The Agent Skills Directory

[PROMPT_INJECTION]: The onboarding flow described in references/onboarding.md creates an attack surface for indirect prompt injection by processing external website content.
Ingestion points: references/onboarding.md instructs the agent to fetch HTML and CSS from a user-provided URL using agent-browser.
Boundary markers: Absent; there are no explicit delimiters or instructions to ignore potential commands embedded in the fetched content.
Capability inventory: The agent uses the fetched content to write design tokens into references/style-guide.md. It does not have access to higher-risk capabilities like arbitrary shell execution or sensitive file reads in this context.
Sanitization: No explicit sanitization or filtering is prescribed for the external content.
[EXTERNAL_DOWNLOADS]: The skill initiates network requests as part of its primary customization feature.
Evidence: references/onboarding.md directs the agent to fetch website data to extract branding information, which is a core functional requirement of the skill's onboarding process.

diagram-design