back to native if the external process fails. API keys are never written to the scan command or report.

If the native LLM request fails, the native static result is still returned and engine.fallback_reason records the reason.

外部引擎集成#

Default auto mode tries a locally resolved external scanner first and falls back to native when it is unavailable or fails. Use --engine external to require the external result, or --engine native to run only the built-in engine.

CMIC passes the target path to a user-configured local tool. The tool runs with the current user's process permissions; trust the external tool and its configuration separately. CMIC only configures the external LLM endpoint when --use-llm is explicitly set; the tool may otherwise bootstrap dependencies or use its own network configuration.

Permissions Required#

ScopeReasonRead files in target pathTo analyze skill source code for patternsWrite to --output-dirTo save scan reports locallyExecute binary / external toolTo run the selected scanner engine; auto may invoke a locally resolved external scannerNetwork (optional)--upload-url sends a JSON report; native --use-llm sends a bounded text packet; external --use-llm follows the external tool's data policy