Description: AdaptaHorts-Engines is a comprehensive repository containing a collection of independent, specialized AI agents developed in Python. Designed to solve complex problems across different domains, this modular framework integrates Claude AI and Python 3.x to execute domain-specific tasks efficiently. This release includes the following core engines: Adaptahorts Agent: An agent focused on adaptation routines and PFBCIA (Problem Formulation, Building, Continuous Integration, and Adaptation) processes. AIGIS Mapping Agent: A dedicated module for handling complex mapping tasks using the AIGIS system. Recommendation Agent: A specialized
Carlos Eduardo Pacheco Lima, Mariana Rodrigues Fontenelle, Marcos Brandão Braga, Ítalo Moraes Rocha Guedes, Juscimar da Silva, Fábio Akiyoshi Suinaga
Research artifact for the paper "No Free Context: Break-Even Arithmetic for Prompt Reduction in LLM Agents", 2026 International Conference on Artificial Intelligence Centric Systems, Services, and Applications (AISSA 2026, IEEE). It contains the code and every simulation record behind the paper's numbers: 2,975 closed-loop tau2-bench simulation records (retail, airline and telecom; Claude Haiku 4.5 and Claude Sonnet 4.5 planners), 240 simulations with per-decision pruning telemetry, and 456 simulations run with Amazon Bedrock prompt caching enabled. bash reproduce.sh rebuilds every number, table and figure in the paper from these records in a
Parikh, Yash Rajeshbhai, Saraiya, Eshaanee Krishnakumar
🚀 Release v1.0.0 — AI Engineering Skills & Instruction Prompts The initial production release of AI Engineering Skills & Prompts , tailored for AI coding assistants (Google Antigravity / Gemini, Cursor Rules, Claude Projects, GitHub Copilot). 🌟 Key Highlights Standard Author Metadata & Frontmatter: Every skill and documentation file now includes structured YAML frontmatter ( author , name , description , version , repository , license , created_at ) for deterministic agent consumption. MIT License with Dedicated AI Grant: Fully permissive open-source license explicitly granting all Artificial Intelligence (AI) agents, LLMs, and automated tool
Saddam Al-Slfi
Source code for the multimodal LLM evaluation pipeline described in "Multimodal Large Language Models for Clinical Scar Assessment: Agreement, Bias, and Reproducibility." Implements a five-stage pipeline (image ingestion, preprocessing, multi-provider API dispatch, JSON schema validation, and Excel/CSV export) used to evaluate five MLLMs (GPT-5.4, Claude Sonnet 5, Gemini 3.1 Pro, Qwen3.7-Plus, GLM-5V-Turbo) against blinded human observer scores on the POSAS Observer Scale. Ships without patient images, API keys, or run outputs; researchers supply their own image dataset and API credentials.
Abdullah, Ahmed
First archived reproducibility release supporting the manuscript Diagnostic Performance of Consumer-Facing Artificial Intelligence Applications for Skin Cancer Across Fitzpatrick Skin Tones . Included Definitive identifiers and analysis-ready data for the 231-image Fitzpatrick17k primary cohort Data for the 252-image DDI external cohort Per-image outputs from AI Skin Scanner, Model Dermatol, ChatGPT, and Claude Diagnostic and benign/malignant adjudication dictionaries Reproducible R analysis scripts, result tables, and publication figures R environment and package-version records The complete analysis was rerun successfully before this releas
Chen, Jason, Yao, Leslie, Borden, Ryan, Kekad, Ziad, Mueller, Kyle, Williams, Andrew Steven, Trinidad, John
Code, patches, git histories, models and paper. Gradient descent changes the values of a network's weights inside an architecture fixed before training. Here an LLM coding agent (Claude Opus 5.5, directed by a person) rewires small trained networks after training instead. It writes patches out of exact identities of the (Leaky)ReLU activations the networks already use, which carry, gate, bound and switch signals, and makes room for them by freeing channels by regression or by growing channels with zero outgoing weights. On four small models the patches do what training the model does not: a coordinate wire and an AND gate give a cat GAN odd-e
Upatising, Voravee, Claude (Anthropic; model Claude Opus 5.5)
mirtverse 0.3.0 adds an AI-in-measurement layer, an open textbook, seven more teaching labs, tools for instructors and an agent that analyzes test data with Claude. The paper describing the package is out as a preprint on PsyArXiv (https://doi.org/10.31234/osf.io/j9ya7_v1). The full list is in the changelog . Highlights AI in measurement, with psychometric checks. train_scorer() and agreement_report() train and judge an automated scoring engine. item_model() , generate_isomorphs() and calibrate_families() generate items from templates and check that each family is equally difficult. item_features() and predict_difficulty() predict difficulty
Bulut, Okan
Schema 1.0.0 (unchanged) In the Atlas at this release: 1,206 entities (+174), 2,409 typed relationships (+242), 148 countries (+34). Schema Multilateral bodies (#511): a default scope rule and the World Bank Group's data programmes (#695) Data Add Jordan (roadmap #480); record Dominica search (#734) Add Saint Kitts and Nevis and Sudan, finishing group 9 (roadmap #478) (#733) Add Antigua and Barbuda and Mozambique (roadmap #478) (#732) Add Eswatini: ESCCOM, Data Protection Act 2022, SzCIRT (roadmap #478) (#731) Countries (#479): Saint Lucia, with the Data Protection Act from the Attorney General Chambers' Revised Laws (#730) Countries (#476):
Daan Steur, Claude
The first integrated DRG validation platform fully authored by a physician. It unites clinical guidelines, coding guidelines, DRG-system analysis, and regulatory law in one connected analytical frame, with utilization and denial management read through a value-based lens. The analysis and every final judgment are the physician's. Artificial intelligence serves in defined supporting roles — research, citation, retrieval-augmented generation over a Zotero library, and maintenance of the codebase and repository by Claude — and takes no part in the final evaluation.
Alattar, Sherif
A registry of 1,574 laws, regulations and bills governing artificial intelligence and its inputs (data, compute, chips, data centers, energy) from 57 countries and blocs, with structured annotations of 770 of them: dated events, thresholds, obligations, sanctions, definitions and relations, each with its provision and a verbatim quote. Annotated by Claude Haiku 5.5 and validated with a blinded protocol (Tier A item precision 93%, recall 96%). Research data, not legal advice.
Covarrubias, Matías
Analysis scripts accompanying the study "Management Agreement and Guideline-Citation Metadata Errors in Large Language Models for Small Renal Mass Management: A Prompt-Sensitivity Evaluation." The scripts reproduce the statistical analyses reported in the manuscript, including inter-rater reliability (Cohen's kappa), model comparisons (Cochran's Q and exact McNemar tests with Bonferroni-adjusted pairwise p values), a scenario-level Monte Carlo permutation test, and an exploratory, hypothesis-generating cluster-robust logistic regression. The study evaluated ChatGPT-5.5, Gemini 3.6 Flash, and Claude Sonnet 5 across 30 standardized small renal
çaviş, mücahit, çaviş, tuğba
Manual benchmark comparing 5 AI models (Claude Sonnet, ChatGPT default, Gemini Flash, DeepSeek unified without "Deep Thinking", and Grok Fast mode) on their free tiers, across 27 tasks spanning 5 difficulty tiers (code, reasoning, and writing), plus 15 timed speed tasks in Tier 1. Overall accuracy: Claude 27/27 (100%), DeepSeek 26/27 (96.3%), Grok 26/27 (96.3%), ChatGPT 25/27 (92.6%), Gemini 25/27 (92.6%). DeepSeek was the fastest model across all three task categories (code, reasoning, and writing). Includes full methodology, evaluation rubric, consolidated results (CSV), comparative charts, data dictionary, and documentation in both Spanish
Daboin Jiménez, Jesús David
