Skill UI

This privileged skill lands self-modification patches approved by users after /meta-optimize. It enforces a privilege boundary by acting as the only skill permitted to mutate the skill corpus. It runs a fresh cross-model jury at landing time to verify patches before applying changes, ensuring safety and preventing self-modification loops. Requires explicit human invocation.

This skill allows users to mine their extensive coding history across various AI agents (Claude, Codex, Gemini, etc.) into a persistent, structured corpus. It surfaces reproducible workflows as evidence and interactively helps the user forge these past practices into reusable, installed skills across their agent ecosystem. It is designed to synthesize judgment from raw transcripts into actionable, portable skills.

This skill provides a comprehensive guide for building scalable Retrieval-Augmented Generation (RAG) pipelines and persistent agent memory using Pinecone as the vector database. It covers index setup, embedding large corpuses, implementing dense and hybrid semantic search, and advanced metadata filtering for real-world production systems. Use this for building searchable knowledge bases and context-aware AI agents.

This skill analyzes historical benchmark run data stored in JSON files. It surfaces key performance drift metrics over time, including win rate, average latency, p99 latency, and LLM-baseline cost. It is crucial for detecting subtle performance degradations or regressions that standard smoke tests might miss, ensuring stability across releases or corpus expansions.

This skill guides the planning and auditing of analysis for linguistic manuscripts to ensure evidence supports theoretical claims. It covers quantitative modeling like mixed-effects models in R, phonetic measurement, and corpus statistics. It helps establish the analytic trail from raw data to generalization without fabricating results. It ensures proper statistical specification, effect size reporting, and handling of uncertainty, aligning with academic standards for credibility in language research.

This skill provides an in-depth, entity-level scorecard for clinical or biomedical Named Entity Recognition (NER) models. It calculates precision, recall, and F1 scores (both strict and partial match modes) using a user-supplied gold corpus. Key features include per-label error breakdown, a confusion matrix, and detailed False Negative/False Positive analysis, essential for debugging model performance in highly sensitive medical domains.

This skill scaffolds a gold-standard annotation project tailored for evaluating clinical NER and de-identification models (OpenMed). It guides users through defining label schemas, writing rigorous guidelines, configuring annotation tools (like BRAT/Label Studio), generating synthetic clinical text, and creating disciplined, non-leaking train/dev/test splits. It ensures that all committed data are synthetic fixtures for reproducible evaluation.

This skill interfaces with the modern v2 REST API of ClinicalTrials.gov, allowing users to search for clinical studies matching specific conditions, interventions, and recruitment statuses. It is essential for building patient-to-trial matching features, generating a corpus of trial records for biomedical NER, or finding open studies for a given diagnosis or drug. It uses cursor pagination and requires no API key.