The Agent Skills Directory

[Unverifiable Dependencies & Remote Code Execution] (HIGH): The skill utilizes npx findskill and npm install -g findskill. These commands download and execute JavaScript from the npm registry. The source 'findskill' is not a verified or trusted entity, posing a significant RCE risk.
[Command Execution] (MEDIUM): The skill explicitly instructs the agent to run shell commands to manage the local environment and install software, including global npm installations and file system writes to ~/.claude/skills/.
[Prompt Injection] (LOW): The skill ingests untrusted metadata from an external registry, creating an indirect injection surface. Ingestion points: Metadata enters via search and info commands (file: SKILL.md). Boundary markers: None are present to distinguish registry data from system instructions. Capability inventory: The skill can execute subprocesses via npx and write to the filesystem. Sanitization: No validation is performed on the downloaded content before the agent is told to follow it.
[Dynamic Execution] (MEDIUM): The workflow involves downloading a SKILL.md from the internet and then asking the agent to read and follow its instructions, representing dynamic instruction loading from untrusted sources.

use-findskill