Skill UI

This toolkit provides six production-ready AI engineering workflows including prompt evaluation, context budget planning, RAG pipeline design, agent security audits, evaluation harness building, and product sense coaching. It transforms AI coding assistants into senior engineering partners with structured methodologies, quantitative scoring, and checklists for consistent results in LLM system optimization and security.

Orchestrates bounded coding tasks to Aider CLI, managing briefs, diffs, and reviews. Supports local models via OpenAI-compatible endpoints, ensuring offline operation and no auto-commits.

Converts requirements into precise, dependency-aware implementation plans. Creates detailed checklists, defines milestones, and maps critical paths for development teams. Ensures tasks are atomic and verifiable before coding begins, bridging the gap between requirements and execution.

Forensic analyzer for AI coding sessions. Identifies root causes of scope drift and rework by classifying user specs, agent errors, and repo friction. Generates evidence-backed reports to improve future prompts and project health.

This skill is designed to prevent AI coding assistants from exhibiting sycophantic agreement bias. When active, it forces a structured, critical assessment of user claims. The process involves extracting the core claim, assessing it independently with evidence (for/against), and concluding solely based on that assessment, ensuring objective and evidence-based responses.

This skill reviews repository architecture using Ontoly Software Graph and MCP capabilities. It helps explain architecture, module boundaries, package topology, service ownership, and architectural risks. It utilizes capabilities like ExplainArchitecture, GraphStatistics, FindCycles, and FindDependencies to provide evidence-based analysis. Suitable for coding agents requiring architectural insights without manual file searching.

Ask Matt is a workflow router designed to help users select the appropriate skill or flow from the mattpocock/skills repository. It guides users through various stages like idea refinement, implementation, and codebase maintenance. The tool manages context hygiene, handles session handoffs, and triages incoming issues, ensuring efficient navigation through complex coding agent workflows.

Design patterns for building autonomous coding agents, including tool calling APIs, permission systems, browser automation, and human-in-the-loop workflows. Inspired by Cline and OpenAI Codex, this skill provides guidance for creating efficient and safe AI agents.

AWT provides AI coding tools with visual perception and interaction capabilities for web applications. It executes declarative YAML test scenarios using Playwright, featuring visual matching via OpenCV and OCR without relying on stable DOM selectors. Supports platform auto-detection for frameworks like React and Vue, includes structured failure diagnosis, and maintains a learning database for failure patterns. Ideal for end-to-end testing workflows requiring visual assertions and AI-assisted debugging.

This skill reconstructs the detailed workflow behind a specific coding artifact (like a feature or PR). By analyzing local `ax` agent traces, it connects commits, sessions, skills, and tool calls into a concise, evidence-grounded narrative. It is used when needing to understand the step-by-step process and key decisions that led to a result.

Before You Build acts as a crucial checkpoint in the AI coding workflow. It prevents premature implementation by rigorously assessing product risk—analyzing market demand, existing user workflows, switching costs, and distribution channels. Instead of writing code, it guides the user to define core assumptions and propose the smallest, most effective validation test (e.g., landing page test, user interview) before committing resources.

This skill enables AI to adapt its operational behavior based on specific task types such as brainstorming, coding, debugging, reviewing, teaching, and deployment. It optimizes performance by switching between distinct modes like BRAINSTORM, IMPLEMENT, and DEBUG, ensuring appropriate communication styles and output formats for each scenario.