The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the ingestion of data from external files (FITS, CSV, HDF5, etc.) and remote URLs, creating a surface for indirect prompt injection where malicious instructions could be embedded in data processed by the agent.
Ingestion points: Data enters the agent context through methods such as Table.read(), fits.open(), and fits.getdata() across SKILL.md and references/data_io_guide.md, as well as download_file() in references/auxiliary_modules.md.
Boundary markers: Absent. The instructions do not define delimiters or provide warnings to the agent about treating external data as untrusted or ignoring embedded instructions.
Capability inventory: The skill provides tools for file system modifications via hdul.writeto(), fits.writeto(), and Table.write(), along with network access capabilities via astropy.utils.data.download_file.
Sanitization: Absent. No evidence of content validation or sanitization is present in the provided examples.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing well-known and trusted Python packages astropy and pytz via pip. It also includes an example of using a built-in utility to download data files from external URLs (astropy.utils.data.download_file).
[COMMAND_EXECUTION]: The skill includes standard shell commands in the documentation for setting up the environment, such as pip install astropy and pip install pytz.

astropy-astronomy