The Cyber Toolchain — issue 079, September 19, 2026

npm granular access tokensaccess tokensA credential that authorizes a client to use a service or API. It usually has limited permissions and a lifetime; anyone holding it may be able to use those permissions. gain a Read and write (stage only) permission option, letting automated workflows stage package versions without direct-publish rights.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ The new npm stage publish command replaces npm publish for these workflows; staged versions require a maintainer to review and approve with 2FA before release, and npm rejects direct npm publish attempts even when bypass-2FA is configured. Stage-only tokens retain other write permissions such as moving dist-tags and deprecating versions. Staged publishing requires npm CLI 11.15.0+ and Node.js 22.14.0+.

$ npm stage publish

The Copilot code review overview comment now groups findings into Open, Resolved since last review, and Previously missed sections, each with severity and a link to the inline comment; every review comment gets a title reused in the overview issue list. Copilot generates a commit title and optional description automatically when committing an eligible batch of Copilot suggestions, including batches that mix in non-Copilot comments. Reviews also now resolve addressed comments during subsequent passes, suggest commit messages when applying suggestions, support shell tools to validate changes, and combine findings from multiple agents in Lite reviews.

⤢ ⤢ ▶ Watch video

Auto model selection gains efficiency, balance, and intelligence tiers, letting users control how Copilot weighs cost, quality, and response time; rolling out in VS Code, Copilot CLI, and the Copilot app.

The Restrict code coverage repository ruleset option, previously UI-only, is now exposed via the REST API, enabling programmatic create, update, and read of minimum line coverage percentage and maximum tolerable coverage drop rules for pull requests.

A Sentry canvas in the Copilot app lets users move from crash report to code fix: review errors, stack traces, and related context, then work with Copilot to investigate, validate a fix, and prepare a pull request.

⤢

Agents in the VS Code Agents window can now run with a project's own tools and dependencies via local Dev ContainerDev ContainerA Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. support, requiring Docker and a supported Dev Container configuration in the repo.

Copilot Business and Enterprise users on usage-based billing can request higher AI credit limits, which owners and billing managers can approve, adjust, or deny in settings; this is now generally available.

An opt-in, preview feature in VS Code automatically marks Agents window sessions as Done once all their pull requests merge, with optional deletion after a configurable grace period.

A Triage with AI control in the Risk Orchestration side panel evaluates eligible SAST findings for reachability, exploitability, and existing sanitization, setting state to Confirmed or Proposed Not Exploitable with AI attribution; a Remediate with AI control generates a suggested code fix surfaced in the Remediation > AI Remediation tab with a what/why/how explanation, and can open a pull request for GitHub Code Repository Integration projects.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Automated AI Triage can run on every completed scan, configurable by project, branch, scanner, risk status, and severity with administrator-controlled overrides, and also runs automatically on pull requests with direct remediation requests via PR comments for GitHub projects. Triage state changes propagate retroactively to identical SAST findings across earlier scans (matched by Similarity ID or Attack Vector ID) and forward into subsequent SCA scans (matched by package version and package manager, without retroactive application). An AI Triage & Remediation dashboard reports total triage and remediation actions, unique developers, estimated developer time saved, activity trends, triage outcomes, remediation results, and top projects, filterable by resource, branch, scanner, time range, tags, and groups. Both actions draw on Checkmarx Credits, with administrator visibility into burn rate and projected depletion; the AI Triage and AI Remediation agents were later upgraded for improved accuracy and performance, and new permissions were added for Analytics — AI Triage and Remediation.

Revamped SCA Scan Reports add a Package Licenses section (CSV, XML, JSON) with per-package license data and a Legal Risks section (CSV, XML, JSON) with license usage risk data, available via the UI and REST APIsREST APIsA convention for building HTTP-based interfaces where resources are addressed by URL and manipulated with standard verbs like GET, POST, and DELETE, giving cyber tools a predictable way to integrate with external services.. Reports also gain license Permissions, Limitations, and Conditions data, plus an option to filter results to show only licenses marked as Effective.

Adds application-level SBOM report generation via the UI at Workspace > Projects, in addition to existing project-level SBOM support, with an option to exclude all licenses not designated as Effective for a package when generating SBOM reports.

The Software Supply Chain Security (SCS) module adds a secrets scanner that identifies 170+ types of login credentials, access tokensaccess tokensA credential that authorizes a client to use a service or API. It usually has limited permissions and a lifetime; anyone holding it may be able to use those permissions., encryption keys, API keys, SSH keys, and webhook URLs across repositories. It was later extended with Custom Secret Detection Rules, letting organizations define, manage, and enforce tailored secret detection logic across codebases and workflows, and with Secret Detection added as a selectable scan type when generating reports.

Supports GitHub App-based authentication for code repository integrations, enabling short-lived scoped tokens, automatic token rotation, and granular permission control as a more secure alternative to Personal Access Tokens and OAuth apps.

Enables changing a risk's state and severity directly through the MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. Server using natural language, without opening Checkmarx One — supports marking risks Not Exploitable or Confirmed across all major scan engines.

Checkmarx One Policy Management gains new policy conditions per scanner: specialized Container Security conditions usable in Condition Groups, and a new status condition for the SCA scanner that triggers policies on New or Recurrent vulnerabilities while filtering out dev and test dependencies.

Cloud Insights can now connect to CrowdStrike's API endpoints via a setup wizard launched from the Cloud Insights page in the Checkmarx One UI, and Sysdig integration onboarding is streamlined by allowing account connection via a Spotlight token.

Enables running a full DAST workflow directly from the CLI via a REST-API-driven script that handles authentication, session creation, scanning, and results retrieval.

Introduces Attack Vector ID, a flow-aware finding grouping that encodes programming language, vulnerability query, and exact dataflow path — providing finer-grained separation than the default Similarity ID.

Package results now distinguish Not Used (no usage detected) from Not Calculated (usage could not be determined), with a reason provided for Not Calculated entries (e.g., unsupported language). Risk results now similarly distinguish Not Found (no Exploitable Path detected) from Not Calculated (exploitability could not be determined), with a reason such as unsupported transitive dependencies.

Adds an Insights engine to DAST that flags conditions affecting scan accuracy — such as authentication failures, blocking mechanisms, or repeated server errors — as non-vulnerability findings; a High-level Insight stops the scan and marks it failed.

Adds SCA dependency scope tracking for Direct dependencies — packages called directly by the application — and adds Java SBTSBTA build tool for Scala and Java projects, maintained by the open-source community, that handles compilation, dependency resolution, and packaging using an incremental build model to speed up iterative development. support (build.sbt, any .sbt file) for SCA dependency scanning.

Adds the lakebase_vector Postgres extension with the lakebase_ann index type for approximate nearest-neighbor vector search, drop-in compatible with pgvector, scaling to over 1 billion vectors on a single index with index builds up to 50-100x faster than HNSWHNSWHierarchical Navigable Small World, a graph index that finds approximate nearest neighbours among millions of vectors in milliseconds. Most vector databases use it by default..⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Adds the lakebase_text extension with the lakebase_bm25 index type for BM25BM25The classic keyword-ranking formula behind traditional search engines. Hybrid search combines it with embedding similarity so exact terms and meaning both count. full-text search over standard tsvector types, adding BM25 ranking and top-K pushdown not available in PostgreSQL's native GIN index. The two can be combined for hybrid searchhybrid searchCombining keyword search with semantic or vector search. The results use both exact term matches and similarity in meaning. returning semantic and exact-term matches in a single ranked list, and indexes are storage-backed, available immediately after scale-to-zero with no warmup in serverless Lakebase projects.

⤢

$ -- After enabling Lakebase Search in project settings:
CREATE EXTENSION lakebase_vector;
CREATE EXTENSION lakebase_text;

-- Create a vector index (lakebase_ann) on an embeddings column
CREATE INDEX ON documents USING lakebase_ann (embedding);

-- Create a BM25 full-text index (lakebase_bm25) on a text column
CREATE INDEX ON documents USING lakebase_bm25 (body_tsvector);

Discover and non-managed dashboards can now run ES|QLES|QLElasticsearch Query Language, maintained by Elastic, a pipe-based query language for filtering, transforming, and aggregating data stored in Elasticsearch, giving cyber tools a concise syntax for log and event analysis. queries and rollups across all linked projects for broad hunting sweeps, using an unqualified FROM that automatically includes newly linked projects with no query edits.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ On Security managed pages, a CPS scope selector shows in-scope projects as read-only, while admins configure default scope per space.

⤢ ⤢

FROM logs-* METADATA _index
| WHERE user.name == "<username>"
| KEEP _index, @timestamp, event.action, source.ip, user.name
| SORT @timestamp DESC

FROM .alerts-security.alerts-* METADATA _index
| STATS count = COUNT(*) BY _index, kibana.alert.severity
| KEEP _index, kibana.alert.severity, count
| SORT count DESC

A new capability lets one origin project run its roughly 2,100 detection rules once and have alerts from up to 100 linked projects consolidated into a single queue on the origin, eliminating per-project rule duplication and the need for analysts to hop between project contexts.

⤢

Timeline search, the Alert flyout, and the Analyzer/Correlations views now extend (read-only) across linked projects, so analysts can follow an attack chain — such as failed console logins followed by AttachUserPolicy and CreateAccessKey — without leaving the origin project.

⤢

Adds threat containment actions — session revocation, model-provider API key deactivation, device isolation, and an agent kill switch (ending an agent's process) — executed through existing EDREDREndpoint detection and response, monitoring computers for suspicious behavior and providing investigation and response controls, such as isolating a device., identity, and model-provider controls, with autonomy tunable per action from analyst-approved to fully autonomous.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

Adds a continuous, agentless inventory of AI apps connected to company accounts, coding agents (Claude Code, Cursor), hosted agents (custom GPTs), MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers, skills, and IDE plugins, tied to the people and permissions behind them — with no new endpoint agent, browser extension, or gateway to deploy.

Adds threat detection that correlates agent actions, agentic-app activity, and prompts with human identities, endpoint data, file access, and code context to detect misuse, sensitive-data exposure, and active threats — including chains of individually legitimate-looking actions.

Adds per-agent and per-skill risk scoring for every agent, skill, MCP server, and plugin, paired with a recommended fix, covering exposures such as excessive permissions and unvetted MCP servers.

Windows Managed OS Library Items now report per-device patch compliance with Pass, Update past due, or Error status.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Expanding a device row reveals its current Windows 11 build versus the required build, plus eligibility, install enforcement, and restart enforcement dates.

Adds MSP Billing in the Partner Portal, giving U.S.-based managed service providers daily, per-tenant spend visibility and the ability to update payment methods, billing information, and view past invoices.

⤢

DELETE /api/v1/secrets/{vault_id} now deletes only the secret and no longer revokes its credential; use the new POST /api/v1/app-secrets-admin/revoke endpoint to queue a credential revoke separately from secret deletion.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Service SDKs rename Revoke to Delete to match.

GET /api/v1/tasks/{task_id}/spend-remedy-status has been removed — migrate to the task APIs. JS/TS clients must handle spend-control revision values as strings, and integrations with a closed list of task action types must handle the new TYPE_SPEND_REMEDY = 7.

Super Admins can now ask C1AI to diagnose a specific stuck, failed, or waiting task; the assistant traces the task's approval history, provisioning logs, access-profile origin, and connector handoff, then recommends whether to approve, deny, or retry it. For example, asking 'why is task #4821 stuck?' returns the approval and provisioning state, access profile origin, app-account access, and connector handoff details.

⤢ ⤢ ⤢

Resource-type filters in Grant Found, Grant Deleted, and revoke criteria now use a faceted, searchable picker with real counts and name disambiguation. A revoke criteria scoped only by a legacy resource-type name now fails the step instead of silently matching every app sharing that name — re-scope any automation relying on name-only matching.

Findings rules can now automatically set severity, add or remove annotations, or assign every matching finding to a specific person as it arrives, via a new Set assignee step in Transformation rules; a person's own manual assignment always overrides a later rule.

The public REST API now exposes classifiers, tool gates, classifier bindings, templates, and the agent classifier policy for managing AI governance programmatically.

Auth0 Actions Modules is now GA, letting you create, manage, and share reusable code across multiple Actions and Triggers within an Auth0 Tenant.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Modules support independent secrets and dependencies decoupled from the Actions that import them, enable cross-trigger reuse of custom functions by importing a single Module into any Action regardless of which Trigger it is bound to, and improve execution performance by moving expensive initialization logic into a Module that is initialized once and reused across Action invocations.

Employees using unapproved password managers can now be nudged in bulk: filter the accounts inventory by password manager observed at login, multi-select the flagged accounts, and send the Use approved password manager nudge.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Each employee receives a single batched nudge listing all their affected accounts instead of one notification per account.

The Terraform module adds Microsoft Teams integration configured via MS_TEAMS_APP_CLIENT_ID, MS_TEAMS_RUNLAYER_TENANT_ID, MS_TEAMS_AZURE_SUBSCRIPTION_ID, MS_TEAMS_AZURE_RESOURCE_GROUP, and MS_TEAMS_APP_URL, with the client secret supplied via ms_teams_app_client_secret (recommended via TF_VAR_ms_teams_app_client_secret); the integration is off unless all five values are set.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Provisioning creates a per-agent Entra app registration and Azure Bot Service resource, requiring a multi-tenant app registration with Graph permissions Application.ReadWrite.All, AppCatalog.ReadWrite.All, and TeamsAppInstallation.ReadWriteForUser.All (admin-consented), a resource group granting Azure Bot Service Contributor, and the Microsoft.BotService provider registered on the subscription. The Redirect URI https://<domain_name>/api/v1/integrations/ms-teams/admin-consent-callback must be registered under Authentication → Web on the app registration or admin consent fails with AADSTS500113. Breaking: the domain name set at agent provisioning time is effectively immutable — each agent's Azure Bot resource stores the messaging endpoint captured at creation and never rewrites it, so changing the domain orphans existing bots until each is deleted and recreated.

The Runlayer Operator gains six environment variables to enable Microsoft Teams bot provisioning: MS_TEAMS_APP_CLIENT_ID, MS_TEAMS_APP_CLIENT_SECRET, MS_TEAMS_RUNLAYER_TENANT_ID, MS_TEAMS_AZURE_SUBSCRIPTION_ID, MS_TEAMS_AZURE_RESOURCE_GROUP, and MS_TEAMS_APP_URL. MS_TEAMS_APP_URL sets the public HTTPSHTTPSHTTP carried over an encrypted TLS connection. It protects traffic in transit and lets a browser verify the server's certificate. base for the Bot Framework messaging endpoint and admin-consent callback, defaults to the operator's own host, and must match the Redirect URI registered on the Entra app or admin consent fails with AADSTS500113; Teams provisioning stays off unless all six variables are set.

MS_TEAMS_APP_CLIENT_ID=<entra-app-client-id>
MS_TEAMS_APP_CLIENT_SECRET=<entra-app-client-secret>
MS_TEAMS_RUNLAYER_TENANT_ID=<entra-tenant-id>
MS_TEAMS_AZURE_SUBSCRIPTION_ID=<azure-subscription-id>
MS_TEAMS_AZURE_RESOURCE_GROUP=<resource-group-name>
MS_TEAMS_APP_URL=https://runlayer.example.com

The Agents Sessions list now shows Slack messages the agent chose to skip alongside runs, with search covering both skipped messages and runs; video artifacts play inline in the run view with the artifact size limit raised to 25 MB; and clicking a trace header with a single tool call now opens its details directly.