vlt 1.5.0 Adds Agent Skills Management and Malware Filtering

vlt 1.5.0, released by Socket, adds native support for managing coding agent skills within JavaScript projects. The update allows developers to link instructions from dependencies directly to agent directories while filtering out known malicious packages. This addresses the manual effort previously required to keep agent instructions current with library versions.

The release introduces the vlt skills command, which discovers skills in installed dependencies and links them into directories read by coding agents. These instructions originate from the package version recorded in the project’s lockfile. According to Socket, this approach ties the instructions to the specific dependency they describe, eliminating the need for developers to copy files by hand.

Context and Background

Agent skills are folders containing a SKILL.md file that instructs coding agents on how to perform tasks or use specific tools. While packages may include these instructions, they typically reside in node_modules, outside the directories agents scan. vlt 1.5.0 bridges this gap by linking skills into standard agent directories such as .agents/skills or .claude/skills, depending on the project structure.

This release follows vlt 1.0, launched in August, which introduced a stable package manager and hosted JavaScript registries. That version separated downloading dependencies from running their install scripts, allowing teams to block known malware by default. The new skills management feature applies a similar separation, treating the exposure of instructions to agents as a distinct, governed action.

Who Is Affected

Developers using coding agents like Codex, Cursor, GitHub Copilot, or Gemini CLI are the primary users of this feature. The tool detects agent folders at the project root and links skills accordingly. Teams can override default destinations using the skills-dir option. The selection is saved in vlt.json, ensuring that subsequent installs and updates maintain the correct links when teammates clone the repository.

Security teams benefit from the integration of Socket’s high-confidence malware alerts. By default, the tool excludes skills from packages flagged as malware. Developers can apply stricter filters using security selectors to exclude packages with typosquatting, vulnerabilities, or obfuscated code. This gives organizations control over which dependencies can supply instructions to their coding agents.

What should you do?

Organizations using coding agents should review how they manage agent instructions. Manual copying of skill files is prone to errors and security risks. Using a package manager that ties instructions to dependency versions can reduce exposure to malicious code. Teams should also audit which agents have access to which skills.

What to do and how to stay safe: vlt

• Audit current projects for manually copied agent skill files and replace them with managed links.
• Configure vlt to exclude skills from packages flagged for malware, typosquatting, or obfuscation.
• Review the allow-skills configuration in vlt.json to ensure only trusted dependencies provide instructions.
• Monitor Socket alerts for new malicious packages that may attempt to supply harmful agent instructions.

Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

How does vlt 1.5.0 handle agent skills?

It discovers skills in installed dependencies and links them into agent directories, updating them automatically as dependencies change.

Can developers filter out malicious skills?

Yes, vlt uses Socket data to exclude skills from packages flagged as malware, typosquatting, or containing obfuscated code.

Which coding agents are supported?

The tool targets directories used by Codex, Cursor, GitHub Copilot, and Gemini CLI, as well as .claude/skills.