Security Toolkit — November 4, 2026

2026-11-04

31 security tools this month across identity, agent governance, detection, recon, AppSec, cloud and the labs to practise on — half of them the defaults practitioners already run, half shipped or relicensed in the last six months.

🔑 Identity & Access

Authelia

Authelia sits in front of Traefik, Caddy, nginx or HAProxy as a forward-auth portal and makes every request prove itself before it reaches the app behind it. You point your proxy at it, define per-domain access rules, and get TOTP, WebAuthn passkeys and Duo push on top of an LDAP or flat-file user store; it is also an OpenID-certified OIDC provider. This is the reference choice for homelab and small-team SSO, and v4.39.20 in May 2026 was a security release worth patching to. Pricing: free and open source (Apache-2.0).

Kanidm

Kanidm is a single Rust binary that does OAuth2/OIDC, attested passkeys, a read-only LDAPS gateway, RADIUS and TPM-backed offline Linux login, so you are not stitching four services together to get an identity platform. Pull kanidm/server:latest, drop a TLS cert into the volume, and follow the book; TLS is mandatory, which tells you the project's priorities. v1.12.0 landed 2026-11-01 with self-service signup, AES-256-GCM token encryption and documented key rotation. Pricing: free and open source (MPL-2.0).

Warpgate

Warpgate is a bastion for SSH, HTTPS, Kubernetes, MySQL, PostgreSQL, RDP and VNC that needs no client agent and records every session for replay. Run ghcr.io/warp-tech/warpgate setup, assign targets per user, and front them with TOTP and OIDC SSO plus in-browser terminals. The October 2026 line added Vault and OpenBao secret backends that fetch target credentials at connect time and issue short-lived signed certificates; v0.30.4 on 2026-11-03 is a security release fixing a high-severity SSRF, so anyone older should move. Pricing: free and open source (Apache-2.0).

🛰️ Agent Identity & Governance

SPIRE

Non-human identity. SPIRE is the CNCF-graduated reference implementation of SPIFFE, and the standard answer to issuing workload identities without static secrets. It attests workloads across Kubernetes, the three big clouds, Docker, Podman and TPM, then hands each one a short-lived X.509 or JWT SVID over the Workload API, with trust-domain federation between estates. Install the hardened Helm chart and run the server, then the agent; v1.15.5 shipped 2026-10-20. Pricing: free and open source (Apache-2.0).

agentgateway

Agentic identity. agentgateway is a Rust data-plane proxy for MCP, A2A and LLM traffic that terminates OAuth and JWT per the MCP authorization spec and applies CEL or RBAC rules per tool call, so an agent presents its own scoped identity instead of a borrowed human token. Install with the one-line script, point your agents through it, and it exchanges tokens toward backends while emitting per-agent telemetry. It joined the Linux Foundation's Agentic AI effort in June 2026; v1.7.0 on 2026-10-27 added OIDC session handling and always-on MCP session encryption, and is breaking in three places. Pricing: free and open source (Apache-2.0); Solo.io sells an enterprise build.

NVIDIA OpenShell

Runtime policy. OpenShell drops an autonomous coding agent into a kernel-enforced sandbox governed by a YAML policy over filesystem, syscalls, processes and network egress, and checks every outbound connection before it leaves. Credentials are injected only on requests to approved endpoints, so the agent never holds the real secret, and policy changes that widen that exposure get flagged for a human. Install with the shell script and run openshell sandbox create; v0.1.7 on 2026-11-03 moved all audit output to OCSF 1.8. It is young at 0.1.x with a long issue queue. Pricing: free and open source (Apache-2.0).

Obot

Agent audit. Obot is an MCP and LLM gateway that gives users one governed endpoint to approved MCP servers and records what passes through it: every gateway request logs the user, provider, requested versus served model, session, IP, token count and duration. Run the container, enforce tool access by IdP group over SSO and SCIM, and export encrypted JSONL audit logs on a schedule to S3, GCS or Azure with a 90-day default. Complete AI audit logs and the Obot Sentry workstation agent, which hooks Claude Code, Codex, Cursor and VS Code tool calls into the same schema, arrived in July 2026; v0.27.0 is the current stable. Pricing: free and open source (MIT); a licence key sets user and device limits.

CISO Assistant

Governance & compliance. CISO Assistant is the open-source GRC platform teams already run for ISO 27001 and SOC 2, and it ships the AI frameworks as loadable libraries: ISO/IEC 42001, the NIST AI RMF, the EU AI Act, MITRE ATLAS and the OWASP LLM checklist, plus a cross-mapping between ISO 42001 and the Secure Controls Framework. Clone it, run the compose script, and you can work an ISO 42001 or AI Act assessment, attach evidence, track an AI risk register and generate audit reports. v4.2.1 shipped 2026-11-01. Pricing: free and unlimited community edition (AGPL-3.0); the enterprise directory is commercial.

⚔️ Offensive & Recon

ffuf

ffuf is the Go web fuzzer most testers reach for first: fast, scriptable, and happy to brute directories, virtual hosts, parameters or POST bodies from a wordlist against a target you are authorized to probe. go install github.com/ffuf/ffuf/v2@latest and you are fuzzing in one command, with matchers and filters to cut the noise down to real findings. It stays the default because it does one job at speed and composes cleanly into larger recon chains. Pricing: free and open source (MIT).

Certipy

Certipy enumerates and abuses Active Directory Certificate Services, the misconfiguration class that quietly hands out domain escalation in a lot of enterprise AD. pip install certipy-ad, point it at a domain you are engaged to test, and it finds vulnerable certificate templates and the paths they open. It is the established pick for AD CS work and was last updated 2026-07-30. Pricing: free and open source (MIT).

PentAGI

PentAGI is a self-hosted multi-agent system that plans and runs pentest tasks inside a disposable Kali container, with a web UI, a GraphQL and REST API, Langfuse tracing and a choice of 13 model providers. Copy the example environment file and docker compose up. v2.2.0 on 2026-10-05 rebuilt the sandbox to close CVE-2026-14784, which exposed the host Docker socket to every worker when the shipped defaults were left in place, so run 2.2.0 or later and keep DOCKER_INSIDE=false. Pricing: free and open source (MIT); a free enterprise-edition key unlocks extra features.

🛡️ Defensive & Detection

Security Onion

Security Onion bundles Suricata, Zeek and the Elastic stack behind one analyst UI for network monitoring, log management and case work, as a single sensor or a distributed grid. Download the signed ISO and run the installer. The 3.x line is a rebuild: the 2.4 branch reached end of life on 2026-10-01, full-packet capture moved to Suricata, and 3.4.0 on 2026-11-02 added initial event correlation. Read the licence before you build a service on it. Pricing: free, source-available (Elastic License 2.0, no hosted/managed-service use); paid "Pro" add-ons.

Malcolm

Malcolm, from CISA and Idaho National Lab, ingests PCAP, Zeek logs and Suricata alerts, enriches them, and indexes everything in OpenSearch with an Arkime session viewer, dashboards, file carving and NetBox asset inventory. It leans toward ICS and OT networks, with industrial protocol parsers most stacks skip. Deploy from the installer ISO or clone and run the setup scripts. Releases are monthly; v26.10.0 on 2026-10-29 moved to Zeek 9.0.1 and several of this year's updates fixed holes in Malcolm itself, so keep it patched. Pricing: free and open source (Apache-2.0).

Tracecat

Tracecat is an open-source alternative to Tines and Splunk SOAR: response workflows and AI agents wired over your SIEM, EDR and ticketing tools, with case management and tables built in. Grab the release bundle and docker compose up. It hit 1.0 GA on 2026-09-16 after a long beta, rebuilt on a pure Temporal engine, and the 1.x line added passkey sign-in, AES-256-GCM database encryption and an nsjail sandbox executor. Pricing: free Open Source edition (AGPL-3.0); Helm, SCIM and air-gap are paid-tier only.

🔒 Privacy & Encryption

libsodium

libsodium is the hard-to-misuse crypto library you reach for when you want authenticated encryption, signatures, key exchange and password hashing with sane defaults and bindings in nearly every language. apt install libsodium-dev or build from the tarball. The reason it earns a fresh slot: 1.0.22 on 2026-04-09 added ML-KEM-768 and the X-Wing hybrid KEM, which the release notes recommend as the default KEM for most applications, bringing post-quantum key exchange into the mainstream library rather than a research fork. Pricing: free and open source (ISC).

liboqs

liboqs collects the NIST post-quantum algorithms behind one C API — ML-KEM, ML-DSA, SLH-DSA, FrodoKEM, HQC and more — with wrappers for Python, Go, Java and .NET and an OpenSSL 3 provider. Build it with CMake and Ninja. 0.16.0 on 2026-07-09 replaced SPHINCS+ with SLH-DSA, updated HQC to the 2025 spec, and fixed a verification out-of-bounds read tracked as CVE-2026-46344. The README is blunt that it is for prototyping and hybrid deployments, not as a drop-in for production secrets. Pricing: free and open source (MIT).

Delta Chat

Delta Chat is a messenger that runs over email transport with OpenPGP end-to-end encryption, so there is no central server to trust and no phone number to hand over; its default chatmail relays issue random addresses you can also self-host. Install it from the download page on Android, iOS, desktop or the usual package managers. The March 2026 "zero metadata" release encrypted the mail headers that used to leak the social graph, randomised the outer date, and stripped key IDs from OpenPGP packets, with weekly releases since. Pricing: free and open source (MPL-2.0 core).

📦 AppSec & Supply Chain

Opengrep

Opengrep is the community-governed fork of Semgrep, started after Semgrep moved features behind a commercial licence, and backed by a group of AppSec vendors. Existing Semgrep-format rulesets run unchanged across 30-plus languages with SARIF output and native Windows support. Install with the shell script and run opengrep scan --config auto .. The tradeoff to know: cross-file dataflow stayed in Semgrep's paid product, so this is pattern-based SAST, not whole-program taint. v1.33.1 shipped 2026-10-29. Pricing: free and open source (LGPL-2.1).

cdxgen

cdxgen is the most complete CycloneDX generator going, producing SBOMs, cryptography BOMs and host BOMs from source trees, lockfiles, container images and live machines, then pushing them to Dependency-Track. brew install cdxgen or run the container. v13 moved the project out of the CycloneDX org and renamed the npm package to @cdxgen/cdxgen, so old install lines need updating; v13.6.0 on 2026-10-31 added an IDE-extension and agent-config inventory built to catch a tampered Claude Code build distributed through Cursor's marketplace. Pricing: free and open source (Apache-2.0).

bumblebee

bumblebee is a read-only scanner from Perplexity that inventories a developer machine — npm, PyPI, Go, RubyGems, Composer and Homebrew metadata, editor and browser extensions, MCP host configs and skill lockfiles — and matches it against exposure catalogs of known supply-chain compromises by exact name and version. It never runs a package manager. go install github.com/perplexityai/bumblebee/cmd/bumblebee@latest, then scan your home directory. Open-sourced 2026-05-22 after the npm and IDE-extension worms of the past year, and the threat catalogs keep landing. Pricing: free and open source (Apache-2.0).

Google Mantis

Mantis is Google's agent toolkit for finding, reproducing and patching vulnerabilities: a set of skills plus a reference harness running a multi-stage pipeline that threat-models, hypothesises, reproduces in a sandbox, then patches with adversarial verification. Clone the reference directory, run the installer, authenticate to a model provider and point it at a codebase. Released 2026-09-02. The repo is explicit that it is a demonstration, not a supported product, so run it in isolation and verify before filing anything it reports. Pricing: free and open source (Apache-2.0); you pay for model inference.

☁️ Cloud & Infra Security

Cartography

Cartography pulls your cloud estate into a Neo4j graph — AWS, Azure, GCP, Okta, GitHub, Kubernetes and more than 40 sources — so you can query attack paths instead of eyeballing consoles. pip install cartography, stand up Neo4j, and sync with read-only credentials; a rules engine flags public RDS, admin static keys, GuardDuty findings and workload-identity issues. It is the CNCF Sandbox default for cloud asset and permission mapping, with v0.144.0 on 2026-10-29. Pricing: free and open source (Apache-2.0).

Inspektor Gadget

Inspektor Gadget is an eBPF toolbox for the moments Falco or Tetragon are more opinionated than you want: individual gadgets like trace_exec, trace_open and trace_sni ship as signed OCI images you run on demand. kubectl krew install gadget && kubectl gadget deploy, or run ig standalone on a host, with Prometheus and OpenTelemetry export. v0.58.0 on 2026-11-04 added SELinux and rename tracing. It needs a 5.10+ kernel with BTF. Pricing: free and open source (Apache-2.0).

kubesplaining

kubesplaining is Cloudsplaining's idea carried into Kubernetes: 104 rules across RBAC over-privilege, pod-escape exposure and multi-hop escalation paths to cluster-admin or node escape, with remediation patches in the output. kubectl krew install kubesplaining and run it against a cluster you own. Created in April 2026, now in the krew index, with v1.3.0 on 2026-09-22 adding EKS access-entry export and the Kubernetes 1.36 mutating-admission injection path. It is single-maintainer and small, so read what it changes. Pricing: free and open source (Apache-2.0).

owLSM

owLSM runs a stateful Sigma rule engine inside the Linux kernel through eBPF LSM hooks and blocks at the hook rather than alerting after the fact, which is the line between detection and prevention. Build it in the provided CI image and run the binary; a compatibility script checks your host first. Cybereason open-sourced it in January 2026, and v0.9.0 on 2026-07-29 added aarch64. It wants a 5.14+ kernel on x86_64, has a slow cold start, and ships no packaged installer yet. Pricing: free and open source (GPL-2.0).

🤖 Security AI & Agents

NVIDIA SkillSpector

SkillSpector statically scans AI agent skills and MCP skill packages before you install them, walking the code for exec, eval and subprocess calls, tracking taint from environment variables to network sinks, and matching 64 patterns across prompt injection, data exfiltration and tool poisoning. It returns findings, a risk score and an install-or-not verdict. uv tool install from the git repo, then skillspector scan ./my-skill/. It broke out in August 2026 and is past 24,000 stars; v2.13.3 on 2026-11-02 is the current release. Install from git, not PyPI. Pricing: free and open source (Apache-2.0).

Cisco AI Defense MCP Scanner

This scanner points at a local or remote MCP server, its config files or its tool code and looks for prompt injection in tool descriptions, tool poisoning, misleading server instructions and vulnerable dependencies, with pluggable analysers. The YARA analyser is fully offline and needs no key; LLM and Cisco API analysers go deeper, and a behavioural command checks whether tool code matches its documented behaviour. uv tool install cisco-ai-mcp-scanner. v4.8.7 on 2026-10-20 migrated to the MCP 2.0 protocol. Pricing: free and open source (Apache-2.0); only the Cisco AI Defense analyser is paid.

Fickling

Fickling, from Trail of Bits, decompiles Python pickle bytecode back to readable code and traces the pickle VM without executing it, so you can see what a .pkl or PyTorch checkpoint will run rather than trusting a yes/no verdict. pip install fickling, then fickling --check-safety -p model.pkl; it can also install a runtime hook that makes pickle.load refuse unsafe files. v0.1.13 on 2026-11-03 is a hardening release that closed four safety-check bypasses and dropped Python 3.10. Pricing: free and open source (LGPL-3.0).

🎓 Labs & Learning

CTFd

CTFd is the platform most capture-the-flag events actually run on: challenge types, hints, team or solo modes, bracketed scoreboards, plugins and themes, and an admin API. docker compose up from the repo and you are hosting an event. It is the established default here because it is the one everyone has already used from the player side; 3.8.9 on 2026-10-22 is a security maintenance release. Pricing: free to self-host (Apache-2.0); hosted plans sold separately.

WebGoat

WebGoat is OWASP's deliberately insecure Java and Spring app, built as guided lessons with a WebWolf attacker helper so you work each vulnerability class with instructions rather than staring at a blank target. docker run it bound to localhost and open the lesson UI. v2026.5 on 2026-10-12 added lessons on command injection, an HMAC timing attack, HTTP parameter pollution and password storage, plus a chained multi-step challenge. Pricing: free and open source (GPL-2.0-or-later).

Ludus

Ludus turns a Proxmox box into a range builder: define a multi-VM environment in YAML — Windows AD, Linux hosts, routers, a SIEM — and it builds them with Packer and Ansible into per-user isolated networks with snapshots and a testing mode that blocks internet access. Install with the one-line script as root on Debian or Proxmox. The 2.x line, starting March 2026, added blueprints, cluster mode and a plugin system, through 2.3.10 in October. It is hardware-hungry, wants 32 GB of RAM and a wired NIC, and the GitHub mirror is archived, so use GitLab. Pricing: free core (AGPL-3.0); some 2.x features are enterprise.