The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The SKILL.md file contains instructions to install system utilities using 'sudo apt-get install -y libreoffice poppler-utils'. This pattern explicitly requests administrative privileges, which is a high-risk practice for agent skills.
[COMMAND_EXECUTION]: The skill utilizes shell commands 'soffice --headless --convert-to pdf --outdir $OUTDIR $INPUT_XLSX' and 'pdftoppm -png $OUTDIR/$BASENAME.pdf $OUTDIR/$BASENAME' for rendering spreadsheets. The use of shell variables to construct these commands introduces a risk of command injection if file names or paths are sourced from untrusted input without strict sanitization.
[INDIRECT_PROMPT_INJECTION]: The skill processes external .xlsx, .csv, and .tsv files, which serves as an ingestion point for untrusted data that could contain malicious instructions designed to influence the agent.
Ingestion points: Reading and analyzing spreadsheet files as defined in SKILL.md and implemented in the example Python scripts.
Boundary markers: No instructions are provided to the agent to use delimiters or specific safety instructions to ignore content within the processed data.
Capability inventory: The skill has the ability to execute shell commands for rendering and write files to the local system, which could be abused if an injection is successful.
Sanitization: There is no evidence of sanitization, escaping, or content validation for the spreadsheet data before it is processed by the agent or libraries like openpyxl and pandas.
[EXTERNAL_DOWNLOADS]: The skill installs standard Python libraries (openpyxl, pandas, and matplotlib) from official package registries. It also directs the installation of system utilities (libreoffice and poppler-utils) via official operating system package managers.

spreadsheet