Guides built on test data · page 4 of 6

The SkillProof blog

Cream illustration comparing an ignored email draft with a one-goal reply-now draft

Jul 10, 2026 · 6 min read

Emails People Actually Reply To: One Goal, No AI Fluff

12 blind email tests, same model: 9/12 got a 'reply now' with the skill vs 3/12 without. One goal, hard word budgets, a fluff scanner. Free, MIT.

Cream illustration of a prompt marked up like a bug report before a minimal rewrite

Jul 10, 2026 · 7 min read

The Prompt Fixer That Diagnoses Before It Rewrites

A Claude prompt optimizer that names the failure first, changes the minimum, and ships an A/B plan — benchmarked, honest about a tiny sample. Free, MIT.

Cream illustration of a README with each line tethered to a source file

Jul 10, 2026 · 7 min read

A README Where Every Claim Traces to the Code

We benchmarked a code-grounded README skill on 3 frozen OSS repos: 18/18 complete sections, preferred 3/3, every command actually run. Free, MIT.

Cream illustration of a code review that must prove each finding before reporting it

Jul 10, 2026 · 8 min read

Code Review That Proves Every Bug It Flags — Benchmarked

Our own bench caught a checklist-guided review missing a live P1. So we rebuilt it: 18.5/21 caught, 0 false positives, a failure path on every finding.

Cream illustration of a source document compressed into a faithful summary with flagged contradictions

Jul 10, 2026 · 7 min read

Claude Summaries That Don't Quietly Rewrite the Facts

6 pre-registered documents, every claim audited against source. The baseline silently resolved 3 planted contradictions; the skill flagged all 3. Free, MIT.

Cream illustration of four rewrite attempts, three crossed out and the fourth winning a blind preference test

Jul 10, 2026 · 9 min read

We Failed 3 Times Before This Humanizer Beat the Baseline

Three humanizer versions lost to the raw baseline. The fourth won blind, 8-4, with facts intact. The full honest arc, published.

Cream-colored illustration of a terminal window with a soft memory wave connecting one session to the next

Jul 8, 2026 · 12 min read

Claude Code Memory: The Complete Guide to Persistent Context

Claude code memory explained: CLAUDE.md, memory directories, memory skills like claude-mem, and plain files: what each solves, and how to set one up.

Cream illustration of a central hub node with lines connecting out to a row of small tool cards

Jul 8, 2026 · 13 min read

Best MCP Servers for Claude in 2026 (Our Picks)

Our engineering picks for the best MCP servers for Claude: filesystem, GitHub, browser automation, Postgres, search, Slack, memory, plus the token cost of each.

Cream illustration of one large agent card fanning out into three smaller worker cards

Jul 8, 2026 · 12 min read

Claude Code Subagents: A Practical Guide (2026)

What Claude Code subagents are, when they beat the main loop, how to write a custom agent definition, and the failure modes we've hit running them daily.

Cream illustration of a friendly terminal window with a document card and a chart card resting beside it

Jul 8, 2026 · 11 min read

Claude Code for Non-Programmers: No Coding Required

Claude Code without coding: real Word and Excel files, research summaries, data cleaning. A 15-minute setup and first-week plan for non-programmers.

Cream illustration of a terminal window with a chain of connected steps representing an event hook

Jul 8, 2026 · 12 min read

Claude Code Hooks: The Complete Guide (2026)

Claude Code hooks explained: what they are, the settings.json schema, 6 working recipes, and when a hook beats a skill for guaranteed behavior.

Cream illustration of a plugin box containing skill cards, an agent card, and a command card

Jul 8, 2026 · 11 min read

Claude Code Plugins: What They Are & How to Use Them

Claude Code plugins explained: how they bundle skills, agents, hooks, and MCP servers into one install, and the security checklist before you install one.

Cream illustration of a terminal window on one side and a code editor with ghost-text autocomplete on the other

Jul 8, 2026 · 12 min read

Claude Code vs GitHub Copilot: The Honest Comparison

Claude Code vs GitHub Copilot compared fairly: autocomplete vs agent workflows, repo integration, extensibility, and who should actually use which.

Cream illustration of a smart-contract code file card with a shield badge overlaid on the corner

Jul 8, 2026 · 10 min read

Claude Skills for Crypto and Web3: What's Actually Safe

Claude for web3 and Solidity work: what skills can do, the line they shouldn't cross, and honest queue status on the crypto skills we're testing.

Cream illustration of a terminal window and a code editor window sitting side by side

Jul 8, 2026 · 13 min read

Claude Code vs Cursor: An Honest Comparison (2026)

Claude Code vs Cursor compared without the marketing: interface, workflow, extensibility, and which one actually fits how you work.