Blog
Useful resources about agents, skills, and AI workflows.
-
Claude Fable 5 and Your Skills: What Changes
Skills run on Fable 5 unchanged; SKILL.md is model-agnostic. What changes is the stakes: a longer-horizon model executes a bad skill for longer.
-
Claude Cowork Computer Use: The Preview, Explained
Cowork can now open files, click, and drive apps on your own screen. What the research preview does, which plans get it, and the honest caveats.
-
Claude Code's /dataviz Skill: What It Does, How to Use It
Claude Code shipped a built-in /dataviz skill for chart and dashboard design. What it does, how to invoke it, and why it is a masterclass in skill design.
-
Claude Cowork Skill Development: Draft, Test, Ship
Skill development for Claude Cowork is a loop, not a one-shot: draft the SKILL.md, run it on real work, fix the misses, version the fix. Here is the loop.
-
What Is Claude Cowork? A Guide for Non-Coders
Claude Cowork is Anthropic's desktop agent for knowledge work: give it a goal and it acts on your files to return a finished deliverable. What it is, who it is for, and how it runs skills.
-
How to Use Claude Cowork: A Setup Guide
Getting started with Claude Cowork as a non-coder: install the desktop app, grant folder access, give it a goal, review the deliverable, and use skills. Plus the context gotcha and recurring tasks.
-
Claude Skills vs Custom GPTs: Honest Comparison
A Claude skill is a portable SKILL.md folder loaded on demand; a Custom GPT is an OpenAI-hosted ChatGPT configuration. The real differences, which to build, and how to move from one to the other.
-
Claude Skills for Small Business: 15 Workflows
Claude for Small Business launched May 13 2026 and runs through Cowork with 15 ready-to-run workflows across finance, ops, sales, and marketing. The built-ins, and how to build your own.
-
Claude Cowork vs Claude Code: Which to Use?
Claude Code is a terminal coding agent for developers; Cowork is a desktop app that brings the same power to non-coders. Both run the same SKILL.md skills. Which surface to pick, and why.
-
Claude Cowork Use Cases: What It Actually Handles at Work
What non-coders actually do with Claude Cowork: organize files, build spreadsheets from receipts, draft reports from templates, synthesize research, and run recurring tasks. Plus where it is weak.
-
Claude Cowork Skills: Build One Without Code
Cowork runs the same SKILL.md skills as Claude Code, and you can build one with no code. How skills work in Cowork, the built-in office skills, and how to make your own.
-
Claude Cowork Pricing: What It Costs
Claude Cowork is part of the paid Claude plans, not a standalone free product. Which plans include it, what they cost, the preview-versus-GA nuance, and how to try skills for less.
-
Claude Cowork Plugins: What They Bundle
A Cowork plugin bundles skills, connectors, and sub-agents into one portable unit that customizes Claude for your role. What a plugin is, the marketplace including private ones, and when to use one.
-
Claude Cowork for Windows: Availability and Install
Claude Cowork has been available on Windows since the April 9 2026 general availability. How to download and install it, which plan you need, and the desktop-app behavior to expect.
-
How to Validate a Claude Skill Before You Ship
Validating a SKILL.md before you ship: the must-pass checks, the real skills-ref validate tool, the load test, and the difference between valid and actually working.
-
How Claude Skills Load: Progressive Disclosure
Progressive disclosure in three levels: the description always loads, the body loads on trigger, bundled files load on demand. How skills stay cheap on context and what makes one fire.
-
Claude Skills vs Tools: Which One Do You Actually Build?
A skill teaches a repeatable procedure and loads on demand. A tool is a callable action the model invokes. What each is, how they work together, and which one to reach for.
-
Claude Skill Versioning: How to Track Versions
There is no built-in version field in SKILL.md. How people actually version skills: git on the folder, a changelog, the metadata version convention, and real semver in plugin manifests.
-
Claude Skill References: Splitting a Long SKILL.md
When a SKILL.md gets long, split detail into referenced files that load only when the model follows the link. How to reference files by path, the references folder, and the one-level-deep rule.
-
Every SKILL.md Frontmatter Field, Explained
Every SKILL.md frontmatter field in one reference: the required name and description with their limits, plus the optional Claude Code fields, and which are core versus extensions.
-
The Claude Skill Description Character Limit, Exactly
The hard limits in one place: description max 1024 characters, name max 64, body under 500 lines, and the separate 1536-char listing cap. What happens if you exceed them.
-
The Agent Skills Standard: What SKILL.md Defines
The open Agent Skills standard defines the SKILL.md format: required name and description, progressive disclosure, and cross-vendor portability. What it is, who governs it, and why open matters.
-
How to Manage Claude Skills: Enable, Disable, and Delete
How to list what is installed, disable a skill without deleting it, remove one for good, and turn it back on. The verified Claude Code methods, settings, and folder moves.
-
Claude Skills for Writing: Workflows Worth Building
A skill turns your style guide, brief format, and editing checklist into a repeatable Claude workflow instead of a prompt you keep re-pasting. The writing and marketing skill patterns worth building, and how a non-coder makes one.
-
Claude Document Skills: PDF, Word, Excel, PowerPoint
Anthropic ships official document skills for pdf, docx, pptx, and xlsx. What each one does, how to install them in Claude Code and the API, and the gotchas worth knowing.
-
The Best Claude Skills for Developers
The categories of developer skills worth installing, where to find and vet real ones, how to judge a skill before you run it, and why the best one is often the skill you build for your own repo.
-
Are Claude Skills Free? Cost, Tokens, and Limits
Yes, the Agent Skills format is free and open, a skill is just a SKILL.md folder. What they actually cost in context tokens, and the practical limit on how many you can load.
-
Can Claude Subagents Use Skills?
Yes, a Claude Code subagent can use skills. The available-versus-invoked distinction, the two ways to combine a skill and a subagent, and when the pairing is worth it.
-
Claude Skills vs System Prompt: Where Guidance Lives
A skill loads on demand through progressive disclosure; a system prompt is always in context. How skills compare to system prompts, custom instructions, Projects, and Cursor rules, and when to reach for each.
-
Claude Skill Best Practices: Rules That Make It Fire
The authoring rules that separate a skill that fires and works from one that does not: description, scope, structure, scripts, and testing. A hub linking each deep dive.
-
How to Build Claude Code Custom Subagents
How to define your own Claude Code subagent: where the file lives, the config fields, scoping its tools, choosing its model, and the patterns worth copying.
-
Claude Agent Teams: What They Are and How to Turn Them On
Agent teams is the experimental Claude Code feature for running a team lead plus independent teammates that message each other. How to turn it on, what delegate mode does, and how it differs from subagents.
-
The SKILL.md Template You Can Copy, Paste, and Fill In Today
A copy-paste minimal SKILL.md plus the frontmatter rules: the required fields, the must-start-with-yaml rule, and what makes a description that actually triggers.
-
How to install a Claude skill
Where a Claude skill folder goes and how to make Claude load it. The project and global paths in Claude Code, installing from a GitHub repo, the plugin route, and how to confirm it loaded.
-
Claude Skill Scripts: When to Bundle Code
Bundling deterministic scripts inside a skill instead of letting the model rewrite code each run. The scripts directory, referencing a script by path, and the security tradeoff.
-
Claude Skill Not Working? A Triage Checklist
A triage checklist for a skill that will not load or fire: wrong location, invalid frontmatter, a weak description, or the listing budget. Check them in order.
-
The Claude Skill Folder Structure, Explained File by File
The anatomy of a SKILL.md folder: the entry file, the optional scripts, references, and assets directories, and how progressive disclosure maps to the layout.
-
What allowed-tools Does in a Claude Skill
The allowed-tools frontmatter field is a permission pre-approval, not a restriction. What it does, the exact syntax, the default, and the fields that actually limit a skill.
-
What Is a Claude Skill? A Plain-English Guide
A Claude skill is a SKILL.md folder that teaches the agent a repeatable job. What that means, how it differs from prompts, commands, plugins, and MCP, and why the open format matters.
-
Claude Skills vs Subagents vs Slash Commands vs Plugins
A decision matrix for the four Claude Code primitives. What each is, when to reach for it, and a short flow that ends the bouncing between four docs.
-
How to Test a Claude Skill: SKILL.md Evals
The two ways a skill fails, how to write test prompts, the draft-test-evaluate-rewrite loop, and the eval tooling that exists.
-
How to Write a SKILL.md Description That Actually Triggers
The description is the only thing loaded at discovery time, so it is the whole trigger. How to phrase it, what to name, the length sweet spot, and how to test that it fires.
-
Claude Plugins Explained: How They Differ from Skills
A plugin bundles skills, commands, agents, hooks, and MCP servers into one installable unit. What a plugin is, the marketplace, and how it differs from a single skill.
-
Claude Code Slash Commands: Build Your Own
Where command files live, the file format with a worked example, arguments and frontmatter, namespacing, and when a slash command is the wrong tool.
-
Multi-Agent Workflows in Claude Code: Skills + Subagents
The composition patterns that turn skills, subagents, and MCP into one pipeline. Serial chains, fan-out, context boundaries, and a worked example.
-
Claude Code Hooks: What They Are and How to Wire Them
Hooks fire your own code at lifecycle events like PreToolUse and PostToolUse. The event list, the config shape, real example hooks, and how to debug them.
-
Claude Code Agent View: Managing Parallel Agents
Agent View is the dashboard for parallel work. The claude agents command, launching background sessions with --bg, monitoring, and when parallel agents help or hurt.
-
How to Reduce Codex Token Usage
Why Codex burns tokens under per-token pricing, how to measure it, and a concrete checklist of reductions from AGENTS.md size to reasoning effort.
-
Gemini CLI Skills: Installing and Writing Agent Skills
Gemini CLI reads the same SKILL.md as Claude Code. Where it stores skills, three ways to install one, and how to write a description that actually fires.
-
Cursor Skills: How to Add SKILL.md Skills to Cursor
Cursor reads the exact same SKILL.md file Claude Code reads. Where it looks for skills, three ways to add one, and why a skill you author once runs everywhere.
-
OpenAI Codex on Windows: Native Setup and App
Both install paths, PowerShell versus WSL, computer use on Windows, the elevated sandbox alpha, and the known Windows gotchas with workarounds.
-
Claude Code Usage and Credits: Limits and Billing
How to read the new /usage command, what actually drives consumption, and what changed when Agent SDK and headless usage moved onto a separate monthly credit.
-
Claude Code Routines: Schedule Prompts on a Timer
Routines run a prompt plus repo on a schedule, an API call, or a GitHub event. The three trigger types, how to set one up, and the sharp edges.
-
Claude Code Effort Levels and Fast Mode
Opus 4.8 ships two dials people keep mixing up. Effort controls how hard Claude thinks; fast mode controls how fast it types. When to reach for each, and what they cost.
-
Claude Code Computer Use: Let Claude Drive Your Desktop
Computer use lets Claude open apps, click, and fill forms on your Mac. Plan and OS requirements, how to enable it, realistic use cases, and the preview caveats.
-
Claude Code /code-review: Cloud Bug-Hunting
The /code-review command runs a fleet of cloud bug-hunters on your pending changes. How to run it, where the findings land, and what it catches that a linter will not.
-
Claude Code Auto Mode: What It Approves and Blocks
Auto mode replaces permission prompts with a safety classifier. What it auto-approves, what it blocks, how to enable it, and how it compares to plan mode and skip-permissions.
-
What Is the Claude Agent SDK? A Plain-English Guide
Claude Code as a library, in Python and TypeScript. What you build with it, how it reads your Skills, and what the June 15 credit change means for your plan.
-
The 30-Minute Cursor to Claude Code Migration
A clean migration takes 30 minutes if your .cursorrules is in shape. Most guides stop at CLAUDE.md. The real payoff is step 3: promoting your repeated Cursor prompts into a SKILL.md.
-
Claude Code Slow? Three Fixes That Solve 95% of Cases
Almost always context bloat. /clear between unrelated tasks. /compact before 60% capacity. Audit your installed skills. Network and Opus-vs-Sonnet are the long tail.
-
Claude Code Skill Not Loading: the Listing Budget
If your SKILL.md is installed but never fires, it isn't a parse error. Claude Code allocates about 1% of context to skill metadata, and the lowest-priority entries get silently dropped.
-
Claude Code May 2026 Update: The /skills Filter
v2.1.108 shipped four changes. The /skills type-to-filter is the one that earns the upgrade. The other three are polish.
-
AGENTS.md in a Monorepo: The Precedence Rules
Codex walks root-down. Claude Code walks cwd-up. AGENTS.override.md trumps AGENTS.md at the same level. The published rules disagree enough that you can ship a correct layout and still hit the wrong file.
-
Where to Publish a Claude Skill in 2026
Seven registries, three tiers, two you should actively push to. anthropics/skills for credibility, skills.sh for reach, your own GitHub repo as the source of truth. The rest follows.
-
SKILL.md vs MCP Server: Which One Should You Publish?
Most authors who reach for an MCP server should have shipped a folder. State, runtime, credentials, distribution: the four axes that actually decide which primitive your idea needs.
-
SKILL.md Security: A 12-Point Audit Before You Install
Snyk found 13.4% of 3,984 audited skills shipped a critical-level issue. A SKILL.md can become shell access in three lines. The audit habit that keeps you out of the next ClawHavoc campaign.
-
The SKILL.md Compatibility Matrix Nobody Published
32 tools implement the Agent Skills spec by name. About six run a vanilla SKILL.md identically. The other 25 break portability in at least one specific way, and you can map every break to one of five axes.
-
How to Use Subagents in Claude Code
Subagents exist for context-window isolation, not parallelism. Reach for one when a task would otherwise blow up the parent context. Reach for a Skill when the work is repeatable.
-
GitHub Copilot Skills and the Cross-Vendor SKILL.md
Three major coding agents now read the same SKILL.md folder format. The author-once-deploy-everywhere pitch from October is finally infrastructure, not a handshake.
-
Codex CLI vs Claude Code: The Honest 2026 Comparison
Six months running both as daily drivers. Claude thinks better, Codex executes more cheaply, and the people who pretend one of them is dead are not actually shipping code.
-
Codex Skills: OpenAI's Take on the Open SKILL.md Standard
A Codex skill is a SKILL.md folder. A Claude skill is a SKILL.md folder. The folders are interchangeable. What changes between vendors is discovery, not the artifact.
-
Claude Code Keyboard Shortcuts: The Six That Matter
Six shortcuts cover ninety percent of daily use. The other forty in the official reference are situational. Learn these six and the three prefixes first.
-
AGENTS.md vs CLAUDE.md: Which One Should You Write?
AGENTS.md is the cross-vendor open standard governed by the Linux Foundation. CLAUDE.md is Anthropic-only. The honest answer for which to write, plus the one-line import that lets you have both.
-
How to Use Claude Skills: Code, Web, and Desktop
Where skills live on disk, how Claude discovers them, the difference between the three surfaces, and the troubleshooting playbook when one won't load.
-
How to Build a Claude Skill Without Writing Code
A Claude skill is a small thing and the format is open. The format is also written for people who already know what a YAML frontmatter block is. This is the walkthrough for everyone else.
-
The Claude Skills Marketplace: Find and Share Skills
There is no single Claude skills marketplace. There are several, plus a long tail of skills living in personal GitHub repos that no marketplace has indexed yet. Where to publish, where to shop, and what to check before installing.
-
Claude Skills on GitHub: Where to Find the Good Ones
The official anthropics/skills repo, the awesome-list community, and the curated registries that fix GitHub's discovery problem.
-
15 Real Claude Skills You Can Install Today
Verified, working SKILL.md folders from Anthropic, Stripe, Vercel Labs, and the community. With links to the source and an honest take on each.