The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to load and process external biological datasets from diverse sources such as AWS S3, Google Cloud Storage, and local filesystems. This creates an attack surface where malicious data could influence agent behavior.
Ingestion points: References in references/core-concepts.md and references/data-management.md detail the loading of data artifacts into the agent's context via ln.Artifact.load(), pd.read_csv(), and streaming APIs.
Boundary markers: The documentation does not specify the use of delimiters or specific instructions to the agent to ignore embedded commands within the processed datasets.
Capability inventory: The skill provides the agent with capabilities to write to local and cloud storage (artifact.save()), manage database instances (lamin init), and perform network operations (requests.get).
Sanitization: Curation and validation workflows focus on biological ontology compliance rather than security sanitization of untrusted data content.
[PRIVILEGE_ESCALATION]: Documentation in references/setup-deployment.md provides instructions for configuring a system-wide cache in multi-user environments, which requires the use of administrative privileges (e.g., sudo mkdir -p /system/settings and sudo nano /system/settings/system.env). While legitimate for deployment, this involves elevated permissions.
[COMMAND_EXECUTION]: The skill heavily utilizes the lamin CLI and shell-based operations for instance initialization, database migrations, and environment configuration (e.g., lamin init, lamin connect, lamin migrate deploy), as documented across the reference files.

lamindb