Claude Skill Selection Criteria: 3 Rules From 82 Skills I Built and 27 I Used
The three criteria I use to pick Claude skills after building 82 and using 27 over 110 days, and where to check each one on GitHub.
Insights
Lists, reviews and decision rules that come from doing the work.
- Insights
- Insights
Why Agent Feedback Stops Working and a Loss-Based Alternative: The Deep Twin Design
Why adding rules to an agent's skill file lowers results, why multi-agent splits compound errors, and an untested loss-based feedback design.
- Insights
GEO Is About Citation Probability, Not Answer Control: A B2B Organic Content Playbook
How B2B brands can raise citation probability in AI answers by building content around customer questions, three content types, and a four-pillar website.
- Insights
Weekly Metrics Framework for a Four-Channel Solo Brand: Definitions and Chart Rules
A three-tier weekly metric system for blog, Threads, LinkedIn, and YouTube, with official definitions, leading-indicator formulas, and 12-week chart rules.
- Insights
Could Aside Become the Second Cursor? An AI Browser Built on Chromium
Aside is an AI browser from a Y Combinator startup founded by three South Korean founders. It reverses browser automation; can it follow Cursor's path?
- Insights
Keeping AI Project Docs Under Control: A Signal-to-Noise Rule for md Files
Rules for keeping AI-generated md files in check: separate living docs from dated records, allow five root filenames, and enforce it with a hook script.
- Insights
Learning Communities: 3 Criteria, 4 Picks, and 5 Ways to Get Results
Choose a learning community using three criteria, see four communities I actively join, and learn five ways to turn membership into real results.
- Insights
Discord grading bot for an n8n challenge: from submissions to similarity scoring
How a Discord bot graded advanced missions for 319 n8n challenge applicants via similarity checks against the book's originals, plus a costly storage mistake.
- Insights
OpenRouter Pay-as-You-Go vs LLM Subscriptions: Measured Costs and a Workload Split
Measured costs for a heavy coding agent: why subscriptions absorb cache reads, how $10 of OpenRouter credit vanished in a week, and how to split workloads.
- Insights
n8n Seoul Meetup 3rd Recap: Three Real-World n8n Adoption Cases
Notes from a talk on adopting n8n at a hospital, a large Korean company, and a solo business, including the failures and the lessons learned.
- Insights
The 4 Stages of AI Automation Adoption in a Company: The Challenge Stage Decides It
Notes from a September 1, 2026 conference on the four stages of company AI automation adoption, and what managers and individuals should do at each.
- Insights
Running an AI Agent in Discord: Threads, Reactions, Model Routing, and Tool Limits
Established Discord practices for an always-on AI agent bot: thread vs. inline replies, reaction status, model routing, tool permissions, and message limits.
- Insights
Does Studying With AI Help? Practice Up 48%, Exam Down 17%
A study of about 1,000 high schoolers using GPT-4 found practice scores rose 48% but exam scores fell 17%. The difference came down to how AI was used.
- Insights
n8n 6-week challenge review: what participants built and what stuck
A record of the n8n 6-week finish-the-book challenge: 319 sign-ups, weekly student projects, a 4.9 review rating, and what changed after six weeks.
- Insights
Where to Get Data for Vibe Coding: 9 Free Public Data API Sources
Six Korean public data portals and three API directories, sorted by use case, plus how to have AI pick the right API for your project.
- Insights
GEO vs SEO: The Three Gates Before an LLM Cites Your Site
Learn why GEO builds on SEO, why client-side rendered pages are invisible to AI crawlers, and how to measure mentions, sources, and citations in LLM answers.
- Insights
AI Agent Architecture Patterns in n8n: How to Choose and Where Each Breaks
Compare nine AI agent architecture patterns, their failure modes, and n8n production fixes for memory, error recovery, and human approval.
- Insights
Choosing an LLM Backend for a Self-Hosted Hermes Agent: Cost, Security, and Policy
Pick an LLM backend for a self-hosted Hermes agent by weighing cost, data jurisdiction, and Claude subscription policy. Includes pricing and risk tables.
- Insights
Agent Knowledge Management Standards: Storage and Retrieval vs. Context Engineering
Learn how agent knowledge storage and retrieval map to CoALA memory types, sleep-time consolidation, and progressive disclosure.
- Insights
Is ChatGPT Plus Enough for Codex? A 72-Day Usage Review
Real Codex usage data from 72 active days: token totals, model mix, and rate-limit peaks. Find out why bursty users should pick Plus $20 over Pro.
- Insights
Building a Solo Board Game Server in One Weekend with Claude Code: A Retrospective
A 36-hour build of an Age of Steam web game with Claude Code: code size, commits, API cost breakdown, and why the LLM hard difficulty was shelved.
- Insights
Claude Code Harness Audit: Finding Silent Failures in Hooks, npx, and YAML
A three-month drift audit of a Claude Code harness. Learn which failures stay silent and how to test hooks, avoid npx squatting, and validate YAML.
- Insights
When Can AI Agents Run Autonomously? A Verifier and Context Test
Autonomous agent loops need a machine verifier and explicit context. Learn the two-axis test, the evidence, and when humans should write rubrics first.
- Insights
LLM Wiki Method vs Google OKF: What Is Shared and What Differs
Compare a markdown-based LLM wiki pipeline with Google's Open Knowledge Format v0.1. Learn where they match, where they differ, and when to align your schema.
- Insights
Why Domain Expertise Matters More Than Coding Skill With AI Coding Agents
Anthropic's analysis of about 400,000 Claude Code sessions shows domain knowledge amplifies agentic coding results more than coding training does.
- Insights
MCP vs SDK: Why Protocols and Building Frameworks Are Different Layers
Why MCP and SDKs are complementary layers, not rivals, and how Google and n8n each handled protocol support versus building a framework.
- Insights
Air-Gapped Automation Delivery Playbook: Lessons from Traditional SI Methods
A four-gate delivery playbook for automation projects in closed networks. Catch tool limits and security constraints in week one, not week three.
- Insights
Fable 5 vs Opus 4.8: Real Session Cost and Agent Behavior Compared
Fable 5 costs twice as much per token as Opus 4.8, but measured sessions cost 1.0 to 1.4 times as much. See the token, cost, and intervention data behind it.
- Insights
Claude Code Hooks: Why Deterministic Control Points Work for Stochastic Agents
Why Claude Code hooks guarantee deterministic control over AI agents, traced from git hooks and Design by Contract to neuro-symbolic guardrails.
- Insights
MCP Ecosystem in 2026: How the Protocol Moved to Shared Industry Infrastructure
Why MCP moved to Linux Foundation governance, how it complements Google's A2A, and what the adoption and security evidence supports.
- Insights
Which IT Term Fits a Paused Service: Teardown vs. Mothballing, Turndown, and Decommission
Teardown means destroying one-off resources, not keeping them for later. Here is which term fits a service you shut down temporarily and plan to restart.
- Insights
Google's Generative AI Search Optimization Guide: AEO and GEO Are Still SEO
Google's first official guide to AI Overviews and AI Mode says AEO and GEO are just SEO. Here is what to keep doing and what to skip.
- Insights
excelkospi Architecture Teardown: Edge Functions, Durable Objects, Supabase
Teardown of excelkospi.pages.dev: vanilla JS on Cloudflare Pages Functions, a Durable Object, and Supabase, plus polling tricks that cut cost.
- Insights
Are Vector Databases Still a Separate Category? History and Market Data
Traces vector search from 1998 research to today's database features, using DB-Engines rankings and funding data to assess standalone vector DBs.
- Insights
What SAP's Investment in n8n Means for Enterprise Automation and Korean Companies
SAP's $5.2B investment in n8n moves it from SMB automation into enterprise AI orchestration. See what changes for buyers, consultants, and self-hosters.
- Insights
Karpathy CLAUDE.md vs Anthropic's Opus 4.7 Guide: What to Add to Your Claude Code Setup
Compare Karpathy's four CLAUDE.md principles with Anthropic's Opus 4.7 guide, find the gaps, and get six concrete changes for your Claude Code setup.
- Insights
OpenClaw vs Hermes Agent: Platform Breadth vs a Self-Learning Operator
Compare OpenClaw and Hermes Agent on messenger channels, tool integrations, memory, token cost, and stability to choose the right chat-driven AI agent.
- Insights
Share a Home Mac mini MySQL With 28 Students: Cloudflare Tunnel and Adminer
Four ways to open a home Mac mini MySQL server to 28 students, and why a Cloudflare Tunnel serving the Adminer web UI needed no client installs.
- Insights
Home Wi-Fi Latency Spikes on 2.4GHz: Neighbor AP Interference and Moving to Channel 1
A step-by-step diagnosis of Wi-Fi latency spikes traced to a neighbor's access point on 2.4GHz channel 11, and how moving the router to channel 1 fixed it.
- Insights
Archiving Threads Drafts vs. Published Posts with a Hybrid Supabase and File Pipeline
Log every Threads draft-to-published edit with a Mac mini pipeline: Supabase for metrics, files for text and diffs, and a manual push gate.
- Insights
Which AI Diagram Tool Should You Use: diagram-design, draw.io, Mermaid, or Excalidraw?
Compare diagram-design, draw.io, Mermaid, and Excalidraw on one n8n workflow. Find out which tool fits lecture slides, editable files, docs, or discussion.
- Insights
Notion API vs Notion MCP: Which to Use for AI Agent Integration
Compare the Notion API and Notion MCP for AI agents, see why Codex missed a Claude Notion connection, and learn to find with MCP and edit with the API.
- Insights
Keeping Multiple AI Agents Consistent with SSOT for Paths and Tokens
Stop AI agents from drifting apart. Keep secrets in one shell file, paths and IDs in one shared guide, and point every agent config file to that guide.
- Insights
How to SSH Into a Home Mac Mini From a MacBook With Tailscale, No Port Forwarding
Learn how Tailscale links a MacBook and a home Mac mini on a private network so you can SSH in without port forwarding, plus a quick way to diagnose timeouts.
- Insights
How to Turn a Mac mini Into a 24/7 Home Server and Connect It with SSH and Tailscale
Set up a Mac mini as an always-on home server, connect it over SSH and Tailscale, and avoid the sleep failures that break laptop-based automation.
- Insights
Why n8n Still Matters When Claude Code Can Write Your Automations
Docker pulls, npm downloads, and GitHub stars show n8n and Claude Code coexisting. Learn the three moats n8n defends and when to use each tool.
- Insights
Claude Code under launchd: debugging five stacked failures in headless automation
Five stacked failures broke a Claude Code briefing run by macOS launchd. Here is how each layer was found and fixed for headless automation.
- Insights
Claude Code Recurring Execution: /loop vs /schedule vs CronCreate Compared
Compare /loop, /schedule, CronCreate, and Routines in Claude Code to pick the right recurring run for local tools, cloud execution, or external triggers.
- Insights
VS Code ripgrep Using 1,100% CPU: Causes and How to Prevent It
Why VS Code's ripgrep can spin on symlinked .venv folders and peg the CPU, plus the three settings.json changes that stop it.
- Insights
Claude API Cost Breakdown: How Cache Hit Ratio Changes What You Pay
One real Claude Opus 4.6 API call broken down by token type, showing why cache writes dominate cost and how cache hit ratio changes the total bill.
- Insights
Fixing n8n Gmail OAuth2 Unauthorized Errors on Railway: Redirect URI Mismatch
Gmail OAuth2 reconnect fails with Unauthorized in self-hosted n8n on Railway. Learn why the callback URL points to the Railway default domain and how to fix it.
- Insights
Claude Code Token Optimization: Why a 2-Hour Session Exhausted the Max x5 Budget
Why parallel lecture editing and PDF indexing exhausted a Max x5 token budget in two hours, plus a per-task token estimate and a delegation strategy.
- Insights
Claude Code + Obsidian: Automatically Capture Sessions and Generate a Daily Work Log
Set up a Claude Code Stop Hook that saves sessions to Obsidian as Markdown, then uses Gemini to draft a daily work log. Setup takes about 15 minutes.
- Insights
Is n8n Dead? What Docker Pulls and Market Data Show in 2026
n8n is not dying. Docker pulls, GitHub stars, and valuation data show the product is growing while tutorial content on YouTube has saturated.