The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install external software packages such as python-pptx (via pip) and playwright (via npm). It also triggers the download of browser binaries through the command npx playwright install chromium. While these tools are well-known, they are requested by a community-provided skill whose source is not a verified trusted organization.
[DYNAMIC_EXECUTION]: In scripts/export-pdf.sh, the skill generates a temporary Node.js script (export-slides.mjs) in the system's temporary directory. This script is then executed to manage a headless browser session for slide capturing. Although the script template is fixed, the practice of generating and executing code at runtime introduces a layer of dynamic execution that requires oversight.
[COMMAND_EXECUTION]: The skill frequently invokes shell commands to perform its core functions, including python3 for PowerPoint data extraction, bash for running export utilities, and node for starting local servers and headless browsers. These operations involve direct interaction with the host operating system.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface in scripts/extract-pptx.py, which parses content from PowerPoint files. This extracted text is then passed to the agent to generate HTML presentations. A lack of explicit sanitization or strict boundary markers for this untrusted data creates a risk where malicious instructions embedded in a presentation could influence the agent's output or behavior.
[DATA_EXFILTRATION]: The PDF export utility launches a local HTTP server (scripts/lib/confined-server.mjs) to render the presentation for capturing. Although the server is limited to the loopback interface and includes path traversal protections, the associated headless browser may still access remote assets or execute embedded JavaScript from the presentation, which could be leveraged to access or exfiltrate content from the designated deck directory.

frontend-slides-frontend-slides