The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The instruction set includes the use of sudo apt-get install to set up system dependencies, which grants root-level access during the installation process.
[COMMAND_EXECUTION]: The skill uses a shell command pdftoppm -png $INPUT_PDF $OUTPUT_PREFIX for rendering. The interpolation of $INPUT_PDF and $OUTPUT_PREFIX variables into a shell string poses a risk of command injection if the agent processes files with malicious names (e.g., using shell metacharacters).
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of Python libraries (reportlab, pdfplumber, pypdf) and system utilities (poppler-utils) from well-known package registries to handle its primary functions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external PDF documents, which are untrusted data sources that could contain instructions intended to manipulate the agent's logic.
Ingestion points: PDF files read for content extraction and visual review (SKILL.md).
Boundary markers: Absent; there are no instructions to the agent to ignore or delimit embedded text within the PDFs.
Capability inventory: Shell command execution (pdftoppm), file system writes to tmp/ and output/ paths, and dependency installation commands (SKILL.md).
Sanitization: No sanitization of PDF content or file metadata is mentioned before processing or rendering.

pdf