Claude Skill Antivirus CLI
A Claude Code skill scanner checks the whole folder, not just SKILL.md, and installs a skill only after your decision.
- 85 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The scanner reads third party files and calls the GitHub API, and installing a skill after the check remains your decision.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/claude-skill-antivirusDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The scanner reads third party files and calls the GitHub API, and installing a skill after the check remains your decision.
This is third-party code. Review the repository files before installing.
What it does
Claude Skill Antivirus checks a third party skill before installation. The whole folder is scanned: `scripts/`, nested directories and also `node_modules/`, `venv/` and `__pycache__/`, so code cannot hide outside `SKILL.md`. Ten engines look for dangerous commands, data exfiltration, external connections, excessive permissions, prompt injection, MCP configuration issues, SSRF, malicious dependencies, sub-agent abuse and binary artifacts. The result is a risk level from informational to critical. The `skill-install ./path/to/skill --scan-only` command only scans, and without the flag it offers installation into `./.claude/skills/` or, with `--global`, into `~/.claude/skills/`.
Who it is for. People who install third party skills into Claude Code from catalogs and repositories and want to see the risks before installing.
Good fit when
- A skill was found in a catalog and must be checked as a whole before installation.
- A repository folder link must be scanned without cloning.
- The team rule says third party skills install only after a risk report.
- A script inside `scripts/`, rather than `SKILL.md` itself, looks suspicious.
Not a fit when
- You need a safety guarantee: the scanner gives a rating and findings, not proof.
- A skill is hard to verify and you cannot rely on static analysis.
- The target is not a Claude Code skill but arbitrary code.
Example request
Scan this skill folder with `skill-install ./skills/my-skill --scan-only` and list the findings by risk level.Limitations
It needs Node.js 22 or newer. Scanning is static: false positives are possible, for example on legitimate password manager integrations, and well hidden code may pass. GitHub links use the API, and without `GITHUB_TOKEN` anonymous rate limits apply. The mass scan results for the SkillsMP catalog in the README concern only `SKILL.md` manifests: scripts were not analyzed there. The outcome stresses manual review of flagged skills.
How to disable. Run `npm uninstall -g claude-skill-antivirus`. Skills installed through `skill-install` live in `.claude/skills/` and are removed separately.
FAQ
How do I scan without installing?
Add the `--scan-only` flag: `skill-install ./path/to/skill --scan-only`. A GitHub link works too.
Can I run it without installing the package?
Yes: `npx claude-skill-antivirus ./path/to/skill --scan-only`.
Where do skills get installed?
By default into the project `./.claude/skills/`, with the `--global` flag into `~/.claude/skills/`.
Which interface languages exist?
English by default and Traditional Chinese via `--lang zh-TW`.
What about symlinks and skipped folders?
Symlinks never lead out of the skill, and skips and truncations are reported as findings so an unusual input cannot pass as safe.
Discussion
No questions or reviews yet.
Related
CLI toolsSecurity scanner for agent skills and MCP servers: finds prompt injection, data exfiltration and supply chain risks before install
A security scanner for agent skills: finds prompt injection, data exfiltration and malicious code before you install
ClawScan runs several security scanners on agent skills and hands their results to a judge for a combined verdict.
A Claude Code skill that finds vulnerabilities in one session: threat model, verified findings, CWE and CVSS report.
The SkillHawk static scanner finds dangerous instructions in SKILL.md, MCP configs and AGENTS.md and runs as a GitHub Action
An OpenClaw skill that audits the agent itself: 230 checks, change watching and an A to F grade.
README in short
A scanner and safe installer for Claude Code skills with ten analysis engines and risk levels. Works on a folder, on a GitHub link and on catalog names.