Skill UI

This skill executes the DeepSWE benchmark, a comprehensive coding-agent evaluation suite. It uses OpenRouter and mini-swe-agent to score various Large Language Models (LLMs) on 113 tasks, providing reproducible results for model comparison, development assessment, and research benchmarking. Use this when a standardized, deep evaluation of LLM coding capabilities is required.

A lightweight CLI tool for systematically comparing various coding AI agents (such as Claude Code, Aider, and Codex) head-to-head. It measures crucial metrics—pass rate, API cost, execution time, and consistency—on reproducible tasks defined via YAML manifests. Use this to objectively select the best agent for your codebase or to validate model updates before production deployment.

This skill helps developers structure and record critical architectural decisions (ADRs) made during coding sessions. It automatically detects moments of decision, capturing the context, alternatives considered, and the rationale (trade-offs and consequences). This maintains a living log that explains 'why' the code base was built the way it is, preventing knowledge loss and improving project maintainability.

This utility executes an autonomous, iterative loop, simulating a comprehensive agent workflow. It systematically manages tasks by checking current status, predicting the optimal next action using specialized tools, executing the required task (e.g., coding, testing), logging the progress, and scheduling the next iteration. It supports task discovery from multiple sources, including team tasks, swarm tasks, and file checklists, ensuring continuous, automated progress towards a defined goal.

This tool serves as a persistent memory and state manager for complex coding projects within the Claude environment. It allows users to initialize new projects with structured metadata, track detailed session progress (including summaries, decisions, and next steps), and save the entire context reliably. By ensuring project continuity across disconnected sessions, it drastically improves the flow and consistency of the development workflow.

A robust framework for orchestrating complex, multi-step coding tasks using multiple AI agents. It functions by first planning the entire project into a mission dependency graph (DAG), then dispatching specialized agents to isolated Git worktrees for parallel development. The system automatically monitors progress, handles merges, and generates structured, comprehensive reports detailing changes, completed tasks, and next steps.

Automatically analyzes unfamiliar codebases to generate a comprehensive, structured onboarding guide. It systematically detects package manifests, framework fingerprints, architecture maps, key entry points, and coding conventions. Ideal for new developers joining a project or setting up a repository for the first time.

Code Health MCP provides structural maintainability feedback and design-level health scores, going beyond standard linting. It is crucial for AI-assisted coding workflows, allowing users to review code quality, verify score deltas after changes, and gate commits or pull requests to prevent codebase regression. Use it when refactoring, updating legacy modules, or ensuring the integrity of large diffs.

This comprehensive guide establishes foundational coding standards and best practices for enhancing code quality, readability, and maintainability across all projects. It covers naming conventions, immutability patterns, error handling, and core design principles like KISS, DRY, and YAGNI, applicable to TypeScript, JavaScript, and React development. Use it when starting a new module or performing code reviews.

This skill analyzes completed code sessions to automatically extract reusable patterns, transforming raw interaction data into structured, learned skills. It allows the AI to continuously improve its domain knowledge by capturing techniques like error resolution, debugging methods, and workarounds, making it effective for advanced coding assistance and knowledge curation.

An advanced system that analyzes coding sessions via hooks to extract atomic, confidence-scored behaviors called "instincts." It tracks patterns, allowing users to evolve these instincts into comprehensive skills, commands, or agents. V2.1 introduces project-scoped learning, preventing knowledge contamination between different codebases and improving development workflow efficiency.

This tool automatically reads usage data from active AI coding sessions (JSONL format). It tallies token usage per model and tier, calculates the total cost in USD using specified pricing models, and persists a structured record into the cost-tracking namespace. Use this after completing a chunk of work or at the end of a session to provide accurate data for cost reporting and optimization.