The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external information during the research phase, creating a potential surface for indirect prompt injection.
Ingestion points: The skill ingests company news, social media profiles, and funding signals from the open web as outlined in references/discovery-call-prep.md and references/outreach-prep.md.
Boundary markers: Execution is governed by platform-level review protocols referenced in ../../shared/output-readiness.md and ../shared/protocol.md which define output validation steps.
Capability inventory: The skill is capable of workspace persistence (via workspace-mutations.md), tool-based record hydration, and generating interactive HTML artifacts.
Sanitization: The skill delegates content sanitization to referenced platform safety modules rather than implementing internal filtering logic.
[EXTERNAL_DOWNLOADS]: The HTML brief architecture in references/brief-html-architecture.md fetches typography assets from Google Fonts' official domains (fonts.googleapis.com and fonts.gstatic.com). These are well-known services used for standard document presentation and are treated as safe resources.

research