{"slug": "gitlab-19-6-release-notes", "title": "GitLab 19.6 release notes", "summary": "GitLab released version 19.6 on November 19, 2026, adding audit events for every action an AI agent takes through the GitLab MCP server, recording who performed the action, what they did, which resource was affected, and when, with events flagged as MCP-originated and viewable under AI > Governance > Audit events. The release also adds beta controls letting top-level group Owners set which GitLab MCP server tools agents may use, enforced for every connecting client, and makes Rapid Diffs generally available, replacing the classic diff view with server-rendered, streamed diffs. GitLab Duo in Slack entered limited availability on GitLab.com, and Agentic Chat gained a /goal command that starts a Developer Flow session which can commit, push, and create merge requests without per-action approval while consuming compute minutes.", "body_md": "On November 19, 2026, GitLab 19.6 was released with the following features.\n\nWe are excited to recognize [José](https://gitlab.com/jmrp), a Level 5 contributor,\nas this month’s [Notable Contributor](https://contributors.gitlab.com/notable-contributors)!\n\nJosé systematically expanded fine-grained personal access token authorization across the GitLab GraphQL API. Fine-grained tokens can now read data like blame, CODEOWNERS validation, available deploy keys, and project security metrics.\n\nBefore you upgrade to GitLab 19.6, review the\n[upgrade notes](https://docs.gitlab.com/update/versions/gitlab_19_changes/#upgrade-to-196)\nand the [deprecations and removals for GitLab 19.6](https://docs.gitlab.com/update/deprecations/#gitlab-196).\n\nEvery action an agent takes through the GitLab MCP server now creates an audit event. You get a record of agent activity without adding any client-side tracking.\n\nEach event records who performed the action, what they did, which resource was affected, and when. Events are also marked as coming from the MCP server, so you can tell agent actions apart from actions people take directly.\n\nTo review these events, go to **AI** > **Governance** in your top-level group and select the **Audit events** tab. Security and compliance teams get a complete view of what external agents do in GitLab.\n\nGitLab Duo in Slack is now in limited availability on GitLab.com. Mention `@GitLab` in a Slack channel or thread\nto turn a conversation into an issue, summarize a discussion, search for issues and merge requests,\nor ask questions about your projects. GitLab Duo reads the conversation as context and posts the result back in Slack.\n\nSetup is simpler too. An Owner installs the GitLab for Slack app on a top-level group. All GitLab Duo sessions from that Slack workspace then run in that group and use its GitLab Duo settings and credits.\n\nUsers with at least the Developer role connect their own GitLab account the first time they mention `@GitLab`.\n\nYou can now control which GitLab MCP server tools AI agents can use, with rules that GitLab enforces for every client that connects.\n\nOwners of a top-level group set each tool to:\n\nThis feature is in beta. To provide feedback, see [issue 633490](https://gitlab.com/gitlab-org/gitlab/-/work_items/633490).\n\nIn previous versions of GitLab, even small merge requests made you wait for the first diff to load. Large ones could slow down or freeze the page.\n\nRapid Diffs, a ground-up rewrite of how GitLab renders diffs, is now generally available and replaces the classic diff view. GitLab renders diffs on the server and streams them to your browser, so reviews start in seconds:\n\nRapid Diffs also supports the review tools you already use:\n\nPreviously, GitLab Duo Agentic Chat in the GitLab UI answered one message at a time and waited for you to continue. For a larger change, like fixing a set of failing tests, you had to keep prompting it.\n\nNow you can start a long-running task with `/goal`.\nAgentic Chat starts a Developer Flow session that keeps working until the task is done, so you don’t have to prompt after each step.\n\nThe flow can commit, push, and create merge requests without asking you to approve each action. Each goal also uses compute minutes.\n\nTo use `/goal`, open Agentic Chat in a project and turn on the Developer Flow for your top-level group.\n\nIn previous versions of GitLab, a GitLab Duo Agent Platform flow in a runner environment ran without isolation if the sandbox was unavailable, and you saw only a warning.\nNow you can [make the job fail instead of running the flow without the sandbox](https://docs.gitlab.com/user/duo_agent_platform/environment_sandbox/#require-flows-to-run-in-the-sandbox).\n\nIn previous versions of GitLab, the Developer Flow and Agentic Chat did not support images as context with self-hosted models.\n\nNow, they can read screenshots and diagrams from your repository and attachments with any self-hosted model that supports image input.\n\nIf your model cannot read images, an administrator can declare it in the AI Gateway model metadata file. The AI Gateway then replaces each image with a text notice before it sends the request.\n\nIn previous versions of GitLab, every flow session cloned your whole repository first, so large projects waited minutes before the agent started. Now you can include a copy of your repository in the image used for GitLab Duo Agent Platform sessions, so each session fetches only recent commits and starts significantly faster.\n\nPreviously, GitLab Duo Agentic Chat couldn’t look up current information. It answered only from the model’s training data and your GitLab context. Answers about things like the latest package version or a security advisory might be out of date.\n\nNow you can turn on web search in a conversation. Chat searches the web and reads pages to find current answers, so you don’t have to leave GitLab. Expand each search in the conversation to see the pages it found. Each search also creates an audit event.\n\nWeb search is off by default. On GitLab.com, a group Owner must allow it for the top-level group. On GitLab Self-Managed and GitLab Dedicated, an administrator must allow it for the instance.\n\nWorkspace subagents, previously released as an experiment, are now generally available.\n\nUse workspace subagents to keep the details of a task, like linter output, out of the context of the Developer Flow. Or use them for a separate agent with its own instructions and limited tools.\n\nDefine subagents as Markdown files in the `.agents/agents/` directory at the root of your repository.\nThe Developer Flow decides when to delegate a task to one, and receives only the result.\n\nIn GitLab Duo CLI 10.2.0 or later, use `/agents` to list subagents and ask the Developer Flow to use one.\nIn an issue or merge request comment, mention the Developer Flow and ask it to list subagents or delegate a task to one.\n\nIn previous versions of GitLab, a flow started only when a person or an event kicked it off. Without one, routine work like issue triage waited.\n\nNow scheduled automations give that work a dependable rhythm. Set the cadence once, from every 15 minutes to monthly, and a flow runs on its own at the times you choose. Triage new issues every morning, or check pipeline health every hour.\n\nEach scheduled automation runs as a dedicated service account with a scoped role, so its actions are attributed to that account, not a person.\n\nTriggers are now called automations. The redesigned **Automations** page brings scheduled\nand event-based automations together, with run history, success rates, and session details.\n\nTo try it, [turn on experiment and beta features](https://docs.gitlab.com/user/duo_agent_platform/turn_on_off/#turn-on-beta-and-experimental-features).\n\nIn previous versions of GitLab, Code Review Flow could only leave comments. It could not approve or block a merge.\n\nNow GitLab Duo can approve or request changes based on its review:\n\nTo use this feature, [turn on experiment and beta features](https://docs.gitlab.com/user/duo_agent_platform/turn_on_off/#turn-on-beta-and-experimental-features),\nthen turn on **GitLab Duo can approve and request changes** for your project, group, or instance.\n\nYou can now turn on security flows across a group the same way you turn on scanners. Apply an [automated triage and remediation profile](https://docs.gitlab.com/user/application_security/configuration/security_configuration_profiles/#automated-triage-and-remediation-profile) from the security inventory or the new **Automated Triage and Remediation** tab in Security configuration. The profile turns on the SAST, secret detection, and dependency scanning flows for the projects you select.\n\nPreviously, the profile was available only through the GraphQL API. Now you pick a preset and apply it in one action. For most flows:\n\nThe security inventory’s new **Capabilities** column shows where flows run, next to where scans run.\n\nAdministrators can now limit the number and size of pipeline variables that can exist in a single pipeline. This helps maintain the stability of your CI/CD infrastructure by blocking users from adding too many large pipeline variables to their pipelines.\n\nThe limits are disabled by default for GitLab Self-Managed and GitLab Dedicated. For GitLab.com, the limits are:\n\nYou can now choose how the Scanner Enablement Wizard applies a [scan profile](https://docs.gitlab.com/user/application_security/configuration/security_configuration_profiles/) to a group. Previously, Quick setup attached the profile to every project, so projects already scanning through CI/CD configuration or a policy ran the scan twice.\n\nQuick setup now asks how to apply it:\n\nYou can now create and customize dependency scanning profiles in the UI, the same way as secret detection and SAST profiles. Set how dependency scanning runs once for a group, then apply it to every project without editing `.gitlab-ci.yml` in each one.\n\nPreviously, you could customize a dependency scanning profile only with the GraphQL API. Now go to **Secure** > **Security configuration**, select the **Scanners** tab, and select **Manage profiles** on the dependency scanning row. Choose whether scans run on merge request pipelines, default branch pipelines, or both, then add settings to override, such as paths to exclude.\n\nApply the profile [from the security inventory](https://docs.gitlab.com/user/application_security/configuration/security_configuration_profiles/#apply-profiles-in-the-ui) or in the [advanced setup of the Scanner Enablement Wizard](https://docs.gitlab.com/user/application_security/configuration/scanner_enablement_wizard/#advanced-setup).\n\nIn previous versions of GitLab, your AI agents needed a separate MCP server connection to query GitLab Orbit.\nNow you can add the [Orbit toolset](https://docs.gitlab.com/user/model_context_protocol/mcp_server/#select-tool-groups-toolsets) to the GitLab MCP server, so your agents get Orbit context from the connection they already use.\n\nService Desk tickets are now their own work item type, so you can:\n\nPreviously, GitLab SAST had no coverage for Rust, so vulnerabilities in Rust code went undetected by your pipeline.\n\nGitLab Advanced SAST now scans Rust code. When you turn on GitLab Advanced SAST, Rust code in your repository is analyzed automatically, with no additional configuration. Incremental scanning is on by default, so only changed code is re-analyzed between pipelines.\n\nAdvanced SAST traces untrusted input across functions and files. Findings such as SQL injection,\ncommand injection, and path traversal include a code flow from source to sink. For the weakness\ntypes covered, see\n[additional languages CWE coverage](https://docs.gitlab.com/user/application_security/sast/advanced_sast_coverage/#additional-languages-cwe-coverage).\n\nRust analysis was added in GitLab 19.5 and is in beta.\n\nIn previous versions of GitLab, false positive detection ran automatically only on vulnerabilities in the default branch. Findings on release branches and tags had to be triaged by hand.\n\nNow, the SAST False Positive Detection Flow runs automatically on new Critical and High\nvulnerabilities in any tracked branch, and in tracked tags that run SAST jobs, so you can filter out noise before you merge or ship.\nIt runs when you turn on false positive detection in your project settings, or when you use the\n[automated triage and remediation profile](https://docs.gitlab.com/user/application_security/configuration/security_configuration_profiles/#automated-triage-and-remediation-profile) set to run automatically.\n\n[Secret false positive detection](https://docs.gitlab.com/user/application_security/vulnerabilities/secret_false_positive_detection/#tracked-branches-and-tags)\non tracked branches and tags is available on GitLab.com in this release.\n\nIn previous versions of GitLab, merge request approval policies treated malicious packages like any other vulnerability. If a finding was dismissed or fell outside a rule’s severity filters, a package built to cause harm could still be merged.\n\nNow you can add a malware rule to your merge request approval policies. When dependency scanning detects a package listed in the GitLab malware advisories, the merge request requires approval, regardless of the finding’s severity or state. The malware rule works alongside your existing scan finding rules, so you can add it to the policies you already have without rebuilding them.\n\nTo use the malware rule, your project must run\n[dependency scanning by using SBOM](https://docs.gitlab.com/user/application_security/dependency_scanning/dependency_scanning_sbom/#turn-on-dependency-scanning).\nMalware advisories cover npm, PyPI, Maven, Go, NuGet, Cargo, and RubyGems packages.\n\nBusiness Logic Security Analyzer is a GitLab Duo Agent Platform flow that reasons about what your code is meant to enforce. It finds vulnerabilities that pattern-based scanners miss:\n\nFindings appear in the vulnerability report and the merge request security widget.\n\nUntil now, the Business Logic Security Analyzer was available only on GitLab.com. Organizations on GitLab Self-Managed and GitLab Dedicated, which often have the strictest requirements for where code is analyzed, couldn’t use it.\n\nNow it runs on both:\n\nTo get started, contact your account team.\n\nEach scan consumes GitLab Credits. For setup, see\n[Business logic scanning](https://docs.gitlab.com/user/application_security/business_logic_scanning/).\n\nDependency scanning auto-remediation opens merge requests that bump the versions of vulnerable dependencies. Previously, Swift projects were not supported, and the job ran only on untagged runners. Merge requests went to a random maintainer, and each project could have at most 10 open at once.\n\nIn GitLab 19.6:\n\n`Package.resolved` file now get\nauto-remediation merge requests.\nFor details, see [dependency scanning auto-remediation](https://docs.gitlab.com/user/application_security/remediate/dependency_scanning_auto_remediation/).\n\nIn earlier versions of GitLab, you could not filter the group vulnerability report by application, business unit, or team. Now, you can filter by the security attributes assigned to your projects.\n\nEach security category, such as **Business impact** or **Application**, appears as its own filter.\nSelect **is one of** to include attributes or **is not one of** to exclude them.\nUse these filters to review every vulnerability for one application, even when it spans many projects.\n\nThis filter requires [advanced vulnerability management](https://docs.gitlab.com/user/application_security/vulnerability_report/#advanced-vulnerability-management).\n\nWe’re also releasing GitLab Runner 19.6 today! GitLab Runner is the highly-scalable build agent that runs your CI/CD jobs and sends the results back to a GitLab instance. GitLab Runner works in conjunction with GitLab CI/CD, the open-source continuous integration service included with GitLab.\n\n**What’s New**\n\n**Bug Fixes**\n\n`gitlab_secrets_manager source group with path 'gitlab-org' not found``--docker-pull-policy` changed and is inconsistent with other flags`FF_CONCRETE` jobs now use the configured shell. Jobs with `shell = \"sh\"` that use bash-only syntax, such as arrays or `[[ ]]`, might now fail.` FF_USE_GIT_NATIVE_CLONE` clones the mutable merge request merge ref instead of the pipeline SHA, breaking merged results pipelines when the target branch advances`file name too long` error in GitLab Runner 19.x`FF_NETWORK_PER_BUILD` is enabled`/proc` mount no longer fail with the Kubernetes executor. The runner now turns on user namespaces for these jobs, which Kubernetes requires for an unmasked mount.`SIGQUIT`) cancels in-flight jobs and job requests, orphaning jobs that GitLab already assigned`FF_USE_PARALLEL_CACHE_TRANSFER` or `FF_USE_PARALLEL_ARTIFACT_TRANSFER` enabled, cache and artifact downloads silently corrupt files when a server or proxy ignores `Range` requests`GIT_STRATEGY` is `fetch`` pod is being deleted``FF_KUBERNETES_HONOR_ENTRYPOINT=true`, the Kubernetes executor ignores service commands` id -u` check for runners without `umask` breaks when the image sets an `ENTRYPOINT`\nThe list of all changes is in the GitLab Runner [CHANGELOG](https://gitlab.com/gitlab-org/gitlab-runner/blob/19-6-stable/CHANGELOG.md).\n\nIn previous versions of GitLab, you could set and view vulnerability due dates only through the GraphQL API.\n\nNow, you can view and set a vulnerability’s due date from the vulnerability details page.\nWhile the vulnerability has the **Needs triage** or **Confirmed** status, GitLab labels the date:\n\nIn previous versions of GitLab, you couldn’t see how quickly your teams fix vulnerabilities over time.\n\nProject, group, and organization security dashboards now include a mean time to remediate (MTTR) chart. It shows the weekly MTTR over the last 12 weeks, with the number of vulnerabilities remediated each week. To find where remediation slows down, group the chart by severity or report type. Compare the trend against your remediation SLAs.\n\nMTTR counts recently detected vulnerabilities that were resolved or are no longer detected. Dismissed vulnerabilities are excluded.\n\nGitLab 19.6 updates the NGINX version bundled in the Linux package from 1.31.4 to 1.31.7. This update fixes five security issues: CVE-2026-15176, CVE-2026-93479, CVE-2026-95101, CVE-2026-97629, and CVE-2026-97205.\n\nThe configuration the Linux package generates is not affected. If you added custom NGINX configuration, two upstream changes might break it:\n\nBefore you upgrade, review your custom configuration. For more information, see the\n[upgrade notes](https://docs.gitlab.com/update/versions/gitlab_19_changes/#nginx-update-might-break-custom-nginx-configuration).\n\nThe GitLab Credits dashboard summary cards and chart include non-billable usage, such as usage covered by a promotion. Previously, the per-user and per-product breakdowns counted billable usage only, so they didn’t add up to the totals. The user list also skipped anyone with only non-billable usage.\n\nGitLab now counts non-billable usage in:\n\nYou can now set a time range for a merge schedule to merge automatically.\nIn previous versions of GitLab, the merge schedule supported only the **After** option,\nwhich prevented a merge before the date you set.\nNow you can select **Between** to define a window: the merge request merges only when the\ncurrent time falls between the two dates you set.\n\nThank you to [Niklas van Schrick](https://gitlab.com/Taucher2003) for these contributions\n([MR 1](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/258024)\n[MR 2](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/258025)\n[MR 3](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/262275))!\n\nAdministrators can now configure a domain allowlist for the GitLab for Jira Cloud app. Jira sites that are not on the allowlist can’t install the app. Upgrading or reinstalling the app also fails for those sites, so add every Jira site that is already connected before you save the allowlist. An empty allowlist allows all domains, preserving the previous behavior. The allowlist applies only to the Connect version of the app, not the Forge version.\n\nThank you to [Roger Meier](https://gitlab.com/bufferoverflow) for this contribution\n([MR](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/256324))!\n\nThe percent rollout and percent-of-users feature flag strategies now accept a custom Group ID. Strategies across flags that share the same Group ID and stickiness value select the same user set at the same percentage, letting you coordinate rollout across multiple flags. Change the Group ID to target an independent set of users instead.\n\nThank you to [Mehdi Youb](https://gitlab.com/myoub) for this contribution\n([MR](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/263973))!\n\nIn previous versions of GitLab, the GitLab MCP server could read a Pages site’s source, but not\nthe site your pipeline built from it. With the `get_pages_content` tool, the GitLab MCP server reads the published content of a Pages URL for you,\nincluding private Pages sites you have permission to view.\n\nThank you to [Mawueli](https://gitlab.com/mawueli) for this contribution!\n\nIn previous versions of GitLab, your AI agents needed a separate request to learn which work item types a project supports.\nNow your agents get a project’s work item types and the fields each type supports in the same request as the [project details](https://docs.gitlab.com/user/model_context_protocol/mcp_server_tools/#get_project).\n\nThank you to [Aditya Parida](https://gitlab.com/adityaswarupparida) for this contribution!\n\nGitLab Workspaces now run native images on `linux/arm64` Kubernetes nodes, including\nApple M-series hardware. Previously, arm64 nodes ran the amd64 image under emulation.\nGitLab pulls the correct image variant for each node automatically.\nCustom devfile images must support the architecture of the node where the workspace runs.\n\nThank you to [Miguel Sacristan](https://gitlab.com/miguel.sacristan1) for this contribution\n([MR](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/264361))!\n\nYou can now add the `archive_subgroup` ability to custom roles. Users with this ability\ncan archive and unarchive subgroups without needing the broader Owner role, which also\ngrants the ability to delete a group. The ability appears as **Archive subgroup** in the\ncustom role editor.\n\nThank you to [Jimmy Spagnola](https://gitlab.com/jspagnola) for this contribution\n([MR](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/257968))!\n\nThe board list REST API endpoints now accept a `status_id` parameter, letting you create\nboard columns that filter issues by a custom status. The API response includes the full\nstatus object so your automation can read back the status it set.\nIn previous versions of GitLab, status-based lists could not be created through the REST API.\n\nThank you to [Daniel Ziegenberg](https://gitlab.com/ziegenberg) for this contribution\n([MR](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/260752))!\n\nCommunity contributors added 10 new capabilities to the GitLab MCP server for 19.6:\n\nGitLab now detects `CITATION.cff` files in repositories.\nProjects with a `CITATION.cff` file show a link to the file in the project overview.\nProjects without one show an option to add a `CITATION.cff` file.\n`CITATION.cff` is an open file format that describes how to cite software.\n\nThank you to [rajasekhar](https://gitlab.com/rajasekharponakala) for this contribution\n([MR](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/254371))!\n\nCommunity contributors made many improvements, including:\n\n`needs:` dependencies now appear in the correct column in the pipeline dependency view. (", "url": "https://wpnews.pro/news/gitlab-19-6-release-notes", "canonical_source": "https://docs.gitlab.com/releases/19/gitlab-19-6-released/", "published_at": "2026-11-19 00:00:00+00:00", "updated_at": "2026-11-19 17:43:10.861569+00:00", "lang": "en", "topics": ["ai-agents", "agent-protocols", "ai-policy", "developer-tools", "ai-products"], "entities": ["GitLab", "GitLab 19.6", "GitLab MCP server", "GitLab Duo", "GitLab Duo Agentic Chat", "Developer Flow", "Rapid Diffs", "Slack"], "also_reported_by": [], "takeaway": "GitLab 19.6 logs every MCP server agent action as an audit event tagged as MCP-originated, and lets top-level group Owners set per-tool permissions that GitLab enforces for every connecting client.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/gitlab-19-6-release-notes", "markdown": "https://wpnews.pro/news/gitlab-19-6-release-notes.md", "text": "https://wpnews.pro/news/gitlab-19-6-release-notes.txt", "jsonld": "https://wpnews.pro/news/gitlab-19-6-release-notes.jsonld"}}