Today's issue: Staff confirm turning off Sync Skills for Cloud Agents copied an old snapshot over a user's skills. Also: old Grok app messages replaying into Bots, and a pre-typed prompt link is coming.

Full issue: Oct 16, 2026

What's new

No new Grok Bot version is on the changelog; it still ends at 0.70.2 from Oct 11. Cursor's status page is All Systems Operational this morning, and last night's Grok model errors in Cursor were resolved at 1:20 AM ET. (confirmed)

Cursor's skill sync can overwrite your local skills folder (confirmed)

What happened: A Mac user reported at 6:15 PM ET Oct 15 that turning off Sync Skills for Cloud Agents copied a stale Oct 14 snapshot over ~/.cursor/skills and "deleted a day of local changes with no warning and no backup." At 4:28 AM ET, Cursor staff member Colin confirmed it and explained: "Today the setting moves your skills rather than mirroring them." He added: "There is no server-side copy of your Oct 15 changes."

Who is affected: Cursor IDE users who turned the setting on, especially if the first sync hung at "Syncing 50 of 50."

What to do now:

1. Before you try the setting again, back up the folder, as Colin suggests: cp -R ~/.cursor/skills ~/cursor-skills-backup.
2. Leave the setting off for now, as Colin advises.
3. If your folder was overwritten, open the Trash in Finder, turn on the Date Added column and sort by it. Colin says Cursor moved each replaced folder to the Trash "rather than deleting it outright," so it should be there unless the Trash was emptied.
4. Use Cancel, not the off switch, to stop a stuck run: Colin says "Cancel stops a run without removing anything."

What to watch: a fix (Colin flagged it to the team; no date), and whether skill names shared with ~/.claude/skills turn out to cause the stall, which staff suspect but haven't confirmed. (full story)

Old Grok app messages keep reaching some Bots (confirmed)

What happened: A user reported at 3:59 AM ET that an old grok.com message keeps arriving in their Bots "as if I’d just sent it," twice while they slept. Colin replied at 4:42 AM ET that these messages come through the link between the Grok app and Grok Bot, which "right now" shows them as if you typed them, and that "A common cause is a scheduled task in Grok that still points at your bots." It's a tracked issue.

Who is affected: anyone who has used grok.com or the Grok phone app with their Bots.

What to do now: open grok.com, go to Tasks, and delete any task that mentions Grok Bot or a Bot, archived ones included. Colin: "Don’t just archive it, since archived tasks can still run." Then check the Grok app on your phone too.

What to watch: a fix for the tracked issue; no date yet. (full story)

A link that opens a Bot with your message typed in is coming (confirmed, staff)

Colin said at 4:23 AM ET that "A new prompt link is coming in an upcoming Grok Bot update." It opens a chosen Bot with text already in the composer, in the form grokbot://app/v1/prompt?id=<bot-id>&text=..., plus a web version for browsers and email. "The message is filled in but never sent automatically," and the text must be one line of up to 1,000 characters. No date yet. (full story)

How people are using it

• Ordering lunch. Nick Sortor posted at 2:14 AM ET: "Shoutout to Grok @Bot for successfully ordering me Chicken McNuggets." He joked that he briefly thought Grok Bot had ordered 110 nuggets. This is one user's report of a completed purchase; we haven't seen the order or receipt. X showed about 24,000 views at 7 AM ET. (reported, one user; several accounts copied the text word for word, which we don't count)
• A build-and-test loop. @the_lord_liam described at 6:42 AM ET a setup where the Grok Build CLI writes programs and Grok Bot tests them: "Grok Bot reports bugs and suggests features; CLI Grok Build fixes and updates them." (reported, one user, small account)
• Start with a brain dump. Dylan Madden suggested at 1:27 AM ET: "dictate exactly what you’re working on to Grok @bot and asking how it can help you," and claim an email address for your Bot. Bot email addresses arrived in v0.70.2. (reported, tip)

Beyond Grok Bot

• Claude Code Review is out of preview for Team and Enterprise (read)
• OpenAI's Rosalind Academic Plan for U.S. university labs (read)
• ChatGPT Sites drops the fixed creation cap for Plus and Pro (read)
• Hark users say it still shops on Amazon after the agent block (read)

Opinion · Friday wrap

This week's column: turn Musk's purchase pledge into a ledger you can file against, plus the desks' build-next picks ranked. (full story)

The digest

• X plugin can read but not write for some accounts. Colin said at 4:37 AM ET that this is "a known issue we’re already tracking." Reads work, but some writes, "including creating bookmark folders and adding bookmarks, are currently refused by X for the connector." The "try again in a few minutes" message is misleading: "retrying, reconnecting, or reinstalling the X plugin won’t clear it." Make bookmarks in the X app for now; your Bot can still read them. (confirmed, staff)
• Why Stop sometimes seems to do nothing. In one case Colin traced at 4:27 AM ET, Stop went through but "took about 30 seconds to take effect," because a finished background helper still showed as running. It's a tracked issue. His advice: after Stop, "give it up to a minute before quitting the app." He also said a second Computer update prompt can appear after a successful one; check the Update row in Settings, and if it says you're on the latest version, you're done. (confirmed, staff; one account)
• New grok_ API keys still can't start My Machines workers. deanrie said at 2:19 AM ET that this is separate from the 503 bug fixed yesterday: the worker's key check "doesn’t recognize the new keys with the grok_ prefix yet." Workaround: unset CURSOR_API_KEY, run agent logout then agent login, and start the worker with browser login, which "refreshes itself." (confirmed, staff)
• Cursor CLI auto-update skips same-day builds. deanrie confirmed at 2:03 AM ET that background auto-update "only moves to a build with a later date," so a second build on the same day waits. Run agent update by hand, or agent about to see if a newer build exists. (confirmed, staff)
• Vulnerability Scanner findings missing from the Flagged list. Colin said at 4:44 AM ET the team confirmed scans run and record findings, but the Flagged vulnerabilities list "isn’t showing findings from the current version of the scanner." Add a Send to Slack action in the scanner's settings to get each finding posted; recorded findings "aren’t lost." (confirmed, staff)
• Handing a creator-run automation to a teammate. There's no one-step way, Colin said at 6:42 AM ET, because changing the owner changes whose accounts it runs with. The path: switch it off, remove MCP and Slack or Teams pieces, set Run as to Service account, use Change owner, then the new owner sets Run as back to themselves and re-adds connections. Webhook triggers need a fresh auth header afterward. (confirmed, staff)
• Turning on Cursor on-demand usage. Colin explained at 4:39 AM ET that the "On-Demand Spending: Disabled" row "is only a status line, not a switch." Use the Monthly Limit dropdown below it on cursor.com/dashboard under Spending, pick Fixed or Unlimited, and Save. On-demand only applies after included usage runs out. (confirmed, staff)
• Organizing work across several Bots. Answering a proposal for task cards, Colin said at 4:23 AM ET that making it "obvious what each Bot is doing and what’s waiting on you, is something the team is actively looking into. Keep an eye on upcoming releases." No details or date. (confirmed, staff; vague)

Anthropic

Opus 5.5 fast mode is out of preview, at double the price. Fast mode for Claude Opus 5.5 is "out of research preview," and "you no longer need to request access," per the Oct 15 release notes. Set speed: "fast" and the same model returns "up to 2.5x higher output tokens per second" at $8 and $40 per million input and output tokens, double standard Opus 5.5. No beta header is needed, and Start, Build and Scale tiers get a default fast mode rate limit. It also runs on Azure-hosted Microsoft Foundry, with Google Cloud "coming soon." It doesn't work with the Batch API, Priority Tier or Claude Platform on AWS, and switching speeds "invalidates the prompt cache." Claude Code fast mode is unchanged. Release notes · Fast mode docs · @ClaudeDevs

Claude Code 2.1.301: a cap on unattended waits and a surprise-credits fix. Claude Code 2.1.301 reached npm at 10:34 AM ET on Oct 15. A new setting, CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS, makes unattended retry mode end a request when a usage limit "resets too far away instead of waiting." OpenTelemetry tool_decision events now carry classifier_verdict and permission_mode, so you can tell auto mode classifier decisions from permission rules. One fix matters for your bill: switching to a model with a smaller context window "could spend usage credits on a summary by the previous model without asking first." VS Code also gets LaTeX math rendering. Update with npm i -g @anthropic-ai/[email protected]. CHANGELOG

Compliance API can now read members' saved Claude instructions. Enterprise compliance teams can now pull the text each member saved as Instructions for Claude in claude.ai and as Global instructions in Cowork. It's in beta for Claude Enterprise organizations. Call GET /v1/compliance/organizations/{org_uuid}/user_preferences with your existing Compliance Access Key and the read:compliance_user_data scope. It follows Oct 8 betas that added chats from the unified Claude experience and Word downloads of Claude Docs documents to the same API. Release notes, Oct 15

Two new Anthropic fellowships for scientists and humanities experts close Oct 28. Anthropic opened applications for STEM Fellows (science and engineering) and Conceptual Reasoning Fellows (philosophy, cognitive science, economics, policy and law), who will work "to help Claude reason well about some of the hardest problems in those fields." The listings say the program runs full-time from January 12 to May 8, 2027, applications are open until October 28, and offers are planned by November 20. The default location is San Francisco, with remote work in the US considered "in certain circumstances," a one-week in-person orientation, and no visa sponsorship. @AnthropicAI · STEM Fellow · Conceptual Reasoning Fellow

Anthropic puts $5 million toward mathematicians working out AI. Anthropic says it has "committed $5 million of funding, compute, and Claude access to help the math community navigate the massive changes that AI brings," with the Simons Foundation managing the award. The foundation says it, Anthropic and OpenAI "will commit $15 million in combined funding and resources," awarded partly through an open call for proposals, for workshops and initiatives that help researchers evaluate AI-assisted results, learn new tools and "shape collective standards for AI use." Anthropic and OpenAI will also give AI tools and compute to Simons-supported researchers, including in basic science. @AnthropicAI · Simons Foundation

OpenAI

Compliance API Costs data still delayed; status now Monitoring. OpenAI's status page shows "Delayed Costs data in the Compliance API," ongoing about 1 day at our morning check, now in Monitoring. "We have deployed a fix and are monitoring recovery. Costs data in the Compliance API remains delayed while we process the backlog. We will provide further updates as processing catches up." Our suggestion: if you pull workspace costs through the Compliance API, hold cost reports until the backlog clears. (confirmed) OpenAI status

Day 11: staff say capacity errors should be rarer and auto-resume. OpenAI's Tibo (@thsottiaux) posted "Day 11/" at 11:11 AM ET on Oct 15: the team "landed a series of changes and more compute to very significantly decrease the occurrence" of "Selected model is at capacity. Please try a different model." When it still happens, he says it "should also eventually just continue by itself and not just block on it." He asks users to "Let us know if you still see it out there in the wild and we'll take a look." (confirmed, staff) Day 11

Native Decisions integration in GPT-Live API (Oct 15). The Platform changelog for Oct 15 says OpenAI "Added native Decisions integration in GPT-Live." Apps can "act while your user is speaking or react to what Live says," with Live calling Decisions "directly through native delegation," or triggering decisions "as input or output transcript deltas arrive." The Decisions API itself shipped earlier (Oct 6 beta); this entry wires it into GPT-Live sessions (gpt-live-1, gpt-6-luna, v1/live/sessions). If you build voice agents on Live, read the changelog entry and the Decisions docs before changing session handlers. (confirmed) Platform changelog

Coming next week

What to watch from Monday, Oct 19 through Friday, Oct 23. Confirmed means a schedule, doc or staff post. Unverified means a hint without a date.

On the calendar

• Monday, Oct 20, 5:00 to 6:00 PM: "Grok Bot for Financial Services: Build your own team of SI analysts," with Hugo Charré. Confirmed, times as listed on Cursor Workshops.
• Tuesday, Oct 21, 5:00 to 5:30 PM: "Grok Bot for Admins," with Jon Eide Johnsen. Confirmed, Cursor Workshops.
• Wednesday, Oct 22, 5:00 to 6:00 PM: "SpaceXSI x Bain: What boardrooms are asking about the SI SDLC," with Mike Saeks. Confirmed, Cursor Workshops.
• Later the same page lists SpaceXSI for Higher Education (Oct 28), Grok Bot for Enterprise Teams (Oct 29) and What's New in Grok Bot (Oct 30, 5:00 to 6:00 PM).

Confirmed deadlines and staff teasers

• Thursday, Oct 23: OpenAI API model shutdowns, including gpt-4-turbo, o1 and o3-mini (deprecations). Confirmed.
• A Grok Bot prompt link that opens a Bot with your message typed in, without sending it, is "coming in an upcoming Grok Bot update," Colin said. Confirmed, staff; no date.

Early signals (unverified)

• Saved web passwords for Bots: Musk replied "Coming very soon." Unverified hint, not a date.
• Written terms or a claim form for Musk's purchase pledge. Nothing on Cursor's help or security pages as of Oct 16; watching only.
• A permanent fix for the shared routine webhook rate limit Colin called a known issue (forum). No date.

Read every story, with sources, at The Daily Bot. Confirmed means we checked a primary source; reported means a credible claim we haven't verified.