The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command: bash "$ENGINE" landscape --prompt "<user subject>". The <user subject> placeholder is intended to be replaced with arbitrary user input. If a user provides a prompt containing shell metacharacters such as backticks (`), semicolons (;), or command substitutions ($(...)), they can execute arbitrary commands with the privileges of the agent process.
[REMOTE_CODE_EXECUTION]: The skill uses a loop to locate an executable script named tzai-image in various hidden directories within the user's home folder (e.g., ~/.agents/skills/..., ~/.claude/skills/...). It then executes the found file using bash "$ENGINE". Executing files from computed paths on the filesystem is risky as it depends on the integrity of those directory structures and can be exploited if an attacker can place a malicious file in one of the searched locations.

tzai-landscape