The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands including find, grep, and head to identify file types and locate local diagnostic scripts. It further uses the uv tool to dynamically download and run Python dependencies.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data in the form of figure files and plotting scripts, which serves as a potential vector for indirect prompt injection.
Ingestion points: Scientific figures (.svg, .png), Python plot scripts (.py), and Jupyter Notebooks (.ipynb) processed in references/figure-qa-procedure.md.
Boundary markers: Absent. There are no delimiters or warnings to ignore instructions embedded in input data.
Capability inventory: Shell command execution (find, grep, ls), Python script execution (uv run), and file system operations.
Sanitization: Absent. Input plot scripts and notebooks are processed without explicit sanitization or filtering.
[DYNAMIC_EXECUTION]: The skill extracts code from Jupyter Notebooks and saves it to a temporary file for subsequent analysis.
Evidence: Use of nbformat to extract and print code cell source into /tmp/extracted.py as seen in references/figure-qa-procedure.md.

figure-qa