The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The slash command template in references/slash-command.template.md documents a pattern for processing untrusted user input that creates a potential injection surface.
Ingestion points: External data is ingested into the agent context through the $ARGUMENTS placeholder defined in the reference template.
Boundary markers: The provided template does not utilize or recommend delimiters (such as XML tags or unique markers) to isolate user-supplied arguments from the agent's instructions.
Capability inventory: The template enables the use of dynamic shell execution (via the ! syntax) and explicitly allows for tool restriction via allowed-tools configuration.
Sanitization: The template does not include instructions or examples for sanitizing, escaping, or validating the input provided in the $ARGUMENTS field before it is used by the agent.

command-optimization