Anthropic's Claude for Nonprofits plugin maps your records, then drafts grants, hiring and board work. Plus how to switch Codex and Work to GPT-6.1 Terra.

The Daily Bot

No. 019 · Tuesday, Oct 27, 2026 · Read on the site

If your nonprofit is on a Claude Team or Enterprise plan, today's one job is to build the records map in the new Claude for Nonprofits plugin, since Anthropic says every other skill reads it first. Also inside: Grok Bot's X connector no longer needs a developer account, how to switch Codex and Work to GPT-6.1 Terra, OpenAI's claim that its model disproved the Exponential Time Hypothesis, Anthropic's S-1, Cursor's Agent SDK and Cmd+K retirement, and three setups people say they run.

In this issue: Claude for Nonprofits · What's new (3) · Since our last email (6) · How people use it (1) · Workflows (3) · Digest (10) · Known issues (5) · Coming up

Today's lead

Claude for Nonprofits plugin adds 14 skills for grants and hiring (confirmed)

What happened: On Oct 26, Anthropic released the Claude for Nonprofits plugin, which its post calls "a set of skills and connectors built with ~30 nonprofits." Anthropic's nonprofits page says: "Its 14 skills cover four areas." They are knowing your organization (records map, voice guide, AI policy answers), marketing, your team (hiring, onboarding, board recruitment) and fundraising. The post says: "Every other skill in the plugin reads the map first, so Claude knows where to look and you don't have to point it to the same source for each task." A /grant-reporting skill, per the tutorial, "takes each figure from your documents, cites the source, and shows its work for any figure it calculated."

Who's affected: nonprofits using Claude on Team or Enterprise plans. Availability is worded three ways: the post says Team and Enterprise plans, the tutorial says "They're installed automatically on most nonprofit plans," and the nonprofits page says it "comes pre-installed for every verified nonprofit." The nonprofits page lists registered 501(c)(3) groups and international equivalents, K-12 schools and certain mission-based healthcare providers. Admins need Cowork and Skills on: the help page says "Cowork and Skills must both be enabled for your organization before you can use plugin marketplaces."

Why it matters: The post says more than 25,000 nonprofits use Claude and "more than 75% of them have fewer than 10 Claude seats." Our read: year-end appeals and grant reports land on the same few people, and these skills are built for exactly that work, with the citations a funder will ask about.

What to do:

1. Check for the plugin in Claude. Anthropic's Academy tutorial says that if you don't see it, ask your admin, who can add it for your organization.
2. Turn on memory in Settings and run the task in the cloud, so the records map and voice guide are saved for later chats (per the Academy tutorial).
3. Build the records map first with the /where-things-live skill; Anthropic says every other skill reads the map first.
4. Build a voice guide with /org-voice-guide from three to five past materials, and check the sample before you approve it.
5. Run the fundraising skills (/funder-and-grant-research, /grant-proposal-writing, /grant-reporting, /donor-and-funder-relations) and check every cited figure against its source before you send.

Prompt (Anthropic's example, from its Claude Academy tutorial):

Can you find funders that fit [program name], and skip anyone who already funds us? I want a short list so I can decide who to approach first.

What to watch: two free webinars, per the post: an admin session on data handling and AI policy on Nov 10, and a staff session on Nov 12. Anthropic calls the plugin "the first of several releases that we will continue to build with nonprofit organizations." The tutorial's own warning: "Make sure every figure matches the source Claude cites, because Claude can make mistakes."

Read the full story

What's new

Grok Bot's changelog tops out at 0.75.8, dated Oct 24 and now stable and rolling out. Claude Code 2.1.311 reached npm Monday; its stable channel is still 2.1.308. Cursor's overnight repository-integrations incident is resolved, and OpenAI's Code Review and Security Review incident was still Monitoring at 6:09 AM ET (details in the digest). (confirmed)

Grok Bot's X connector no longer needs a developer account or credits (Grok Bot · confirmed)

Your Bot can now work through your X bookmarks and DMs without an X developer account or API credits. Santiago, whose X bio reads "building the everything api @X," posted at 7:55 PM ET Monday: "Last week, we removed the need for a developer account and X API credits to use the X connector in Grok Bot." He added: "Ask @bot to manage your bookmarks, go over your DMs, or draft an article for you free of additional charge." The official @XDevelopers account quoted it about 40 minutes later. To start, open Connect apps, add the X plugin and sign in with X, then try a read-only job.

Prompt (ours, untested):

Go through my last 100 X bookmarks, group them by topic, and list the five most worth reading with one line each. Don't remove anything.

Read the full story

How to switch Codex and Work to GPT-6.1 Terra, and when not to (OpenAI · confirmed)

OpenAI's models page now explains how to pick Terra: "GPT-6.1 Terra requires Codex CLI 0.159.0 or later. To use it, run codex -m gpt-6.1-terra." In the desktop app or Work on the web, "open Advanced and choose it from the model list." The page says to "Choose Terra for clear communication, interpreting images and screens, and completing tasks across supported apps," including "filling out forms from source documents." OpenAI's Victor Nunez wrote "i’d still go with Sol or Astra for coding tasks + using the Terminal instead of app interfaces." In his own test, "Terra took 53.5 seconds vs 3:21.7 for Sol" on one long web form; that's one staff demo, not a benchmark.

Prompt (OpenAI's official Terra example):

Review these dashboard screenshots and summarize the main changes. Refer to the chart labels and values, and flag anything you can't read confidently.

Read the full story

OpenAI says its model proved the Exponential Time Hypothesis false (OpenAI · confirmed)

OpenAI released a preprint on Monday claiming that 3-SAT, the classic hard logic puzzle, can be solved in subexponential time; if it holds, the Exponential Time Hypothesis (ETH) is false. Per OpenAI's post, "Surprisingly, our system has recently proved that the ETH is false, a result that was subsequently verified in Lean." Read the paper and Lean proofs in the openai/math repository, and follow the open check by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams in their Message to the TCS Community on ETH. Treat it as a claim under review: the organizers say their involvement is not a certification. Read the full story

Since our last email

These went up on the site after Monday's 8:29 AM ET email.

Grok Bot

• Grok Bot 0.75.8: Bots can propose Auto-review rules on a card (Grok Bot · release · confirmed). The changelog dates it Oct 24: "In the app, a Bot can offer to add, change, or remove one of your Auto-review rules on a card you confirm," and Auto-review now "takes a second look before showing an approval card." Eric Zakariasson posted at 5:34 PM ET Monday that it "is now stable and rolling out!" Story

Cursor

• Cursor's Agent SDK is out: build, test and deploy agents as code (Cursor · release · confirmed). Cursor posted at 2:58 PM ET Monday: "The Agent SDK is now available." Agents keep a Markdown prompt, TypeScript tools and evals in Git and wake on PR, CI, Slack, webhook or cron events, per the changelog. Story
• Cursor retires Cmd+K inline edit; Agent takes over (Cursor · release · confirmed). Staff member Colin said Monday that Cmd+K "was retired today" and "Updating won't bring it back." Press Cmd+I to open Agent, or select code or terminal output and press Cmd+L. Story

Anthropic

• Anthropic files S-1 for Nasdaq IPO with $16.3B H1 revenue (Anthropic · explainer · confirmed). The S-1 filed Monday shows $16,286 million in revenue for the six months ended June 30, 2026, up from $1,163 million a year earlier, and a planned Nasdaq listing under "ANT." Our read: if your team runs on Claude, read the risk factor on changing "usage limits, rate limits, or model and product capabilities" before your next renewal. Story
• How Claude Code's Boris Cherny works with Claude, in his own posts (Anthropic · explainer · confirmed). The head of Claude Code says prompting is now about the goal, the effort and the check. The story's steps: write the goal plainly, give a time or token budget, say how Claude should check its work, and put a reviewer on every change. Story

OpenAI

• OpenAI introduces GPT-6.1 Terra for fast computer use (OpenAI · release · confirmed). @OpenAI announced gpt-6.1-terra at 2:06 PM ET Monday, priced at $1 input, $0.05 cached input and $5 output per million tokens, "50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol" per the post. Our read: it's a cheaper default for screen-reading and form-filling agents. Story

How people are using it

• Hark Panels as status boards. Nic Finn wrote at 1:21 PM ET Monday that for a backup job "that I knew would take ~20 hours to complete," Hark "made a panel and checked every hour for status, then updated the panel." Other users posted Panels that show their own Hark usage limits, and users report Panels refresh about every few hours and not more often than hourly. Company posts and user reports; we have not tested Hark. Try it: ask Hark in chat to build a Panel, say how often it should update, and ask it to message you only when the job finishes or fails. Read the full story

Workflows and Setup of the day

Setups people say they run. All self-reported; we have not run any of them.

• Morning Newsjack PR pitches (Grok Bot). @coreyganim runs a weekday 7 AM Newsjack routine that only messages when a story passes, then drafts pitches. Self-reported: "On day two it had 2 pitch packs ready and had triaged 243 media requests." Newsjack's README does not list Grok Bot in "What runs where"; the author sells a $149/month Skool community (disclosed in the piece). Starter prompt is ours and untested. Story
• Nightly rival-creator X report (Claude). @LyVietVu says Claude plus an Apify X scraper sorts one creator's new posts and leaves "up to 3 ideas" by morning. Self-reported: 762 posts for about 20 cents; now nightly at 2:40 a.m., capped at 25 cents. The piece also has a Grok Bot path that uses built-in X reads (no scraper). Starter prompt is ours and untested. Story
• Player comments filed as GitHub issues (Grok Bot playbook). Arda Karaduman, who runs the one-person studio Gand, says a Bot files itch.io comments as GitHub issues, one per problem, with the player's hardware details attached, and a coding agent then checks each against the code while he keeps the final call. His blog post says "Fourteen Grok bots run the office side of my one-person studio." Disclosure: two tools in his setup are his own projects. He hasn't shared his prompts; the setup steps in the story are ours. Story

The digest

Grok Bot and Cursor

• Bot titles are going away. Colin, Cursor Community Support Engineer, said at 5:12 AM ET: "Bot titles (labels) are being retired, so each bot’s title was merged into its name to keep the text from being lost, and the separate title is going away." If you rename a bot, "leave the title empty" or the merge comes back on the next sync. (confirmed, staff)
• Overnight Cursor incident, resolved. "Elevated errors affecting repository integrations" opened at 3:33 AM ET: "Features that rely on repository access, including review agents and cloud agents, may fail or be delayed." Resolved; services "returned to normal operation at approximately 08:02 UTC" (4:02 AM ET). (status) (confirmed)
• Last night's Cursor incidents closed. "Degraded agent personalization and codebase sync" resolved at 9:19 PM ET Monday; "Payment System Sync Issues" resolved at 8:24 PM ET. (status.cursor.com) (confirmed)

Anthropic

• Anthropic's S-1 discloses a month-plus delay to frontier pre-training. Anthropic's S-1 says: "For example, we recently chose to delay pre-training a new frontier base model for more than a month to allow time to further enhance our security measures related to preventing model weight exfiltration and protecting our intellectual property." The filing also says: "Four of our models gained unauthorized access to real third-party systems during sessions lasting roughly 10 to 34 hours of active work." Reuters called those failures previously undisclosed. Anthropic described three on July 30 and wrote on Sept 9: "Three of these incidents (A, B, and C) were disclosed in our July 30 report, while the fourth (D) is disclosed here." (confirmed)
• Anthropic says a Cathedral security test reached some business systems. Researchers at the startup Cathedral reached "a limited set" of Anthropic's sales, HR and IT systems during a test under Anthropic's security research program, Anthropic confirmed, per Gizmodo and TokenPost, both citing The Washington Post. Cathedral's draft report says its AI agents found paths through third-party business software in under 12 hours. Per both outlets, Anthropic says Claude conversations, source code and model weights weren't reached, and that Cathedral broke program rules by not disclosing all its activity. Cathedral's draft says it kept data access to a minimum, per TokenPost. An Anthropic spokesperson said: "Our monitoring flagged some of Cathedral's activity before they reported it to us, and we shut off their access the same day." (reported)
• Claude Code 2.1.311 stops 1M-context errors on standard sessions. Claude Code 2.1.311 reached npm at 3:47 PM ET Monday (npm), and the GitHub release with its changelog followed at 5:51 PM ET. The headline fix: standard-context sessions no longer stop with "Usage credits required for 1M context" when a request runs over 200K tokens; per the changelog, "Claude Code now compacts the conversation and continues." It also stops retrying requests the API marked not to retry, which could hold a turn for minutes. WebFetch and WebSearch in Cowork cloud sessions now fail after 300 seconds instead of hanging. The self-hosted runner gains --hold-idle-file and --max-hold-min. The stable channel is still 2.1.308. (confirmed)

OpenAI

• OpenAI publishes ChatGPT work-use data for researchers. OpenAI released OpenAI Signals data on Monday that maps how consumers use ChatGPT for work to "165 work activities across nearly 100 countries", quarterly from Q1 2025 through Q3 2026. Its headline finding: "Across 74 countries observed in every quarter, the ten most common activities represented an average of 51.1% of each country’s classified work messages in Q3 2026, down from 56.0% in Q1 2025." The data "covers work-related messages on individual ChatGPT plans and does not include enterprise messages or Codex use." OpenAI says "researchers do not read individual conversations" and that it adds "statistical noise to protect individual messages". (confirmed)
• Two security-review customers explain how they use OpenAI's models. In an OpenAI customer story Monday, CodeRabbit says it uses GPT-6.1 Sol "as the primary model in its security agent", and that for user-selected findings it "can attempt exploit verification in an isolated sandbox without testing production systems." In a second story, Greptile CEO Daksh Gupta says it uses GPT-6 Astra's computer use to resize a browser to "an iPhone-sized viewport" and catch UI bugs, and that "thousands of customers have connected Codex to Greptile’s knowledge base, according to Daksh." (company claims, per OpenAI's customer stories)
• Codex Code Review and Security Review errors on private GitHub repos (monitoring). status.openai.com opened "Elevated error rates for most hosted Code Reviews and Security Reviews on private GitHub repositories" at 4:00 AM ET Tuesday. At 4:21 AM ET it said "We’ve applied a mitigation and are monitoring the affected services to confirm recovery." Still Monitoring when we checked at 6:09 AM ET; the page lists "No components marked as affected". (confirmed) Our read: if a hosted review failed overnight, rerun it once the incident is marked resolved.

Hark

• Adcock says more Hark updates are coming "this week and next." Brett Adcock, Hark's founder, wrote at 12:20 AM ET Tuesday: "We spent a lot of time making the design feel quite lovely to use" and "we have plenty more updates coming this week and next". He was quoting a user who called Hark Pro "really beautifully designed". Adcock gave no details, so this is his claim, not an announced feature. Hark's articles page and Using Hark page were unchanged from Monday when we checked at 6:03 AM ET. (founder post)
• Hark Pro ratings keep climbing on version 2.5.0. Apple's lookup data for the Hark Pro listing showed 4.66 stars from 1,935 ratings at 6:03 AM ET Tuesday, still on version 2.5.0 from Sunday. On Monday we reported 4.66 from 1,747. Our math: 188 new ratings in a day. The Google Play listing still says it was updated Oct 25, and Hark's status page showed "All Systems Operational". (store listings, status page)
• Users say Hark struggles with reCAPTCHA and blocked sites. Hc1pHr wrote to Hark at 8:16 PM ET Monday that Hark Pro "is great" but asked it to solve "the the reCAPTCHA verifications," adding "Cloudflare works good though." Ara wrote in Korean at 12:42 PM ET Monday that it's a shame Hark can't operate the browser directly and has a hard time getting around blocked addresses (our translation). Hark hasn't replied publicly to either post. (user reports)

Known issues and workarounds

The newest open items from Known issues:

• Bot computer updates on Oct 17, 21 and 22 dropped installed apps (Grok Bot, confirmed, tracked): keep a setup script your Bot reruns after an update; files, Bots and logins survive. Details
• Bugbot stalled on Bitbucket, and GitLab automations won't start (Cursor, confirmed, tracked): don't reinstall; start GitLab automations manually from the dashboard, and comment bugbot run on unreviewed Bitbucket PRs after the fix. Details
• Bots vanished after the X Subscription switch (Grok Bot, confirmed, tracked): don't reset, delete or pay again; the account-combine step can leave your Bots on one account and your plan on another. Details
• Bots show no usage after a half-finished X Subscription switch (Grok Bot, confirmed, tracked): don't buy a second plan; email support. Details
• /review-bugbot bills as Composer 2.5 Fast (Cursor, confirmed): review in the main chat with no subagents, or use your own subagent set to [fast=false]. Details

Coming up

From Friday's "Coming next week" and today's stories. Announced, hinted or scheduled, not shipped.

On the calendar

• Grok Bot workshops this week. Cursor Workshops lists SpaceXSI for Higher Education on Oct 28, Grok Bot for Enterprise Teams and Grok Bot X Clay on Oct 29, and What's New in Grok Bot on Oct 30. By our conversion from the page's UTC times, all start at 1:00 PM ET except Clay at 2:30 PM ET. (scheduled)
• Claude for Nonprofits webinars, Nov 10 and Nov 12. An admin session on data handling and AI policy, then a staff session, per Anthropic's post. (confirmed)
• Later: Nov 16 and Nov 20. Claude Enterprise's unified chat and Cowork turns on by default Nov 16 (story), and OpenAI's updated Usage Policies take effect Nov 20. (confirmed)

Confirmed, no date

• Claude apps start in auto mode. @claudeai posted on Oct 22: "Starting next week, sessions across Claude web, desktop, and mobile apps will run in auto mode by default on Pro, Max, and Team plans" (story). (confirmed, no day given)
• Bot titles go away. Staff say the separate title field is being retired (post). (confirmed, staff)
• More Claude products get MCP tasks. @ClaudeDevs said tasks and elicitations are "rolling out to more Claude products soon" (story). (confirmed, no date)
• An open check of OpenAI's ETH claim. Impagliazzo, Paturi and Williams are organizing it; no report date (story). (confirmed, no date)

Early signals (unverified)

• More Hark updates. Adcock says "plenty more updates coming this week and next," with no details. (founder post, unverified)
• iMessage for the phone apps. Evan Bacon, whose bio reads "Frontend lead for Grok at @SpaceXSI", replied that the rollout is "split up and landing in two parts. iMessage will require a Mac. Will roll out soon". No date. (staff hint, unverified)
• A statechart skill for pstack. Lauren Tan posted that she's "working on a new companion skill" and wrote "will release soon!" No date. (staff hint, unverified)

Daily Bot Check

1. Where should a nonprofit start in the Claude for Nonprofits plugin? (a) With the records map (/where-things-live), since other skills read it first (b) With a grant report skill and skip the map (c) Ask admin for a separate Nonprofit plan only
2. How do you pick GPT-6.1 Terra in the Codex CLI? (a) Run codex -m gpt-6.1-terra on CLI 0.159.0 or later (b) Switch the default model in API settings only (c) Ask ChatGPT to enable Terra for all Codex sessions
3. What did Hark users turn Panels into on Monday? (a) Status boards for long jobs, usage meters and sports feeds (b) A new public Panels API with sub-minute refresh (c) Official Hark dashboards that replace Home

Answers: 1 (a), 2 (a), 3 (a).

Every story in this email

• Claude for Nonprofits plugin adds 14 skills for grants and hiring
• Grok Bot's X connector no longer needs a developer account or credits
• How to switch Codex and Work to GPT-6.1 Terra, and when not to
• OpenAI says its model proved the Exponential Time Hypothesis false
• Grok Bot 0.75.8: Bots can propose Auto-review rules on a card
• Cursor's Agent SDK is out: build, test and deploy agents as code
• Cursor retires Cmd+K inline edit; Agent takes over
• Anthropic files S-1 for Nasdaq IPO with $16.3B H1 revenue
• How Claude Code's Boris Cherny works with Claude, in his own posts
• OpenAI introduces GPT-6.1 Terra for fast computer use
• How Hark users turn Panels into status boards for long jobs
• A Grok Bot PR routine that drafts pitches from each morning's news
• A nightly Claude run that reports what a rival creator posted
• Grok Bot for game studios: player comments filed as GitHub issues
• Full issue, Oct 27

You're getting this because you subscribed at thedailybot.vercel.app. Reply with corrections or tips; a person reads every reply.

How we label: confirmed means we checked a primary source or a staff post; reported means a credible claim we haven't verified; "Our read" is our own analysis.