The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were detected in the analyzed skill files. The logic of the skill aligns perfectly with its stated purpose of PDF design and generation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted HTML and user-provided data. It includes robust internal instructions to mitigate risks: the agent is explicitly told to treat input as untrusted data (never as instructions) and to remove active content before rendering. The command-line instructions for PDF generation include --blink-settings=scriptEnabled=false, which is a critical security control that prevents JavaScript execution within the headless Chromium browser.
[DATA_EXFILTRATION]: The skill handles remote uploads to user-specified destinations securely. It mandates using platform-integrated tools (like Google Drive) to manage OAuth credentials and requires the agent to ask for explicit user confirmation before any upload. It also forbids the inclusion of private context or unrelated local files in the documents.
[OBFUSCATION]: The provided HTML template contains a Base64-encoded image for a logo. This is a standard method for embedding assets and is not used for malicious purposes. The skill's documentation explicitly notes the presence of large Base64 strings as a known performance issue for the agent's tools, demonstrating transparency.

pdf-design