The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources that could contain malicious instructions or malformed content designed to exploit the agent's capabilities.
Ingestion points: User-provided template URLs and the metadata fields of downloaded YAML files (SKILL.md).
Boundary markers: The instructions do not define boundary markers or instruct the agent to ignore embedded instructions in the processed data.
Capability inventory: The skill utilizes several powerful shell commands including curl, grep, mkdir, mv, and git (SKILL.md).
Sanitization: Although the skill suggests converting template names to kebab-case, the initial extraction and use of the template code from the URL are not sanitized.
[COMMAND_EXECUTION]: Shell commands are constructed using variables derived from external/user input, creating a potential vector for command injection.
The {TEMPLATE_CODE} variable is interpolated directly into a curl command (SKILL.md).
The {service-name} variable, derived from the template's YAML content, is used in mkdir and mv commands (SKILL.md).
[EXTERNAL_DOWNLOADS]: Fetches template configuration files from the developer's official domain at zeabur.com (SKILL.md).
[REMOTE_CODE_EXECUTION]: Instructs the agent to execute the latest version of the vendor's CLI tool using npx zeabur@latest, which involves downloading and running code from a remote registry at runtime (SKILL.md).

zeabur-template-backup