The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill interacts with the OpenRouter API (https://openrouter.ai) to facilitate LLM-driven optimization. OpenRouter is a well-known service for accessing various AI models and is used here for legitimate experimental parameter suggestions.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests measurements from external hardware (camera images and balance data) which are then interpolated into LLM prompts.
Ingestion points: Camera images processed via scripts/image_processing.py and gravimetric data from a mass balance described in references/viscosity-optimization.md.
Boundary markers: Not present in the prompt templates within scripts/optimizers.py.
Capability inventory: Automated liquid handling protocol execution on Opentrons OT-2 hardware and local command execution via uv run.
Sanitization: Measurements are validated and cast to numeric types (int/float) before being formatted into strings for prompt interpolation.
[COMMAND_EXECUTION]: The documentation describes using the uv package manager to run a local bridge service (edge/balance.py) to handle serial communication with a mass balance. This is a standard operational requirement for the skill's laboratory use case.
[SAFE]: The skill demonstrates a high security posture by utilizing environment variables for the OPENROUTER_API_KEY and ROBOT_IP, and it includes explicit rules preventing the agent from asking the user to reveal secrets in the chat.

bears-workflows