The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the interpolation of user-supplied arguments in swarm mode orchestration.\n
Ingestion points: The $ARGUMENTS variable in the Task prompt within SKILL.md captures user input from the /edd command invocation.\n
Boundary markers: No delimiters (such as XML tags or triple quotes) or isolation instructions are used to separate user-provided input from the skill's coordinator instructions.\n
Capability inventory: The subagents are configured with the 'general-purpose' type, granting them broad access to system-level tools and the file system.\n
Sanitization: User input is not validated, escaped, or filtered before being interpolated into the subagent's execution context.

edd