The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill processes untrusted external data (bank statements in PDF/CSV format) to generate analysis and interactive HTML reports. A lack of sanitization in this pipeline creates a vulnerability surface.
Ingestion points: scripts/extract_pdf_data.py (reads PDF statements) and scripts/analyze_finances.py (reads CSV/JSON transaction data).
Boundary markers: Absent. The scripts do not implement delimiters or instructions to ignore malicious content within the processed data.
Capability inventory: The skill performs local file system operations, including reading sensitive financial files and writing reports (transactions.csv, analysis.json, report.html).
Sanitization: Absent. Review of the sample report.html indicates that transaction descriptions are injected directly into JavaScript data structures, which could allow a malicious transaction entry to execute arbitrary script in the user's browser.
[EXTERNAL_DOWNLOADS] (SAFE): The skill requires the installation of pdfplumber and pandas from PyPI and loads Chart.js from cdn.jsdelivr.net in the generated reports. These are well-known, trusted sources.
[Metadata Analysis] (SAFE): The skill metadata correctly describes its financial analysis purpose, and no obfuscation or deceptive instructions were found in the SKILL.md file.
[Missing Component]: The script scripts/generate_report.py is referenced in the documentation and required for the workflow but was not provided in the skill files, preventing a full audit of its implementation details.

finance-manager