AI Toolkit

Curated prompts, skills, configs, and patterns for working with LLMs. This is an experimental area I'm working on as I collect prompts and play around with asking Claude to generate new prompts and skills.

Technique

Root Cause Analysis Technique

A systematic technique for diagnosing bugs you can't reproduce locally — production errors, intermittent failures, and gradual performance degradation. Reasoning backwards from evidence through hypothesis elimination.

Prompt

Dependency Evaluation Prompt

A structured prompt for evaluating whether to add a new dependency. Goes beyond GitHub stars — analyzes API surface, bundle cost, maintenance signals, supply chain risk, and alternatives to prevent bad long-term decisions.

Technique

Codebase Archaeology Technique

A systematic technique for understanding unfamiliar codebases using Claude Code. Builds a layered mental model from entry points inward — architecture, conventions, data flow, and the implicit decisions nobody documented.

Prompt

Explain Diff Prompt

A prompt for narrating a branch diff for PR reviewers. Turns a wall of code changes into a clear, file-by-file explanation of what changed and why — ready to paste into a PR description.

Prompt

Security Scan Prompt

A prompt for OWASP-style security review of a web codebase. Scans for XSS, injection, exposed secrets, insecure headers, and common vulnerabilities — with concrete fixes, not just findings.

Technique

Migration Assistant Technique

A technique for using Claude Code to plan and execute framework or library migrations. Scans your codebase for affected patterns and produces a step-by-step plan with incremental, verifiable changes.

Prompt

Performance Audit Prompt

A prompt for identifying performance bottlenecks in a web project. Analyzes bundle size, rendering patterns, data loading, and asset delivery to produce a prioritized list of optimizations.

Technique

Impact Analysis Technique

A technique for tracing the blast radius of a code change before you make it. Maps imports, exports, and call sites to show exactly what pages, components, and tests would be affected.

Prompt

Test Generation Prompt

A prompt for generating tests that match your project's existing conventions. Reads your test suite first, then writes tests that look like a human on your team wrote them.

Prompt

Dependency Audit Prompt

A prompt that audits your project's dependencies for outdated packages, security vulnerabilities, unused imports, and bloated bundles. Produces an actionable summary instead of raw CLI output.

Prompt

Changelog Generation Prompt

A prompt for generating clean, grouped changelogs from git history. Turns a range of commits into a human-readable summary organized by type — features, fixes, and chores.

Prompt

Commit Message Prompt

A prompt for generating meaningful commit messages from diffs. Focuses on the 'why' not the 'what', with structured format for both simple and complex changes.

Technique

Error Handling Audit

A technique for systematically auditing error handling across a codebase. Finds silent failures, swallowed errors, and missing boundary validation.

MCP Config

MCP Filesystem Server Config

A starter MCP server configuration for giving Claude desktop access to local files. Includes the filesystem server setup with scoped directory access.

CLAUDE.md

How to Write a Toolkit Entry

A template and checklist for writing AI toolkit entries. Covers the frontmatter schema, content structure, and quality bar for each category.

Technique

Design It Twice

A technique for generating multiple radically different interface designs using Claude Code sub-agents, then comparing trade-offs. Based on the principle from A Philosophy of Software Design.

Prompt

Documentation Generation Prompt

A prompt for getting Claude to write useful project documentation — README sections, architecture docs, and change summaries — by reading the actual code instead of guessing.

Skill

Git Guardrails for Claude Code

A Claude Code hook that blocks dangerous git commands — push, reset --hard, clean, branch -D — before they execute. Copy the script and config to set up in minutes.

Prompt

Refactor Planning Prompt

A structured prompt for planning refactors with Claude Code — interview, scope, test coverage check, then a plan of tiny commits that each leave the codebase working.

Technique

TDD with Claude Code

A technique for test-driven development with Claude Code — vertical slices, behavior-based tests, and one RED-GREEN cycle at a time. Prevents the 'write all tests then all code' anti-pattern.

CLAUDE.md

CLAUDE.md — Real-World Pattern

A battle-tested CLAUDE.md structure based on a production Next.js blog. Covers commands, architecture, data flow, testing, CI/CD, and tech stack.

Prompt

PRD to GitHub Issues

A structured workflow for breaking a PRD into vertical-slice GitHub issues using Claude Code. Each issue is a thin, demoable end-to-end slice — not a horizontal layer.

Technique

Critical Product Review Prompt

A prompting technique that turns Claude into a sharp product reviewer — finds UX bugs, performance issues, and dead-end interactions that you'd miss after staring at your own code.

Prompt

Code Review Prompt

A structured prompt for getting thorough, actionable code reviews from an LLM. Separates real issues from style nits and forces concrete fixes.

Skill

Debug with Context

A Claude Code slash command that forces context gathering before debugging. Prevents the 'guess and patch' anti-pattern by requiring root cause analysis first.

CLAUDE.md

CLAUDE.md Starter Template

A starter CLAUDE.md template with sections for commands, architecture, and conventions.