The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs shell commands using string interpolation of user-supplied data. Specifically, the command ga <branch-name> "<prompt>" main incorporates a user-provided description and context within double quotes. An attacker providing input with shell metacharacters (such as backticks, semicolons, or unbalanced quotes) could execute arbitrary commands in the host environment.
[PROMPT_INJECTION]: The skill accepts untrusted data from users and interpolates it into a prompt for a subsequent AI session. • Ingestion points: The user's task description, file paths, and error messages are directly inserted into the prompt template in SKILL.md. • Boundary markers: No delimiters or instructions are used to separate the user-controlled content from the skill's instructions. • Capability inventory: The skill executes shell commands via the ga tool and manages terminal sessions via tmux. • Sanitization: User input is not sanitized or escaped before interpolation, allowing the data to potentially hijack the downstream agent's logic.

ga