The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains an embedded Windows executable binary at servers/win32/msasxpress.dll. Shipping binary blobs in a skill is a critical security concern because the code is not human-readable or auditable, posing a threat of arbitrary code execution when the DLL is loaded by local processes.
[COMMAND_EXECUTION]: Several automation scripts (create_direct_lake_model.py, download_workspace.py, execute_dax.py) use the Python subprocess module to execute the fab CLI. These scripts take user-provided arguments (workspace names, item paths, and DAX queries) and pass them to the shell, which could lead to command injection if malicious inputs are provided.
[EXTERNAL_DOWNLOADS]: The skill documentation and scripts specify dependencies on external libraries such as azure-storage-file-datalake, azure-identity, and requests. While these are legitimate packages from well-known providers, they represent a dependency on external code that is fetched at runtime.
[PROMPT_INJECTION]: The SKILL.md file includes instructions to the agent to "Invoke this skill automatically" and provides directives to bypass interactive modes ("interactive mode doesn't work with coding agents"). These instructions attempt to control the underlying agent's behavioral framework.

fabric-cli