The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents capabilities for ingesting and processing data from a wide variety of external file formats, including FITS, CSV, HDF5, and VOTable. This represents a vulnerability surface where maliciously crafted scientific data files could contain instructions intended to influence agent behavior.
Ingestion points: The skill uses fits.open(), Table.read(), CCDData.read(), and download_file() to bring external data into the environment.
Capability inventory: The library has access to the file system (read/write), network operations for data retrieval, and the ability to interact with the system browser.
Boundary markers: The documentation examples do not include explicit prompt delimiters or "ignore embedded instructions" warnings for the agent when processing these files.
Sanitization: No custom sanitization or validation logic is provided for the data content beyond standard library usage.
[EXTERNAL_DOWNLOADS]: The documentation describes features that perform network requests to resolve astronomical object names (SkyCoord.from_name), access cloud-hosted data (S3/HTTP URI support in FITS), and automatically download Earth rotation parameter updates (IERS data) for precise time calculations. These are standard research functionalities of the Astropy library.
[COMMAND_EXECUTION]: The skill documents the Table.show_in_browser() functionality, which allows for local visualization of data by launching the system's default web browser.

astropy