Claude — Konrad Wright

What Claude can do and how to drive it, as of October 2026.

Expanded Capabilities

Ask for any of these in a chat and Claude makes it as a live page you can edit, comment on and share.

Motion New · Beta

Short animated explainers built from code, not a video model, so every word, number and timing stays editable. Export as MP4.

Ask for an animation, or pick Output › Motion in the message box.

Team & Enterprise only, not Pro or Max yet

Docs

Writing you and Claude edit together. Download as Word or PDF.

Every plan

Slides

Decks you present from Claude or download as PowerPoint.

Every plan

Design

Posters, social posts, pages and app screens. Replaces claude.ai/design, which closes December 14.

Every plan

Dashboards Beta

Charts on live data from your connected sources; the numbers refresh every time it opens.

Paid plans

AI inside artifacts

An artifact can call Claude itself, like a question box on a shared dashboard.

Settings › Capabilities

Power User Tools

In a chat

/btworCmd ;: Side question about the session without interrupting it. Not kept after you close the app.
Esc+Escor/rewind: Go back to an earlier message, undoing what came after.
/recap: One line on where a session stands.
/output-style: Make replies shorter or plainer.
Cmd+/: Every keyboard shortcut.

Across sessions

Option+click: Open a session in split view, beside the current one.
@name: Message another session, or ask to hear when it finishes.
Remote Control: Carry on from your phone. Settings › Claude Code makes it the default.
Pin: Pin one session as a hub and ask it what the others are waiting on.

Cost and setup

/advisor: Run a cheaper model that calls Opus or Fable for the big decisions.
/skill-doctor: What each skill costs in context, and how often it's used.
/doctor: Check your setup and apply fixes.
/insights: A report on how you use Claude Code, with features to try.

Mac keys shown. On Windows, Cmd is Ctrl and Option is Alt.

Parallel Work

Two ways to run several jobs at once.

Nested sessions

Work that needs your computer

Each job is its own chat under the one that started it; open any to steer it
Runs on your computer, with your files, skills and tools
Stops if the computer sleeps
Every plan

Projects Beta

Long jobs that run while you're away

A coordinator chat runs worker threads with one shared memory and instructions
Runs in Anthropic's cloud; anything you add is stored there
Keeps going with your computer off
Pro and Max, with a waitlist

Planning the Work

Two ways to set the work up before Claude starts. They agree on one thing: the useful structure is files Claude can read.

Folders, not agents

Any job you repeat

One folder for prompts and instructions, one for tools and when to use them, one for data connections
Claude Code or Codex reads them with plain shell commands, so the folders are the architecture
Ask a question and it wanders the folders and becomes the agent that question needs
A coded agent per workflow is wasted effort about 99% of the time, he says

Elephant-Goldfish

Big features that are costly to get wrong

/egm-bootstrap writes a README for every code folder. Point CLAUDE.md at them, or Claude won't read them
/egm-brainstorm questions you until the idea is clear; /egm-draft writes the design doc a section at a time
/egm-review hands it to fresh subagents that never saw your chat, so they can't just agree with it
/egm-drift-report later checks the code against those docs
Free plugin from Google's GitHub, not an official Google product. In Claude Code every command takes an egm: prefix, as in /egm:egm-review. Skip it for small fixes

The goldfish review in /egm-brainstorm, after AI LABS' walkthrough and the repo: one subagent starts six more, each reading the plan through a different lens, and folds their notes into a single report. /egm-review then runs its own five stages, one after another.

If you install EGM, read the CLAUDE.md it writes. In AI LABS’ test, Claude added a line from memory saying subagents can’t start subagents, which is no longer true and stops the review from running as designed.

Writing Skills

Rules from Anthropic’s updated skill guide. Newer models need less hand-holding and more structure.

01
Keep SKILL.md under 500 lines
Split the rest into files linked from it, so Claude loads each one only at the step that needs it.
Anthropic docs
02
Link reference files one level deep
A file reached through another file may only be previewed, roughly its first 100 lines.
Anthropic docs
03
Give long reference files a contents list
Anything over 100 lines opens with a table of contents, so even a preview knows where to jump.
Anthropic docs
04
Match freedom to fragility
Plain instructions where many approaches work, a template where the shape matters, an exact script where a mistake is costly. One skill can mix all three.
Anthropic docs
05
Test on every model you'll run it with
Enough guidance for Haiku, clear and efficient for Sonnet, not over-explained for Opus. Skills written for older models can be too prescriptive for new ones.
Anthropic docs
06
Cut what Claude already knows
Keep your prices, terms and internal rules; drop the paragraph explaining what an invoice is. Write the description in the third person.
Anthropic docs
07
Hand long jobs a checklist
Steps Claude copies into its reply and ticks off, with a go-back line: if a total doesn't match, return to step 2.
Anthropic docs
08
Close the loop
Run it, check it against a validator or a style guide, fix, repeat, and only proceed when it passes.
Anthropic docs
09
Use a template, examples or a fork
When the output must be defined: a template (strict or a sensible default), input and output pairs, or a conditional workflow.
Anthropic docs
10
Rules that can never break go in a hook
A hook is code Claude Code runs at a set moment whether or not the model follows the skill. Capital letters work most of the time; a hook works every time.
the video

Paste a skill to check it against the rules.

Live · paste a SKILL.md, get the rules applied

SKILL.md contents

Turn on JavaScript to run the checks.

The checks are the ones a script can make from the text alone: the frontmatter limits, the 500-line ceiling, first- or second-person descriptions, missing contents lists, capital-letter rules that want to be hooks, and step lists with no checklist. It can't test your skill on three models for you. Nothing you paste leaves this page.

Skill Shelf

Skills I’d reach for again, kept here so I don’t have to find them twice. The claims on each card are their makers’, not my tests.

BAMA (Build Any Mac App)
A native Mac app built just for me: a tool, a utility, a clone of an app I pay for.
How it works
Skills grouped into start, plan, code, design and ship, with a /bama router that picks the next one. Builds Swift apps with local data, can rebuild a design from screenshots of an app you like, has a measure-then-optimise loop for heavy work, and can add a command line so agents can drive the app.
Point Claude at the repo and ask it to install the plugin. After that, asking for a Mac app is enough.
by Dave Ebbelaarfree
/generate
Images and video for a project without a Higgsfield-style subscription.
How it works
Routes each job to the cheapest model aggregator that's up (Kie.ai first, then fal.ai, then WaveSpeed), writes the prompts, works to a budget you set, logs every prompt, and keeps every file on your machine. You pay per generation.
There's no public repo. You build your own from his guide, which needs API keys from the aggregators. Give it a hard rule to quote the cost before it submits anything.
by Jay E · RoboNuggetspay per use
Thermo-Nuclear Code Review
A brutally strict review of a branch before it merges: did this change delete complexity, or just move it around?
How it works
Hunts for "code judo" that makes a change dramatically simpler, and flags files pushed past 1,000 lines, conditionals bolted onto unrelated flows, thin wrappers, needless casts and logic in the wrong layer. It leaves a few high-conviction comments instead of a pile of nits, and won't approve a structural regression.
It's a single SKILL.md in Cursor's team kit plugin, in the same format Claude Code reads. Drop it in your skills folder; it only runs when you ask for it by name.
by Cursorfree

BAMA’s maker says its speed loop cut his video editor’s export time tenfold, and the /generate video puts GPT Image 2 at 5 cents for a 2K image on Kie.ai, against roughly 31 to 34 cents through Higgsfield’s plans. Kie.ai is the cheapest but sometimes unreliable, so he defaults to fal.ai and uses WaveSpeed for models fal doesn’t have. The thermo-nuclear review is the one I’d run on this site: its 1,000-line limit and “delete complexity, don’t move it” are my own code standards, said louder.

Mods

Add-ons that change Claude Code itself. Describe one to Claude to build it; it lasts one session unless you ask Claude to save it as a plugin. Needs Claude Code 2.1.287 or newer, in the terminal or the desktop Code tab.

Savvy Progress

A progress bar above the prompt, plus a live panel of every subagent with its model, progress and rough cost.

Claude Skins

Re-themes how Claude Code shows its work: icons per tool call, colors for edits. Ships with seven themes.

Open source

You Should Know

A second agent reads Claude's replies and pops up anything important you might skim past, like a cost.

AnthropicUses tokens

File Tree

A file panel where files shimmer while Claude touches them and turn green once the change is committed.

No tokens

Cache Tax

Warns before a message that would reload a cold prompt cache, with an estimated cost, and can keep the cache warm while you're away.

Blast Radius

Stops risky commands like deleting a folder, lists exactly what would go, and waits for a person to press proceed.

Anthropic

Reflect

When you correct Claude, offers to save the correction as a permanent rule in CLAUDE.md or a skill.

Terminal Browser

A real web browser beside the terminal chat, for reviewing pages and pull requests without switching windows.

Replay Theater

After a turn, steps through every file edit Claude just made, for code or for a draft.

Anthropic

Start with Blast Radius and Reflect.

Changelog

October 16, 2026: first version.
October 27, 2026: rebuilt as a cheat sheet: capabilities, power-user tools and parallel work added; background explainers cut.
October 28, 2026: Planning the Work added, from two videos and the EGM README.
October 30, 2026: Skill Shelf added, with BAMA and /generate, then redesigned as cards, plus Cursor’s thermo-nuclear code review.

Sources

Checked against Anthropic’s own docs and help pages where they cover it. The mods, “folders, not agents” and the Elephant-Goldfish walkthrough come from videos and are their claims, with the EGM command names and the not-an-official-product note checked against the repo’s README, which credits Dave Rensin’s essay for the idea. The skill shelf’s speed and price figures are the videos’ own, from October 2026; I checked BAMA’s repo exists and is free, read the code review’s SKILL.md, and haven’t run any of them yet.

Docs01 / 14