A vulnerability was identified in Linksys E5600 1.1.0.26. Affected by this vulnerability is the function strcpy of the file /cgi-bin/mesh_login.cgi of the component JSON Request Parser. The manipulation of the argument username/password/token/source/cn/action leads to out-of-bounds write. The attack may be initiated remotely. The exploit is publicly available and might be …

About the Microsoft Decision Model In October 2026, Microsoft announced Microsoft-Decision-1, the latest model for fast decision-scoring available on Microsoft Foundry. Unlike LLMs that generate textual outputs, decision models return structured outputs that allow software to immediately act on, without having to first interpret the response. The model accepts a state value which can be text or JSON. There are three question types based on the decision we need: Question Type Result Yes/No, true/false questions noul A probability from 0 through 1 Multiple Choice Questions choice One selection option and the probability for each option Questions related to ranking or grading score A value on an ordered scale and the probability for each level. Starts from 0 Use Case We will be building a simple application that decides whether a software package needs upgrading. An agent searches trusted sources (via Grounding with Bing Custom Search resource) for release notes and security advisories. A decision model then reads that evidence and says whether to upgrade, why, and how urgently. Pre-Requisites Azure Subscription Microsoft Foundry resource deployed in a resource group. Deploy the Microsoft-Decision-1 model, and gpt-4.1 model to power the agent A Grounding with Bing Custom Search resource Adding Sources to Bing Custom Search Go to the Bing Custom Search resource Click on Configurations in the left pane under the Resource Management section Create a new configuration and name it ‘documentations’. This is the Domain Name Add the following web addresses: https://docs.djangoproject.com/en/stable/releases https://www.djangoproject.com/weblog https://www.djangoproject.com/download https://pypi.org/project/django Save the configuration Add Bing Custom Search as a Tool to Foundry In the previous section, we made the tool. Now, we will add it to Foundry Steps: Click on Management Centre Click on Connected resources Click on + New Connection Choose Grounding with Bing Custom Search It will automatically show all the Bing Custom Search resources. Connect the one you created for this tutorial by choosing ‘Add Connection’ Click on the connection, and save the Id Creating the Foundry Agent Set up a .env file and populate the following information: ENDPOINT=<https://<your-foundry-resource-name>.services.ai.azure.com/providers/microsoft> (This is the decision model’s endpoint) PROJECT_ENDPOINT=https://<your-foundry-resource-name>.services.ai.azure.com/api/projects/proj-default DEPLOYMENT_NAME=<decision-model’s deployment name> CONN_ID=<Bing Custom Search Id we saved in the previous step> DOMAIN_NAME=documentations We create a foundry agent using the following code snippet and deploy it to Foundry. The file is called create_agent.py. Run this file to deploy the agent to foundry. import os from azure.identity import DefaultAzureCredential, InteractiveBrowserCredential from azure.ai.projects import AIProjectClient from azure.ai.projects.models import ( PromptAgentDefinition, BingCustomSearchPreviewTool, BingCustomSearchToolParameters, BingCustomSearchConfiguration, ) from dotenv import load_dotenv load_dotenv() PROMPT = """ You are a software upgrade research agent. Given a package name and installed version, use Bing Custom Search to find official release notes, security advisories, and support information. Summarize: - Security issues affecting the installed version and available fixes. - Whether the version is still supported. - Relevant breaking changes and upgrade requirements. Use only retrieved evidence and include source URLs. If applicability is unclear, say so. Do not treat missing evidence as proof that a version is safe. Return a JSON object with: - package - installed_version - summary - source_urls """ conn_id = os.getenv("CONN_ID") project_client = AIProjectClient( endpoint=os.getenv("PROJECT_ENDPOINT"), credential=DefaultAzureCredential() ) tool = BingCustomSearchPreviewTool( bing_custom_search_preview=BingCustomSearchToolParameters( search_configurations=[ BingCustomSearchConfiguration( project_connection_id=conn_id, instance_name=os.getenv("DOMAIN_NAME"), ) ] ) ) with project_client: agent = project_client.agents.create_version( agent_name="extract-news-agent", definition=PromptAgentDefinition( model="gpt-4.1", instructions=PROMPT, tools=[tool], ), ) print(f"Agent created (id: {agent.id}, name: {agent.name}, version: {agent.version})") Setting up the Decision Model The following code sets up the decision model. The file is called model.py import os from azure.identity import DefaultAzureCredential, get_bearer_token_provider from dotenv import load_dotenv from typesafe_sdk import TypeSafeClient load_dotenv() ENDPOINT = os.getenv("ENDPOINT") DEPLOYMENT_NAME = os.getenv("DEPLOYMENT_NAME") def assess_upgrade(state: str): with DefaultAzureCredential() as credential: token_provider = get_bearer_token_provider( credential, "https://ai.azure.com/.default" ) with TypeSafeClient( api_key=token_provider(), base_url=ENDPOINT, model=DEPLOYMENT_NAME, timeout=60, ) as client: return client.system_one( state=state, questions={ "upgrade_needed": { "type": "noul", "instructions": ( "Does the supplied evidence indicate that the installed " "version needs an upgrade? Use only the supplied evidence. " "Missing evidence is not proof that no upgrade is needed." ), "criteria": { "true": ( "Evidence identifies an applicable security fix, " "end-of-support, or required compatibility upgrade." ), "false": ( "Evidence establishes that the installed version is " "supported and no applicable upgrade requirement exists." ), }, }, "reason": { "type": "choice", "instructions": ( "Choose the primary upgrade reason. If multiple apply, " "prioritize security-fix, then end-of-support, then " "compatibility. Choose insufficient-evidence if the " "installed version cannot be assessed reliably." ), "criteria": { "security-fix": ( "An official advisory explicitly affects the installed " "version and identifies a fix." ), "end-of-support": ( "The installed version no longer receives support." ), "compatibility": ( "An upgrade is required for the stated environment " "or application requirements." ), "routine-update": ( "Only optional improvements or routine fixes apply." ), "no-upgrade-needed": ( "Evidence supports retaining the installed version." ), "insufficient-evidence": ( "Evidence is missing, conflicting, or does not " "establish applicability to the installed version." ), }, }, "urgency": { "type": "score", "instructions": ( "Rate upgrade urgency using only verified applicability " "and impact in the supplied evidence. Do not assume an " "older version is vulnerable." ), "criteria": [ "Routine: no established urgent upgrade requirement.", "Planned: upgrade in a normal maintenance cycle.", "Urgent: applicable material security risk, unsupported " "version, or significant compatibility impact.", "Immediate: confirmed exposure to a critical issue or " "active exploitation relevant to the stated deployment.", ], }, }, ) def print_upgrade_decision(response): needed = response.nouls["upgrade_needed"] reason = response.choices["reason"] urgency = response.scores["urgency"] print(f"Upgrade-needed signal: {needed.noul:.1%}") print(f"Reason: {reason.choice} (confidence: {reason.confidence:.1%})") if reason.choice == "insufficient-evidence": print("Urgency: not actionable without more evidence") action = "gather more evidence" else: print(f"Urgency: {urgency.score:.2f} of 3") if needed.noul >= 0.5 and urgency.score >= 2: action = "prioritize upgrade review" elif needed.noul >= 0.5 or reason.choice != "no-upgrade-needed": action = "schedule upgrade review" else: action = "monitor" print(f"Action: {action}") We leverage all three types of questions in this code. “upgrade-needed” has a noul type associated to it, we return either true or false based on the agent’s output. “reason” has a choice type associated to it. We choose a reason for the upgrade. “urgency” has a score type associated to it. We define a criterion to decide the urgency. The score starts from 0 and can be a decimal. The instructions field tells the decision model how to judge the question, while criteria field defines the best possible answers. The print_upgrade_decision() function handles the output formatting: We print if the upgrade is needed (noul type) We print the reason (choice type) We print the action needed based on the urgency score (score type) Calling the Agent In this segment, we will watch the workflow in action! The following code snippet sends a query to the agent and returns a response. The file is called fetch_news.py from azure.identity import DefaultAzureCredential from azure.ai.projects import AIProjectClient from model import assess_upgrade, print_upgrade_decision import json import os from dotenv import load_dotenv load_dotenv() endpoint = os.environ["PROJECT_ENDPOINT"] agent_name = "extract-news-agent" user_message = "We use Django 4.2.0. Find official security advisories affecting this version and identify the fixed versions." # change this as per your needs with ( DefaultAzureCredential() as credential, AIProjectClient( endpoint=endpoint, credential=credential, allow_preview=True, ) as project_client, ): # Bind the OpenAI client to the existing agent endpoint. openai_client = project_client.get_openai_client(agent_name=agent_name) # GitHub Copilot preview agents require a conversation for every Responses turn. conversation = openai_client.conversations.create() response = openai_client.responses.create( conversation=conversation.id, input=[{"role": "user", "content": user_message}], ) raw_output = response.output[-1].content[0].text print(raw_output) news_item = json.loads(raw_output) summary = news_item["summary"] # Classify each news summary with the decision model decision = assess_upgrade(summary) print_upgrade_decision(decision) # print the decision The response contains a JSON object with these fields (we had defined this in our prompt while creating the agent): - package - installed_version - summary - source_urls In the snippet, we send the summary to the decision model and print the final decision. Sample output: Run the fetch_news.py file to get the output! Next Steps: Try it yourself! Clone the repository Extend the use case by: Scanning a real dependency file (requirements.txt, pyproject.toml, package.json) and ranking all packages by urgency. Outputting a report (Markdown or JSON) with the decision, reason, confidence and source URLs. Running it as a scheduled job or CI check that opens an issue or PR comment when an upgrade is urgent. References: Deploy and use Microsoft-Decision-1 in Microsoft Foundry - Microsoft Foundry | Microsoft Learn Microsoft-Decision-1: Our model for fast decision-making

JSON Web Token (JWT) is a compact URL-safe means of representing claims to be transferred between two parties. The claims in a JWT are encoded as a JSON object that is digitally signed using JSON Web ...

I work on OnePassApps, so this is our own product, and everything below is a read of our own public pages - fetched and measured, not a guess. Search engines get a narrow view of a site like ours: a sitemap, a robots file, and a handful of HTML pages. Here is exactly what `app.onepassapps.com` hands them, measured by fetching every URL it names. **How this was measured:** `GET https://app.onepassapps.com/sitemap.xml`, `GET https://app.onepassapps.com/robots.txt`, then one request per URL in the sitemap. For each page I recorded the HTTP status, the `<title>`, the meta description, the `<link rel="canonical">`, `og:title`, and the JSON-LD graph. ## The sitemap: 16 URLs, and nothing else `https://app.onepassapps.com/sitemap.xml` answers `200` with a plain `<urlset>` (not a sitemap index) containing 16 `<url>` entries. Each entry carries a `<loc>` and nothing else: no `<lastmod>`, no `<priority>`, no `<changefreq>`. The 16 URLs, in the order the file lists them: Group | URLs ---|--- Site | `/`, `/voting`, `/plans`, `/terms` Studios | `/studios/image`, `/studios/audio`, `/studios/files`, `/studios/text`, `/studios/data`, `/studios/icons`, `/studios/colour` Apps | `/apps/newsflow`, `/apps/movie-tracker` Guides | `/guides`, `/guides/newsflow`, `/guides/movie-tracker` That is the whole indexable map: four site-level pages, the seven browser studios, the two live apps, and three guide pages. ## robots.txt: five rules and one sitemap pointer User-agent: * Allow: / Disallow: /api Disallow: /api/ Disallow: /account Disallow: /sign-in Disallow: /sign-up Sitemap: https://app.onepassapps.com/sitemap.xml `/api` and `/api/` are both listed, so the API stays closed however you write it. The account surfaces are `/account`, `/sign-in` and `/sign-up` - note those three names, because they are not the same as the names you would guess (see below). There is no `Crawl-delay`, no per-bot block and no second sitemap declaration. ## Every URL in the sitemap has a canonical that points at itself All 16 URLs returned `200`. Each one carries a `<link rel="canonical">` whose value is exactly its own URL - no trailing-slash variant, no `www` host, no query string, no session id. There is no canonical on the site that disagrees with the URL it sits on, so there is nothing for a crawler to reconcile. ## Titles: one pattern, one exception Fifteen of the sixteen titles end with an em dash and `OnePassApps`: * `/voting` - "Vote on future apps — OnePassApps" * `/plans` - "Plans — OnePassApps" * `/studios/text` - "Text Studio — OnePassApps" * `/guides/newsflow` - "How to read briefs and listen with NewsFlow — OnePassApps" * `/guides/movie-tracker` — "How to track movies and TV in a private watchlist — OnePassApps" The exception is the home page, whose title is "OnePassApps — PDF, writing, data and design tools" - the brand first, and a description of the tool categories instead. ## Descriptions: sixteen of them, all different Every page in the sitemap carries its own meta description, all 16 distinct, and `og:title` matches the `<title>` on all sixteen. Two of them, verbatim: * Home: "Organise PDFs, build a résumé, chart CSV data and create designs in your browser. Use local tools without an account and explore OnePassApps membership." * Text Studio: "Clean, compare, convert and inspect text in your browser." ## The JSON-LD: one shared identity, then a per-page graph Each of the 16 pages ships exactly one `application/ld+json` block, and each one is a `@graph` that starts the same way. The `Organization` and `WebSite` nodes are identical on all sixteen pages: the brand name, the URL, the logo at `/icon-512.png`, the description "A growing collection of practical apps for everyday work, creativity, and life, with one account and one membership.", a customer-support `contactPoint` at `info@onepassapps.com`, and the `WebSite` node marked `inLanguage: en-GB`. After those two, the graph changes per page, and the shape tells you what the page is: * The **home page** adds a `WebApplication` node (`applicationCategory: UtilitiesApplication`) plus two `ItemList` nodes: "OnePassApps apps" (NewsFlow, MovieMatcherAI) and "OnePassApps browser-local Studios" (the seven studios, Image first, Colour last). * The **seven studios and the two apps** add a `WebPage` node, a second `WebApplication` node for that app, and a `BreadcrumbList` of two steps: Home, then the studio or app. * The **guides index** is a `CollectionPage`, with a `BreadcrumbList` and an `ItemList` naming its two guides. * The **two guide pages** are `WebPage` nodes with a three-step `BreadcrumbList`: Home, App guides, then the guide. * `/voting`, `/plans` and `/terms` are plain `WebPage` nodes with a two-step `BreadcrumbList`. So the site describes its identity once and then tells a crawler what each individual page is, rather than stamping one generic block on everything. ## Three routes that are not in the sitemap * `/pricing` returns `200` with the same `<title>` as `/plans`, and its canonical points at `https://app.onepassapps.com/plans`. It is a second name for the plans page, resolved by canonical rather than by a redirect. * `/support` returns `200` ("Support and feedback — OnePassApps") but has no canonical tag at all, and it is not in the sitemap. * `/login` and `/signup` both return `404` with the title "Page not found — OnePassApps". The account routes that robots.txt keeps out are `/sign-in` and `/sign-up`; the more obvious guesses are simply not routed. ## If you are linking here * Link the canonical URL of the page you mean. For plans that is `/plans`, not `/pricing`. * `https://app.onepassapps.com/support` works even though it is not in the sitemap. * Do not link `/login` or `/signup` - they 404. The account pages are `/sign-in` and `/sign-up`. * Every studio URL answers with the tool page itself, no sign-in redirect in front of it. ## Check it yourself curl -s https://app.onepassapps.com/robots.txt curl -s https://app.onepassapps.com/sitemap.xml | grep -c '<loc>' curl -s https://app.onepassapps.com/studios/text | grep -o '<link rel="canonical"[^>]*>' No login, no uploads, nothing to install - the same browser tools the guides walk through, described in one page each. Start with the Text Studio, or browse the app guides.

Claude Code 2.1.318 affiche la date de fin d'une connexion claude.ai enregistrée, dans claude auth status et les résultats JSON de claude -p, avec un avertissement dix jours avant l'échéance. Elle ajo