The Agent Skills Directory

[Unverifiable Dependencies & Remote Code Execution] (HIGH): The skill facilitates the installation and execution of remote code through the 'npx skills add' command.
Evidence: Explicit instructions in Step 4 to run 'npx skills add owner/repo@skill -g -y'.
Risk: The '-y' flag bypasses confirmation prompts, creating a silent path for remote code execution. While 'vercel-labs/agent-skills' is a trusted source, the skill's search-and-install nature allows it to target any arbitrary repository.
[Indirect Prompt Injection] (HIGH): The skill exposes a high-risk attack surface because it ingests untrusted search results from an external ecosystem and possesses the capability to execute them.
Ingestion points: Search results and package metadata from the 'npx skills find' command.
Boundary markers: Absent; the agent is not instructed to treat search descriptions as untrusted data.
Capability inventory: Full shell execution for package installation ('npx skills add').
Sanitization: None; the agent passes search results directly into installation commands.
[Privilege Escalation] (MEDIUM): The instruction to use the '-g' (global) flag for installations increases the blast radius of any malicious package by applying changes at the system or user level rather than within a restricted local directory.

find-skills