The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill documents the use of the !command syntax to execute shell commands and inject their output into agent prompts. It specifically demonstrates interpolating user-supplied arguments (e.g., $1, $ARGUMENTS) into these shell strings, which facilitates command injection. Examples include !find . -name "$1" -type f in references/template-syntax.md and !npm run deploy:$1 in examples/deploy-command.md.
[INDIRECT_PROMPT_INJECTION]: The skill describes an attack surface for arbitrary file disclosure by interpolating user-provided arguments into the @file reference syntax without sanitization.
Ingestion points: Command arguments $1, $2, and $ARGUMENTS in SKILL.md and reference files.
Boundary markers: None provided or recommended for templates.
Capability inventory: Shell execution (!) and file reading (@) are documented as core features.
Sanitization: No validation, escaping, or filtering instructions are provided for user-supplied arguments used in shell commands or file paths.
[COMMAND_EXECUTION]: The documentation provides numerous examples of executing system-level commands such as git, npm, ls, and find using the documented ! shell execution syntax across all reference and example files.

command-creation