The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill automates the production of complex visualizations by generating and executing Python scripts based on pre-defined templates (e.g., bar_memevolve.py, line_aime.py). This dynamic behavior is the core mechanism of the skill, used to adapt generic plotting logic to specific user datasets.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided CSV files and uploaded images, which serves as a potential vector for instruction injection. Ingestion points include the data profiling step in advise.md and image analysis in from-image.md. The capability inventory includes the ability to write files (plots/scripts) and execute Python code. No specific sanitization of CSV content or boundary markers for external data were identified in the instructions.
[DATA_EXFILTRATION]: Local preference management is implemented via the academic_figure_pref.py script, which reads and writes a configuration file at ~/.config/my-claude-skills/academic-figure.json. This access is localized to the agent's environment and is used to persist settings like preferred libraries and journal styles.
[COMMAND_EXECUTION]: The skill provides documentation and workflow guidance for using external CLI tools such as AgentFigureGallery and pubfig. These are treated as optional dependencies and the skill does not automatically install or execute them without user direction.

academic-figure