The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external content, presenting an indirect prompt injection surface.\n
Ingestion points: External HTML content is loaded via page.goto(url) in scripts/export_to_pdf.py.\n
Boundary markers: No delimiters or instructions are used to prevent the agent from following instructions embedded in the HTML.\n
Capability inventory: The skill can launch a browser, create directories (os.makedirs), and write PDF files (page.pdf).\n
Sanitization: No sanitization or filtering is performed on the content of the URL or file being rendered.\n- [DATA_EXFILTRATION]: The script can read local files to perform conversions, which could be exploited if an attacker influences the input path.\n
Evidence: scripts/export_to_pdf.py resolves local file paths and loads them using the file:// protocol for browser rendering.\n- [EXTERNAL_DOWNLOADS]: The skill involves downloading dependencies and browser binaries from well-known sources.\n
Evidence: The installation instructions in SKILL.md require the playwright package and the Chromium browser binary.\n- [COMMAND_EXECUTION]: The skill performs filesystem operations and launches sub-processes.\n
Evidence: The Python script uses os.makedirs for directory creation and playwright to launch and manage a Chromium browser instance.

html-to-pdf