The Agent Skills Directory

[COMMAND_EXECUTION]: The gql function in scripts/linear.sh constructs a curl command using double quotes for the data payload (-d "{\"query\": \"$query\"}"). This configuration allows the shell to perform variable expansion and subshell execution (e.g., $(...) or backticks) on the contents of the $query variable. Because this variable includes unvalidated user-supplied strings like ticket titles and comments, an attacker could trigger arbitrary command execution on the agent's host system by including shell metacharacters in Linear data.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to the way it retrieves and displays data from external sources.
Ingestion points: Ticket descriptions (via the issue command) and Notion research pages (via notion-search and notion-fetch).
Boundary markers: The skill does not define delimiters or protective instructions to separate fetched content from the agent's system prompt or current task instructions.
Capability inventory: The agent can execute the linear.sh script (which has network and file access) and use curl to download files to /tmp.
Sanitization: While the script attempts to escape double quotes, it does not sanitize against shell command expansion sequences.
[EXTERNAL_DOWNLOADS]: The skill provides guidelines for the agent to download image assets from Linear's official infrastructure (uploads.linear.app) using curl. This is a standard operation using a well-known service associated with the skill's functionality.

linear