Skill UI

Guides setting up Constitutional AI by combining self-critique SL with RLAIF training to align generative models without human labels and reduce harmful outputs in systems like Claude.

Builds molecular property prediction and MoleculeNet workflows with DeepChem: SMILES and graph featurization, scaffold or grouped holdouts, masked labels, graph models, and pretrained encoder transfer for ADMET, toxicity and solubility ML.

Utilize zero-shot NER to extract custom entity types (e.g., Drug, Symptom, Device) from complex clinical or biomedical text without requiring any labeled training data or fine-tuning. OpenMed wraps GLiNER/GLiNER2, allowing users to define labels on the fly. The workflow involves building a model index and then running inference via CLI or Python API, making it ideal for rapid prototyping or unstable schemas.

Adds one optional external Codex critique after the council workflow has produced a synthesis draft. Use for ambiguous, high-consequence decisions where a separate model should attempt to break the consensus. Requires explicit user consent before sending a compact review packet to OpenAI, labels cross-provider vs same-provider honestly, sandboxes the external call, and marks the review absent if the adapter is unavailable.

Design LLM-as-judge rubrics, direct scores and pairwise comparisons. Use for judge models, evals, position bias and calibration against human labels.