The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python CLI tool (scripts/hpa_cli.py) using the uv run command. This is a standard execution pattern for skills utilizing the uv package manager and does not involve arbitrary or unsafe command execution.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an interface to the Human Protein Atlas, which is an external data source. This creates a potential surface for indirect prompt injection if malicious data were present in the database and subsequently interpreted as instructions by the agent. However, HPA is a reputable scientific resource, and the skill uses standard parsing methods for the retrieved data.
Ingestion points: Data is retrieved from https://www.proteinatlas.org/ via JSON and XML search APIs in scripts/hpa_cli.py.
Boundary markers: The skill specifies that output should be written to the /tmp/ directory in JSON format, which provides a level of structural isolation.
Capability inventory: The skill performs network GET requests to the HPA domain and writes result files to disk.
Sanitization: The script uses json.loads() and xml.etree.ElementTree for parsing, which are standard practices for processing structured data.

human-protein-atlas-database