Agent Skills Hub

autoresearch

Autonomous multi-team codebase improvement agent with specialized modes: narrow (goal-directed), broad (hypothesis-divergent), and sweep (quality-focused).

research-review

Get deep, critical, NeurIPS/ICML-style peer reviews of your research, paper drafts, and experimental setups using external LLMs via Codex MCP.

evaluating-code-models

Evaluate code generation models using BigCode Evaluation Harness. Benchmarks include HumanEval, MBPP, and MultiPL-E with pass@k metrics for multi-language coding models.

golang-cli-review

Expert automated code review for Go CLI applications, focusing on Cobra/urfave patterns, security, performance, idiomatic Go, and robust error handling.

python-plan-optimization

6-phase read-only Python analysis workflow that identifies design principle violations, code smells, and modernization opportunities based on specific project types (POC to Open Source).

eval-harness

Official evaluation framework for AI agent sessions, implementing Evaluation-Driven Development (EDD) principles to ensure reliability.

orchestration-patterns

Framework for orchestrating long-running agentic tasks, evidence-based delivery, and automated QA gates following Simon Willison's iterative loop.

file-todos

A structured file-based system for tracking todos, managing technical debt, and coordinating code review workflows directly within your repository.

codebase-analyzer

Automates the creation and maintenance of CLAUDE.md files. It monitors codebase evolution and keeps project memory in sync with file changes, structure, and build commands.

update-llm-model-list

Audit and synchronize the supported LLM model list in assets.py against the authoritative litellm registry.

core-principles

Architectural governance and project standards for React 19 SPA development, ensuring consistency in stack integration, project structure, and agent execution rules.

multi-llm-advisor

Fetches expert perspectives from OpenAI Codex and Google Gemini for architecture, code reviews, and debugging, with transparent LLM synthesis.