The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill uses web_fetch to retrieve content from user-provided URLs to extract design cues such as tone, vocabulary, and layout density. This represents an attack surface where a malicious website could host instructions designed to influence the agent's behavior during the slide creation process.
Ingestion points: Content retrieved via web_fetch from external URLs in SKILL.md.
Boundary markers: Absent; there are no instructions to ignore embedded commands within the fetched content.
Capability inventory: Shell execution, file writing (HTML/CSS), and network access.
Sanitization: Absent; the fetched text is processed directly to influence the visual style and copy of the generated slides.
[DYNAMIC_EXECUTION]: The skill generates HTML and CSS files at runtime based on user preferences and then executes a local script (scripts/export_pdf.py) that uses a headless browser to render the content. This workflow involves assembling and executing code generated from external inputs.
[PRIVILEGE_ESCALATION]: The instructions in SKILL.md advise the user to run playwright install-deps chromium, a command which typically modifies system-level packages and may require administrative privileges.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill fetches the playwright and PyMuPDF packages from standard registries. It also manages the download of the Chromium browser binary required for the PDF export functionality.

slide-creator