The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill provides instructions for system-wide configuration that require elevated privileges. Specifically, in references/setup-deployment.md, it suggests using sudo to create directories (e.g., /system/settings) and modify file permissions and ownership (chmod, chown) when setting up a shared multi-user cache.
[INDIRECT_PROMPT_INJECTION]: The skill processes biological datasets and metadata from various untrusted external sources, including files, cloud storage, and REST APIs, creating an attack surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through methods like requests.get(), pd.read_sql(), and ln.Artifact creation from external files as documented in references/integrations.md and references/core-concepts.md.
Boundary markers: The documentation emphasizes using schemas and curators to define boundaries and validate data structure.
Capability inventory: The skill has capabilities to write to the local filesystem and cloud storage (S3/GCS), connect to databases, and track lineage via ln.track() across all reference files.
Sanitization: The skill explicitly instructs to validate and sanitize external content using ln.curators.DataFrameCurator and ln.Schema definitions before registration.
[EXTERNAL_DOWNLOADS]: The skill documentation includes commands for installing the LaminDB suite and its dependencies from public registries using uv pip install. This includes the core lamindb package, the bionty ontology module, and various optional integration extras.

lamindb