The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to install or update the zilliz-cli by fetching a shell script from the vendor's official GitHub repository (https://raw.githubusercontent.com/zilliztech/zilliz-cli/master/install.sh) and piping it directly to bash. Although this is the vendor's official installation method, it constitutes remote code execution.
[COMMAND_EXECUTION]: The skill heavily uses shell commands via the zilliz utility to perform operations such as creating clusters, managing collections, and searching vectors.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied vector data and filter expressions by interpolating them into shell commands. This creates a surface where maliciously crafted data could potentially influence command execution.
Ingestion points: The --data, --filter, and --body flags in zilliz vector and zilliz collection commands.
Boundary markers: The instructions lack explicit delimiters or guidance to treat user-provided data as untrusted.
Capability inventory: The skill uses subprocess execution of the zilliz CLI for all operations.
Sanitization: No specific input validation or sanitization steps are defined for the agent to follow when handling external data.

zilliz