The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructs the agent to perform system-level operations using high-privilege commands like apt-get install for package management, which typically requires root access.
[COMMAND_EXECUTION]: The instructions mandate the use of the Bash tool to execute CLI utilities such as pdftotext and a local Python script (scripts/sha256_wiki.py) to process files.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from various sources including PDFs, website URLs, and entire source code repositories. There is no evidence of robust sanitization or hard boundary markers to prevent embedded malicious instructions in these sources from influencing agent behavior.
Ingestion points: PDF files, web URLs (via WebFetch), and project source files (SKILL.md).
Boundary markers: The 'CWD Boundary Rule' is mentioned as a textual instruction but is not programmatically enforced.
Capability inventory: Uses Bash tool, WebFetch, and local script execution.
Sanitization: No input sanitization or validation mechanisms are implemented.
[REMOTE_CODE_EXECUTION]: The skill requests the installation of external system packages (poppler-utils) from remote repositories during runtime to enable its core functionality.

tutor-setup