PAGES
Subscribe Get it delivered — five to ten releases we picked each morning from the tools you run, with every release in the issue on the web. Compare plans →

The Cyber Toolchain — issue 141, November 16, 2026

THE CYBER TOOLCHAIN NO. 141
Tail
THE DAILY RELEASE FIREHOSE
PUBLISHED NOVEMBER 16, 2026 · EVERY DAY
EDITIONS tail grep head ls diff uniq

The daily firehose — everything the toolchain shipped today, already filtered.

We read every release from the 1325 tools on our watchlist at the source — GitHub and GitLab release notes, vendor release pages and changelogs, project blogs and feeds, vendor press releases, and the source code behind the tag. Bug-fix-only releases and non-product newsroom noise are dropped; what's left is summarized down to the new capability, how to try it, and any screenshots or videos the release itself published. Every entry links to the sources it was built from.

VIEW
ISSUE VIEW scroll
INFORMATION DENSITY everything
ISSUE WIDTH reading width
Do you prefer scroll?
ADDITIONAL FILTERS
$ tct list   # 24 tools matched
Same issue, same prompt, two writers:

What stands out across today's releases, grouped by what it lets you do. Every tool named links to its entry below.

Three unrelated things stand out. Socket now traces CVEs to actual call chains across six ecosystems and ships a socket fix command to close them. StepSecurity surfaces which environment secrets exist, who uses them and when they were rotated. Horizon3 NodeZero gives MSSPs one console across client tenants and licenses instead of logging in per customer.

  • protect

    Cut a dependency CVE list down to the ones your code can actually reach

    Socket's reachability analysis now spans Ruby/Bundler, .NET/NuGetNuGetMicrosoft's package manager for .NET, hosting a public registry of reusable libraries so tools can pull in dependencies without bundling or manually maintaining third-party code., Go, JVMJVMThe Java Virtual Machine, a runtime maintained by Oracle (and implemented by others) that executes Java bytecode on any OS, giving cyber tools a portable, garbage-collected environment with a large ecosystem of libraries., Python and JavaScript, so an advisory in a transitive package only lands on the backlog if a call chain gets to it — and socket fix plus SCASCASoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself. Patches attempt the upgrade rather than just filing it. Veracode's Reporting API can now filter for ageless SCA findings, the long-tail items that no longer age into an SLASLAA formal agreement between a service provider and customer that defines expected uptime, response times, and remedies for failures, giving cyber tools a contractual baseline for availability and support commitments. bucket and quietly disappear from reports.

    Socket · Veracode

  • identify

    See which environment secrets exist, who consumes them, and when they were last rotated

    GitHub environment secrets are normally write-only once set, so teams track rotation in a spreadsheet and guess which workflows still consume a given credential. A permission-check endpoint plus an Environment Secrets tab with per-environment rotation and usage tracking turns that into something you can query before deciding whether a leaked token matters.

    StepSecurity

  • govern

    Run one console across client tenants instead of logging in per customer

    NodeZero's Client and License Management console gives MSSPs a single place to see seat allocation and client state, which is the difference between pentest capacity being manageable at twenty accounts and being a billing problem. Okta extends agent identity governance to third-party IdPs via an open protocol plus agent discovery, so service-account-shaped agents created outside your primary directory stop being invisible.

    Horizon3 NodeZero · Okta for AI Agents

  • detect

    Query threat intel, alerts and cases conversationally during triage

    An 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 exposes TI, alerts, cases and SOL data to assistants like Claude Code, so an analyst pivoting from an IOCIOCAn indicator of compromise, an observable clue that may point to malicious activity, such as a suspicious file hash or domain. It is evidence to investigate, rather than proof by itself. to related cases does it without hand-writing SOL queries. The risk is the usual one — the assistant gets whatever read scope you grant it — but the pivot cost during live triage drops sharply.

    Sekoia.io

Does Opus 5 read better?
Offensive Security
◆  Exploitation & C2

NetSPI OCInferno

Sources Commits → changes since 2026-11-13 Code

OCInferno adds --no-group-hop, --no-default-admin-group-hop, and --threads flags, and now models dynamic-group rule modification as an attack pathattack pathA 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. edge⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ TRY IT
Enumerate 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. while skipping hops through groups, to surface only direct principal-to-resource escalation routes.
$ ocinferno process_og_attack_paths --no-group-hop
Filter out default-admin-group hops when reviewing privilege-escalation paths, narrowing results to non-admin-group pivots.
$ ocinferno process_og_attack_paths --no-default-admin-group-hop
  • Filter attack paths that hop through any group node with --no-group-hop on process_og_attack_paths.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
  • Filter attack paths that hop through default admin groups with --no-default-admin-group-hop on process_og_attack_paths.
  • Parallelize IDD credential scans with --threads <N>; defaults to 1 (sequential) so parallelism is opt-in.
  • OCInferno now models a principal's ability to rewrite dynamic-group matching rules as the attack-path edges OCI_UPDATE_DYNAMIC_GROUP_RULES and OCI_CAN_MODIFY_DG_RULES, surfacing dynamic-group injection as a privilege-escalation path.
  • Pathfinder now loads all graph edges by default without requiring --include-containment.
+1 more show less
  • Adds progress counters for large enumerations (100+ items) and labeled progress output for IDD credential scans.
  • Filter attack paths that hop through any group node with --no-group-hop on process_og_attack_paths.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
  • Filter attack paths that hop through default admin groups with --no-default-admin-group-hop on process_og_attack_paths.
  • Parallelize IDD credential scans with --threads <N>; defaults to 1 (sequential) so parallelism is opt-in.
  • OCInferno now models a principal's ability to rewrite dynamic-group matching rules as the attack-path edges OCI_UPDATE_DYNAMIC_GROUP_RULES and OCI_CAN_MODIFY_DG_RULES, surfacing dynamic-group injection as a privilege-escalation path.

2 more in Everything

└──▷ BREAKING ON UPGRADE
  • !The --include-containment flag has been removed; containment edges are now always loaded by the pathfinder.
  • !The --exclude-service-agents flag has been removed as it was not applicable to OCI.
Was this useful?
◆  Hacking Hardware

JustCallMeKoko LLC ESP32 Marauder v8

Sources Vendor store → back in stock · Nov 15, 2026 Store

ESP32 Marauder v8 is the official handheld ESP32-C5 board for the ESP32 Marauder firmware, with a touchscreen, dual-band Wi-Fi 6, Bluetooth 5 and GPS for Wi-Fi and BLE auditing and wardriving.

  • Marauder v8 (marauder-v8) back in stock at justcallmekokollc.com on 15 Nov 2026.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
Was this useful?
Application Security
◆  Secrets Detection & Management

passdetective

Sources Release notes → 1.1.0 Notes Rank Why this rank top features scored 35/25 — 30 How ranking works →

PassDetective is a command-line tool that scans shell command history to detect mistakenly written passwords, API keys, and secrets. Using regular expressions, it helps prevent accidental exposure of sensitive information in your command history.

PassDetective 1.1.0 adds support for scanning fish shell history and broadens its regex-based detection of secret patterns.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 2 FEATURES 0 OF 2 FEATURES most completely described first
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.

01 00
Fish shell history scanning new 35

PassDetective can now scan fish shell command history for accidentally typed passwords, API keys, and secrets, in addition to previously supported shells.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

— Named shell support but no mechanism or command details 1.1.0
02 00
Expanded secret detection regexes improved 25

PassDetective now detects a wider range of secret patterns through expanded secret regexes, improving its ability to catch mistakenly written passwords and API keys in command history.

— States improvement but no specific patterns or regex examples given 1.1.0
Was this useful?
◆  Code Security

Mend.io Mend Platform

Sources Release page → Version 26.10.2 (15-November-2026) Notes Rank Why this rank top features scored 68/67/10, +15 breaking change, +6 runnable example — 69.33 How ranking works →

Mend Platform's November release focuses on AI-assisted workflows, adding a Claude Connector for its MCP ServerMCP ServerA process that exposes tools, resources, and prompts over Anthropic's Model Context Protocol, letting an AI agent call external APIs or local services through a standardised interface. and a guided onboarding wizard for SASTSASTStatic application security testing, analyzing source code or compiled code for vulnerabilities without running the application. It helps developers find problems before deployment. scanning, while removing the Forge tab.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

Mend Platform is a traditional application security misses AI risk. Standalone AI security tools miss your code.

Mend Platform's November release focuses on AI-assisted workflows, adding a Claude Connector for its MCP ServerMCP ServerA process that exposes tools, resources, and prompts over Anthropic's Model Context Protocol, letting an AI agent call external APIs or local services through a standardised interface. and a guided onboarding wizard for SASTSASTStatic application security testing, analyzing source code or compiled code for vulnerabilities without running the application. It helps developers find problems before deployment. scanning, while removing the Forge tab.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 3 FEATURES 0 OF 3 FEATURES most completely described first
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.

01 00
Guided SAST onboarding wizard new 68

A new guided onboarding wizard activates SAST scanning for repositories directly from the Mend Platform, covering quick repository selection, pilot scan execution, and team onboarding via email invitations.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ To use it, go to SAST, click Activate SAST, select a repository, run a pilot scan, then use Invite via email to delegate onboarding to a colleague who owns the repository.

Activate SAST scanning for a repository and invite a colleague to complete the setup when you do not own the repository.
📍In the Mend Platform, go to SAST, click Activate SAST, select a repository, run a pilot scan, then use Invite via email to delegate onboarding to a colleague.
— Describes each wizard step with exact UI navigation Version 26.10.2 (15-November-2026)
02 00
Claude Connector for Mend MCP Server new 67

A new Claude Connector enables one-click installation and discovery of the Mend 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 within the Claude environment, with permission-gated access to Mend findings and fixes. Installation is done from within Claude by going to Connectors, searching for Mend, and clicking Install on the Mend MCP Server connector.

Connect the Mend MCP Server to the Claude assistant without manual configuration to scan repositories and review findings in one place.
📍In the Claude environment, go to Connectors, search for Mend, and click Install on the Mend MCP Server connector.
— Clear mechanism and exact UI steps, but no detail on permission scopes Version 26.10.2 (15-November-2026)
thinner coverage below
03 00
Forge tab removed breaking 10

The Forge tab has been removed from the platform. No further detail on its prior function or migration path is given.

— Only a bare statement of removal, no context or mechanism Version 26.10.2 (15-November-2026)
└──▷ BREAKING ON UPGRADE
  • !
    The Forge tab has been removed from the platform.
Was this useful?

Mend.io Mend SCA

Sources Release page → Version 26.10.2 (15-November-2026) Notes Rank Why this rank top features scored 50 — 50 How ranking works →

Mend SCA added a Claude Connector for its Mend MCP ServerMCP ServerA process that exposes tools, resources, and prompts over Anthropic's Model Context Protocol, letting an AI agent call external APIs or local services through a standardised interface., letting users install and scan repositories directly from within Claude.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

Mend SCA is a traditional application security misses AI risk. Standalone AI security tools miss your code.

Mend SCA added a Claude Connector for its Mend MCP ServerMCP ServerA process that exposes tools, resources, and prompts over Anthropic's Model Context Protocol, letting an AI agent call external APIs or local services through a standardised interface., letting users install and scan repositories directly from within Claude.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 1 FEATURE 0 OF 1 FEATURE most completely described first
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.

01 00
Claude Connector for Mend MCP Server new 50

Mend SCA now adds a Claude Connector for the Mend 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, giving one-click installation and discovery within Claude for repository scanning, remediation, and organization-wide security posture reporting without manual configuration.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

— Names integration and capabilities but no setup steps or config. Version 26.10.2 (15-November-2026)
Was this useful?
◆  Software Supply Chain Security

Socket

Sources Blog / feed → New alert covering dual use tools (2026-11-13) Blog Rank Why this rank top features scored 100/95/95, +15 breaking change, +10 product media, +6 runnable example — 127.67 How ranking works →

Socket fights vulnerabilities and provides visibility, defense-in-depth, and proactive supply chain protection for JavaScript, Python, and Go dependen...

Socket's biggest addition this window is Full Application Reachability analysis, extended across Ruby/Bundler, .NET/NuGetNuGetMicrosoft's package manager for .NET, hosting a public registry of reusable libraries so tools can pull in dependencies without bundling or manually maintaining third-party code., Go, JVMJVMThe Java Virtual Machine, a runtime maintained by Oracle (and implemented by others) that executes Java bytecode on any OS, giving cyber tools a portable, garbage-collected environment with a large ecosystem of libraries., Python, and JavaScript to narrow CVE lists down to exploitable call chains, alongside a new socket fix auto-remediation command and SCASCASoftware composition analysis, identifying third-party dependencies and checking them for known vulnerabilities and license issues. It covers software a team uses rather than writes itself. Patches for code-level fixes.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Socket Firewall also gained webhook/Splunk decision streaming, Redis-backed verdict caching, metadata filtering, Cooldown mode for newly published packages, DNS-override and Registry Mode routing, plus smaller platform updates to policy propagation speed, liveness checks, the Threat Feed API, and PR scan search.

└──▷ WHAT SHIPPED · 16 FEATURES 2 OF 16 FEATURES most completely described first
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.

01 01
Full Application Reachability analysis across ecosystems new 100

Socket adds Full Application Reachability analysis, which builds a cross-dependency call graph to identify which CVEs have an exploitable call chain from application code to the vulnerable function, across Ruby/Bundler, .NET/NuGet, Go Modules, JVM (Gradle/Maven/Scala/sbt), Python (Poetry/uv), and JavaScript (npm/pnpm/Yarn) projects.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Run it with socket scan create --reach ., or locally without uploading via socket scan reach, which saves output to .socket.facts.json. Tune large scans with --reach-analysis-timeout (default 10m), --reach-analysis-memory-limit (default 8GB, 16–32 GB recommended for large projects), and --reach-concurrency; skip workspaces with no source via --reach-continue-on-no-source-files, and keep the intermediate facts file with --reach-retain-facts-file. For builds without a lock file, resolve the dependency graph per build with --dynamic-sbom-inference (scoped with --dynamic-sbom-inference-ecosystems), or generate manifests independently with socket manifest dynamic-sbom-inference, socket manifest gradle, socket manifest maven, or socket manifest scala, or via CycloneDXCycloneDXAn open standard for describing software components and their dependencies, vulnerabilities and other supply-chain information. Tools use it to exchange bills of materials and related evidence. using socket cdxgen -t gradle|maven|sbt -o socket-gradle.cdx.json --install-deps --lifecycle build. Set the org with --org <org-slug> or socket login, and authenticate headlessly via SOCKET_SECURITY_API_TOKEN; run the same analysis in CI via socketcli --reach --reach-analysis-memory-limit --reach-analysis-timeout. Source code and dependency graphs are analyzed locally, not uploaded. In the Dashboard, filter results by reachability status or CVE Reachability to compare Reachable, Unreachable, and Potentially reachable counts, view the CLI's Noise% column (((Unreachable + NoSup + NoReach) / Total Vulns) × 100), and override unreachable low-impact CVEs to suppress them from future scans.

Compare baseline vs. reachability scan to see how many CVEs are actually exploitable in your Ruby app.
$ socket scan create .
# note total CVE count, then:
socket scan create --reach .
Run reachability analysis offline on a large monorepo with extended timeout and memory, without uploading results.
$ socket scan reach
# results saved to .socket.facts.json
Tune reachability analysis for a large Bundler project where the default 10-minute timeout and 8 GB memory limit are insufficient.
$ socket scan create --reach \
  --reach-analysis-timeout 20m \
  --reach-analysis-memory-limit 32GB \
  --reach-concurrency 4 \
  .
Scan a .NET solution in CI and get a noise-reduced CVE list showing only exploitable paths, with the dependency graph resolved per build.
$ socket scan create --reach --dynamic-sbom-inference .
Tune a large NuGet monorepo scan that exceeds the default memory and timeout limits.
$ socket scan create --reach \
  --reach-analysis-timeout 20m \
  --reach-analysis-memory-limit 32GB \
  --reach-concurrency 4 \
  .
Enable lock file generation in a .csproj so Socket can produce the most accurate dependency graph.
xml
<PropertyGroup>
  <RestorePackagesWithLockFile>true</RestorePackagesWithLockFile>
</PropertyGroup>
Establish a baseline CVE count, then run a reachability scan to filter out non-exploitable findings in a Go Modules project.
$ socket scan create .
# note baseline CVE count, then:
socket scan create --reach .
Analyze a large Go monorepo locally without uploading, with extended timeout and higher memory and concurrency limits.
$ socket scan create --reach \
  --reach-analysis-timeout 30m \
  --reach-analysis-memory-limit 32GB \
  --reach-concurrency 4 .
Run reachability analysis offline and inspect results before deciding whether to upload.
$ socket scan reach
Run a full reachability scan on a Gradle or Maven project to identify only the CVEs that are actually reachable from your application code.
$ socket scan create --reach --dynamic-sbom-inference .
Analyze reachability locally without uploading a scan result — useful for sensitive codebases where you want to inspect .socket.facts.json before sharing.
$ socket scan reach --dynamic-sbom-inference
Tune reachability analysis on a large multi-module project to avoid timeouts and memory exhaustion.
$ socket scan create --reach --dynamic-sbom-inference --reach-analysis-timeout 20m --reach-analysis-memory-limit 32GB --reach-concurrency 4 .
Identify which CVEs in a Poetry project are actually reachable from your application code, to cut triage noise before a release.
$ socket scan create --reach --reach-analysis-memory-limit 32GB --reach-concurrency 4 .
Run reachability analysis on a uv project in CI without uploading results, then inspect the output file before promoting to a full scan.
$ socket scan reach
Authenticate in a headless CI environment using an API key instead of interactive login.
$ export SOCKET_SECURITY_API_TOKEN=sktsec_<your_key>
socket scan create --reach --org <your-org-slug> .
Compare total CVE count before and after enabling reachability to measure noise reduction in a large npm project.
$ npm install
socket scan create .
# note baseline CVE count, then run with reachability:
socket scan create --reach --reach-analysis-memory-limit 32GB --reach-analysis-timeout 20m --reach-concurrency 4 .
Run reachability analysis locally on a pnpm workspace without uploading results, then inspect the output file before committing.
$ pnpm install
socket scan reach
cat .socket.facts.json
Authenticate via API key in CI and run a reachability scan for a Yarn project without an interactive login.
$ export SOCKET_SECURITY_API_TOKEN=sktsec_<your_key>
socket scan create --reach --org <your-org-slug> .
Scan an npm app with full reachability to collapse Potentially reachable CVEs into definitive Reachable/Unreachable verdicts, isolated on its own branch so it does not pollute the Alerts baseline.
$ npm ci
socket scan create --org=your-org --repo="your-app" --branch="socket-reach" --reach --no-interactive .
Scan a large app in CI with raised memory and timeout limits to avoid out-of-memoryout-of-memoryAn out-of-memory condition: a program requests more memory than the system or device can provide. The allocation fails or the operating system may stop the process. or timeout failures, running two parallel analysis workers.
$ socket scan create --org=your-org --repo="your-app" --branch="socket-reach" --reach --no-interactive \
  --reach-analysis-memory-limit 32GB \
  --reach-analysis-timeout 30m \
  --reach-concurrency 2 .
Run reachability in a CI pipeline using socketcli, constrained to the default branch only so pull requests stay on the faster standard scan.
$ socketcli --target-path . --reach --reach-analysis-memory-limit 16GB --reach-analysis-timeout 30m
— Every flag, default, and command named across all six ecosystems. product docs
02 00
SCA Patches apply code-level CVE fixes new 95

Socket adds SCA Patches, which applies code-level CVE fixes without changing package versions or lockfiles. Automate reapplication after every npm install or npm ci with socket patch setup --yes, which adds "postinstall": "socket patch apply --silent --ecosystems npm" to package.json; query reachable, patchable CVEs across all repos via GET https://api.socket.dev/v0/orgs/your-org/historical/alerts with filters.cvePatchStatus=patch_available and filters.alertReachabilityType=reachable; and add --offline to socket patch apply to fail rather than fetch a missing patch from .socket/blobs/. Socket Patch CLI 5.1.0 adds recursive monorepo scanning via socket-patch scan run from the repo root, Go workspace support, and rebuilds gems with native extensions (e.g. nokogiri) after patching, with hosted mode replacing already-installed gems while keeping Gemfile declarations intact.

Find all reachable, patchable CVEs across your org's repos in the last 30 days to prioritize which packages to patch first.
$ curl -s -H "Authorization: Bearer $SOCKET_CLI_API_TOKEN" "https://api.socket.dev/v0/orgs/your-org/historical/alerts?range=-30d&per_page=100&filters.cvePatchStatus=patch_available&filters.alertReachabilityType=reachable" | jq -r '[.items[]? | "\(.repoSlug) \(.artifact.name)@\(.artifact.version) \(.alert.props.cveId)"] | unique[]'
Scan a project for available patches and apply one by name without changing the manifest or lockfile.
$ cd /path/to/your-app && npm ci && socket patch scan --json > patches.json && socket patch get minimist && npm test
Gate CI so it fails if patches were not applied — catching cases where the installed version no longer matches a saved patch.
$ out=$(socket patch apply --json)
echo "$out" | jq -e '.status == "success" and (.patchesApplied + .alreadyPatched) > 0' >/dev/null \
|| { echo "Socket patch apply: $(echo "$out" | jq -c '{status, patchesApplied, alreadyPatched, failed}')"; exit 1; }
Scan an entire monorepo recursively from the root rather than running socket-patch per workspace.
$ socket-patch scan
— Named endpoint, flags, and commands across both releases. Socket Patch CLI 5.1.0 (2026-11-11)
03 00
`socket fix` automated remediation and PR status new 95

Socket adds a socket fix command that calculates the smallest set of upgrades to clear a vulnerability — including bumping a parent package when needed — and applies them to the manifest and lockfile; it supports npm, pnpm, Yarn, uv, Poetry, Gradle, dotnet, Cargo, Bundler, and Maven. Skip major-version bumps with --no-major-updates, or omit it for CVEs where a major upgrade is the only path, and enable auto-merge on fix PRs with --autopilot so they merge once required checks pass. The Open Socket Fix PR button is now disabled for unsupported ecosystems — npm, Maven, NuGet, Go, Cargo, PyPI, and RubyGems are supported — and Owners/Admins see a setup notice with a Settings link when Fix PRs aren't yet enabled for the org, while other members see no PR button until it's turned on.

Audit what socket fix would change across all vulnerabilities before touching any files — useful for a first-pass review in a new project.
$ socket fix --all --no-apply-fixes --output-file fixes.json
jq '.fixes | length' fixes.json
Schedule weekly automated fix PRs in CI, capped at 5 per run and restricted to package versions older than one week to reduce supply-chain risk.
yaml
name: socket-fix
on:
  schedule:
    - cron: '0 6 * * 1'
permissions:
  contents: write
  pull-requests: write
jobs:
  socket-fix:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: '22'
      - run: npm ci
      - name: Run Socket Fix
        env:
          SOCKET_CLI_API_TOKEN: ${{ secrets.SOCKET_CLI_API_TOKEN }}
          SOCKET_CLI_GITHUB_TOKEN: ${{ secrets.SOCKET_FIX_GITHUB_TOKEN }}
        run: npx --yes socket fix --pr-limit 5 --minimum-release-age 1w
Fix a single reachable CVE locally, verify tests still pass, then commit — keeping the blast radius to one advisory at a time.
$ socket fix --id GHSA-xxxx-xxxx-xxxx
npm ci
npm test
git diff
— Flags, supported ecosystems, and UI logic all named. Clearer Availability and Status for Socket F…
04 00
Webhook delivery of firewall decisions new 95

Socket Firewall adds a webhook to POST package decisions as JSON or NDJSONNDJSONA text format where each line is a self-contained JSON object, making it easy to stream, append, and parse large datasets record-by-record without loading the whole file into memory. to any HTTPSHTTPSHTTP carried over an encrypted TLS connection. It protects traffic in transit and lets a browser verify the server's certificate. endpoint. Enable it with WEBHOOK_ENABLED=true and WEBHOOK_URL in .env, with an optional WEBHOOK_AUTH_HEADER sent as the Authorization header; toggle which decision types are forwarded with WEBHOOK_ON_BLOCK, WEBHOOK_ON_WARN, WEBHOOK_ON_IGNORE, and WEBHOOK_ON_MONITOR. Enable batched NDJSON delivery (application/x-ndjson, one event per line) with WEBHOOK_BATCH_ENABLED=true and WEBHOOK_BATCH_SIZE, and disable TLS verification for local testing with WEBHOOK_SSL_VERIFY=false. Documented routing targets include Microsoft Sentinel or Defender XDRXDRExtended detection and response, combining signals and investigation across endpoints, networks, identities or cloud services. It helps analysts connect activity that separate tools might miss. (via an Azure Function/Logic App), syslog-only SIEMsSIEMsSecurity Information and Event Management platforms that collect, normalize, and correlate log data across an infrastructure so cyber tools can detect threats and trigger alerts from a central point. (via Logstash), and ticketing systems such as Jira or ServiceNow (filtering on decision = blocked).

Forward all firewall decision types to a SIEMSIEMSecurity information and event management, collecting logs from many systems and correlating events to find suspicious activity. Analysts use it to investigate alerts and incidents. or log collector so your SIEM can show allowed installs, warnings, and blocks, not just blocks alone.
ini
WEBHOOK_ENABLED=true
WEBHOOK_URL=https://collector.example.com/socket-firewall
WEBHOOK_AUTH_HEADER=Bearer your_receiver_token
WEBHOOK_ON_BLOCK=true
WEBHOOK_ON_WARN=true
WEBHOOK_ON_IGNORE=true
WEBHOOK_ON_MONITOR=true
WEBHOOK_SSL_VERIFY=true
Apply the new .env webhook settings — environment variables are only read at container startup, so a force-recreate is required.
$ docker compose up -d --force-recreate
— Every env var and routing target named. product docs
05 00
Metadata filtering removes blocked versions from listings new 95

Socket Firewall adds metadata filtering, which strips blocked versions from package metadata before the client resolves a version range. Enable it with METADATA_FILTERING_ENABLED=true in .env, or configure it via a top-level metadata_filtering: block in socket.yml with enabled, filter_warn (default false), max_versions (default 100 newest versions), include_unchecked_versions, and cache_ttl. Block unscanned versions from both listings and downloads with socket.fail_open_unscanned: false or SOCKET_FAIL_OPEN_UNSCANNED=false. Coverage spans npm, Go, RubyGems, Cargo, Conda, OpenVSX, NuGet, PyPI, and Maven, each filtered at the appropriate per-version or per-file granularity, and filter activity is logged as decision events with event_category: "proactive" and decision: "blocked".

Turn on metadata filtering for a fresh firewall deployment so blocked versions are pruned from registry responses before npm resolves a semver range.
ini
METADATA_FILTERING_ENABLED=true
Verify that a blocked version is absent from the firewall's metadata listing before testing an install.
$ diff <(curl -s https://registry.npmjs.org/is-odd | jq -r '.versions | keys[]') \
     <(curl -sk https://sfw.example.com/npm/is-odd | jq -r '.versions | keys[]')
— Every config key, default value, and covered ecosystem named. product docs
06 00
Redis-backed distributed verdict caching new 95

Socket Firewall adds Redis-backed distributed verdict caching so allow/warn/block decisions are shared across instances and survive restarts. Enable it with REDIS_ENABLED=true, REDIS_HOST, REDIS_PORT, and REDIS_PASSWORD in .env; control freshness with cache.ttl (default 600s) and redis.ttl (default 86400s), and set re-check behavior for expired allow verdicts with cache.confirm_allow_mode (default wait). Enable TLS for managed Redis (ElastiCache, Azure Cache for Redis) with REDIS_SSL=true and REDIS_SSL_VERIFY=true, plus optional REDIS_SSL_CA_CERT, REDIS_SSL_CLIENT_CERT, REDIS_SSL_CLIENT_KEY, and REDIS_SSL_SERVER_NAME, or equivalently a redis: block in socket.yml using ssl_ca_cert, ssl_client_cert, ssl_client_key, and ssl_server_name. The cache holds verdicts only — package bytes always stream from the upstream registry — and multiple instances share verdicts by pointing at the same REDIS_HOST. With Redis enabled, restarting the firewall no longer clears verdicts; run FLUSHDB via redis-cli to force fresh verdicts after a policy change or waiver.

Add Redis to a Socket Firewall deployment so verdicts survive restarts and are shared across all firewall instances.
env
REDIS_ENABLED=true
REDIS_HOST=redis
REDIS_PORT=6379
REDIS_PASSWORD=change_me_to_a_long_random_string
Connect a Socket Firewall instance to a managed Redis service (ElastiCache, Azure Cache for Redis) with mutual TLSmutual TLSMutual TLS, an encrypted connection where both client and server present certificates to prove their identity. It helps services verify who they are communicating with..
env
REDIS_ENABLED=true
REDIS_HOST=redis.example.com
REDIS_PORT=6379
REDIS_PASSWORD=change_me_to_a_long_random_string
REDIS_SSL=true
REDIS_SSL_VERIFY=true
REDIS_SSL_CA_CERT=/etc/redis-tls/ca.pem
REDIS_SSL_CLIENT_CERT=/etc/redis-tls/client.pem
REDIS_SSL_CLIENT_KEY=/etc/redis-tls/client-key.pem
REDIS_SSL_SERVER_NAME=redis.example.com
Force fresh verdicts after a policy change or waiver when Redis persistence is active, since a firewall restart alone no longer clears the cache.
$ rc() { docker compose exec -T redis redis-cli --no-auth-warning -a "$(grep '^REDIS_PASSWORD=' .env | cut -d= -f2-)" "$@"; }
rc FLUSHDB
— Every env var, default TTL, and TLS option named. product docs
07 00
Cooldown mode for recently published package versions new 95

Socket Firewall adds Cooldown mode: set a Recently Published Alert Threshold in Dashboard → Settings → Alerts → Scans to auto-block new package versions until they age past it, holding back newly published versions for a configurable number of days. Allow recent versions through for selected ecosystems by adding recently_published_enabled_ecosystems to socket.yml; listed ecosystems receive a 200 with x-socket-block-reason suppressed, while unlisted ecosystems stay blocked. Script cooldown waivers via the Policies API — POST https://api.socket.dev/v0/orgs/{org}/alert-policies/{id}/rules or POST https://api.socket.dev/v0/orgs/{org}/alerts/resolutions — using tokens scoped to alert-policy:list, alert-policy:create, or alert-resolution:create. In SIEM pipelines, count cooldown blocks on the decision field rather than action, since ecosystems allowed via recently_published_enabled_ecosystems log decision: allowed with action: error.

Allow recent npm packages through the cooldown while keeping other ecosystems blocked — add this to your existing socket.yml and recreate the container.
yaml
socket:
  api_url: https://api.socket.dev
  fail_open: true
  recently_published_enabled_ecosystems:
    - npm
Programmatically grant a one-version waiver for a specific package in the new Policies system, so a pinned release clears the firewall's 403 without disabling the cooldown for other versions.
$ curl -s -X POST -H "Authorization: Bearer $SOCKET_API_KEY" -H 'Content-Type: application/json' \
  "https://api.socket.dev/v0/orgs/$ORG/alert-policies/$DEFAULT_ID/rules" \
  -d '{"name":"Cooldown waiver rollup 4.25.0",
       "vigil_selector":{"artifact.type":"npm","artifact.name":"rollup","artifact.version":"4.25.0",
                          "finding.alertType":"recentlyPublished"},
       "action":"ignore",
       "note":"Waiver TICKET-123, approved by NAME"}'
Confirm Socket flags a freshly published package with a recentlyPublished alert before testing it through the firewall.
$ curl -s -X POST "https://api.socket.dev/v0/purl?alerts=true" \
  -H "Authorization: Bearer $SOCKET_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"components":[{"purl":"pkg:npm/rollup@4.25.0"}]}' \
  | jq -c 'select(.name!=null)|[.alerts[]?|select(.type=="recentlyPublished")|.action]'
— Dashboard path, config key, and scoped API endpoints all given. product docs
08 02
Registry Mode Docker firewall proxy for npm and PyPI new 95

Socket Firewall's Registry Mode runs as a Docker container (socketdev/socket-registry-firewall:latest) that proxies npm and PyPI requests and returns HTTP 403 for blocked packages. Configure per-ecosystem routes in socket.yml via path_routing.routes entries, each with a path, upstream registry URL, and registry type (npm or pypi); set fail_open: true/false to control behavior when the Socket API is unreachable, and path_routing.use_incoming_domain: true to preserve the client's host/port in rewritten download URLs. Authenticate with SOCKET_SECURITY_API_TOKEN (requires packages:list and entitlements:list scopes, distinct from the CLI/CI token), tune verdict-cache lifetime with cache.ttl, and control log verbosity with SOCKET_IGNORE_LOG_LEVEL. The quickstart guide adds Configure registry: dev and Configure registry: CI/CD instructions for pointing developer machines and CI/CD pipelines/Docker builds at the firewall.

Bootstrap the firewall for npm and PyPI in a strict posture so any Socket API outage blocks installs rather than allowing them through.
yaml
socket:
  api_url: https://api.socket.dev
  fail_open: false
  ports:
    http: 8080
    https: 8443
  cache:
    ttl: 600
  path_routing:
    enabled: true
    domain: sfw.example.com
    rewrite_scheme: https
  routes:
    - path: /npm
      upstream: https://registry.npmjs.org
      registry: npm
    - path: /pypi
      upstream: https://pypi.org
      registry: pypi
Set required environment variables, including the firewall-specific API token, before starting the container.
env
SOCKET_SECURITY_API_TOKEN=your_api_key_here
SOCKET_FAIL_OPEN=true
SOCKET_IGNORE_LOG_LEVEL=warn
Start the firewall and confirm it can reach Socket and load verdicts before routing real traffic through it.
$ docker compose up -d
for i in $(seq 1 30); do curl -sk https://sfw.example.com/health && break; sleep 2; done
docker compose logs socket-firewall | grep 'Resolved org slug from API'
— Docker image, config keys, and token scopes all named. product docs
09 00
SFW registry CI/CD integration via package-manager env vars new 90

Socket adds a CI/CD integration guide for the SFW registry, routing npm, pnpm, Yarn 1, and pip installs through the firewall proxy by setting NPM_CONFIG_REGISTRY to https://sfw.example.com/npm/ and PIP_INDEX_URL to https://sfw.example.com/pypi/simple/ in CI environment variables. Trust the firewall's CA for Node via NODE_EXTRA_CA_CERTS (adds to Node's existing trust list) and for pip via PIP_CERT (replaces pip's entire bundle, so point it at the system bundle /etc/ssl/certs/ca-certificates.crt after adding the firewall CA with update-ca-certificates); bake these into a container image with ENV NPM_CONFIG_REGISTRY, PIP_INDEX_URL, NODE_EXTRA_CA_CERTS, and PIP_CERT. Verify the integration by installing lodahs on a test branch and confirming the build fails with E403 ... Blocked by Security Policy.

Wire a GitHub ActionsGitHub ActionsGitHub's built-in CI/CD system that runs automated workflows triggered by repository events, letting cyber tools automate testing, building, and releasing without managing separate pipeline infrastructure. job to block malicious packages at install time for both npm and pip without changing your existing install steps.
yaml
jobs:
  build:
    runs-on: ubuntu-latest
    env:
      NPM_CONFIG_REGISTRY: https://sfw.example.com/npm/
      PIP_INDEX_URL: https://sfw.example.com/pypi/simple/
    steps:
      - uses: actions/checkout@v4
      - name: Install firewall CA
        run: |
          printf "%s\n" "$SFW_CA_PEM" | sudo tee /usr/local/share/ca-certificates/sfw-ca.crt > /dev/null
          sudo update-ca-certificates
          echo "NODE_EXTRA_CA_CERTS=/usr/local/share/ca-certificates/sfw-ca.crt" >> "$GITHUB_ENV"
          echo "PIP_CERT=/etc/ssl/certs/ca-certificates.crt" >> "$GITHUB_ENV"
        env:
          SFW_CA_PEM: ${{ secrets.SFW_CA_PEM }}
      - run: npm ci
Bake SFW registry and CA trust into a custom CI container image so every pipeline that uses it is protected with no per-job configuration.
dockerfile
COPY sfw-ca.pem /usr/local/share/ca-certificates/sfw-ca.crt
RUN update-ca-certificates
ENV NODE_EXTRA_CA_CERTS=/usr/local/share/ca-certificates/sfw-ca.crt \
    PIP_CERT=/etc/ssl/certs/ca-certificates.crt \
    NPM_CONFIG_REGISTRY=https://sfw.example.com/npm/ \
    PIP_INDEX_URL=https://sfw.example.com/pypi/simple/
— Exact env vars, CA handling, and a runnable verification step. product docs
10 00
Firewall decision-event schema and log routing new 80

Socket documents the firewall decision-event data model, routing events to Splunk HECSplunk HECAn HTTP endpoint built into Splunk that accepts JSON-formatted log and event data pushed directly from applications or agents over HTTPS, letting cyber tools ship telemetry to Splunk without requiring a forwarder agent on the sending host., a webhook (Sentinel, Defender XDR, Elastic, ticketing workflows), or container stdout via environment variables. Every event is JSON with event_type: firewall_package_encountered and fields including timestamp, client_ip, registryFqdn, repo, purl, artifact_purl, action, decision, block_reason, warn_reason, blocked_alerts, request_id, and cached. Log every decision — including clean allows hidden by default — by adding SOCKET_IGNORE_LOG_LEVEL=warn to .env, then filter with docker compose logs socket-firewall | grep SOCKET_DECISION. Filter event_category: "proactive" events out of install-block counts, since each version listing generates one proactive event per version up to max_versions. Use decision rather than action to determine what the client actually received (some allowed downloads carry action: error under Cooldown). Detect the real requester behind a load balancer via client_ip.header and client_ip.trusted_proxies in socket.yml. Dedupe PyPI events on purl plus client_ip, since pip downloads produce two events with the same purl but different source_path.

Capture every firewall decision including clean allows to stdout so you can ship the full install-decision stream to your SIEM.
$ # Add to .env
SOCKET_IGNORE_LOG_LEVEL=warn

# Recreate the container and tail decision events
docker compose up -d --force-recreate
docker compose logs socket-firewall | grep SOCKET_DECISION
— Full field list and log-level variable given, no new endpoint. product docs
11 00
Domain-based DNS override routing new 80

Socket Firewall adds domain-based DNS override routing so clients reach the firewall without any package manager reconfiguration. Configure it in socket.yml using registries[].domains and registries[].upstream keys so the firewall answers directly for registry.npmjs.org, pypi.org, and files.pythonhosted.org. Generate a TLS certificate covering all registry hostnames with -addext 'subjectAltName=DNS:registry.npmjs.org,DNS:pypi.org,DNS:files.pythonhosted.org', and prove it out on one machine by adding firewall IP entries to /etc/hosts before rollout. Track client coverage by matching the client_ip field in firewall decision events against MDMMDMMobile device management, centrally configuring and managing enrolled devices. Administrators use it to apply settings, distribute applications and enforce device security policies., DHCP, or VPNVPNA virtual private network, an encrypted tunnel between devices or networks. It carries traffic across another network as though the endpoints had a private connection. records.

Switch Socket Firewall from path routing to domain-based routing so npm and pip clients need no reconfiguration — back up the existing config first.
$ cp socket.yml socket.yml.path-routing
Domain-based routing config that makes the firewall intercept npm and PyPI traffic transparently using real registry hostnames.
yaml
socket:
  api_url: https://api.socket.dev
  fail_open: true
  ports:
    http: 8080
    https: 8443
  registries:
    npm:
      domains: [registry.npmjs.org]
      upstream: https://registry.npmjs.org
    pypi:
      domains: [pypi.org, files.pythonhosted.org]
      upstream: https://pypi.org
Validate that a single test machine is fully covered before org-wide rollout — no real registry addresses should appear.
$ getent ahosts registry.npmjs.org
— Config keys and a test command both given. product docs
12 00
DNS-aware liveness endpoint for Kubernetes new 80

Socket Registry Firewall adds a GET /health/live endpoint that checks DNS resolution across every nginx worker, so Kubernetes can restart the firewall when a worker fails its DNS check.

Wire the new DNS-aware endpoint into a Kubernetes liveness probe so the pod restarts automatically when any nginx worker loses DNS resolution.
yaml
livenessProbe:
  httpGet:
    path: /health/live
    port: 80
  initialDelaySeconds: 10
  periodSeconds: 15
— Exact endpoint and a ready-to-use probe config. New: DNS-Aware Liveness Checks in Socket Reg…
13 00
`displayName` field in Threat Feed API new 65

Threat Feed API responses and webhooks now include a displayName field containing the registry-native spelling of a package name (for example, the crates.io spelling of a Cargo crate), making it easier to match a threat feed entry to the package under investigation; the field may be null for Cargo, browser extension, and ClawHub entries and should be treated as optional.

When consuming the Threat Feed API, read displayName to get the exact crates.io (or other registry) spelling of a flagged package, then cross-reference it directly in your registry or dependency manifest.
$ curl -s -H 'Authorization: Bearer <token>' 'https://api.socket.dev/v0/threat-feed' | jq '.[] | {displayName: .displayName, name: .name, severity: .severity}'
— Field name and nullability conditions given, no new endpoint. New: displayName in Socket threat feed API r…
14 00
Search and date filters for PR Scans new 60

Socket adds search by repository, branch, PR number, title, or author on the Scans > PRs tab (case-insensitive, minimum two characters), and a Filter → Created option that replaces the old Last 90 days dropdown with presets and custom date ranges. Repositories → Labels can now be filtered by Security policy or License policy status (Enabled or Disabled), and search/filter state is reflected in the page URL for bookmarking and sharing.

Find all PR scans by a specific author within a custom date range and share the resulting view with a teammate.
📍In the Socket dashboard, go to Scans › PRs. Type the author name in the search box, then open Filter → Created and set a custom start and end date. Copy the URL from the browser address bar and share it.
Quickly audit which labels in your org have Security policy disabled so you can triage policy gaps.
📍In the Socket dashboard, go to Repositories → Labels. Set the Security policy filter to Disabled to list all labels where the policy is off.
— UI navigation given but no API or config surface. New: Find PR Scans Faster with Search and Da…
thinner coverage below
15 00
Stale `.socket.facts.json` uploaded as SBOM breaking 55

If .socket.facts.json is present when socket scan create is run without --reach, Socket uploads it as an SBOMSBOMA software bill of materials, an inventory of the components and versions inside an application. It helps teams find affected software when a dependency has a vulnerability or licensing issue. carrying stale reachability results, affecting NuGet, Go, Python, and JavaScript scans; delete the file first to avoid this.

— Clear cause and fix, but no mechanism beyond that. product docs
16 00
Faster policy propagation to Socket Firewall improved 45

Hosted Socket Firewall now propagates security policy, license policy, and triage rule changes in about 5 seconds, down from a 10-minute cache for decisions and up to an hour for package version listings.

— Concrete before/after timing but nothing for a reader to do. New: Policy changes reach Socket Firewall in…
└──▷ BREAKING ON UPGRADE
  • !
    Delete .socket.facts.json before running socket scan create without --reach; if the file is present it is uploaded as an SBOM carrying stale reachability results.
  • !
    Delete .socket.facts.json before running socket scan create without --reach; otherwise the file is uploaded as an SBOM carrying stale reachability results.
  • !
    If .socket.facts.json is present and you run socket scan create without --reach, the file is uploaded as an SBOM with stale reachability data — delete it first to avoid this.
  • !
    Delete .socket.facts.json before running socket scan create without --reach; if the file is present it is uploaded as an SBOM and carries stale reachability results.
  • !
    With Redis enabled, restarting the firewall no longer clears verdicts — run FLUSHDB via redis-cli to force fresh verdicts after a policy change or waiver.
Was this useful?
Reverse Engineering, Malware & Forensics
◆  Digital Forensics & Incident Response

ElcomSoft

Sources Changelog → changelog · Nov 16, 2026 Changelog

ElcomSoft is a try professional password recovery, data decryption, mobile and cloud forensic tools from a manufacturer with 30+ years of expertise, providing tools and training to law enforcement, financial and intelligence agencies.

Was this useful?

WithSecure chainsaw

Sources Release notes → v2.18.0 Notes Rank Why this rank top features scored 40/35, +15 breaking change — 52.5 How ranking works →

Chainsaw v2.18.0 expands windash expansion with permutation support and pulls in a newer upstream EVTXEVTXWindows XML-based binary log format used by the Event Log service to store system, application, and security events; cyber tools parse it to extract forensic artifacts from live systems or offline disk images. library that introduces breaking changes.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

Chainsaw is a rapidly Search and Hunt through Windows Forensic Artefacts

Chainsaw v2.18.0 expands windash expansion with permutation support and pulls in a newer upstream EVTXEVTXWindows XML-based binary log format used by the Event Log service to store system, application, and security events; cyber tools parse it to extract forensic artifacts from live systems or offline disk images. library that introduces breaking changes.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 2 FEATURES 0 OF 2 FEATURES most completely described first
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.

01 00
Permutation support in windash expansion new 40

The sigma-to-tau parser's windash expansion now supports permutations, broadening detection coverage for Windows command-line variations.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

— Explains mechanism but no concrete command/flag examples v2.18.0
02 00
Upstream EVTX library update breaking 35

Chainsaw updates its upstream EVTX library to a newer version, which introduces breaking changes (see issue #248); existing integrations or workflows depending on EVTX parsing behaviour may require adjustment.

— Notes breaking change and issue reference, lacks specifics on behaviour change v2.18.0
└──▷ BREAKING ON UPGRADE
  • !
    Updating the upstream EVTX library introduces breaking changes (see issue #248); existing integrations or workflows depending on EVTX parsing behaviour may require adjustment.
Was this useful?
◆  Malware Analysis

CAPEv2

Sources Commits → changes since 2026-11-13 Code

CAPEv2 is an open-source malware sandbox that automatically executes and analyzes suspicious files to extract configurations, payloads, and behavioral indicators.

CAPEv2 adds a bulk report download API, payload dump-trigger attribution, and a Linux process dump option⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ TRY IT
Download reports for a range of tasks as a SHA256SHA256A cryptographic hash function that turns data into a fixed 256-bit fingerprint. It is used to check integrity; hashing does not encrypt the original data.-named ZIP archive for bulk triage or archival.
$ curl -s 'https://<cape-host>/apiv2/tasks/get/reports/bulk/101-105/json/' -o reports_101_105.zip
Enable bulk report downloads and raise the per-request cap in the API config before exposing the endpoint.
conf/default/api.conf.default
[taskreport]
enabled = yes
bulk_download = yes
bulk_max_reports = 10
  • Download multiple analysis reports at once via the new /apiv2/tasks/get/reports/bulk/ endpoint (and /apiv2/tasks/get/reports/bulk/<task_ids>/<report_format>/), accepting task ID ranges or lists (e.g.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ 101-103 or 101,102,105), a report_format of json, litereport, html, htmlsummary, pdf, maec, maec5, metadata, protobuf, or parti, and an optional package filter; results are streamed as a .zip archive with entries named <sha256>.<ext>.
  • Gate bulk report downloads with the new bulk_download and bulk_max_reports keys in conf/default/api.conf.default (defaults: bulk_download = no, bulk_max_reports = 3), alongside the existing [taskreport] enabled guard.
  • CAPEv2 now shows the specific debugger breakpoint (bp0–bp3 / br0–br3 / action0–action3) or Dump-on-API trigger that produced each payload via the new extract_dump_triggers() and CAPE._resolve_dump_trigger() functions, displayed as a Dump Trigger row in the Payloads tab when file.dump_trigger is present.
  • CAPEv2 now supports procdump as an analysis option for Linux guests.

4 details in Everything

Was this useful?
Network Security
◆  Network Analysis & Monitoring

Ettercap

Sources CVE Disclosure → CVE-2026-129432 CVE

Ettercap is an open-source network security tool for man-in-the-middle attacks, enabling live packet sniffing, filtering, and protocol analysis on local networks.

  • CVE-2026-129432: Heap out-of-bounds read in the mountd dissector's remote-directory copy (CVSSCVSSCommon Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 6.5, affects <= 0.8.5.1)⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
Was this useful?

Netcap

Sources Release notes →Source code → v0.16.7 Notes Code

Netcap is a framework for secure and scalable network traffic analysis - https://netcap.io

Netcap v0.16.7 adds --pcapng-merge-sections, per-run local API authentication, directoryMatches in Hosts.json, and a 3x speed improvement for large 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.-flow workloads.

└──▷ GET THIS VERSION
$ git clone --branch v0.16.7 https://github.com/dreadl0ck/netcap.git
# already have the repo? check out this version:
$ git checkout v0.16.7
  • Enable continuous-interface PCAPNG processing across concatenated sections with --pcapng-merge-sections (or env var NC_PCAPNG_MERGE_SECTIONS)
  • Offline web UI is now opt-in; per-run authentication, strict Host/Origin checks, same-origin CORSCORSA browser mechanism that lets a server declare which foreign origins may read its HTTP responses, enforced by the browser to block unauthorized cross-site requests from JavaScript., an HttpOnly SameSite cookie for browser bootstrap, and 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. backend authentication protect the local API
  • Print the local web UI access tokenaccess tokenA 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. to stderr at startup when the web UI server starts
  • Add directoryMatches field to Hosts.json entries, listing LDAPLDAPA protocol for querying and modifying directory services, standardized under RFC 4511, commonly used to look up users, groups, and attributes in corporate identity stores like Active Directory. computer objects whose name equals a host's observed name
  • Index service-flow membership to reduce processing time for large UDP-flow workloads, cutting the 40k-UDP-flow case from 3.21 s to 1.09 s
+1 more show less
  • Enable YARA in Linux DPIDPIDeep Packet Inspection: a network analysis technique that examines packet payloads beyond headers, letting cyber tools classify traffic, detect patterns, or extract content at wire speed. and nodpi builds

6 details in Everything

└──▷ BREAKING ON UPGRADE
  • !Offline capture runs (--read) no longer start the web UI by default; pass --http <addr> explicitly to enable it.
  • !Invalid or non-capture additional positional arguments now return a fatal error (unexpected capture argument) instead of being silently skipped.
  • !The --gen-config flag now returns an error ('capture --gen-config is not implemented; use NC_* environment variables or command flags') instead of printing a stub message.
Was this useful?
Cloud & Container Security
◆  CNAPP

Plerion

Sources Release page → See the tools your agents define: Notes Rank Why this rank top features scored 60/60, +6 runnable example — 66 How ranking works →

Plerion's Agent security now surfaces the tools an Amazon Bedrock AgentCore runtime actually defines, and adds an Access tab showing reachability across runtimes, gateways, and agents.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

Plerion is a cloud security platform that provides comprehensive risk management, vulnerability detection, and threat prioritization across cloud environments and workloads.

Plerion's Agent security now surfaces the tools an Amazon Bedrock AgentCore runtime actually defines, and adds an Access tab showing reachability across runtimes, gateways, and agents.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 2 FEATURES 0 OF 2 FEATURES most completely described first
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.

01 00
Tool-definition visibility for AgentCore runtimes new 60

Agent security now shows the tools an Amazon Bedrock AgentCore runtime defines in its code for agents built with LangChain, LangGraph, or Strands.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ This is read automatically from the code zips and container images Plerion already scans, requiring no additional setup — select a runtime and open the Tools view to see every defined tool.

Review what tools an AgentCore runtime exposes before approving it for production, without any additional setup if Plerion already scans its code.
📍In Agent security, select an AgentCore runtime and open the Tools view to see every tool defined in its LangChain, LangGraph, or Strands code.
— Names frameworks and source but no limits or output format See the tools your agents define:
02 00
Access tab for AgentCore runtimes, gateways, and agents new 60

Every AgentCore runtime, gateway, and Bedrock agent in Agent security now has an Access tab showing what each one can reach and what can reach it, displayed as a graph or a list, with IAMIAMIdentity and Access Management — a framework of policies and services for controlling who can authenticate to systems and what resources they are authorized to use, letting cyber tools enforce least-privilege and audit access. and resource policies applied.

Audit the blast radius of a Bedrock agent — what it can reach and who can reach it — using IAM and resource policies.
📍In Agent security, select any AgentCore runtime, gateway, or Bedrock agent, then open the Access tab and toggle between the graph and list views.
— Clear UI path and mechanism, but no CLI/API surface See the tools your agents define:
Was this useful?
◆  Container & Image Security

Mend.io Mend Container

Sources Release page → Version 26.10.1 (01-November-2026) Notes

Mend Container is a traditional application security misses AI risk. Standalone AI security tools miss your code.

Was this useful?
Detection & Response
◆  Detection Engineering

Tenzir

Sources Blog / feed → Tenzir Skills v4.1.0 (2026-11-15) Blog Rank Why this rank top features scored 45 — 45 How ranking works →

Tenzir shipped a small update to Tenzir Skills, mirroring agent documentation locally for offline parity with the Platform.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

Tenzir is a data pipeline engine for security teams that collects, transforms, enriches, and routes telemetry.

Tenzir shipped a small update to Tenzir Skills, mirroring agent documentation locally for offline parity with the Platform.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 1 FEATURE 0 OF 1 FEATURE most completely described first
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.

01 00
Local mirroring of agent map and docs improved 45

The tenzir skill now mirrors the agent map and documentation pages published on tenzir.com, so agents using the skill locally read the same documentation as agents connected to the Tenzir Platform via 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..⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

— Explains what changed but no config key or command given Tenzir Skills v4.1.0 (2026-11-15)
Was this useful?
◆  SIEM

graylog2-server

Sources CVE Disclosure → CVE-2023-41044 CVE Stable

Graylog2-server is a free and open log management

  • CVE-2023-41044: Partial path traversal vulnerability in Support Bundle feature (affects >= 5.1.0)⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
1 more release in this issue · 2023-07-05
CVE-2023-41041 CVE Stable
  • CVE-2023-41041: User session is still usable after logout (affects >= 1.0 (?))⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
Was this useful?
Exposure & Vulnerability Management
◆  Endpoint Management (UEM/RMM)

Microsoft Intune

Sources Release page → Week of October 26, 2026 (Service release 2610) Notes

Microsoft Intune is a cloud-based unified endpoint management solution built on Zero Trust. Empower IT to manage, secure and govern devices, apps and endpoints.

Intune 2610 adds STIGSTIGA configuration standard published by DISA that specifies hardening requirements for a given OS, application, or device, giving cyber tools a concrete checklist to audit or enforce against. audit baselines, certificate connector health monitoring, Apple Enforce Routes VPNVPNA virtual private network, an encrypted tunnel between devices or networks. It carries traffic across another network as though the endpoints had a private connection. support, Administrator protection settings, and faster Windows script delivery via push notifications.

  • New compliance frameworkAudit Windows devices against STIG recommendations in commercial tenants using the new STIG audit baselines feature, which evaluates per-setting and per-device compliance without changing device settings — requires Intune Advanced Analytics.
  • New supported platformRetired supported platformConfigure Apple VPN profiles on iOS/iPadOS and macOS to now support Apple's Enforce Routes capability, with routing behavior defined via Include all networks and Exclude local networks settings.
  • View certificate issuance health signals directly in the Microsoft Intune admin center via the Certificate Connector, with indicators for CA connectivity failures, service account permission issues, and PKCS template mismatches.
  • Configure two new Administrator protection settings — Administrator Approval Mode type and elevation prompt behavior — in endpoint security Account protection policies for supported Windows 11 versions.
  • Configure the same two Administrator protection settings in the settings catalog for Windows 11 versions 24H2 and 25H2.
+3 more show less
  • Use the Microsoft Defender for Endpoint agent runtime protection endpoint security template, now generally available, including a new AI agent runtime inspection setting.
  • Add *.trouter.communications.svc.cloud.microsoft to firewall rules to improve push notification connectivity for Device actions, Win32 apps, PowerShell scripts, and Remediations on Windows.
  • Windows scripts now use push notifications for faster delivery of admin-initiated and service-side changes, reducing deployment delays compared with the normal polling interval.

8 details in Everything

└──▷ BREAKING ON UPGRADE
  • !Setup Assistant (legacy) authentication for iOS/iPadOS and macOS automated device enrollment with user affinity is no longer available when creating or editing profiles; existing profiles are automatically migrated to Setup Assistant with modern authentication.
  • !Legacy Apple MDMMDMMobile device management, centrally configuring and managing enrolled devices. Administrators use it to apply settings, distribute applications and enforce device security policies. software update workloads have been removed from Intune following Apple's retirement of the underlying commands and payloads in OS 27; admins must switch to Apple declarative device management update policies.
Was this useful?
Identity, Access & Data Protection
◆  Secure Access (VPN / SASE / ZTNA)

Twingate

Sources Changelog → changelog · Nov 16, 2026 Changelog Rank Why this rank top features scored 71/56 — 63.5 How ranking works →

Twingate expanded its Privileged Access ManagementPrivileged Access ManagementBrokering administrator sessions through a gateway instead of handing out standing credentials. Access is granted for one session, to one target, and recorded. offering with SSH Resource registration for servers and VMs and added support for protecting internal HTTP web apps.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

Twingate is an identity-based access for users, services, and AI agents that deploys in minutes, scales to every resource, and finally lets you retire your VPN.

Twingate expanded its Privileged Access ManagementPrivileged Access ManagementBrokering administrator sessions through a gateway instead of handing out standing credentials. Access is granted for one session, to one target, and recorded. offering with SSH Resource registration for servers and VMs and added support for protecting internal HTTP web apps.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 2 FEATURES 0 OF 2 FEATURES most completely described first
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.

01 00
SSH Resource registration in PAM new 71

Servers and VMs can now be registered as SSH Resources in Twingate, letting users authenticate via their identity provideridentity providerA service that authenticates users and supplies identity information to other applications. It is the sign-in authority behind many single sign-on systems. while the Twingate Gateway handles certificate-based authentication.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Every session is recorded to stdout in asciicast format for SIEMSIEMSecurity information and event management, collecting logs from many systems and correlating events to find suspicious activity. Analysts use it to investigate alerts and incidents. forwarding or audit replay.

— Names mechanism and format but no exact command or UI path changelog-20261116-0fce4e7f
thinner coverage below
02 00
HTTP web app protection in PAM new 56

Twingate Privileged Access Management now supports HTTP web apps, enabling MFAMFAMulti-factor authentication, requiring proof from different categories, such as a password and a hardware key, to sign in. It reduces reliance on a password alone., JITJITA runtime compilation strategy where code is translated from an intermediate form to native machine instructions at execution time, giving tools the speed of compiled code without sacrificing dynamic flexibility. access, security policies, and access logging for internal web apps, data portals, CI/CD infrastructure, and AI-coded business apps, without requiring users to re-authenticate at the app.

— Lists capabilities and use cases but no config or endpoint details changelog-20261116-0fce4e7f
Was this useful?
Governance, Risk & Compliance
◆  Integrated & IT Risk Management

CISO Assistant

Sources Release notes → v4.3.2 Notes Rank Why this rank top features scored 60 — 60 How ranking works →

CISO Assistant's only update this window is a set of new 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. tools for managing documents directly through the MCP interface.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

CISO Assistant is a one-stop-shop GRC platform for Risk Management, AppSec, Compliance & Audit, TPRM, BIA, Privacy, and Reporting. It supports 200+ global frameworks with automatic control mapping, including ISO 27001, NIST CSF, SOC 2, CIS, PCI DSS, NIS2, DORA, GDPR, HIPAA, CMMC, and more.

CISO Assistant's only update this window is a set of new 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. tools for managing documents directly through the MCP interface.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 1 FEATURE 0 OF 1 FEATURE most completely described first
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.

01 00
Document management via MCP tools new 60

Documents can now be managed through the MCP interface using three new tools: get_documents, create_document, and update_document.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

— Names exact tool calls but no detail on parameters or behavior v4.3.2
Was this useful?
Govern
◆  AI Security Posture & Discovery

Mend.io Mend AI

Sources Changelog → changelog · Nov 16, 2026 Changelog Rank Why this rank top features scored 55/50/35, +15 breaking change — 61.67 How ranking works →

Mend AI's reporting layer gained two new report types—one that traces chat AI system prompts to their code location and an expanded AI inventory covering the full agent/tool/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. stack—plus the ability to run any report at the organization level.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

Mend AI is a traditional application security misses AI risk. Standalone AI security tools miss your code.

Mend AI's reporting layer gained two new report types—one that traces chat AI system prompts to their code location and an expanded AI inventory covering the full agent/tool/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. stack—plus the ability to run any report at the organization level.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 3 FEATURES 0 OF 3 FEATURES most completely described first
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.

01 00
AI Inventory report expanded and renamed breaking 55

The AI Inventory report now includes one row per detected AI entity across models, system prompts, agents, agent configurations, tools, and MCP servers.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ The previous AI Inventory report, which covered AI models only, has been renamed to Models Inventory.

— Names entity types and the renamed report but no migration steps. changelog-20261116-605c5559
02 00
Conversational System Prompts report new 50

A new report type that locates the system prompts behind chat-based AI applications and shows exactly where each one is defined in code.

— Describes mechanism and output but no UI path or command given. changelog-20261116-605c5559
03 00
Organization-scope report generation new 35

All Mend AI reports can now be generated at the organization scope, in addition to the existing application and project scopes.

— States the new scope but no further detail on usage. changelog-20261116-605c5559
└──▷ BREAKING ON UPGRADE
  • !
    The previous AI Inventory report, which covered AI models only, is renamed to Models Inventory.
Was this useful?
◆  AI Red Teaming & Testing

Giskard

Sources Release notes → giskard-scan/v1.0.2 Notes Rank Why this rank top features scored 88/78/70, +15 breaking change, +6 runnable example — 99.67 How ranking works →

Giskard tests and evaluates LLM applications for performance, bias, and security issues.

Giskard shipped a modular v3 scanner that replaces the old v2 LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. scan and RAGET, adding OWASPOWASPA nonprofit foundation that publishes open standards, guides, and ranked vulnerability lists—most famously the OWASP Top 10—giving cyber tools a vendor-neutral benchmark to test or report against.-aligned red-teaming, RAGRAGRetrieval-augmented generation: looking up relevant documents and adding them to the prompt before the model answers. It grounds responses in private or current data without retraining. knowledge-base quality evaluation, and synchronous execution helpers for agent testing.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 8 FEATURES 1 OF 8 FEATURES most completely described first
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.

01 01
v3 LLM vulnerability scanner with custom generators new 88

Giskard v3 introduces a modular vulnerability scanner that runs automated red-teaming across OWASP LLM Top-10 threat categories — prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools., harmful content, stereotypes, and misinformation — via vulnerability_scan(target, description, languages) from giskard.scan.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Scans can be extended with custom adversarial generators by passing ScenarioGenerator instances to generate_suite, or by registering them on vulnerability_suite_generator_registry.

Red-team a customer support agent against OWASP LLM Top-10 threat categories including prompt injection and harmful content.
python
import asyncio
from giskard.scan import vulnerability_scan

async def my_agent(inputs: str) -> str:
    # Replace with your real agent call
    return f'Echo: {inputs}'

async def main() -> None:
    await vulnerability_scan(
        target=my_agent,
        description='A customer support chatbot for an e-commerce platform.',
        languages=['en'],
    )

asyncio.run(main())
— Full function signature, threat categories and runnable example included giskard-scan/v1.0.2
02 00
v2 LLM scan and RAGET deprecated in favor of v3 breaking 78

Giskard v2 is no longer actively maintained; v3 (giskard-scan) replaces the v2 LLM scan and RAGET for agentic systems. The v2-only automatic tabular scan (giskard.Model + giskard.Dataset) and the giskard.testing ML test suite are not planned for v3 — pin pip install 'giskard[llm]>2,<3' to keep using them.

— Names deprecated surfaces and gives migration pin command giskard-scan/v1.0.2
03 00
Synchronous run_sync() for scenarios and suites new 70

Call run_sync() directly on a scenario or suite to run evals from a plain script or CI pipeline without managing an event loop or calling asyncio.run().

Run a Giskard scenario in a plain script or CI pipeline without needing asyncio.run() or an existing event loop.
python
from giskard.checks import Scenario, Groundedness

def get_answer(inputs: str) -> str:
    return "Paris"

scenario = (
    Scenario("test_france_capital")
    .interact(inputs="What is the capital of France?", outputs=get_answer)
    .check(
        Groundedness(
            name="answer is grounded",
            context="France is in Western Europe. Its capital is Paris.",
        )
    )
)
result = scenario.run_sync()
result.print_report()
— Runnable example demonstrates exact method usage giskard-core/v1.0.4
04 00
Scanner installation and provider extras new 60

Install the scanner with pip install 'giskard[scan]' or pip install giskard-scan; LLM provider extras such as giskard[openai] or giskard[anthropic] are available separately.

— Exact install commands given, nothing further to explain giskard-scan/v1.0.2
05 00
Telemetry opt-out via environment variables new 60

Opt out of telemetry by setting DO_NOT_TRACK=1 or GISKARD_TELEMETRY_DISABLED=1 as environment variables or keys in a .env file in the working directory.

— Exact variable names and config location given giskard-scan/v1.0.2
thinner coverage below
06 00
RAG quality evaluation with quality_scan and KnowledgeBase new 50

Giskard adds quality_scan and KnowledgeBase, the v3 successor to RAGET, for evaluating RAG knowledge-base quality in agentic systems.

— Names the new function and class but shows no usage giskard-scan/v1.0.2
07 00
Agent prompt template validation new 45

Validate agent prompt templates before execution with the new validate_template function in giskard.agents.

— Names function but no example or mechanism given giskard-core/v1.0.4
08 00
Graceful handling of unavailable Hub checks improved 35

Checks that are unavailable on the Hub now warn and skip rather than failing the entire run.

— Describes behavior change without naming a surface giskard-core/v1.0.4
└──▷ BREAKING ON UPGRADE
  • !
    Giskard v2 is no longer actively maintained; v3 (giskard-scan) replaces the v2 LLM scan and RAGET for agentic systems. The v2-only automatic tabular scan (giskard.Model + giskard.Dataset) and giskard.testing ML test suite are not planned for v3 — pin pip install 'giskard[llm]>2,<3' to keep them.
Was this useful?
◆  AI Runtime Defense & Guardrails

doberman-core

Sources Release notes → v0.21.0 Notes Rank Why this rank top features scored 90/75/50, +15 breaking change, +6 runnable example — 92.67 How ranking works →

doberman-core's v0.21.0 release introduces Windows HelloWindows HelloA Microsoft authentication framework built into Windows 10 and later that lets users sign in via biometrics or PIN instead of passwords, giving cyber tools a phishing-resistant credential layer to build on. as a TPMTPMA dedicated hardware chip, standardized by the Trusted Computing Group, that stores cryptographic keys and measurements in tamper-resistant silicon, giving cyber tools a root of trust that software alone cannot fake.-backed 2FA factor for AI agent action approvals, replacing TOTPTOTPA one-time password algorithm (RFC 6238) that generates short-lived codes from a shared secret and the current time, giving cyber tools phishing-resistant second-factor authentication without a server round-trip. secrets, alongside two breaking packaging and hook changes.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

doberman-core's v0.21.0 release introduces Windows HelloWindows HelloA Microsoft authentication framework built into Windows 10 and later that lets users sign in via biometrics or PIN instead of passwords, giving cyber tools a phishing-resistant credential layer to build on. as a TPMTPMA dedicated hardware chip, standardized by the Trusted Computing Group, that stores cryptographic keys and measurements in tamper-resistant silicon, giving cyber tools a root of trust that software alone cannot fake.-backed 2FA factor for AI agent action approvals, replacing TOTPTOTPA one-time password algorithm (RFC 6238) that generates short-lived codes from a shared secret and the current time, giving cyber tools phishing-resistant second-factor authentication without a server round-trip. secrets, alongside two breaking packaging and hook changes.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 3 FEATURES 0 OF 3 FEATURES most completely described first
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.

01 00
Windows Hello 2FA mode for agent approvals new 90

Windows Hello can now be enrolled as the sole 2FA factor with doberman 2fa hello enroll (run from an admin terminal), which deletes the TOTP secret and pins a TPM-backed key so no 2FA secret is stored on disk.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Approval and loosening commands now accept a Windows Hello gesture instead of a TOTP code, and each approval is stored as a Hello-key signature tied to the specific action for later audit. doberman doctor now reports Windows Hello enrollment status alongside other health signals.

Replace TOTP-based 2FA with a Windows Hello gesture on a machine with a TPM, so no 2FA secret is ever stored on disk.
$ doberman 2fa hello enroll
Confirm that Windows Hello mode is active and check overall guardrail health after enrolling.
$ doberman doctor
— Full enroll command, mechanism and audit behaviour all named. v0.21.0
02 00
Cursor hook timeout updated to 660 seconds breaking 75

After upgrading, users must re-run doberman install-hooks --host cursor so the Cursor hook picks up the new 660-second wait timeout; hooks installed before the upgrade keep using the old value until reinstalled.

— Exact command and new timeout value given, no old value. v0.21.0
thinner coverage below
03 00
winhello pip extra now empty breaking 50

The winhello pip extra is now empty; any install that pinned doberman-core[winhello] pulls no extra dependencies.

— States the change but not what replaces the removed dependency. v0.21.0
└──▷ BREAKING ON UPGRADE
  • !
    Cursor users must re-run doberman install-hooks --host cursor after upgrading, or the hook will use the wrong timeout (the new value is 660 seconds).
  • !
    The winhello pip extra is now empty; any install that pinned doberman-core[winhello] pulls no extra dependencies.
Was this useful?
◆  Model & Agent Protection

Minimal Software Research minimal

Sources Release notes → release-32b1f303 Notes Rank Why this rank top features scored 85/85/65, +6 runnable example — 84.33 How ranking works →

This release hardens box security with sealed-box loadout restrictions and a hash-chained audit trail, adds a GitHub App-based sign-in flow for min login, and introduces a daily update check reported via min.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

Minimal is a secure, declarative runtimes for Claude Code, Gemini, Codex, Pi and more, so you can run multiple agents safely on your stack from a verified software supply chain.

This release hardens box security with sealed-box loadout restrictions and a hash-chained audit trail, adds a GitHub App-based sign-in flow for min login, and introduces a daily update check reported via min.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 4 FEATURES 0 OF 4 FEATURES most completely described first
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.

01 00
GitHub sign-in via `min login` new 85

min login is now the GitHub sign-in entry point, completing sign-in on a real page with repository choice.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ Sign-in happens under a published GitHub App, with keys held in the host keychain, the Secure Enclave (macOS), TPMTPMA dedicated hardware chip, standardized by the Trusted Computing Group, that stores cryptographic keys and measurements in tamper-resistant silicon, giving cyber tools a root of trust that software alone cannot fake. (Linux), or Secret Service (Linux hosts with no usable TPM).

— Names exact command and the platform-specific key-storage backends. release-32b1f303
02 00
Hash-chained audit trail and JSON spec inspection for boxes new 85

min box audit --parent mybox reviews the full hash-chained audit trail for a box, merging its parent chain, to verify no decisions were edited. min box show -o json mybox inspects a box's declared spec and egress in machine-readable JSON for scripted policy checks.

Review the full hash-chained audit trail for a box, merging its parent chain, to verify no decisions were edited.
$ min box audit --parent mybox
Inspect a box's declared spec and egress in machine-readable form for scripted policy checks.
$ min box show -o json mybox
— Exact commands and flags make this immediately runnable. release-32b1f303
03 00
Loadout restrictions on sealed boxes new 65

A sealed box now refuses a named loadout, a loadout that sets loadout_mode, and a contradicted constraint; a sealed box reads no default loadouts.

— Names exact rejected conditions but no runnable example. release-32b1f303
thinner coverage below
04 00
Daily update check via `min` new 35

minimal now checks once a day whether the channel has a newer release and reports it via min.

— Describes behavior but no command or config shown. release-32b1f303
Was this useful?
Other / Uncategorized
◆  UNCATEGORIZED

vigolium

Sources Release notes → v0.5.7 Notes Rank Why this rank top features scored 88/86/72, +15 breaking change, +6 runnable example — 103 How ranking works →

Vigolium - High-fidelity vulnerability scanner fusing agentic AI with native speed, modularity, and precision

Vigolium v0.5.7 adds safe-mode scanning, scan resume, baseline diffing and OASTOASTOut-of-band Application Security Testing: a technique for detecting vulnerabilities by having a target system make DNS or HTTP callbacks to an attacker-controlled server, revealing flaws that produce no visible in-band response. session persistence, while reworking redirect handling, finding identity, and numerous CLI command defaults with breaking changes across fuzzing, replay, and export workflows.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 16 FEATURES 0 OF 16 FEATURES most completely described first
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.

01 00
Scan control flags new 88

Adds new scan control flags: --safe limits scans to read-only HTTP methods (GET/HEAD), skips forms, uploads and logins, caps request pace and enforces a 15-minute budget (with a method allow-list settable via the scope.methods config key); --resume-scan <uuid> resumes a scan from its coverage ledger, avoiding re-crawling already-covered paths; --max-time or the scanning_pace.scan_max_duration config key caps total scan duration (defaults to 1 hour); --baseline and --ignore-file (pointing to a .vigolium-ignore.yaml file) suppress pre-existing findings, with --fail-on and the new scan diff command counting only new, un-ignored findings for CI gating; and --require-scope or the VIGOLIUM_REQUIRE_SCOPE=1 environment variable forces a scan to refuse running against a scope that names no hosts.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

Run a low-impact scan during business hours that stays read-only and times out automatically — safe for production targets.
$ vigolium scan https://app.example.com --safe --max-time 10m
Pick up a scan that was interrupted by a timeout or Ctrl-C without re-crawling already-covered paths.
$ vigolium scan --resume-scan a1b2c3d4-e5f6-7890-abcd-ef1234567890
In CI, compare a feature branch scan against a saved baseline so only regressions fail the build.
$ vigolium scan https://staging.example.com --baseline baseline.jsonl --ignore-file .vigolium-ignore.yaml --fail-on high
— Named flag, config key, and exact behavioral limits given. Flag and mechanism named, no detail on ledger format. Flag, config key and default value given, no mechanism beyond cap. Multiple named flags, file format, and CI use case specified. Flag and env var named, behavior is simple. v0.5.7
02 00
Credential redaction in reports and raw views new 86

Credentials can be redacted in reports and notifications via the output.redact_evidence config key (suppressed with --no-redact), and masked in raw request/response views under -j using VIGOLIUM_REDACT or --redact/--no-redact; the JSON envelope now carries a redacted field reflecting this state.

— Multiple named config keys, env var, and envelope field given. v0.5.7
03 00
Redirect-mode default scoped to in-scope hosts breaking 72

--redirect-mode now defaults to in-scope; cross-host redirects no longer carry session or credential headers, and raw requests no longer follow redirects.

— Exact flag and before/after behavior named. v0.5.7
04 00
Scan commands attack only direct input breaking 70

scan-url, scan-request, and run -t now attack only their direct input; pass --include-stored to include stored records as before.

— Commands and flag named with before/after behavior. v0.5.7
05 00
Header fuzzing scope control improved 68

Header fuzzing now defaults to only headers that can sink a vulnerability class via --header-fuzz gated; passing --header-fuzz all restores fuzzing of all headers.

— Exact flag values given, limited further explanation. v0.5.7
06 00
`fuzz` requires explicit insertion point breaking 68

fuzz now requires an explicit insertion-point position and defaults to --rate-limit 100 and --max-requests 10000.

— Command, flags and defaults all named. v0.5.7
07 00
Dry-run preview for mutating commands new 64

The --dry-run flag previews any mutating or traffic-sending command without sending requests, including agent autopilot and agent swarm.

— Flag and affected commands named, no further mechanism. v0.5.7
08 00
`replay -i` sends each record once breaking 62

replay -i now sends only once; URL positionals and multi-record matches without --all-matches are refused.

— Flags and refusal conditions named, limited mechanism. v0.5.7
09 00
Finding identity v2 schema (v6) breaking 60

Finding identity now keys on location and insertion point under schema v6; older vigolium binaries cannot open databases written by this version.

— Schema version and migration risk named, no migration path given. v0.5.7
thinner coverage below
10 00
OAST session persistence after scan ends new 58

OAST sessions now stay alive after a scan completes, and late out-of-band callbacks can be retrieved afterward with oast poll.

— Names the subcommand but little mechanism detail. v0.5.7
11 00
`--intensity` replaces strategy/profile flags breaking 58

--strategy and --scanning-profile are moved to Advanced presets; --intensity becomes the public dial for scan behavior.

— Flags named, no detail on intensity levels. v0.5.7
12 00
Default `db export` format changes to JSONL breaking 56

db export --format nuclei was the previous default output; the new default format is --format jsonl.

— Command and both formats named explicitly. v0.5.7
13 00
OAST public fallback disabled by default breaking 54

oast.fallback_to_default now defaults to false; public OAST fallback is no longer active unless explicitly enabled.

— Config key and new default named, little else. v0.5.7
14 00
JSON envelope `ok` field tied to exit code breaking 48

ok in the JSON envelope is now true exactly when the exit code is 0, 3, or 4.

— Exact field and codes named, no further context. v0.5.7
15 00
`--ci-output-format` hidden and errors breaking 46

--ci-output-format is now hidden from help output and exits with code 2 when used.

— Thin: names the flag and exit code only. v0.5.7
16 00
Unreadable config file now fatal breaking 46

An unreadable config file is now fatal, raising config_invalid, rather than being silently ignored as before.

— Error code named, minimal mechanism. v0.5.7
└──▷ BREAKING ON UPGRADE
  • !
    --redirect-mode now defaults to in-scope; cross-host redirects no longer carry session or credential headers, and raw requests no longer follow redirects.
  • !
    Finding identity v2 now keys on location and insertion point (schema v6); older vigolium binaries cannot open databases written by this version.
  • !
    --ci-output-format is hidden and now exits 2.
  • !
    fuzz now requires an explicit insertion-point position and defaults to --rate-limit 100 and --max-requests 10000.
  • !
    replay -i now sends only once; URL positionals and multi-record matches without --all-matches are refused.
  • !
    ok in the JSON envelope is now true exactly when the exit code is 0, 3, or 4.
  • !
    An unreadable config file is now fatal (config_invalid) rather than silently ignored.
  • !
    scan-url, scan-request, and run -t now attack only their direct input; pass --include-stored to include stored records.
  • !
    oast.fallback_to_default defaults to false; public OAST fallback is no longer active unless explicitly enabled.
  • !
    --strategy and --scanning-profile are moved to Advanced presets; use --intensity as the public dial.
  • !
    db export --format nuclei is the old default output; the new default format is --format jsonl.
Was this useful?

Sandbox0 s0

Sources Release notes → v0.10.6 Notes Rank Why this rank top features scored 91 — 91 How ranking works →

s0 introduced optional preauthorized memory burst for sandboxes, letting sandboxes in the ali-ue1 region expand memory by up to 50% on demand without restarting processes.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

s0 is the open-source CLI for Sandbox0, a hosted execution environment for long-running AI agents.

s0 introduced optional preauthorized memory burst for sandboxes, letting sandboxes in the ali-ue1 region expand memory by up to 50% on demand without restarting processes.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

└──▷ WHAT SHIPPED · 1 FEATURE 0 OF 1 FEATURE most completely described first
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.

01 00
Dynamic memory burst for sandboxes new 91

Sandboxes in the ali-ue1 region can now be given preauthorized memory burst at creation by setting config.resources.memory_burst_percent, allowing up to 50% additional memory (e.g.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠ a 4 GiB base can expand to 6 GiB) without restarting processes. The feature is disabled by default (resources.memory_burst_percent defaults to 0), is limited to ali-ue1, and base plus burst must fit the platform maximum. Billing follows the effective memory limit at each point in time (4 GiB before expansion, 6 GiB while expanded, 4 GiB after shrinking), while CPUCPUThe main general-purpose processor in a computer. It runs application instructions and coordinates work with memory and other hardware. remains based on the original base memory specification.

— Names exact config key, region, limits and billing mechanics v0.10.6
Was this useful?
Saved in this browser. Sign in to sync your tools.