The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing the getdot CLI from remote scripts hosted on the vendor's domain. \n
Evidence: SKILL.md directs agents to download from https://app.getdot.ai/install.sh and https://app.getdot.ai/install.ps1. \n
Context: These are official resources from the vendor, Snowboard-Software. \n- [REMOTE_CODE_EXECUTION]: The installation process involves a 'curl-pipe-bash' command sequence, executing remote code in the local environment. \n
Evidence: curl -fsSL https://app.getdot.ai/install.sh | sh found in SKILL.md. \n
Context: While common for developer tools, this pattern relies on the integrity of the vendor's delivery infrastructure. \n- [COMMAND_EXECUTION]: The skill allows the AI agent to execute shell commands to query databases through the CLI and manage local output files like PNG charts and CSV data. \n
Evidence: Shell command patterns such as dot catalog and dot "..." in SKILL.md. \n- [PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection (shell injection) by instructing the agent to place untrusted user questions directly into shell commands. \n
Ingestion points: User questions provided to the dot command in SKILL.md. \n
Boundary markers: Use of double quotes is suggested, which provides only minimal protection against malicious shell characters. \n
Capability inventory: The agent is granted shell access to run the getdot tool and read results from temporary directories. \n
Sanitization: There is no requirement or logic provided for the agent to sanitize user input for shell-sensitive metacharacters before execution.

dot