Radar, November 9, 2026: ten signals from X on Claude Code secrets, Codex plans and OpenClaw
In short: The standout signal of November 9: according to Apiiro, public Claude Code settings files contain hundreds of secrets, because 'don't ask again' stores the whole command. Alongside it, ten short signals on subscriptions, coding agents, an OpenClaw patch release and open models, each with its source on X.

Illustration: AI-generated (OpenAI image model via Codex): a radar screen with ten signals.
1.1,122 public Claude Code settings files contain secrets
Confidence: medium (mechanism confirmed in documentation; figures via third parties)A summary of Apiiro research reports that 1,122 of 104,185 public copies of .claude/settings.local.json hold at least one secret. That file stores commands you approve with 'don't ask again', including any password in them. We could not read the research itself. Check whether the file is in your repo and rotate what was in it.
Source on X: @andyweb_3 · @TheHackersNews
Context in the briefing: Security in the November 9 briefing
Read also: Secrets in Claude Code settings · Claude Code 2.1.319 · Claude Code news
3.Karpathy builds 'TristanTube' with Claude Code in three minutes
Confidence: high (own post)Andrej Karpathy turned Tristan's favourite YouTube channels into a web app of his own with 47,000 videos and a TV mode. According to his post, it took three minutes with Claude Code. It shows how low the bar for a personal app has become; the app itself is not public.
Source on X: @karpathy
Context in the briefing: Labs in the November 9 briefing
Read also: Claude Code news
4.Codex: Pro plans without a 5-hour limit, says the Codex team
Confidence: medium (staff member on X; plans page not read)Tibo from the OpenAI Codex team says all Pro plans no longer have a 5-hour limit: Pro 100 gives 10 times, Pro 200 20 times and Pro 500 50 times the usage of Plus. He also polls whether a 'Pro 50' makes sense; that is a poll. Plan on the official plans page, not on a post.
Source on X: @thsottiaux · @thsottiaux (verdubbeling)
Context in the briefing: Subscriptions in the November 9 briefing
Read also: Codex news
5.Copilot app sessions open straight in VS Code
Confidence: high (own announcement)GitHub says a session from the Copilot app opens directly in VS Code, with task, plan, conversation and code changes. A task you started on your phone or in the app no longer needs explaining again in the editor.
Source on X: @github
Context in the briefing: Subscriptions in the November 9 briefing
Read also: Skills and plugins
6.Musk promises Grok Bot fixes: leaner on tokens
Confidence: medium (announcement; effect not yet measured)Elon Musk says fixes for Grok Bot target token efficiency and smarter behaviour, and that users should notice a 'dramatic improvement' on Monday. One user writes that he runs heavy tasks outside Grok Bot because usage runs out fast. Whether the fixes solve that remains to be seen.
Source on X: @elonmusk
Context in the briefing: Subscriptions in the November 9 briefing
7.OpenClaw 2026.11.2: a patch that says it does nothing new
Confidence: high (release notes read)The OpenClaw 2026.11.2 release notes open with 'This is a patch release with no new features.' The version fixes stuck updates, scheduled reminders that vanished when they contained the word heartbeat, and duplicate chat widgets, plus a series of permission and log fixes. Update for the fixes, not for a feature.
Source on X: @lcorneliussen
Read also: OpenClaw 2026.11.2 · MCP and agent tooling
8.Qwen4-27B: two release times on X, no confirmation from Alibaba
Confidence: low (no confirmation)Two accounts give different release times for Qwen4-27B, the open-weight model announced on September 22. Alibaba has not confirmed a date. Many overnight repackagings of older Qwen models circulate on Hugging Face; they are not a new release.
Source on X: @MartinSzerment
Context in the briefing: Open weights in the November 9 briefing
9.Kimi K3.1 Pro on Amazon Bedrock dates from November 6
Confidence: high for the date (AWS page)A post today presented Kimi K3.1 Pro on Amazon Bedrock as fresh news. According to the AWS what's-new page, the model has been there since November 6, with a 1 million token context. Useful if you want the model without running the weights yourself; the licence is not Apache.
Source on X: @zabracloud · AWS what's new
Context in the briefing: Open weights in the November 9 briefing
10.GPT-6.2 'Astra' remains a rumour
Confidence: low (rumour)On November 8 and 9 there was no OpenAI post about GPT-6.2. The only fresh trace is a community post about 'Astra' with a context length of over a million tokens that nobody has confirmed. A name in a benchmark table is not a release.
Source on X: @cslelouch
Context in the briefing: Unconfirmed in the November 9 briefing


