The Agent Skills Directory

Command execution through shell interpolation: The skill constructs and executes shell commands such as npm run plugin:new using inputs provided by the user for plugin and skill names. While typical for scaffolding utilities, this introduces a potential command injection surface if the inputs are not properly sanitized before being passed to the shell. This pattern appears in the scaffolding steps within SKILL.md.
Indirect prompt injection surface: The skill ingests untrusted user input (plugin and skill names) which is then interpolated into shell commands and file paths. 1. Ingestion points: User prompts in SKILL.md. 2. Boundary markers: None present. 3. Capability inventory: Shell execution via npm run and file system writes. 4. Sanitization: No sanitization or validation of the input strings is mentioned in the instructions. This creates a surface where malicious input could influence agent actions.

onboard-plugin