The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the astropy package (v8.0.1) and its optional dependencies from the official Python Package Index (PyPI). It correctly identifies the risk of unpinned transitive dependencies when using extras like [recommended] or [all] and suggests using lockfiles for production environments.
[COMMAND_EXECUTION]: The skill provides standard shell commands for package management using the uv tool to install the library in an isolated environment.
[DATA_EXFILTRATION]: The documentation covers several functions that perform legitimate network operations required for astronomical research, which may transmit data to third-party services:
SkyCoord.from_name() transmits celestial object names to remote name-resolution services (Sesame/SIMBAD/NED).
EarthLocation.of_address() transmits address strings to Nominatim or Google Geocoding APIs.
fits.open() with use_fsspec=True enables remote data access via S3 or HTTP.
download_file() fetches data from arbitrary URLs.
Automatic IERS updates download Earth rotation data from established scientific mirrors.
Note: These findings are classified as safe because the skill provides prominent warnings (e.g., in SKILL.md and references/review.md) advising users to avoid disclosing sensitive or proprietary identifiers to these services.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process astronomical data from external sources, presenting an inherent surface for indirect injection via metadata or table content.
Ingestion points: Data enters the context via fits.open(), Table.read(), and various coordinate query methods.
Boundary markers: The code examples do not utilize explicit boundary delimiters, which is standard for library documentation, but the prose documentation emphasizes data provenance.
Capability inventory: The library provides capabilities for file system writes (writeto), network connectivity, and extensive data manipulation.
Sanitization: No explicit sanitization of FITS headers or table metadata is demonstrated, as the focus is on scientific integrity rather than adversarial data processing.

astropy