CMIC Skill Scanner (macOS ARM64)

使用内置 Rust 引擎审计待安装的 skill 包或归档，并可选桥接外部 scanner。

Security Scan

The skill's description mostly matches a local scanner, but the package and instructions contain contradictory/omitted artifacts (it claims a bundled binary that is not present in the manifest and points to remote sources), which is an incoherence you should resolve before trusting or running anything.

Purpose & Capability

The stated purpose (local Rust-based skill scanner for macOS ARM64) is coherent with the instructions to run a local binary or build from source. However the SKILL.md repeatedly references a bundled binary at assets/bin/skillscan and checksum files, while the provided file manifest lists only INSTALL.md, SKILL.md, agents/openai.yaml, and assets/build/build-info.json — the actual binary and checksum file referenced in the docs are missing. Additionally the registry metadata earlier said 'No code files present' despite the presence of build-info and installation docs; these contradictions are unexplained and could indicate packaging errors or a deliberate omission.

Instruction Scope

Runtime instructions tell the agent/user to execute a local binary (./assets/bin/skillscan) or to clone a remote repo (gitee) and build. Running a binary not included in the package would require fetching remote artifacts — the SKILL.md does not supply a vetted release URL (it points to a repo) and the package lacks an installation spec. The doc claims uploads and external engines are disabled by default, which is good, but the presence of optional --upload-url and --engine external means a user or agent could enable network behavior; instructions do not require or declare any environment variables or credentials.

Install Mechanism

The skill has no formal install spec in the registry, yet INSTALL.md and SKILL.md describe a precompiled binary bundled at assets/bin/skillscan and a packaged sha256 file. The manifest does not include that binary or the sha256 file. This mismatch (claims of a bundled executable without it being present) is the primary install-related risk: to follow the docs a user/agent would need to download or fetch the binary from external sources, which increases risk unless the source is verified. The suggested build-from-source path points at a gitee repo rather than an official release host (GitHub/GitLab/GitHub releases), which is less standard but not inherently malicious.

Credentials

The skill declares no required environment variables, credentials, or privileged config paths. The permissions described (read target path, write output-dir, execute the binary, optional network only when --upload-url is configured) are proportionate to a local scanner. There is no request for unrelated cloud credentials or secrets. That said, the optional upload feature would transmit structured findings and an instance-id if enabled — avoid configuring upload-url or supplying instance identifiers until you validate the binary/source.

Persistence & Privilege

The skill is not always-enabled and is user-invocable. It does not request persistent privileges or modifications to other skills' configs. Autonomous invocation is allowed (default), which is normal for skills; this is not in itself a distinguishing risk here.

What to consider before installing

Do not run or download any binary referenced by this SKILL.md until you resolve the packaging contradictions. The SKILL.md claims a bundled binary at assets/bin/skillscan and a checksum file, but the package manifest does not include them. Actions to take before trusting this skill: - Ask the publisher to provide the missing assets (binary and checksum) or a reproducible build artifact. Do not rely on an undocumented remote download. - If you need to use it, build from source yourself and verify the repository and commit SHA; prefer locally-built artifacts. - If you must run a prebuilt binary, verify its SHA-256 exactly matches the reported checksum and confirm the checksum file is present in the package. - Keep --upload-url and --engine external disabled. Do not provide an instance-id, credentials, or allow network uploads until you confirm what data the binary actually sends. - If the publisher cannot explain why the binary is omitted, treat the package as untrusted and avoid execution. These inconsistencies could be an innocent packaging mistake, but they materially increase risk — treat this as suspicious until clarified.

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 operates locally and requires user trust in the binary you run. Always verify the checksum after downloading. For maximum security, build from source (recommended).

Binary Included

Property	Value
Location	assets/bin/skillscan
Version	v0.8.0
Platform	macOS ARM64
SHA-256	3d0e50040dbcb8e9ffa24433587796f61f3c94926ee7e8a87b3359b9e2ae1130

Verify locally before running:

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:

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

前置条件

• 默认不需要任何外部依赖
• --upload-url 和 --engine external 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:

Approach	Trust Requirement	Verification
Build from source	None (you control everything)	Manual code review
Bundled/downloaded binary	You trust the release host	SHA-256 checksum

What the tool does NOT do by default:

• Does NOT upload data anywhere
• Does NOT connect to the network
• Does NOT access credentials, SSH configs, or environment variables
• Does NOT execute external tools unless you explicitly configure --engine external

工作流程

1. 调用 skillscan：

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

2. 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.

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

外部引擎集成 (默认禁用)

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

Delegates pattern-matching to a user-configured local tool. This runs locally — no remote calls are made.

Permissions Required

Scope	Reason
Read files in target path	To analyze skill source code for patterns
Write to --output-dir	To save scan reports locally
Execute binary	To run the scanner engine
Network (optional)	Only if --upload-url is explicitly configured