The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The references/setup-deployment.md file contains multiple instructions to use sudo for system-level modifications. Specifically, it advises users to run sudo mkdir -p /system/settings, sudo nano /system/settings/system.env, and sudo chmod/sudo chown on cache directories to configure multi-user environments. While these are documented for administrative setup, they represent high-privilege operations.
[METADATA_POISONING]: There is a discrepancy in the skill's authorship metadata. The SKILL.md frontmatter identifies the author as 'K-Dense Inc.', whereas the system context identifies the author as 'eturkes'. This mismatch is potentially misleading regarding the skill's provenance.
[INDIRECT_PROMPT_INJECTION]: The skill documents workflows that ingest untrusted data from various external sources, including local CSV/AnnData files (pd.read_csv, ad.read_h5ad) and remote REST APIs (requests.get). This creates an attack surface where maliciously crafted biological data or API responses could influence the agent's behavior. The skill mitigates this by emphasizing the use of ln.curators for schema validation and curation.
[CREDENTIALS_UNSAFE]: The documentation in references/setup-deployment.md explicitly mentions sensitive file paths such as ~/.lamin/ where API keys are stored and references the use of environment variables for cloud credentials (e.g., AWS_ACCESS_KEY_ID, LAMIN_DB_URL). Although it follows best practices by using placeholders like <redacted> and <set-in-secret-manager>, the mention of these sensitive storage locations provides a target for data exposure.

lamindb