The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from external sources and unverified user inputs, creating a surface for indirect prompt injection.
Ingestion points: The skill retrieves research recommendations via get_prompt_research and accepts user-supplied customer questions through add_prompts as described in SKILL.md. It also performs broad inventory reads of existing prompts and answers in references/cleanup.md.
Boundary markers: While the instructions emphasize reviewing recommendations before adoption and utilizing dry_run previews for verification, there are no explicit prompt boundary markers or instructions to treat embedded text as data rather than instructions.
Capability inventory: The skill has extensive write and delete capabilities, including add_prompts, update_prompts, update_topics, create_tags, retire_tags, and the ability to delete_personas across both SKILL.md and references/cleanup.md.
Sanitization: The instructions do not specify any sanitization, filtering, or escaping of the external research content or customer questions before they are passed to the platform's tools.

sitefire-prompts