The Agent Skills Directory

Indirect Prompt Injection (LOW): The scripts read_excel.py and read_pptx.py extract content from office documents and output it directly to the agent's context without isolation.\n
Ingestion points: filepath parameter in read_excel.py and read_pptx.py.\n
Boundary markers: Absent. The data is printed directly to stdout as raw text, markdown, or JSON.\n
Capability inventory: read_excel.py and read_pptx.py have read access to the filesystem; build_pptx.js has write access for generating the PPTX output.\n
Sanitization: None. Text and table data are extracted verbatim from the files and passed to the agent.\n- Dynamic Execution (LOW): build_pptx.js uses playwright for handling HTML content. If the input HTML is sourced from an untrusted location, rendering it in a browser context exposes the system to browser-based attacks such as Local File Disclosure or SSRF via the browser instance, although no explicit malicious code is present in the provided scripts.

data-office-pro