This Week's Highlights — issue 013, September 25, 2026
The best tooling updates that shipped this week.
The week in numbers · Sep 18–24
Changes per week, last 12 weeks · Sep 18–24
Changes we read in that week's dailies, including docs and API changes. A gap means no daily was published.
Category moves this month
Every feature release from the 1325 tools on our watchlist goes into the daily newsletter. Once a week we read the whole field side by side and choose our top picks, judged on two questions: how deep and complete is the single best capability in the release, and how much it changes what you can actually do.
A tool is judged on everything it shipped that week, so a project that releases daily gets credit for the sum — and still only takes one slot. We would rather run a short list than a padded one.
Focus: What features shipped and what problems they solve, plus launches, packaging and where the market is moving.
Leaves out: Command-line flags, API-only deltas, usage examples, and releases that are only refinements.
Focus: New security controls, new APIs, and product improvements you can put to work, plus anything an upgrade breaks.
Leaves out: Launch and marketing news, governance, pricing, and documentation changes with no operational effect.
Focus: Dashboards, reporting, and anything you can quantify: metrics, coverage, risk scores, compliance impact, vendor moves.
Leaves out: Settings, flags, API and command-line detail, library bumps, and minor releases with no reporting or posture angle.
Focus: API, SDK and MCP changes, new libraries, breaking changes in what you import, and cyber tooling.
Leaves out: Governance, SOC operations, and vendor launches with no developer surface.
This issue came out before editions were ranked, so it shows the Standard edition in issue order.
The picks in full
Trusted by 7M+ developers, SonarQube is the AI code review and code verification platform that catches quality and security issues before they reach production.
SonarQube Hunter Agent adds AI-driven detection of business-logic, broken access control, and authentication/session flaws that static analysis and SCA usually miss. It lands findings as normal SonarQube issues with severity, file, line, triage, scheduling, and an auditable evidence trail, making it immediately usable in existing AppSec workflows.
SonarQube Cloud added first-class Rust analysis, including ClippyClippyA Rust crate, maintained by the Rust project, that adds extra lint checks to catch common mistakes, non-idiomatic code, and suspicious patterns before they reach production.-based rules, Rust metrics, coverage, duplication detection, quality gates, and scanner properties for CargoCargoRust’s package manager and build tool, maintained by the Rust project. Cyber tools use it to fetch dependencies, compile crates, run tests, and produce reproducible builds. and Clippy control.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ The window also brought Hunter Agent AI verification, native R analysis, and zero-configuration GitLab automatic analysis.
what's the number?
Each feature carries 0–100 for how completely the vendor documented it — not how big or important the work is. A major capability described in eight words scores low, and that is the finding.
- depth0–40 what it does and how it works or what changed
- specificity0–30 names real surfaces — APIs, flags, formats, limits, numbers
- actionability0–30 enough to go use it — a named endpoint, flag, or config key tops this; a UI path is a starting point
Code and config rank first by construction: a documented endpoint, flag or config key scores at the top of specificity and actionability, so it sorts above a feature described only in prose. Under 60 the notes go thin — everything below the “thinner coverage” line is thinner documentation, not smaller work. Hover any meter for that feature's three sub-scores.
SonarQube Cloud now supports Rust as a first-class language with 122 curated Clippy rules enforced through quality profiles, plus library- and framework-aware Rust rules for tokio async handling, serde serialization, and actix-web misuses.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ It calculates Cognitive Complexity and Cyclomatic Complexity, imports LCOVLCOVAn open-source front end to GCC's gcov coverage tool that collects line and branch hit counts from instrumented C/C++ code and renders them as HTML reports, giving cyber tools a standard way to measure test coverage. or Cobertura coverage into the issue view, detects duplicate Rust code with CPD, analyzes Rust in the same multi-language project view as C/C++, Java, and JavaScript, and supports quality gate enforcement for Rust with pass/fail thresholds scoped to new code for CI blocking. The scanner adds sonar.rust.cargo.manifestPaths for non-default Cargo.toml locations and sonar.rust.clippy.enabled, which can be set to false to import an external Clippy JSON report instead of running cargo clippy automatically.
$ sonar-scanner -Dsonar.rust.cargo.manifestPaths=services/myservice/Cargo.toml
$ sonar-scanner -Dsonar.rust.clippy.enabled=false
Hunter Agent adds an AI-driven verification layer that detects business-logic, broken access control, and authentication/session vulnerabilities through structured multi-stage playbooks. Findings appear as standard SonarQube issues with severity, file, and line in the existing triage, assignment, and tracking workflow, are summarized on a dedicated Hunter Agent page, can run manually or daily against long-lived branches, and produce an auditable evidence trail mapped to DORA, NIS2NIS2EU directive that expands mandatory cybersecurity requirements to more sectors and member states than its predecessor, giving cyber tools a compliance framework to map controls and reporting obligations against., PCI DSSPCI DSSThe Payment Card Industry Data Security Standard, security requirements for protecting payment account data. It covers organizations that store, process or transmit that data and relevant service providers. 4.0.1, HIPAAHIPAAA US federal law enacted in 1996 that sets privacy and security requirements for protected health information, giving cyber tools a compliance baseline to enforce or audit against., GDPRGDPREU regulation enacted in 2018 that governs how organizations collect, store, and process personal data of EU residents, giving cyber tools legal obligations around data handling and breach notification., EU CRAEU CRAEU Cyber Resilience Act, a European Union regulation requiring manufacturers and publishers of products with digital elements to meet mandatory cybersecurity standards before selling into the EU market., NYDFS, and FDA 21 CFR Part 11 requirements.
SonarQube Cloud extends zero-configuration automatic analysis, previously available for GitHub and Azure DevOps, to GitLab. It scans eligible GitLab repositories without .gitlab-ci.yml edits, covers 20+ languages including C#, Java, and Python, and starts analysis after users link GitLab, import repositories, and select eligible repositories.
SonarQube adds native R support on Cloud and Server, analyzing R source files and embedded R code inside R Markdown documents with 82 rules covering key sections of the lintr rule set plus Sonar-developed rules. It imports lintr problem reports, adds Cobertura test coverage import, extends secrets detection, code metrics, syntax highlighting, and copy-paste duplication analysis to R, and applies existing quality profiles, quality gates, PR decoration, and branch analysis to R codebases.
Palo Alto Networks is a cybersecurity platform that provides advanced firewalls, cloud security, and AI-driven threat detection to protect networks and enterprises.
Unit 42 Continuous Frontier AI Defense adds always-on agentic offensive testing that continuously finds, validates, and remediates exposures across web apps, APIs, cloud infrastructure, source repositories, and network assets. The Advanced Adversary Simulation validates end-to-end attack paths, closing a practical gap between asset scanning and proven exploitability.
Palo Alto Networks launched Unit 42 Continuous Frontier AI Defense, an always-on agentic offensive cyber service for continuously finding, validating, and remediating enterprise exposures.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ The service includes continuous estate testing, multi-model AI routing, and adversary simulation across web apps, APIs, cloud infrastructure, source code repositories, and network assets.
what's the number?
Each feature carries 0–100 for how completely the vendor documented it — not how big or important the work is. A major capability described in eight words scores low, and that is the finding.
- depth0–40 what it does and how it works or what changed
- specificity0–30 names real surfaces — APIs, flags, formats, limits, numbers
- actionability0–30 enough to go use it — a named endpoint, flag, or config key tops this; a UI path is a starting point
Code and config rank first by construction: a documented endpoint, flag or config key scores at the top of specificity and actionability, so it sorts above a feature described only in prose. Under 60 the notes go thin — everything below the “thinner coverage” line is thinner documentation, not smaller work. Hover any meter for that feature's three sub-scores.
The proprietary Multi-Model AI Harness routes offensive security tasks to the model best suited for each job.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Palo Alto Networks says this addresses findings that no single model catches more than 40% of vulnerabilities and that leading models share less than 10% overlap in identified exposures.
Unit 42 Continuous Frontier AI Defense is a new always-on agentic offensive security service that continuously finds, validates, and remediates enterprise exposures. It uses proprietary multi-model AI harnesses with Anthropic Claude Mythos 5 and OpenAI GPT-5.6-Cyber.
The Continuous Testing Engine performs a full-estate baseline scan followed by always-on testing as an organization's environment changes. Coverage includes web apps, APIs, cloud infrastructure, source code repositories, and network assets.
Advanced Adversary Simulation validates end-to-end attack pathsattack pathsA possible sequence of steps an attacker could take to reach a valuable target. It connects weaknesses, permissions and reachable systems rather than treating each finding separately. to prove real-world exploitability. It works across first- and third-party web apps, APIs, cloud infrastructure, source code repositories, and network assets.
Semgrep is a static analysis tool that scans code for bugs, security vulnerabilities, and policy violations using customizable rules.
Semgrep Supply Chain adds zero-day malware detection that rescans customer repositories within minutes of a supply-chain incident and reports affected repos, projects, findings, and scan progress. The new org-wide response policies and Malware Firewall extend that from detection into automated alerting and pre-install blocking.
Semgrep Supply Chain added zero-dayzero-dayA vulnerability for which defenders have had no prior opportunity to apply a fix, often because it was previously unknown. A zero-day exploit takes advantage of that gap. malware detection and response automation, including org-wide policies, incident rescans, early blast-radius notifications, and a private-beta Malware Firewall.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Semgrep also added CSV export for Agentic Workflows issues from the Issues page and public API.
what's the number?
Each feature carries 0–100 for how completely the vendor documented it — not how big or important the work is. A major capability described in eight words scores low, and that is the finding.
- depth0–40 what it does and how it works or what changed
- specificity0–30 names real surfaces — APIs, flags, formats, limits, numbers
- actionability0–30 enough to go use it — a named endpoint, flag, or config key tops this; a UI path is a starting point
Code and config rank first by construction: a documented endpoint, flag or config key scores at the top of specificity and actionability, so it sorts above a feature described only in prose. Under 60 the notes go thin — everything below the “thinner coverage” line is thinner documentation, not smaller work. Hover any meter for that feature's three sub-scores.
Semgrep added a Malware Firewall in private beta, embedded in Semgrep Guardian via Agent admin.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ It intercepts dependency requests at the network level between the package manager and public registries and blocks malicious packages before they reach developer machines, with no per-device install.
Semgrep Supply Chain now supports org-wide policies that trigger automated responses when a malware finding is generated, including opening a Jira ticket, sending a Slack alert, or calling a webhook with no human in the loophuman in the loopHuman in the loop, involving a person in an otherwise automated workflow. The person can review, approve, correct or supply information before work continues..
Customers can opt in to early notifications that proactively report their blast radius via Slack or webhooks when a new supply chain incident is detected. Notifications cover affected repos, projects, findings, and scan progress.
Semgrep added zero-day incident detection that rescans all customer repositories with AI-assisted, expert-verified advisory rules within minutes of a supply chain incident. Findings are posted to the Semgrep dashboard, organized by advisories and rolled up into incidents.
Semgrep now supports exporting Agentic Workflows issues as CSV from the Issues page and through the public API.
Docker Sandboxes adds Cloud Sandboxes and an OCI-native Kit format that lets teams run coding agents in managed microVMs with declared network and credential authority. Kits, proxy-managed secrets, network policies, and authority-diff gating make agent execution governable rather than just isolated.
Docker Sandboxes launched managed Cloud Sandboxes and the v3 OCIOCIOpen Container Initiative standard, maintained by the Linux Foundation, defining the image format and runtime spec that container tools share so images built once run anywhere OCI-compliant. Kit format for packaging sandbox authority.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ The window also added model-provider configuration, 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. transport controls, network-policy hardening, richer CLI/JSON surfaces, and several breaking CLI behavior changes.
what's the number?
Each feature carries 0–100 for how completely the vendor documented it — not how big or important the work is. A major capability described in eight words scores low, and that is the finding.
- depth0–40 what it does and how it works or what changed
- specificity0–30 names real surfaces — APIs, flags, formats, limits, numbers
- actionability0–30 enough to go use it — a named endpoint, flag, or config key tops this; a UI path is a starting point
Code and config rank first by construction: a documented endpoint, flag or config key scores at the top of specificity and actionability, so it sorts above a feature described only in prose. Under 60 the notes go thin — everything below the “thinner coverage” line is thinner documentation, not smaller work. Hover any meter for that feature's three sub-scores.
Docker Sandbox Kit Specification v3 defines a portable OCI image format, open source under Apache 2.0 at docker/sandbox-kit-spec, for declaring what authority a sandbox agent may exercise.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Kit declarations live in the OCI manifest annotation vnd.docker.sandbox.kit.descriptor, build with docker buildx build, pull with docker pull, and are pinnable by digest; v3 kits separate workload Kits, which supply the root filesystem, from mixin Kits, which overlay a CLI, network rule, or credential binding. The spec defines versioned capability types including com.docker.sandbox/network-policy@2 and com.docker.sandbox/credential@1, plus conformance suites for artifacts and runtimes.
$ cd examples sbx run ./hello --kit ./gh
/repos/**, keeping a token that can open PRs from deleting repositories.capabilities:
- type: com.docker.sandbox/network-policy@2
config:
runtime:
allow:
- github.com
- hosts: [api.github.com]
methods: [GET, HEAD, POST, PATCH, PUT, DELETE]
deny:
- hosts: [api.github.com]
methods: [DELETE]
paths: [/repos/**]
- type: com.docker.sandbox/credential@1
optional: true
config:
service: github
phase: runtime
apiKey:
name: GH_TOKEN
proxyManaged: true
inject:
- {domain: api.github.com, header: Authorization, format: 'Bearer %s'}
Cloud Sandboxes run sandboxes on Docker-managed cloud compute with sbx --cloud run <kit> and support moving local work with sbx move --to cloud. Size options range from Micro (1 vCPU / 2 GiB, $0.07/hr) to XL (16 vCPU / 32 GiB, $1.12/hr), sessions can run up to 24 hours, and cloud support requires sbx version 0.45.1 or later.
$ sbx --cloud run claude
$ sbx move my-project --to cloud
V3 Kit capabilities can scope network access by HTTP method and path, declare install-phase network access and credential use, and specify where an agent reads shared skills. Network-policy capabilities support explicit allow and deny rules scoped to hosts, methods, and paths, with deny taking precedence, while credential capabilities support proxyManaged: true, injecting real tokens into requests to named domains and exposing only a sentinel inside the sandbox. Authority-diff gating reduces every descriptor to a normalized grant set so a runtime may auto-apply updates that stay within the recorded set, but any widening, including removing a deny rule, requires explicit approval.
/repos/**, keeping a token that can open PRs from deleting repositories.capabilities:
- type: com.docker.sandbox/network-policy@2
config:
runtime:
allow:
- github.com
- hosts: [api.github.com]
methods: [GET, HEAD, POST, PATCH, PUT, DELETE]
deny:
- hosts: [api.github.com]
methods: [DELETE]
paths: [/repos/**]
- type: com.docker.sandbox/credential@1
optional: true
config:
service: github
phase: runtime
apiKey:
name: GH_TOKEN
proxyManaged: true
inject:
- {domain: api.github.com, header: Authorization, format: 'Bearer %s'}
Local network allow rules gained --protocol, allowing UDPUDPA network protocol that sends individual messages without guaranteeing delivery or order. Applications trade built-in reliability for less overhead and can add their own recovery. to be selected explicitly alongside the default TCPTCPA network protocol that delivers a reliable, ordered stream of bytes between two endpoints. It detects lost data and retransmits it. coverage. Experimental outbound UDP now follows sandbox network policy.
$ sbx policy allow network --protocol udp 8.8.8.8:53
Reverse-DNS lookups are now restricted to destination IPs already authorized by policy, closing a previous policy bypass. DNS resolution is blocked when no network rule permits it, and sbx policy allow network, sbx policy deny network, --allow-network, and --deny-network now reject malformed patterns before saving them.
Destructive commands gained --force for non-interactive confirmation, and sbx kit builder history rm also gained --yes and -y. Commands that remove resources now prompt for confirmation and return a non-zero exit code when the prompt is declined.
sbx mcp catalog has been removed. Register a remote MCP server with sbx mcp add and then authorize it with sbx mcp auth instead.
sbx run --model now supports any OpenAI- or Anthropic-compatible endpoint configured through the new model.providers setting. For sbx run opencode --model, the model's supported thinking levels are exposed as OpenCode variants selectable with Ctrl+T.
$ sbx run --model my-anthropic-proxy/claude-3-5-sonnet 'summarize the repo'
sbx mcp add --disable-http2 disables HTTP/2HTTP/2The second major version of the HTTP protocol, standardized by the IETF, adding multiplexed streams, header compression, and binary framing over a single TCP connection to cut latency and reduce overhead. for a remote MCP transport. The flag is intended for servers whose HTTP/2 handling stalls long-lived streams.
$ sbx mcp add --disable-http2 my-mcp-server https://mcp.example.com/sse
sbx secret rm now returns an error on stderr when the requested secret does not exist, where it was previously silent. sbx mcp rm now returns an error when the requested MCP server is not registered, where it was previously silent.
sbx kit validate now rejects malformed API-key declarations, including invalid names, missing injection domains, invalid format placeholders, and Basic-auth usernames containing a colon. It also warns for no-op or out-of-allowlist declarations.
sbx exec -d and --detach now fail immediately for unsupported detached execution instead of silently running in the foreground.
Kit sets let authors combine a workload and mixins, pin component versions, and publish the result as a single OCI reference with its own capabilities, lifecycle hooks, instructions, and arguments. A kind: set descriptor names a collection of Kits, runs coherence checks at build time, and merges them into one ordinary Kit so incoherence is caught at the author's build rather than at launch.
sbx ls --json and sbx inspect --json now include created_at, CPUCPUThe main general-purpose processor in a computer. It runs application instructions and coordinates work with memory and other hardware. limits, and memory limits in their output.
Cloud Sandboxes introduced pre-configured, pre-built Kits for leading coding agents, launchable with sbx --cloud run <kit>. Named Kits include Claude Code, Codex, Copilot, Antigravity, Open Code, and Hermes.
$ sbx --cloud run claude
sbx policy ls gained the --created-via filter. The command also now shows how each rule was created.
MCP server, secret, skill, template, volume, and policy-profile list commands gained --quiet and -q for name-only output.
macOS distributions now ship as a single signed Sbx.app bundle. Homebrew and tarball PATH installs work through a symlink into the bundle.
Cloud Sandboxes added a secrets proxy that injects stored keys or tokens per request so agents do not hold the actual secret value. Separately, local and cloud sandboxes maintain separate secrets, templates, and network policies, so existing local workflows must be reconfigured before moving to cloud.
A centralized MCP gateway lets agents running in cloud, locally, or in other clients such as ChatGPT desktop share one set of connected MCP servers. The launch names Jira, Linear, Grafana, incident.io, and any streamable HTTP endpoint as supported server targets.
Multiple OAuth-backed agents can now be composed in the same sandbox with credentials scoped to the individual kits that request them. OAuth refreshes are coordinated across sandboxes that share credentials, preventing simultaneous refreshes from forcing another sign-in.
V3 workloads and mixins must be used together. They cannot be combined with v1 or v2 kits.
The OpenCode kit now configures GitHub Copilot from the account's stored GitHub credential, enabling Copilot models without a separate device login.
Cloud Sandboxes added network policies that define which endpoints agents can reach. Policies are configured once and applied across all cloud sandboxes, but local and cloud sandboxes maintain separate network policies and require reconfiguration before moving workflows to cloud.
A web console can spin up and manage Cloud Sandboxes without using the CLI. The console includes Kit selection and a Run button.
Adding, updating, or removing global service secrets now updates existing local sandboxes without a restart.
- !Local and cloud sandboxes maintain separate secrets, templates, and network policies — existing local workflows must be reconfigured before moving to cloud.
- !
sbx mcp cataloghas been removed; register a remote MCP server withsbx mcp addand then authorize it withsbx mcp authinstead. - !
sbx secret rmnow returns an error on stderr when the requested secret does not exist (previously silent). - !
sbx mcp rmnow returns an error when the requested MCP server is not registered (previously silent). - !V3 workloads and mixins must be used together; they cannot be combined with v1 or v2 kits.
- !Commands that remove resources now prompt for confirmation and return a non-zero exit code when the prompt is declined; use
--forceor--yes/-yin non-interactive workflows. - !
sbx exec -d/--detachnow fails immediately for unsupported detached execution instead of silently running in the foreground.
Cycode’s Agentic Development Security Platform unites security and development teams with actionable, code-to-runtime context to identify, prioritize, and fix the software risk that matters.
Cycode Workstation Protection intercepts package install commands on developer machines in real time, enforcing cooldown policies and blocking known malicious packages before installation. That closes a supply-chain control gap upstream of CI by applying policy at the workstation regardless of package manager version or local config.
Cycode added Workstation Protection to stop risky package installs directly on developer workstations before code enters the pipeline.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ The release adds real-time install interception, cooldown policy enforcement, malicious package intelligence, and MDMMDMMobile device management, centrally configuring and managing enrolled devices. Administrators use it to apply settings, distribute applications and enforce device security policies. deployment without changing developer install workflows.
what's the number?
Each feature carries 0–100 for how completely the vendor documented it — not how big or important the work is. A major capability described in eight words scores low, and that is the finding.
- depth0–40 what it does and how it works or what changed
- specificity0–30 names real surfaces — APIs, flags, formats, limits, numbers
- actionability0–30 enough to go use it — a named endpoint, flag, or config key tops this; a UI path is a starting point
Code and config rank first by construction: a documented endpoint, flag or config key scores at the top of specificity and actionability, so it sorts above a feature described only in prose. Under 60 the notes go thin — everything below the “thinner coverage” line is thinner documentation, not smaller work. Hover any meter for that feature's three sub-scores.
Workstation Protection intercepts package install commands in real time on developer workstations to enforce cooldown, or release-age gating, policies and block known malicious packages before installation.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ It checks every install against a continuously updated threat intelligence feed, including typosquattingtyposquattingUsing a name close to a trusted domain or software package to catch spelling mistakes. Attackers use it to lure people or installers to a malicious lookalike., slopsquattingslopsquattingPublishing malicious packages under names invented by AI coding tools. A developer who installs a hallucinated dependency may then retrieve the attacker's real package., and persistent compromised maintainer releases, and enforces policies at the device level regardless of package manager version or local config correctness.
Workstation Protection is deployed via Mobile Device Management, with no new console, no new infrastructure, and no change to developer install workflows.
Cycode extended ADLC Security coverage upstream to the developer workstation by unifying workstation package enforcement with existing AI Bill of MaterialsBill of MaterialsA structured list of every component, library, and dependency that makes up a piece of software or hardware, used by cyber tools to track what is inside a product and surface known vulnerabilities in its parts., prompt guardrails, pipeline integrity, and secret scanningsecret scanningChecking files, code or other data for exposed credentials such as passwords, API keys and tokens. Findings help teams remove the secret and revoke or rotate it. into the Context Intelligence GraphIntelligence GraphA graph database structure linking entities such as hosts, IPs, domains, and threat actors by relationship, letting cyber tools traverse connections to surface context that flat tables cannot express..
Where each pick stands · Sep 18–24
Tools with changes in each pick's category over the last 12 weeks, including the pick itself.
Each pick's release rhythm · Sep 18–24
Changes per week for the last 12 weeks. The bright bar is this week; a dash marks a week with no daily.
Which categories moved · Sep 18–24
The thick bar is this week. The thin bar is the average of the four weeks before, counting only weeks with a daily.
Who shipped the most · Sep 18–24
Changes per tool this week. Volume is not a quality score.