The Agent Skills Directory

[COMMAND_EXECUTION]: The skill schema for the tasks tool includes a verification_commands field designed to hold arbitrary shell command strings that worker agents are expected to execute to verify their work.
[PERSISTENCE]: The agents tool allows the definition of worker agents with a schedule parameter using cron syntax, enabling automated and recurring execution of tasks that could be leveraged for maintaining persistence.
[DATA_EXFILTRATION]: The docs registry and the tasks media attachment feature allow for the registration and access of files via absolute local paths. This provides a mechanism for an agent to read sensitive files from the host system if specified in a task's configuration.
[DYNAMIC_EXECUTION]: The workflow relies on worker agents dynamically interpreting and executing instructions, phases, and verification commands provided within the task data at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill involves agents processing complex data structures like task descriptions and acceptance criteria stored in an external daemon, which lacks explicit sanitization or boundary markers to prevent malicious content from influencing the worker agent's behavior.

zenf-tools