The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external URLs via browser rendering tools. This content is then summarized or processed by the AI, creating a surface for indirect prompt injection where a website could contain hidden instructions to manipulate the agent's behavior.
Ingestion points: Web content retrieved via browser_markdown, browser_json, and browser_links in SKILL.md.
Boundary markers: No specific delimiters or "ignore instructions" warnings are configured to isolate external web content from the agent's instruction context.
Capability inventory: The skill possesses file-write capabilities (Write) to the workspace and network access through the browser tools as defined in SKILL.md.
Sanitization: No explicit sanitization or filtering of the retrieved HTML/Markdown content is performed before interpolation into the prompt.
[EXTERNAL_DOWNLOADS]: The skill is designed to fetch and render content from arbitrary external URLs provided by the user. It utilizes a well-known service (Cloudflare Browser Rendering) for these operations.
Evidence: Uses mcp__cf-browser tools and WebFetch to access external web resources as specified in the frontmatter of SKILL.md.

deep-read