Reusable Claude skills and new Gemini connections
Anthropic's public skills library shows how to make repeatable Claude tasks reusable. Gemini is adding connections to apps including Airtable and PandaDoc; we also cover agent memory, voice tools, and the permissions worth checking before you link a business account.
The Big Story
GitHub Trending
skills: Public repository for Agent Skills
Anthropic's public Agent Skills repository shows how to give Claude reusable instructions, scripts and resources for tasks such as document creation, web-app testing and company communications. Each skill lives in a folder with a SKILL.md file describing what it does and when Claude should use it; examples can be used through Claude Code, paid Claude.ai plans or the Claude API. Many examples carry an Apache 2.0 license, but the Word, PDF, PowerPoint and spreadsheet creation skills are source-available rather than open source, and Anthropic cautions that actual behavior may differ from the demos.
If staff repeat the same Claude task, start with a narrowly scoped skill for a documented process (such as formatting a standard report), then test its output before using it for critical work.
For Everyone
Google Gemini Blog
A new wave of Connected Apps is rolling out to Gemini.
Google is rolling out more Connected Apps in Gemini, including Airtable, Linear, monday.com, PandaDoc, Zoho, Adobe, Squarespace, and Webflow. Users can connect services in Gemini settings or mention one in a chat to work with it without switching tabs. The announcement doesn't specify pricing or which plans get each connection.
If your team already uses one of these apps, check whether its Gemini connection is available on your plan and review the data permissions before linking business accounts.
For Builders
GitHub Trending
hindsight: Hindsight: Agent Memory That Learns
Hindsight is an open-source memory system that lets AI agents store information, retrieve relevant details, and form longer-term observations rather than relying only on conversation history. It offers Docker and embedded deployments, client libraries for Python, JavaScript and Go, and integrations with coding agents, n8n and Zapier. The hosted Hindsight Cloud option has usage-based billing, free starting credits and no fixed monthly or per-seat fee; its README cites independently reproduced results on a conversational-memory benchmark, though those results don't establish how well it will work for a particular business.
If an existing AI workflow repeatedly forgets customer or project context, ask its developer to test Hindsight with non-sensitive data and check hosting, data retention and usage costs before connecting business records.
Google Gemini Blog
Gemini 3.8 text-to-speech says hello
Google is rolling out two text-to-speech models: Gemini 3.8 Flash TTS for custom voices and directed performances, and Flash-Lite TTS for higher-volume audio such as dubbing and voice agents. Developers can try both in Google AI Studio and the Gemini API; Flash is also going into Gemini Notebook and Flash-Lite into Google Vids, while Gemini Enterprise API access is still coming. Google says Flash offers over 2,000 ready-made voices and voice replication from a 30-second sample, subject to matching verbal consent from the voice owner.
Try a short sample for training videos or customer audio, then compare quality, language coverage, usage costs, and voice rights before committing to production.
GitHub Trending
claude-plugins-official: Official, Anthropic-managed directory of high quality Claude Code Plugins.
Anthropic maintains an official directory of Claude Code plugins, separating its own plugins from third-party submissions. Users can browse the directory through Claude Code’s `/plugin` interface, and plugins may add commands, agents, skills or connections to external tools. Anthropic warns that it doesn’t control all software included in listed plugins or guarantee that it will continue to behave as expected.
If your team installs Claude Code plugins, review each plugin’s permissions, external connections and maintainer before enabling it (an official listing isn’t a blanket security guarantee).
AI TLDR
Whiteboard — an open-source app for reviewing what coding agents built
Whiteboard is an MIT-licensed desktop app that lets coding agents such as Claude Code and Codex explain their work on a shared canvas. Its diagrams link to the underlying code, while a semantic diff viewer filters changes and a decision log connects requirements to the agent’s choices. It runs against local checkouts, but users can’t edit files in the app, and shared reviews don’t update automatically.
If your business uses coding agents, try Whiteboard for reviewing their changes, but keep your editor handy for fixes and re-share reviews after updates.
AI TLDR
Codex CLI 0.157.0 — GPT-6 Sol and Luna reach Amazon Bedrock users
OpenAI’s Codex CLI 0.157.0 adds GPT-6 Sol and Luna to its model catalog, including for Amazon Bedrock, and prompts users to migrate from older models. The release also starts its background server automatically in eligible interactive sessions and adds a shortcut to fork conversations open in another app. Fixes cover proxy routing, file-upload retries and enforcement of network restrictions across redirects and active connections.
If your development team uses Codex through Bedrock, check which model is selected after updating and verify that existing network policies still behave as expected.
GitHub Trending
univer: The Office Harness for AI Agents — Spreadsheets, Docs, Slides, Canvas, Relational Tables, and PDF in one runtime.
Univer offers an open-source Office SDK for embedding spreadsheets, documents and presentations in other products, with a shared browser and Node.js runtime that AI agents can use to edit and inspect files. Its self-hostable Univer Workspace example shows agents building interactive spreadsheet dashboards and working alongside people on Office content. Collaboration, shared revisions and isolated agent drafts depend on specific Web SDK features whose availability and licensing vary; the README still lists PDF support as coming soon.
This is mainly for businesses building their own internal tools or software products, not a ready-made replacement for Google Workspace or Microsoft 365. Ask a developer to check feature licensing and test human review of agent edits before adopting it.
Worth Watching
TechCrunch AI
Google tests letting Gemini call businesses for you
Google is testing Call for Me, which lets Gemini phone businesses to check stock, book reservations, reschedule appointments, or put items on hold. The initial rollout is limited to U.S. Pixel 11 owners with a paid Gemini subscription and the beta Android Phone app. Calls use the owner's phone number, and users approve information to share, follow a live transcript, and can take over at any point.
If customers start calling through AI agents, clear phone menus and accurate availability information will matter even more. For now, this is a limited test, not a change every business needs to make.
Google Gemini Blog
Introducing Gemini 3.8 Live with Live Avatar
Google has made Gemini 3.8 Live with Live Avatar available in Gemini Enterprise. It pairs a conversational AI agent with a generated on-screen character that can listen, see, speak, and fetch information in the background while a conversation continues. Google says it supports 97 languages and watermarks generated audio and video; custom avatars made from reference images require enterprise allowlisting, and the announcement gives no price.
For customer-facing use, wait for pricing and test a real support or check-in workflow before committing. Ask how the avatar handles consent, mistakes, and handoff to a person.
GitHub Trending
starnet: A living pixel-art station where real AI agents do real work. Local-first desktop agent harness - bring your own key, watch your crew actually run.
StarNet is an early-release desktop app that lets users organize multiple AI agents in a pixel-art station where rooms, handoffs and objects map to actual permissions and work. It supports scheduled tasks, messaging through services including Slack and Telegram, and local records of agent activity and spending; users connect their own AI provider accounts or run a local Ollama model. Windows and macOS downloads are available, but the project says Windows is better tested, Linux isn't a supported public release, and cloud model requests leave the machine even though transcripts and other station records are stored locally.
Treat this as an experiment, not an unattended business workflow. If testing it, start with low-risk tasks, cap spending and permissions, and check what each connected model or messaging service receives.
TechCrunch AI
Ando wants to take on Slack with a team messaging app that lets humans and agents work together
Ando has launched a team messaging app where AI agents get their own identities and inboxes alongside employees. According to founder Sara Du, agents can join channels and calls, follow conversations without being tagged, and message colleagues when they spot a task or overlap between discussions. The startup raised $20 million and says it has customers in 15 countries, but Slack and Teams already offer agent integrations.
Before moving a team off its current messaging tool, compare a small Ando pilot against the AI features you already have, especially who can see conversations and when an agent can act without approval.