The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill is designed to process external laboratory protocols and deck layouts which can be loaded from JSON files. This creates a high-risk surface for physical side effects.
Ingestion points: Ingests deck layouts and state management data through JSON serialization/deserialization as described in the Resource Management section.
Boundary markers: No explicit boundary markers or instruction-guarding mechanisms are mentioned for data parsed from JSON.
Capability inventory: Full control over laboratory robots (Hamilton, Opentrons, Tecan), including movement, pipetting, heating, and centrifugation via the LiquidHandler and various hardware backends.
Sanitization: No evidence of sanitization or validation for protocol instructions parsed from external sources.
[External Downloads] (MEDIUM): The Quick Start guide instructs the user to install the pylabrobot package using uv pip install.
Evidence: The package is hosted on PyPI and GitHub (PyLabRobot/pylabrobot), which are not within the defined 'Trusted Source' scope for automated safety downgrades.
[Command Execution] (HIGH): The core functionality involves executing commands that result in physical hardware movement.
Evidence: Operations like lh.setup(), lh.aspirate(), and pr.read_absorbance() translate high-level Python commands into hardware-level instructions. Subversion of these commands through malicious input could lead to laboratory accidents.

pylabrobot