The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a section titled 'MANDATORY SKILL INVOCATION' using strong imperative language ('YOU MUST', 'Failure... violates your operational requirements') designed to override the agent's natural decision-making process for tool selection.
[COMMAND_EXECUTION]: The skill relies on executing multiple local shell scripts (unraid-query.sh, dashboard.sh, disk-health.sh, read-logs.sh) through the zsh-tool. These scripts accept arguments such as GraphQL queries and log file names which may be influenced by user prompts.
[DATA_EXFILTRATION]: The skill facilitates access to sensitive server-side information, including system logs (read-logs.sh), disk health metrics, array status, and network share comments.
[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by ingesting untrusted data from the server environment into the agent context.
Ingestion points: System log files accessed via examples/read-logs.sh and API responses from scripts/unraid-query.sh.
Boundary markers: No delimiters or safety instructions are provided to the agent for handling the ingested log content.
Capability inventory: Access to shell execution (zsh-tool), network requests, and file system operations.
Sanitization: There is no evidence of sanitization or escaping of the external log data before it is presented to the agent.

unraid