Ship Shit

Design and operate LLM-as-a-Judge evaluation systems using direct scoring, pairwise comparison, rubric calibration, evaluator bias mitigation, confidence scoring, and automated quality assessment. Use when building LLM-as-judge systems, comparing model responses, calibrating rubrics, debugging inconsistent evaluations, or designing A/B tests for prompt or model changes.

Automates browser interactions for web testing, form filling, screenshots, and data extraction. Use when the user needs to navigate websites, interact with web pages, fill forms, take screenshots, test web applications, or extract information from web pages.

Granular feedback on drafts without rewriting. Generates highlighted HTML with click-to-reveal inline comments. Use when user says "comment on this", "leave comments on", "give feedback on", or asks for feedback on a draft. Supports multiple lenses—editor feedback, POV simulation ("as brian would react"), or focused angles ("word choice only", "weak arguments"). A granular alternative to rewrites that lets users review feedback incrementally without losing their voice.

Recognize, diagnose, and mitigate patterns of context degradation in agent systems. Use when context grows large, agent performance degrades unexpectedly, or debugging agent failures.

Explain or reason about foundational context engineering concepts: what context is, the anatomy of a context window, attention mechanics, the U-shaped attention curve, why context quality matters more than quantity, and the mental models needed to interpret context-engineering decisions. Use for conceptual explanation, onboarding, and background reading. Route operational work to context-degradation for attention failures and context-optimization for token-efficiency work.

Improve context efficiency through context budgeting, observation masking, prefix or KV-cache strategy, partitioning, token-cost reduction, retrieval scoping, and extending effective context capacity without lowering answer quality. Use when token costs or context budgets constrain a task, tool outputs are verbose, cache hit rate is low, or context must be partitioned across agents.

Build evaluation frameworks for agent systems. Use when testing agent performance, validating context engineering choices, or measuring improvements over time.

Creates MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Activates on: "build an MCP server", "create MCP tools", "integrate API via MCP", "write a FastMCP server", "add an MCP server to an agent", or any request to wrap an external API as LLM-callable tools in Python or Node/TypeScript.

Design and implement memory architectures for agent systems that persist state across sessions, maintain entity consistency, and reason over structured knowledge. Use when building agents that persist knowledge across sessions, choosing between memory frameworks, maintaining entity consistency, or designing memory architectures for production.

Design multi-agent architectures for complex tasks. Use when single-agent context limits are exceeded, when tasks decompose naturally into subtasks, or when specializing agents improves quality.

Expert guide on prompt engineering patterns, best practices, and optimization techniques. Use when user wants to improve prompts, learn prompting strategies, debug agent behavior, or design content generation prompts.

Regenerate catalog counts, layout claims, and README summaries from canonical sources. Run after adding, removing, or renaming skills, commands, or bundles.

Audit the skills library for duplicates, stale content, spec violations, and structural issues. Run periodically or before releases.

Validate SKILL.md files against the Agent Skills spec and Claude Code extensions. Run on new or modified skills before committing.