The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation includes instructions to install an external binary via go install github.com/pjtf93/weathercli/cmd/weathercli@latest and provides a link to binary releases at https://github.com/pjtf93/weathercli/releases. These resources originate from an unverified third-party account ('pjtf93') not associated with the skill author ('sundial-org') or a trusted vendor list.
[COMMAND_EXECUTION]: The skill operates by invoking a local binary weathercli. If the binary installed from the unverified source is malicious, it could lead to full system compromise upon execution.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection and shell command injection. User-supplied input for the <location> variable is placed directly into a shell execution string.
Ingestion points: Location strings used in current, forecast, and search commands.
Boundary markers: The commands wrap the location in double quotes in examples, but this does not prevent injection via subshells (e.g., $(...)) or quote-breaking characters.
Capability inventory: The skill documentation shows the agent executing shell commands with user-controlled parameters.
Sanitization: No input validation or character escaping is specified to mitigate command injection risks.

weathercli