The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains an indirect prompt injection attack surface where untrusted data from external sources is processed and used to influence subsequent agent actions.
Ingestion points: The skill ingests untrusted data from the user (user_message) in the brief step and from external search engines in the research and storyline steps within SKILL.md.
Boundary markers: While the skill uses xml_escape for the initial user input, it lacks explicit boundary markers or instructions to ignore embedded commands for the data retrieved from search engines in later steps.
Capability inventory: The skill has the ability to perform network operations (via multi-search-engine) and write to the filesystem (via the pptx tool execution in pptx_artifact).
Sanitization: Partial sanitization is present via xml_escape on user-provided messages, but no validation or sanitization is performed on the results gathered from the web before they are synthesized into the final presentation storyline.

meta-research-to-slide-deck