
Same-day release notes for Claude Code, Codex, Grok Build, Cursor and open models: what changes for you, with the source link. By @KHASSINX
- Tweets15.1K
- Following39
- Followers536
- Likes95

ALT How-to card: Sandbox your Remote Control sessions. Step 1: start the server with its own settings file: claude remote-control --settings rc.json; needs Claude Code 2.1.321 or later. Step 2: in rc.json set "sandbox": { "enabled": true }; it applies to every session the server starts. Step 3: limits: put --settings after remote-control, not before, or Claude Code refuses to start; --sandbox was removed in v2.1.300; after editing the file, restart the server. Source: Claude Code docs, Remote Control, and changelog 2.1.321.

ALT How-to card: Stop Claude Code from waiting on prompts nobody will answer. Step 1: in ~/.claude/settings.json add "permissions": { "denyInsteadOfPrompting": true }; needs Claude Code 2.1.319 or later. Step 2: what still runs: calls approved by an allow rule, your permission mode, or auto mode's classifier; everything else that would prompt is denied and the turn continues. Step 3: limits: user, managed or --settings only, a repo can't set it; it also removes AskUserQuestion and the plan-mode tools. Source: Claude Code docs, Settings reference.
![How-to card: Keep .env out of Claude Code. Rules run deny, then ask, then allow; the first match decides. Step 1: in settings.json add "permissions": { "deny": [ "Read(./.env)", "Read(./.env.*)" ] }. Step 2: run /status to confirm the file loaded; strict JSON, no comments or trailing commas. Step 3: limits: .claudeignore has no effect, and a script that opens files itself isn't covered, so use the sandbox. Source: Claude Code docs, Settings and Permissions.](https://pbs.twimg.com/media/HWzl77QXUAAGOBw.jpg?name=small&format=webp)
ALT How-to card: Keep .env out of Claude Code. Rules run deny, then ask, then allow; the first match decides. Step 1: in settings.json add "permissions": { "deny": [ "Read(./.env)", "Read(./.env.*)" ] }. Step 2: run /status to confirm the file loaded; strict JSON, no comments or trailing commas. Step 3: limits: .claudeignore has no effect, and a script that opens files itself isn't covered, so use the sandbox. Source: Claude Code docs, Settings and Permissions.

ALT Card: Codex CLI 0.167 release, Nov 5. Headline: A command center for your agents, in the terminal. Code panel: codex agents --no-daemon (unread results marked); switch backends in the session picker with drafts kept; local image preview while composing; GCP Vertex via gcp_setup_wizard under Sign in another way. Source: github.com/openai/codex/releases/tag/rust-v0.167.0.

ALT How-to card: Give your agent Xcode. Three steps from Apple's docs: 1) In Xcode Settings, Intelligence, allow external agents Always or While Xcode is Open. 2) Add the bridge: claude mcp add --transport stdio xcode -- xcrun mcpbridge, or codex mcp add xcode -- xcrun mcpbridge. 3) Open the project in Xcode first, then prompt; check with claude mcp list or codex mcp list. Source: Apple Developer Documentation.

ALT Card for Claude Fable 5.2: headline says 20% cheaper and up to 30% faster. Model id claude-fable-5-2, 1M context, 128k output. Panel lists $8/$40 per million tokens (down from $10/$50), fast mode with speed "fast" at up to 2.5x for $16/$80, and a warning that editing the prompt prefix and resending 5.2 thinking blocks returns a 400. Source: Anthropic, Nov 9, 2026.

ALT Card: Kimi K3.1 Pro by Moonshot AI. Headline: The weights are now public on Hugging Face. Mixture-of-experts, 2.8T total parameters with 104B active, 1M-token context; free license, API-hosting businesses over $20M a year need a separate deal. Panel: repo moonshotai/Kimi-K3.1-Pro, 896 experts with 16 per token, vLLM recipe and SGLang cookbook. Source: huggingface.co/moonshotai/Kimi-K3.1-Pro.

ALT Card: Claude Code 2.1.318 release, Nov 3. Headline: It now tells you when your stored sign-in expires. Code panel: claude auth status shows loginExpiresAt; mods get an effort option on $.model.fork for low-effort side jobs; cloud sessions can start with No repository. Source: github.com/anthropics/claude-code/releases/tag/v2.1.318.

ALT Dark AK× card. Kicker: What you get. Mark AK×, AI / dev-tools · release notes. Headline: Same-day release notes. What changes for you. Note: Claude Code, Codex, Grok Build, Cursor, and open models — with the link. Right panel, Coverage, daily + Sunday: 01 Claude Code · Codex · Grok Build · Cursor; 02 Open-model drops, with source links; 03 Sunday recap of the week; 04 How to turn it on, step by step. Footer: By @KHASSINX.

ALT Card titled Message threads, Claude Messages API public beta. Left: send the conversation once, then only what's new; the API holds the thread in memory, encrypted, for 5 minutes; median time to first token drops about 35% at 500K to 1M tokens; token usage stays the same. Right: code showing the beta header and thread create and continue fields.

ALT Card: DeepSeek dsh-superagent, open source MIT research prototype. Headline: A coding agent for the whole repo, not one person. Note: it listens to GitHub; when an allowed teammate asks, it opens a work item, works in its own session and answers in the thread, for example with a pull request. Right: terminal with the README setup prompt, Read SETUP.md in this repository and set up dsh-superagent with me, plus notes that it runs on DeepSeek Harness, has a web panel and acts only when asked or a team rule allows. Footer cites github.com/deepseek-ai/dsh-superagent-prototype.

ALT Card for Claude Code 2.1.317: headline says it now asks before your account email leaves the machine; code panel lists onFailure: "block" for Agent SDK hook callbacks, Ctrl+C no longer stopping background agents, and Read returning content again for unchanged files. Source: Claude Code v2.1.317 release notes on GitHub.

ALT Card: SemiAnalysis GDPx benchmark, Nov 2. Left: kicker "Benchmark · Nov 2", title "GDPx", meta "SemiAnalysis · 206 tasks · 11 domains", headline "The best model solves 12%." and a note on domains, 8-hour tasks, native harnesses, and finance and legal scoring nearly 0. Right: leaderboard with Claude Opus 5.5 12.0, Gemini 4 Carbon 10.9, GPT-6 Astra 9.6, Muse Wonder Full 9.0, Claude Fable 5.1 8.5, GPT 6.1 Sol 8.4 and Grok 4.7 3.9. Footer cites gdpx.semianalysis.com.
![Premium how-to card: Codex CLI profiles. Left: kicker “How to · Codex CLI”, product “Profiles”, meta “--profile · layered config”, headline “One flag. Whole setup switches.” Right: terminal example “codex --profile deep-review” and a deep-review.config.toml snippet with model, reasoning effort, and approval_policy. Cap notes that since 0.134, [profiles.x] tables are ignored. Footer cites developers.openai.com/codex/config-advanced.](https://pbs.twimg.com/media/HWJhEb2WwAA2Uqp.jpg?name=small&format=webp)
ALT Premium how-to card: Codex CLI profiles. Left: kicker “How to · Codex CLI”, product “Profiles”, meta “--profile · layered config”, headline “One flag. Whole setup switches.” Right: terminal example “codex --profile deep-review” and a deep-review.config.toml snippet with model, reasoning effort, and approval_policy. Cap notes that since 0.134, [profiles.x] tables are ignored. Footer cites developers.openai.com/codex/config-advanced.
