The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill incorporates a comprehensive security module (security.py) that prevents Server-Side Request Forgery (SSRF). It validates that all target URLs are public and resolves hostnames to verify that they do not point to private or reserved IP addresses before performing network operations.\n- [COMMAND_EXECUTION]: In scripts/notify_human.py, the skill uses subprocess.run to call the macOS osascript utility. This is used solely for displaying user-facing dialogs and employs safe argument passing via a list, which effectively prevents shell injection attacks.\n- [EXTERNAL_DOWNLOADS]: The skill retrieves academic content from well-known and reputable services such as ArXiv, Semantic Scholar, and Unpaywall. All downloaded files undergo structural PDF validation (checking for headers and EOF markers) to ensure data integrity.\n- [DYNAMIC_EXECUTION]: The skill uses importlib.import_module in artifact.py as a runtime check to ensure the pypdf dependency is available. This is a standard and safe implementation for managing optional library dependencies.\n- [PROMPT_INJECTION]: The skill ingests external PDF files, creating a potential surface for indirect prompt injection. This risk is mitigated through structural parsing and strict metadata extraction.\n
Ingestion points: External PDFs are processed using the pypdf library in scripts/paper_fetch/artifact.py.\n
Boundary markers: The skill returns structured JSON output to the agent, providing clear separation between data and instructions.\n
Capability inventory: The skill has capabilities for file system writes, network requests, and system dialog execution via osascript.\n
Sanitization: The skill performs structural PDF validation and SHA-256 integrity verification on all downloaded artifacts.

academic-paper-download