The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill is designed to interpolate untrusted user data into text templates, which could be used to inject malicious instructions if the output is processed by other agents or tools without proper escaping.
Ingestion points: The template and variables inputs defined in SKILL.md.
Boundary markers: Absent; there are no instructions to the agent to treat interpolated content as strictly data.
Capability inventory: No dangerous capabilities (such as file system access, network operations, or command execution) are present in this skill.
Sanitization: The instructions do not define any sanitization or validation of the interpolated content beyond type conversion to string.

render-template