The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads installation scripts and driver artifacts from downloads.cdata.com. These are official vendor resources and are considered safe for the skill's primary purpose.
[REMOTE_CODE_EXECUTION]: The skill provides commands to download and pipe shell scripts directly into bash, sh, or iex (PowerShell) for CLI installation. While this is a high-risk pattern, the sources are official vendor domains (downloads.cdata.com), and this is a standard installation method for the tool.
[COMMAND_EXECUTION]: The skill frequently uses cdatacli and other shell commands (e.g., mkdir, curl, jq) to perform its tasks. These are necessary for the intended functionality of managing data connections and executing queries.
[DATA_EXFILTRATION]: The skill is designed to read data from enterprise systems (Salesforce, Snowflake, etc.) and run SQL queries. While this involves data movement, the operations are user-directed and performed through a local CLI tool. The skill explicitly warns users against entering sensitive credentials like API keys or passwords directly into the AI session, recommending OAuth or out-of-session configuration instead.
[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data from external sources (database query results) and supports a dynamic skill generation feature (cdatacli drivers skill). This creates a surface for indirect prompt injection if a driver were to contain malicious instructions in its metadata. However, the skill provides clear context for these operations and the impact is limited to the local environment.

cdata-cli