Analysis updated 2026-05-18
Check a Claude Code plugin from a GitHub link for hidden hooks before installing it.
Review an MCP server package for credential access or data exfiltration risks.
Get a file-and-line breakdown of suspicious code in a skill before adding it to your agent.
Decide whether to trust a marketplace add-on with a simple go or no-go recommendation.
| proluct/skanna | 0-bingwu-0/live-interpreter | 010zx00x1/faresnipe | |
|---|---|---|---|
| Stars | 2 | 2 | 2 |
| Language | — | Python | Python |
| Setup difficulty | easy | moderate | easy |
| Complexity | 2/5 | 2/5 | 2/5 |
| Audience | vibe coder | general | general |
Figures from each repo's GitHub metadata at analysis time.
Installed as a plugin through the Claude Code marketplace or by copying files, no API key, dependencies, or per-scan cost required.
skanna is a security scanner for Claude Code skills, plugins, and MCP servers. These skills and plugins are code and instructions that you give to an AI agent with access to your shell, your files, and your credentials. The problem skanna addresses is that people install these add-ons from marketplace links and GitHub URLs without checking them first, which creates a supply chain with familiar attack patterns like hooks that run automatically, install scripts fetched from the internet, and hidden instructions that hijack the agent. The tool reads the entire package contents, including SKILL.md files, scripts, hooks, and manifests. It returns a verdict of SAFE, CAUTION, or DANGEROUS, along with specific findings that cite the exact file and line number. It gives a go or no-go recommendation. The name "skanna" comes from the Swedish word for "scan." skanna checks for eight categories of danger. These include hooks that run code before you review it, install scripts that fetch remote code, access to credentials like SSH keys and browser cookies, data exfiltration through outbound network calls, obfuscated code, prompt injection that turns the agent against its user, destructive operations like mass deletion, and persistence mechanisms that survive reboots. Each rule describes what signals to look for, why they are dangerous, and common safe patterns that should not trigger false alarms. The scanner never executes anything from the target it reviews. Remote targets are cloned to a temporary directory outside your working files and removed afterward. The developers are honest about limitations: this is a static review by a language model, not a sandbox or a formal proof. A DANGEROUS verdict means do not install. A SAFE verdict means nothing was found, not that the package is certified. Installation is done as a plugin through the Claude Code marketplace or by manually copying files. There is no API key, no dependencies, and no per-scan cost. The scanner uses the model you already run, directed at the right things with the right suspicions.
A security scanner that checks third-party Claude Code add-ons (skills, plugins, MCP servers) for hidden risks like credential theft or data exfiltration, giving a clear go or no-go verdict.
No license information was provided in the explanation, so the terms of use are unclear.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly vibe coder.
This repo across BitVibe Labs
Verify against the repo before relying on details.