The Agent Skills Directory

[Prompt Injection] (SAFE): The files contain technical documentation and code patterns for data analysis. No attempts to override agent instructions or bypass safety filters were found.
[Data Exposure & Exfiltration] (SAFE): No hardcoded credentials, sensitive system paths, or unauthorized network communication patterns were detected. The data paths referenced (e.g., 'homo_sapiens') are standard for the genomic database.
[Obfuscation] (SAFE): The content is clear and uses standard markdown and Python formatting. No hidden characters, Base64 encoding of commands, or homoglyphs are present.
[Unverifiable Dependencies & Remote Code Execution] (SAFE): The skill references well-known scientific libraries (cellxgene_census, scanpy, torch, pandas, pyarrow). No remote script downloads or piped bash executions are included.
[Privilege Escalation] (SAFE): No commands involving sudo, chmod, or administrative level modifications were found.
[Indirect Prompt Injection] (SAFE): The skill defines patterns for ingesting scientific data from a public census. While it processes external data, it lacks capabilities (like shell execution of data content) that would make it vulnerable to indirect injection.
[Dynamic Execution] (SAFE): Although the TileDB-SOMA framework uses a DSL for filtering, the examples provided are standard library usage and do not involve unsafe eval() or exec() calls on untrusted input.

cellxgene-census