The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The Python script in SKILL.md is designed to read a sensitive Google Drive token file located at /home/jamditis/.claude/google/drive-token.json. This file contains OAuth access tokens, refresh tokens, client IDs, and client secrets, the exposure of which could lead to unauthorized account access.
[DATA_EXFILTRATION]: The skill implements a workflow to read local files and transmit them to external Google Drive folders (1lKTdwq4_5uErj-tBN112WCdJGD2YtetO and 1e5dtKOiuvk0PPrFq3UyNI2UAa6RFiom3). While targeting a well-known service, the ability for an agent to programmatically upload local data to predefined remote storage represents a significant exfiltration vector if subverted.
[COMMAND_EXECUTION]: The skill frequently invokes shell commands via os.system or subprocess equivalents, including chromium-browser, pdftoppm, and pdfinfo. It also executes arbitrary Python code using a heredoc in SKILL.md. These patterns increase the risk of command injection if user-provided filenames or paths are not strictly validated.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8).
Ingestion points: User-provided descriptions of changes for PDF design (SKILL.md).
Boundary markers: None. No instructions are provided to ignore embedded instructions in the user's design requests.
Capability inventory: Shell command execution, local file reading/writing via Chromium, and Google Drive upload capabilities (SKILL.md).
Sanitization: None. User input is directly interpreted to modify HTML content which is subsequently rendered by a headless browser, creating a path for local file disclosure (LFD) or cross-site scripting (XSS) within the PDF generation environment.

pdf-design