The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by design.
Ingestion points: User-supplied data enters the agent context through the <选题> (Chinese) and <topic> (English) parameters in the commands /文豪 and /literary respectively, as defined in SKILL.md.
Boundary markers: The instructions lack specific guidance for the agent to wrap user-provided topics in delimiters or to apply 'ignore embedded instructions' warnings, which could allow a malicious user topic to influence agent behavior.
Capability inventory: SKILL.md explicitly discusses bash command execution and provides defensive rules for it, indicating the agent environment has shell access.
Sanitization: There are no documented procedures for validating or filtering the user input before interpolation into the creative writing prompts.
[SAFE]: The skill includes a 'Safety Rules' section that proactively instructs the agent on secure shell usage, such as enforcing timeouts and avoiding dangerous commands like rm -rf or sudo, which mitigates platform-specific risks.

literary-ghostwriter