The Agent Skills Directory

[COMMAND_EXECUTION]: Th e skill instructs th e agent t o perform shell-based operation s using code -d an d git show t o compare file versions.
[PROMPT_INJECTION]: Th e skill exhibit s a vulnerabilit y s u r f a c e f o r indirect prompt injection through th e us e o f unvalidate d file path s i n shell commands. Ingestion points: File path s an d commit hashes are interpolated into shell command template s i n SKILL.md. Boundary markers: N o delimiter s o r safet y instruction s are use d t o separate user-provided dat a from th e command context. Capability inventory: Shell command execution with file system redirectio n capabilities. Sanitization: N o inpu t validatio n o r s a n i t i z a t i o n routine s are specifie d f o r th e externa l dat a used i n commands.

vscode