Claude Code (Anthropic)
Anthropic's agentic coding CLI that anchors this library's AI-coding highlights — vanilla setups, copy-paste recipes, plugins/skills, and prompting tips for Opus 4.5.
Claude Code is Anthropic's terminal-based agentic coding tool, and it recurs across this library more than any other single AI product — not as a subject of essays but as a set of practical recipes the reader flagged: install commands, copy-paste prompts, plugin names, and the official prompting guide for the Opus 4.5 model that powers it. The through-line of these highlights is a bet that the agent works well out of the box, that it can bootstrap its own configuration by reading a codebase, and that plugins and skills extend it into design, presentations, and self-testing tools. A single skeptical article sits alongside them as counterweight, insisting the whole category is "automation" wearing a smarter costume.
The creator's "vanilla" thesis
The anchor highlight comes from Boris Cherny, who introduces himself simply: "I'm Boris and I created Claude Code."1 Asked repeatedly how he uses it, his answer undercuts the entire cottage industry of customization tips: his setup is "surprisingly vanilla," because "Claude Code works great out of the box, so I personally don't customize it much," and "there is no one correct way."1 This is the load-bearing claim the rest of the collected recipes negotiate with — most of them add a little, none of them argue the defaults are broken.
The self-configuring agent
A recurring idea in the highlights is that the agent can set itself up by reading. Brian Lovin's flagged prompt tells Claude Code to "Read the README on this repository... Explore the code base to learn about best practices and patterns for using Claude Code effectively" and then "bring it back into the context" — the tool internalizing its own operating manual.2 This mirrors Anthropic's own prompting advice for the underlying model: when Opus 4.5 proposes changes without inspecting the code first, tell it "ALWAYS read and understand relevant files before proposing edits. Do not speculate about code you have not inspected."3 Reading-before-acting is the common denominator between the community recipe and the vendor guidance.
Recipes, plugins, and skills
Much of what the reader saved is executable — near-identical quickstart sequences that reflect how the tool is actually adopted. The pattern is consistent enough to tabulate:
| Source | The move | Command / skill |
|---|---|---|
| Cat Wu | Beautiful greenfield apps | /plugin install frontend-design@claude-code-plugins4 |
| Eric Buess | Full quickstart to Opus + plan mode | /model → opus, install frontend-design, shift+tab until plan mode5 |
| Nityesh | Prettier slide decks | forked Anthropic's pptx skill because default output "can be quite bland"6 |
| Alex Albert | Migrate an app to Opus 4.5 | official migration Claude Code plugin3 |
Eric Buess's recipe is the fullest expression: update, switch the model to Opus, add the marketplace, install the frontend-design plugin, tab into plan mode, then "Use the frontend-design skill to [describe what you want to build]."5 Cat Wu, who ships the same skill, frames its purpose plainly — "try the frontend design skill to make beautiful green field apps!"4 Nityesh's forked pptx skill shows the extensibility going the other direction: the primitives ship, users remix them toward their own aesthetic standards.6
flowchart TD
CC["Claude Code (agentic CLI)"]
CC --> M["Model: Opus 4.5<br/>effort parameter, vision, plan mode"]
CC --> P["Plugins & Skills"]
CC --> A["Agentic loop:<br/>read code → edit → self-test"]
P --> FD["frontend-design<br/>(Cat Wu, Eric Buess)"]
P --> PPTX["pptx skill, forked<br/>(Nityesh)"]
P --> MIG["Opus 4.5 migration plugin<br/>(Alex Albert)"]
A --> PW["Playwright self-testing<br/>(Simon Willison)"]
A --> README["Self-configure from README<br/>(Brian Lovin)"]
CC --> V["'Vanilla' default setup<br/>(Boris Cherny)"]
Self-testing agents and the single-file tool
Simon Willison's highlights place Claude Code in a concrete workflow: building small, dependency-light HTML tools. His decisive advantage for coding agents is that "Coding agents such as Claude Code and Codex CLI have the advantage that they can test the code themselves while they work on it using tools like Playwright."7 The surrounding constraints keep the agent on a short leash — "Avoid React, or anything with a build step... I prompt 'no react' and skip that whole rabbit hole entirely,"7 load dependencies from a CDN, and favor paste-transform-copy flows. The lesson embedded here is that agents produce more reliable output when the target is simple enough for them to verify unassisted.
Prompting the model underneath
The Opus 4.5 prompting guide the reader saved is effectively the operator's manual for the intelligence inside Claude Code. Its most notable shifts:
- The effort parameter — a single dial that "controls approximately how many [tokens] Claude will use for an output," letting you "trade off intelligence for cost/latency."3
- Dial back the shouting — because Opus 4.5 is more responsive to system prompts, aggressive language over-triggers tools; reduce "CRITICAL: You MUST use this tool" to just "Use this tool when..."3
- Stop over-engineering — the model "can at times overengineer and add extra files, abstractions"; counter with "Only make changes that are directly requested. Keep solutions simple and focused."3
These map neatly onto Cherny's vanilla ethos: the best results come from less prompting theater, not more.
Democratization — and the skeptic's asterisk
Two saved items frame the stakes. Every reports that "Two hundred people joined our Claude Code for Beginners course... Most had never written code. Eight hours later, they'd each deployed a working project"8 — the strongest evidence in the library that the tool lowers the floor for who can ship software, echoed by Dmitri Volkov's advice to watch a Claude Code masterclass "instead of watching a 2-hour movie."9
Against this enthusiasm the reader deliberately shelved a dissent. Linguist Emily Bender refuses even the term, calling these systems "stochastic parrots" — machinery "for haphazardly stitching together sequences of linguistic forms it has observed in its vast training data... without any reference to meaning."10 She insists there is "no view from nowhere," that models trained on society's mess "reinforce" rather than merely replicate its biases.10 Notably, Bender names Anthropic directly among the makers of these tools.10 Keeping her beside the recipes is the collection's own hedge: the workflows are real, and so is the caveat about what the thing underneath actually is.