Young Tries AI

Everything on Claude Code

26 mentions, newest first.

  1. Claude Code v2.1.326 allows Code Mode with agent teams enabled

    Code Mode can now be turned on while agent teams are enabled, and sessions with teams on start in Code Mode if you already turned it on. /status gets a row showing when a provider or an environment variable such as DISABLE_TELEMETRY turns features off. A failed MCP headersHelper no longer makes other sessions start without that server.

    Usable today
  2. Claude Code team shares how it verifies agent-written code

    In a post discussed on Hacker News, the Claude Code team says about 90% of its code is written and reviewed by Claude. In internal testing on structured file generation, a separate grader checking Claude's work against a rubric improved task success by up to 10 points over a standard prompting loop.

    Usable today
  3. Reddit test: a 1h+ break makes a long Claude Code chat cost 4x a fresh one

    A Reddit user measured each "save tokens" trick on a Pro account, using the same prompt and file, 3 runs each, with numbers from Claude Code's JSON output (Sonnet unless noted). A fresh chat cost $0.038 per message. A long chat (~65k context) with no break cost $0.007 to $0.015. After a 1h+ break it cost $0.157.

    Usable today
  4. Free skill rewrites text in controlled English for Claude Code, Codex and Cursor

    A Reddit user released a free, open-source skill for Claude Code, which also works with Codex and Cursor. It is based on a tip to ask LLMs to write 80% of the way to ASD-STE100, the controlled English of aircraft maintenance manuals. It works in two passes: remove the 26 AI tells from Wikipedia's list, then rebuild the text in STE style.

    Usable today
  5. Openbar shows Claude Code and Codex permission prompts in the MacBook notch

    A Reddit user built Openbar after losing time to parallel Claude Code and Codex sessions waiting on permission prompts in tabs they weren't watching. Every session shows in the MacBook notch, or an island on Windows and Linux, where the user can read the diff and allow it. Each session can get its own git worktree. It is free and open source.

    Usable today
  6. Claude Code projects enter beta with parallel cloud threads

    Redesigned Claude projects are in beta in Claude Code for select Pro and Max subscribers. You set a goal plus a repo or context, and Claude scopes the request, delegates to parallel threads, reviews the outputs and assembles the result. Each thread is a cloud session on its own branch and copy of the repo.

    Partly usable
  7. Claude Code v2.1.325 patches a zstd security issue and subagent model bugs

    Claude Code v2.1.325 fixes a security issue in how zstd-compressed responses are decoded. It also fixes subagents with their own model being moved to the plan-mode model when the model setting is opusplan or haiku, and sessions on many third-party gateways and Foundry stopping with a context length error on every turn.

    Usable today
  8. Claude Code v2.1.323 adds an experimental Code Mode and plugin version limits

    Code Mode is an experimental opt-in in which Claude writes JavaScript to call its tools. Plugins can now declare a Claude Code version range in plugin.json, and claude plugin validate type-checks a plugin's TypeScript hooks module. Fixes include /compact summarizing away an unanswered message.

    Usable today
  9. A Claude Code skill draws pictures stroke by stroke with no image model

    A Reddit user reports a Claude Code skill called drawing. Put a picture in a folder as subject.png and ask Claude to draw it. Claude types every stroke as coordinates in a Python script and a headless tldraw canvas renders them. The demo replays about 350 strokes for one house drawing.

    Partly usable
  10. Klad, a 2D painting app that Claude Code can paint in, ships as source

    According to the Hacker News post, Klad is a 2D painting and animation app shared with a coding agent (Claude Code, Codex, or anything that can run commands). The agent paints through the klad command line with the same brushes while you paint with a pen. It has no image model.

    Partly usable
  11. Modern Web Guidance gives coding agents current web platform practices

    Modern Web Guidance is a set of skills that puts web platform best practices and browser compatibility data into coding agents, steering them away from legacy patterns. It is stewarded by the Google Chrome, Microsoft Edge, Mozilla Firefox and MDN teams and covers 203 real-world developer use cases.

    Usable today
  12. A Reddit user builds a Claude Code mod that maps agent activity live

    A Reddit user built a Claude Code mod that shows the agent's work as a live map in a pane beside the chat. The agent sits in the centre with ten stations: you, model, memory, files, shell, network, browser, subagents, skills and hooks. Each model request, tool call and hook sends a pulse to its station and back.

    Partly usable
  13. Claude Code adds Fable 5.2 as the default Fable model with 1M context

    Claude Code v2.1.321 added Claude Fable 5.2 (claude-fable-5-2) as the default Fable model, with 1M context and $8/$40 per Mtok. TechCrunch reports responses are 30% faster across effort levels. Artificial Analysis scores it 65 on its Intelligence Index (median 40), at 46.7 minutes per task versus a 13.0-minute median.

    Usable today
  14. Apiiro finds thousands of secrets in Claude Code settings files on GitHub

    Apiiro researchers found thousands of secrets, including database passwords, API tokens and root logins, in Claude Code settings files in GitHub repositories. Choosing the don't-ask-again option stores the command in .claude/settings.local.json, often as written. Apiiro found over 100,000 cases of the file on public GitHub; about 1,100 from 63 organizations contained secrets.

    Not yet
  15. Blogger runs Claude Code and Codex from a markdown ticket queue in git

    A blogger describes running Claude Code or Codex work as a priority queue, with every ticket a markdown file in the git repo. An agent researches new tickets, builds those a human tagged safe, runs tests and a browser check, and never deploys to production. Items waiting on the human collect in an owner inbox on one Claude artifact page rebuilt from git each cycle.

    Usable today
  16. Blogger retires Claude Code for Pi, citing VSCode limits and prompt changes

    A blogger says they are retiring Claude Code as their main harness for Pi. The author says only one Claude Code session can connect to the VSCode IDE at a time, and that the VSCode GUI extension lacks half of the built-in slash commands and custom status line support. The author counts 244 Claude Code releases this year, over 3/4 of which changed the system prompt or tool descriptions.

    Partly usable
  17. Claude Code v2.1.319 changes auto mode and adds MCP background tasks

    With telemetry or feature flags off, including behind LLM gateways, auto mode now asks about or denies calls the server can't check, with no fallback to the built-in classifier. Bedrock, Vertex and Foundry are unchanged. MCP tasks are supported in the terminal and move to the background. A new permissions.denyInsteadOfPrompting setting denies calls that would wait on a prompt.

    Usable today
  18. GitHub issue: Claude Code put a user's email in its search User-Agent

    A GitHub issue on Claude Code 2.1.318 (macOS) reports that the tool put the user's email address in the User-Agent header during web searches. A reply says a built-in mod called cc-plugin-privacy, introduced in v2.1.317 and later, deterministically blocks this. It is enabled by default and can be turned off via /plugins.

    Usable today
  19. Stroq audits Claude Code, Codex and Cursor sessions for leaked keys

    Stroq reads your Claude Code, Codex and Cursor sessions from the last 30 days and shows which commands came from text the agent read and which keys left the machine. It prints key names, never values, sends nothing, and needs no account. It is Apache-2.0 and needs Node.js 22 or newer.

    Usable today
  20. Reddit user says background agents in Claude Code found the AirPods 3 protocol

    A Reddit user reports using Claude Code to work out how AirPods 3 send head tracking data, to control a mouse on Linux. The user says it took a lot of failed tries and that AirPods 3 use a different protocol than the Pro models.

    Usable today
  21. Claude Enterprise adds a security center, inference hooks and new admin controls

    A security center page in Organization settings compares settings for chat, Cowork and Claude Code with Anthropic's recommended configuration. Prompt hooks and pre-tool-call hooks are now generally available and send prompts and tool calls to data security tools. A parent organization gets its own admin console at admin.claude.com.

    Partly usable
  22. Claude Code v2.1.318 fixes --continue history loss and adds login expiry dates

    The release fixes --continue sometimes losing the earlier conversation after a malformed tool call's retries ran out, a regression in 2.1.314. It also fixes macOS stalls with many running Bash commands. claude auth status now shows when a stored claude.ai sign-in ends.

    Usable today
  23. GitHub issue claims Postman desktop installed plugins into Claude Code and Codex

    A user-filed GitHub issue reports that after opening Postman desktop 12.33.4 on a Mac, the reporter saw recurring MCP authentication warnings and Codex routing API work through Postman skills. The reporter cites a main.log entry from 2026-10-24 showing claude-code:installed and codex:installed.

    Not yet
  24. Claude Code v2.1.317 lets failing hook callbacks block actions

    Agent SDK hook callbacks get onFailure: "block", so a callback that fails blocks the action instead of letting it run. /schedule now warns when a new cloud scheduled task gets connectors you did not ask for. A new check asks before a web or shell tool call sends your account email address to the network.

    Usable today
  25. Claude Code on Opus 5.5 now steers away from cloning specific games

    Ars Technica reports Anthropic added a rule to the Claude Code prompt on or around October 27: build original games, games in a genre and games from the user's own designs and assets, not a recreation of a specific existing game, even if renamed. XDA independently verified the rule in Claude Code version 2.1.314, released October 30. Anthropic software engineer Addy Osmani confirmed on social media that two sentences were added to the prompt on Opus 5.5.

    Usable today
  26. Claude message threads enter public beta on the Messages API

    Message threads is in public beta on the Claude Platform through a thread parameter. The first request sends the full conversation; later requests send only new messages and the ID of the previous response. For requests of 500k to 1M tokens, median time to first token drops by about 35%.

    Usable today