The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/ensure_publication_tooling.py uses the subprocess module to execute shell commands for package installation.
[EXTERNAL_DOWNLOADS]: The skill is configured to automatically download and install third-party Python packages (pubfig and pubtab) from public registries if they are missing from the active environment. These packages are not associated with trusted organizations or well-known services.
[REMOTE_CODE_EXECUTION]: The skill instructions in SKILL.md explicitly mandate an "auto-install" behavior for its dependencies. This allows the agent to modify the host system by installing and subsequently executing third-party code without requiring explicit user confirmation for each installation.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it processes untrusted user data (Excel workbooks and LaTeX tables) to generate publication artifacts.
Ingestion points: Data enters the context via the pubtab and pubfig tools when processing files like results.xlsx or benchmark.xlsx as described in references/pubtab-recipes.md.
Boundary markers: There are no instructions defining delimiters or "ignore embedded instructions" warnings for data processed from external files.
Capability inventory: The skill has capabilities to write to the file system (via save_figure and batch_export) and execute shell commands via the installation helper.
Sanitization: There is no evidence of content sanitization or validation of the data files before processing.

publication-chart-skill