Module trust 

Prompt-boundary helpers for skill-provided resources.

Skill metadata and instruction files are discovered from user- and repository-controlled locations. They are useful model context, but they are not host policy. Callers should place the rendered value between their trusted policy sections and keep enforcement in the tool/runtime policy.

Constants

MAX_UNTRUSTED_SKILL_INSTRUCTIONS_BYTES

Maximum encoded body size for skill instructions placed in a prompt.

Functions

render_activation_for_skill

Render activation instructions for a loaded skill.

render_untrusted_skill_instructions

Render skill instructions as bounded, escaped untrusted resource content.

render_untrusted_skill_instructions_with_context

Render skill instructions with directory and resource context.