OpenAI Releases LifeVal to Measure Model Progress on Everyday Life Tasks
On November 2, 2026, OpenAI released LifeVal, based on U.S. and multi-country time-use surveys and a representative survey of 3,995 people covering scenarios such as parenting, government services, finance, and shopping, to track ChatGPT’s annual improvements in agent capabilities including browser and computer use. Officials provided examples of workflow completion rates for tasks such as school enrollment, flood relief volunteering, and apartment hunting (approximately 90%–100%), shifting the evaluation focus from academic knowledge benchmarks to executable daily-life tasks.
Read MoreAnthropic Open-Sources FlashProve to Advance Provable Reasoning in Large Models with zkSNARK
On November 2, 2026, the Anthropic Frontier Red Team released a breakthrough in provable inference and open-sourced the FlashProve prototype. By utilizing zkSNARKs, it demonstrates that outputs originate from a committed model without revealing the underlying weights. Using Llama 4 Maverick (400B MoE, 4-bit quantization), the system achieves a proof generation speed of approximately 110 tokens/s on a single H100 GPU. Proofs for 1024-token requests are approximately 19 MB in size, requiring roughly 70–95 times the GPU time compared to vLLM inference for the same prompt. Concurrently, the team released a technical report, protocol specifications, prover/verifier code, and Lean formalizations under the Apache 2.0 license.
Read MoreSalesforce Agrees to Acquire AI Customer Service Automation Company Cresta
Around November 3, 2026, Salesforce announced it had agreed to acquire Cresta Intelligence; the company did not disclose the financial terms, while The Information reported the deal value at approximately $2.7 billion. Cresta provides multi-agent-orchestrated customer-service automation, Deputy for handling complex tickets, and real-time assistance for agents, as well as Opera fine-tuning and Synthetic Customers testing tools. This marks Salesforce’s second billion-dollar-scale customer-service acquisition of the year, following its $3.6 billion purchase of Fin, while its largest single deal this year is the $5.66 billion acquisition of Braze. Both the Braze and Cresta transactions are expected to close in fiscal 2028, which begins in February 2027.
Read MoreGenerative Media Inference Platform Fal Completes $1.3 Billion Financing, Reaching a Valuation of $16 Billion
According to a TechCrunch report on November 3, 2026, San Francisco-based inference hosting company fal has raised approximately $1.3 billion in a Series F round led by Bessemer, valuing the company at $16 billion—roughly double its $8 billion valuation in April. Other participants include Nvidia, Sound Ventures, and T. Rowe Price. The company simultaneously disclosed that the combined Series E and F financing totals approximately $1.65 billion, with Sequoia leading the Series E. Demand is being driven by open-source models facilitating video, image, and audio inference. Bloomberg reports that fal's annualized revenue has already surpassed $800 million. Competitors in the same sector, including Baseten, Fireworks AI, and Modal, are also expanding. However, the report cites sources claiming that inference companies still maintain thin gross margins due to high compute costs.
Read MoreOpenAI Announces Progress in the Application of GPT-6 Astra to Mechanical and Electrical Engineering Design
On November 3, 2026, OpenAI published a blog post detailing how its frontier models are being used in CAD, simulation, PCB design, and interdisciplinary systems engineering. It demonstrated Astra completing a prototype of a conversational Codex without direct human editing during testing, while acknowledging that manufacturing readiness still lacks DFM and tolerance considerations. Hardware team token usage has increased about 23x since February; on BenchCAD 2.0, newer internal models scored 0.875, while GPT-6 Astra achieved up to 0.594 in reasoning. The company also announced research collaboration with Mazda on aerodynamics and crash safety simulation, and expanded its scan-to-CAD workflow with Chip Ganassi Racing.
Read MoreQuantitative Football Prediction: Running Elo, Poisson, and Monte Carlo in Agent
This article introduces the "Football Quantitative Analysis & Prediction" Skill (`@user_fa4d9372/football-match-analysis`, v1.0.0, categorized under data-analysis) released by the SkillHub community. It unifies football match analysis under a quantitative framework: employing Elo ratings to characterize strength differentials, Poisson distribution to model goal distributions, and Monte Carlo simulations to address format-level uncertainties such as advancement scenarios. It also includes odds discrepancy detection and team attack-defense comparisons. Trigger conditions include analyzing matches, predicting outcomes, viewing advancement probabilities, comparing team strengths, detecting betting value, and checking World Cup schedules and odds. The provided documentation does not include installation commands or configuration options, confirming only the `requires_api_key=false` tag, which indicates no API key is required. It is important to note that the final analysis for this Skill is performed by the current Agent. The page indicates 0 installations, and while the rating is listed as 100,000, this is a page annotation value; the timeliness of the data source requires practical verification.
Read MoreOfficial Documentation Formatting Tool: Automatically Reformats Misaligned and Scattered Documents into Standard DOCX Files According to Official Document Conventions
The layout rules for official documents are complex and fragmented; manual adjustment is time-consuming and difficult to batch process. This Skill (SkillHub Name: "Official Document Format Conversion", @user_60300dde/doc-format, Version 1.0.5, Community Source, Category: office-efficiency, no API Key required) accepts doc/docx/wps/txt/md documents and automatically formats them according to official document layout conventions, outputting standard docx files. Verified capabilities include: recognition of four hierarchy levels ("一、（一）1.（1）"), distinction between titles and subtitles, attachment formatting, figure/table title recognition, page margins/page numbers/line spacing, font types and sizes, TXT/MD blank line patterns, custom configuration, and batch processing for table of contents. Trigger keywords cover official document layout, document formatting, font settings, heading hierarchy, format unification, wfp_cli, etc. The official page does not provide installation commands, and any unverified aspects are omitted here. It is applicable for scenarios requiring batch unification of official document formats and conversion of txt/md to standardized docx. Please note that the output is limited to docx, and title recognition relies on numbering styles; it is recommended to verify default configurations with a sample draft before use.
Read MoreIntroduction to the Enterprise Custom Edition Skill for AI Bid Writing System
This article introduces the AI Bid System Enterprise Customization Skill. Published by Jinlicheng Information, its coordinate is @org-unhar74t/ai-bid-enterprise-customization, version 1.0.0, categorized under business-ops, with tags including requires_api_key=true. The Skill is designed to screen candidate projects based on enterprise business data and templates, interpret tender documents, assist in drafting and reviewing bid documents, and list enterprise customization requirements that need confirmation. For the announcement, winning bid, procurement intention, and enterprise monitoring data from the Shipao Bidding Data API, actual service coverage and permissions shall prevail, with no automatic monitoring implemented. Queries default to the top 10 nationwide listings from the past 30 calendar days, while analyses default to the past 6 months with disclosed samples and limitations. Development, integration, deployment, automatic push notifications, and acceptance of enterprise systems require separate confirmation; the Skill does not equate to a delivered customized system.
Read MoreShuoruan · Cross-border E-commerce Product Selection Assistant: Compressing the Product Selection Funnel into a Single Conversation
The cross-border e-commerce product selection assistant by Shuoruan is released on SkillHub under the business-ops category, version 1.1.3, with no API key required. By entering the platform and category, one conversation generates an Excel product selection summary: one product per row, including product image, dimensions, price band, five-dimension ratings, estimated gross margin, 1688 sourcing link, and product testing recommendations. Ratings adjust weights based on platform characteristics (Amazon emphasizes competition, TEMU emphasizes profit, TikTok Shop emphasizes content differentiation), with red/yellow/green coloring to distinguish rush/trial/watch/abandon. Purchase prices include direct 1688 links and use the median of tiered pricing; if no reliable source is available, it is truthfully marked. Data is sourced from rankings, front-end prices, reviews, and Google Trends, with each item marked for source and confidence. It provides two outputs: quick scan and in-depth report. Applicable to product selection and competitor research for Amazon / TEMU / Shein / TikTok Shop sellers.
Read MoreAI Bid System SaaS Private Deployment: Integrating Bidding Assistance into Your Own Agent
Agent Skill for Bid Assistance: Assists in identifying candidates, analyzing scoring and disqualification criteria, drafting technical and commercial proposals, and auditing documents against tender requirements. Data is sourced from the Shibo Bid Data API, covering announcements, awards, procurement intentions, and corporate monitoring; actual coverage is subject to the service, and automatic monitoring is not implemented. Queries default to the first 10 results nationwide from the last 30 calendar days; analysis defaults to the last 6 months with sample limitations disclosed. Published by Jinlicheng Information (Version 1.0.0, Category: business-ops); requires an API key. Note that the Skill is not equivalent to a deployed SaaS; intranet deployment, push notifications, knowledge bases, and security isolation require separate validation. An external API does not imply that data remains within a private network.
Read MoreMr. Da Feng Class Teacher Workbench Generator (Single-file HTML): Put Class Management into an HTML File
This is a skill designed to generate a single-file HTML Homeroom Teacher Workbench. It covers 11 modules: Dashboard, Student Management, Discipline Statistics, Assignment Management, Grade Analysis (Ranking & Progress), Leave Management, Work Trails, Interview Records, Home-School Communication (including a Parent Group Notification Generator), Class Activities, and Todo/Memos (supporting recurring cycles). It features built-in subject presets for elementary, middle, and high schools. Users can declare their teaching stage and subjects in "Settings," making it suitable for homeroom teachers of various disciplines. Data is stored in the browser's localStorage, ensuring offline availability, zero external dependencies, and usability via a simple double-click on the single file. The top bar features a "Homeroom Teacher Workbench" brand title with a graduation cap icon, and the bottom right corner of the main page is signed "Created by: Feng Menggang / WeChat Official Account 'Digitalized Teaching: Teacher Feng'." It is ideal for K-12 homeroom teachers seeking a class management platform or efficiency tool, or those looking to publish and reuse this workbench.
Read MoreOfficial Writing · Lingxi: Introduction to the Skill for Creating and Reconstructing Government Documents
This article introduces the community Skill "Institutional Writing · Lingxi": It supports a revision mode for restructuring original source materials and a creation mode based on themes, outlines, and past materials. The Skill outputs Word documents compliant with the GB/T 9704-2012 standard, and includes the official link along with usage notes.
Read MoreProgramming Expert Pro: Engineering Topics & Multi-Platform Development Integrated Derived Skill
Programming Expert Pro is a dev-programming skill sourced from the SkillHub community (Version 1.1.0, located at @user_28275a78/dev-expert-pro) that requires no additional API Key. It integrates engineering topics from Programming Expert 2.0.3 with development resources from "Programming God" regarding HarmonyOS ArkTS/ArkUI, WeChat Mini Programs, and Mini Games. It is designed for project implementation, root-cause debugging, code review, refactoring, testing, architecture performance analysis, and multi-platform development verification. Invocation supports natural language and @topic identifiers; it accesses in-package knowledge based on the task, while answering conceptual questions directly. This article introduces its positioning, verified capabilities, installation and locating methods, and typical usage. It also outlines considerations regarding its derivative integration: it represents an integration of upstream materials rather than an independent, original knowledge source; furthermore, since official documentation does not provide installation commands or details on client-side differences, the official pages should be treated as the source of truth.
Read MoreTeacher Da Feng's Quality Courses PPT Courseware Generator: Batch Generate Coursewares Based on Templates
This is an Agent Skill based on the SKILL.md format, maintained by community publishers and categorized as "office-efficiency" (Version 2.1.3). By inputting a topic and outline, it automatically generates high-quality courseware in PPTX format with a unified structure and consistent fonts and sizes, while visually inheriting the template's background, theme, and color scheme. Each set of courseware strictly includes five slide types: cover, learning objectives, content, summary, and assignments. Cover text is white, content is positioned in the left two-thirds of content pages, and the right one-third is left blank for an instructor's portrait. It is suitable for scenarios such as batch production of teaching slides, public lecture PPTs, and training handouts based on templates. The article introduces core features, installation and activation methods, typical usage, and applicable limitations, noting that the data source is the SkillHub page.
Read More