Together with
👋 Hey builders,
Your best repeated prompt is not a chatbot. It is a process: when it runs, which facts it needs, what good looks like, and when it must stop. Google’s Gemini Skills make that process inspectable.
🗺️ TODAY'S SITE MAP
Why Google is moving from Gems to reusable Skills
Build a reusable task card in five minutes
Copy the portable Skill starter
Find three free packages that teach agents better habits
Watch the code-agent skill wave and web-access experiments
🔨 BREAKING GROUND
Google is turning Gemini Gems into Skills →. Users can save reusable instructions, call one from the prompt bar with /, attach reference files, and combine multiple Skills in one task. Google says Gemini can also suggest or run a matching Skill automatically.
The rollout is staged. Skills are in Gemini Spark, while The Verge reports Workspace starts October 5 and the Gemini app starts October 13. Google plans to retire individual Gems in November and migrate them. Availability still depends on rollout and account eligibility.
The shift is not the slash command. It is moving from a named helper to a reusable recipe. Define its trigger, inputs, checks and stop rule once; reuse it without rebuilding the assistant.
Why it matters: Repeated prompts become shared operating procedures. That makes them easier to improve—and easier to accidentally over-trust.
My take: A Skill is valuable only when it says when not to run. Start with a narrow task and a visible output check. Do not hide a broad permission, a fuzzy goal, or confidential context inside a cute helper name.
Hype Scale: 🔨🔨🔨 (3/5 — practical workflow shift; rollout and portability remain uneven)
From Our Partners
How 2M+ Professionals Stay Ahead on AI
What’s the secret to staying ahead of the curve in the world of AI? Information.
Luckily, you can join 2,000,000+ early adopters reading The Rundown AI — the free newsletter that makes you smarter on AI with just a 5-minute read per day.
A CRM so smart, it updates itself.
HubSpot's CRM is so smart, it now updates itself. Calls get logged and summarized the moment they end. New leads get researched and a first outreach drafted, automatically. Deals move forward on their own, with next steps flagged before you have to ask.
Even the one repetitive task you've been meaning to fix can now run itself, no code required. Customers using it are already seeing the difference.
This isn't the CRM you have to work hard to figure out. It's the one already working before you log in.
🏗️ AROUND THE SITE
Reflection unveils Beam, an open-weight model still waiting on weights → The company says its 501B-parameter MoE activates 23B parameters per task; full weights and technical details are promised this month, while performance claims await independent checks.
Cohere ships North 2 with agent memory and cost controls → Cohere says the enterprise platform adds a new harness, libraries, Skills, and granular spend governance; treat its reliability and efficiency language as vendor claims.
GitHub lets Copilot operate desktop apps in preview → The feature can click, type and navigate GUI-only software, but asks for approval and can be disabled by organizations.
Google pauses part of its open-source bug bounty over AI submissions → Product-vulnerability reports are paused while Google says it works through a surge of mostly invalid automated filings.
⚡ THE 5-MINUTE BUILD
Turn one repeated prompt into a task card
Pick one narrow repeat. Choose a task you run weekly, such as turning meeting notes into decisions—not “help with operations.”
Write the trigger. Say exactly when it runs and name one nearby task where it must stay off.
Set the contract. List the inputs it may use, the three checks it must pass, and the output format.
Add a stop rule. Make it ask you before sending, publishing, deleting, spending, or using missing facts.
Save and test twice. Put the card in a Markdown file, run one good example and one bad-fit example, then tighten the trigger.
The rule: If your Skill cannot explain when to stop, it is not ready to reuse.
📋 THE BLUEPRINT
Save this as SKILL.md inside a folder with the same lowercase, hyphenated name. It follows the free Agent Skills specification; tools support the format differently, so you can also paste it into any chat as a task card.
---
name: [task-name-in-kebab-case]
description: [WHAT IT DOES]. Use when [SPECIFIC TRIGGER]. Do not use when [OUT-OF-SCOPE CASE].
---
# [TASK NAME]
## Inputs
Use only: [APPROVED NOTES, FILES, OR FACTS].
Ask for [MISSING FACT] before continuing.
## Workflow
1. [FIRST CHECK OR TRANSFORM]
2. [SECOND CHECK OR TRANSFORM]
3. [THIRD CHECK OR TRANSFORM]
## Output
Return: [EXACT FORMAT AND LENGTH].
Show: [EVIDENCE OR SOURCE FOR EACH IMPORTANT CLAIM].
## Stop rule
Do not [SEND / PUBLISH / DELETE / SPEND / ASSUME].
Pause and ask when [ESCALATION CONDITION].🧭 HIDDEN GEMS
Give a coding agent a real development playbook. Superpowers is a free, open set of planning, testing and review workflows—not just another prompt.
Borrow production-grade engineering gates. Addy Osmani’s Agent Skills is a free MIT-licensed collection for specs, tests, reviews and launch checks.
Make the agent challenge the plan first. Skills for Real Engineers offers free installable flows for triage, testing, research and handoffs.🧰 THE TOOLBOX
🧰 THE TOOLBOX
📡 SITE RADAR
Breakout: “do less” becomes a code-agent package → Ponytail is a reusable Skill that pushes an agent toward smaller, native solutions; its own benchmark claims should be treated as project-reported. Open-source trend reports logged roughly +1,900 stars in a day on October 5. Signals: the GitHub project, GitHub trend tracking, and a weekly trend digest.
Emerging: an agent gets “internet eyes” → Agent-Reach claims one CLI can set up access to public web, video, social and code sources, often without API fees. A trend report recorded +980 stars on October 5. Signals: the project’s public repo, daily tracker, and weekly tracker. Inspect auth, terms and data handling before installing it.
📐 ONE TERM A DAY
Today’s term: progressive disclosure
Plain English: Pack a day bag, not your whole closet. An agent sees a Skill’s tiny label first, then opens the full instructions only when the task fits.
Why you should care: Smaller, focused instructions are cheaper to load and simpler to review than one giant forever-prompt.OURCES
1,000 Claude AI prompts, 40 templates, 6 frameworks. Free Starter Pack and Premium Edition.
The checklist your agent is probably failing. OWASP's Top 10 for Agentic Applications, 2026 edition, is the free security standard for AI agents — and almost nobody shipping agents has read it. → genai.owasp.org
A CLI that logs every prompt you've ever sent. Simon Willison's
llmwrites each request and response to a SQLite file as you go, so your entire AI history becomes something you can query in SQL. One install, free forever. → llm.datasette.io
21 free lessons, first prompt to working agent. Microsoft's Generative AI for Beginners covers RAG, agents and prompt engineering with code that actually runs. No signup, no paywall. → github.com/microsoft/generative-ai-for-beginners
💬 Reply CARD with the repeated task you would turn into a Skill first.
Tools down.
— Swati, ByteBuilders