Claude Code

Oct 27

Roast Room, an MIT-licensed repo with about 1,200 stars, turns Claude Code into a chair that runs seven simulated advisors through a debate and a scored verdict. Here is how it works, how to run it, what it costs in plan usage, and where the idea breaks down.

Oct 26

Oct 25

Oct 24

Cindy is an open-source agent client that sits on top of Claude Code, Codex and Pi, carrying memory, skills and workspace across them. It gained about 470 GitHub stars in a day. Here is what the README promises, what it costs, what it collects, and what to check before giving it your real files.

Oct 24

imozuru, a new Claude Code skill from developer MaryCache, gained 1,173 GitHub stars in about eleven days. It maps an unfamiliar field, follows citations to original sources, saves them verbatim and builds a page of checked extracts instead of a summary. Here is how it works, how to install it, and what to watch for.

Oct 22

Anthropic Fellows and researchers attacked Claude Code Auto Mode and OpenAI Codex Guardian with an Opus 5 agent. Production monitors missed more than half the attacks. The new Auto Mode ++ design cuts attack success by over 50 points, but multi-context attacks remain open.

Oct 22

A post on X on October 22, 2026 said Claude Code exposed personal data in HTTP headers because sites it fetched asked agents to identify their users. GitHub issues back up part of that: agents have copied the signed-in email into a header sent to Liquipedia and Wikipedia. Here is what is confirmed, who is at risk, how attackers could abuse it, and how to stop it in five minutes.

Oct 22

luongnv89/claude-howto gained 428 stars in a day and sits near 42,600 total. It is a ten-module, example-driven guide to Claude Code with copy-paste templates for commands, memory, skills, subagents, MCP, hooks and plugins. Here is what it contains, how to use it safely, and what to check before you trust it.

Oct 20

Simon Willison published Battle Drones, a browser real-time strategy game built over 12 iterations and 321 commits, mostly by coding agents. Two phones can play each other by sending sound. Here is how it was made, the engineering tricks Claude came up with, and what the author admits he has not reviewed.

Oct 20

Oct 20

Oct 20

earshot, a trending GitHub repo with about 1,500 stars, turns audio into evidence a coding agent can reason about: descriptions, match scores, and a real-versus-synthesized flag, all on CPU. Here is how to install it, what it claims, and where its own README says it falls short.

Oct 20

A new MIT-licensed GitHub project, Free My TV, swaps the sponsored Google TV home screen for a clean shelf of your own apps. It was built with Claude Code and gained about 1,500 stars in days. Here is how it works, how to install it, and the risks the README itself lists.

Oct 20

A GitHub repo with about 1,300 stars publishes fifteen ten-second motion-design films, each produced by Claude Opus 5.5 writing code from a single prompt, plus the prompts, sources, and a headless-Chrome renderer. Here is what is inside, how to reuse it, and what does not reproduce.

Oct 19

A new GitHub repo, tv-vibehack, hands Claude Code a prompt and a set of ADB tools to strip ads and telemetry from an Android TV and install a clean launcher. Here is how it works, what it touches, and what to check before you point an agent at your TV.

Oct 18

Folkbench is an open-source project that publishes rankings of third-party AI API relay stations and ships Folkbench Switch, a desktop app that rewrites the endpoint and key in Claude Code, Codex, Gemini CLI and others. It gained about 1,664 stars in two weeks. Here is how it works, what the rankings show, and the terms-of-service and data risks the README does not dwell on.

Oct 18

Oct 17

Two Claude Code releases in two days change defaults that affect cost and safety: a setting for the auto mode classifier model, an idle compaction switch, fork subagents on by default in more places, and a cap on concurrent hooks. Here is what the release notes say and what to check.

Oct 16

Anthropic published how the Claude Code team reviews code when Claude writes about 90% of it. Merged PRs more than doubled from June to September while reverts fell by more than 40%. Here is the process and how to copy it.

Oct 15

Oct 15

Anthropic is pitching the Claude Code desktop app as a workbench for many agents at once. Here is what the docs confirm about parallel sessions, Git worktree isolation, the status sidebar and cross-session messaging, and what we could not confirm.

Oct 15

Christian Schaller describes how Claude Code, a KVM Windows VM and usbmon traces produced a working Linux driver for a film scanner and patches for a ThinkPad MIPI camera. The method is a diff between a known-good and a failing capture.

Oct 15

ECC, the agent harness optimization system from affaan-m, gained over 1,000 stars in a day and sits near 280,000 in total. It bundles 71 subagents, 311 skills, hooks, instincts and a memory vault for Claude Code, Codex and others. This guide covers what is in it, how to install it with the least blast radius, and what the README admits does not work yet.

Oct 15

Ruflo, the project formerly known as Claude Flow, is an agent "meta-harness" around Claude Code and Codex with 74,800 GitHub stars. It promises swarms, vector memory, federation and hundreds of MCP tools. This guide shows how to install it, separates the README's vendor-reported numbers from what you can check, and lists the limitations the project states itself.

Oct 13

free-claude-code is a GitHub trending repo that runs a local proxy so Claude Code, Codex, OpenCode and others talk to 61 other providers. It does not give you free Claude. Here is how it works, the AGPL license, and the plain terms-of-service and data risks.

Oct 13

pstack-claude, a GitHub repo gaining about 393 stars this week, translates a Cursor skill stack into plugins for Claude Code, Codex, GitHub Copilot, Pi, OpenCode and Gemini CLI. Here is what poteto-mode does, how to install it, and what to check before you trust it.

Oct 12

LightOn's ColGrep-Agent-2B is an open 2-billion-parameter model that runs locally, searches your repository with semantic ColGrep, and hands file paths and line ranges to Claude Code, Codex or OpenCode. In the authors' tests it cut API tokens by 34% but made questions slower.

Oct 12

Lemo-Opuscar, a new open-source Claude Code skill, makes short films without any video-generation model: 43 style prompts, each backed by a demo film that Claude Opus 5.5 drew, scored and mixed entirely in code. Here is how it works, what it costs in time and tokens, and where it falls short.

Oct 12

Linus Torvalds published a branch called noref-filemap-read-with-llm and wrote that the code, test programs and review were done with Claude Code under his direction. Here is what the patch does, the numbers reported so far, and what nobody has checked yet.

Oct 11

Anatomy, a new MIT-licensed Claude Code skill by wheresryan22, passed 2,100 GitHub stars in about three days. You name an idea, and Claude invents a machine whose moving parts behave like it, then builds it as an interactive SVG, WebGL or 3D figure. Here is how to install it, what it costs in time, and where it breaks.

Oct 9

When an agent hits a step only a human may do, it usually prints "please click Allow" in a terminal you are not watching. bigarrow, a small open-source macOS CLI with an agent skill, lets it point at the button instead. Here is how it works, how to install it, and what it will not do.

Oct 9

An independent researcher says Claude Code helped him find two transit-like signals around a nearby K dwarf in NASA TESS data. TESS has approved follow-up observations. The signal is a candidate, not a confirmed planet. This post separates the claim, the checks, and the open risks.

Oct 7

On October 6, 2026, developer Zohaib Ansari argued that the suggested next message Claude Code now pre-fills in the prompt box is really a cheap way to collect preference data. The post reached the Hacker News front page with 190 points, and an Anthropic engineer replied that suggestions are not used to collect preference signals. Here is the argument, the strongest objections, what is and is not verifiable, and what to do if you want it off.

Oct 7

On October 7, 2026, Anthropic's Lydia Hallie announced that you can now ask Claude Code to run subagents at a specific effort level, on version 2.1.292 or later. Her example: find every payments API call with low-effort subagents, then have a high-effort one check the error handling. Here is what that enables, how to use it well, what we could not verify, and the cost trade-offs.

Oct 4

On October 3, 2026 an AI systems engineer wrote that the terminal is the wrong interface for coding agents because tabs are ephemeral while context is persistent. The post passed 800K views and drew a reply from Claude Code's creator. We unpack the claim, the best counterarguments, and a practical way to choose between terminal, desktop app and orchestrator.

Oct 3

Anthropic opened Claude Code to mods on October 1, 2026. Within days we published awesome-claude-code-mods: 50 independent, MIT-licensed mods for session diagnostics, Git, repository viewing, workspace notes, utilities and workflow control. Here is how to install them, which ones are worth trying first, and what they can and cannot do.

Oct 3

Mods work in the Claude Code desktop app's Code tab as well as the terminal, but the steps differ: a plugin browser instead of slash commands, a gap for local folders, and desktop-only elements like SVG. This guide covers installing, having Claude build a mod, loading your own folder, and building a tested usage-ring mod for the desktop.

Oct 3

Claude Code mods are plugins of function hooks that can draw UI, add commands and react to tool calls. This tutorial builds one from scratch, a Touched Files mod that shows how many files Claude has edited, and shows real screenshots of it running, plus the validate and test output.

Oct 2

On October 1, 2026 Addy Osmani published Anthropic's official getting-started guide for Claude Code mods on claude.dev. This is the GA how-to: version 2.1.287+, JavaScript or TypeScript modules inside plugins, observe/rewrite/answer hooks, the $ API, Token Weather, Blast Radius, Replay Theater, and how you share a mod with /plugin.

Sep 29

On September 28, 2026, Lance Martin published Anthropic’s playbook for eval design and hillclimbing in the claude-api skill. /claude-api build-eval interviews you, samples production traces, and validates graders. /claude-api hillclimb then patches one change per round against a train/test split. One internal support benchmark cut cost to about one-fifth while lifting held-out accuracy from 78.6% to 90.5%.

Sep 27

Sep 27

Sep 26

Sep 24

Sep 24

Sep 20

Sep 19

Sep 18

Sep 17

Sep 15

Sep 15

Sep 15

Sep 12

Sep 11

Sep 10

Sep 10

Sep 5

Sep 4

Sep 4

Sep 3

Sep 3

Aug 30

Aug 29

Aug 29

Aug 27

Aug 26

Aug 26

Aug 25

Aug 25

Aug 23

Aug 22

Aug 22

Aug 22

Aug 22

Aug 22

Aug 22

Aug 22

Aug 21

Aug 21

Aug 20

Aug 20

Aug 20

Aug 20

Aug 18

Aug 18

Aug 18

Aug 14

Aug 14

Aug 13

Aug 11

Aug 9

Aug 8

Aug 8

Aug 8

Aug 8

Jul 31

Jul 20

Jul 17

Jul 17

Jul 16

Jul 16

Jul 16

Jul 15

Jul 13

Jul 12

Jul 9

Jul 7

Jul 7

Jul 7

Jul 5

Jul 4

Jul 4

Jul 2

Jul 2

Jul 1

Jun 29

Jun 29

Jun 28

Jun 28

Jun 28

Jun 27

Jun 27

Jun 27

Jun 27

Jun 25

Jun 25

Jun 24

Jun 23

Jun 22

Jun 21

Jun 21

Jun 19

Jun 17

Jun 14

Jun 12

Jun 12

Jun 12

Jun 12

Jun 12

Jun 12

Jun 12

Jun 12

Jun 12

Jun 12

Jun 11

Jun 11

Jun 11

Jun 9

Jun 6

May 29

May 28

May 27

May 15

May 12

May 11

May 10

May 7

May 6

Apr 28

Apr 25

Apr 24

Apr 13