The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes unverified data from external sources during the research phase.
Ingestion points: The skill fetches external data via the company-intel and person-lookup tools, specifically reading LinkedIn profiles, recent posts, and company news.
Boundary markers: No delimiters or instructions to 'ignore embedded instructions' are present in the 'Briefing Generation' or 'Assemble Briefing' steps to separate untrusted external data from the skill's logic.
Capability inventory: The skill has read access to sensitive internal files like memory/pipeline.json and MEMORY.md, and it can execute commands via orth run for calendar management.
Sanitization: There is no evidence of sanitization, filtering, or validation of the content retrieved from LinkedIn or external web searches before it is interpolated into the briefing template.

meeting-prep