Skill UI

Audit experiment integrity before claiming results. Uses an independent cross-model reviewer to detect fake ground truth, score normalization fraud, phantom results, and insufficient experimental scope. Invoke after experiments complete or whenever experimental claims need honest verification.

A comprehensive security skill designed to detect and mitigate indirect prompt injection attacks. It scans untrusted external content—including web pages, PDFs, and images—for hidden, obfuscated instructions before an LLM agent processes them. The pipeline combines content extraction (HTML, OCR), normalization, and dedicated ML models (e.g., LLM Guard, Prompt Guard 2) to ensure data safety.

A comprehensive guide for researchers submitting to major speech conferences (like INTERSPEECH). It details how to harden the reproducibility of speech results, covering crucial 'decay channels' such as corpus versions, text normalization rules (WER), objective metrics (EER), subjective protocols (MOS), and model stochasticity. The guide emphasizes pinning specific toolkits, commit hashes, configurations, and hardware details to ensure scientific rigor and verifiability.

This skill constructs a comprehensive, chronologically sorted patient timeline from raw clinical entities extracted from medical notes. It specializes in temporal normalization, resolving both absolute and relative dates (e.g., 'two days after admission') against a defined anchor date. The resulting sorted sequence of diagnoses, medications, and procedures is ready for immediate visualization or seamless export to FHIR resources.

This tool automates the cleanup and standardization of complex legal documents written in French (Word format). It distinguishes between deterministic corrections (typography, spacing, obvious errors) applied automatically, and context-sensitive revisions (judgment rewrites) that require intelligent human review. It generates a detailed, reversible registry of all changes, ensuring high accuracy for professional legal writing.