v1.2.1

Improvements

• Richer tool definitions for list_tools and manage_remote_hosts — fuller descriptions and Literal enum types for method / action, so the input schema every MCP client sees is self-documenting and validated.

Maintenance

• Add glama.json (Glama maintainer metadata) and skip image rebuilds for doc-only / glama.json changes.

Full changelog: v1.2.0...v1.2.1

v1.2.0

Features

• MCP server: crypto solver venv, install-aware advisors, and AI-directive sanitizing.

Fixes

• MCP registry manifest is now accepted — the OCI package no longer carries a version field, and the image ships the required io.modelcontextprotocol.server.name label.
• Resolve HOME before set -u reads; install uv in the prerequisite phase.

Docs & maintenance

• Document DeepSeek Harness (dsh) as an MCP client; correct the PUBLISHING guide; add the Glama score badge.
• Image builds skip doc-only commits via paths-ignore.
• Dependency bumps: uv, CodeQL action, harden-runner.

Full changelog: v1.1.1...v1.2.0

v1.1.1

Patch release.

Distribution & discoverability

• Prebuilt container image on GHCR — no local build needed: docker run --rm ghcr.io/26zl/cybersec-toolkit --profile ctf
• MCP registry manifest (server.json) for io.github.26zl/cybersec-toolkit
• GitHub Discussions enabled with Q&A / Show-and-tell templates and issue routing

Maintenance

• make bump VERSION=x.y.z syncs every release surface in one step; the release-version validator now also covers CITATION.cff and server.json
• Rolls up prior fixes landed on main since v1.1.0

Full changelog: v1.1.0...v1.1.1

v1.1.0

Highlights

MCP server security hardening

• Blocked command-execution options on allowlisted utilities that reached RCE with CYBERSEC_MCP_ALLOW_SCRIPTS=0 (sed e/r/w/s///e, zip --unzip-command/-TT, exiftool -if/-p, wget --use-askpass, openssl -engine).
• Guarded scanner output flags (nmap -oN/-oA, tcpdump/tshark -w) against sensitive-path clobber.
• Timeout now kills the whole process group so forked children die; DNS resolution runs in a bounded pool; credential-bearing long flags (--api-token, --client-secret) are redacted from the audit log.

Installer: provenance and safe uninstall

• Tools already present before install are recorded as existing, so --rollback and remove.sh never uninstall what the user already had.
• Rollback now covers source/special/snap/docker and system packages, and only forgets a .versions entry once the tool is actually gone.
• Backup restore is per-target transactional with an archive-bomb guard and a find-availability check.

Cross-distro correctness

• lib/distro_compat.tsv package names audited against live apt/dnf/pacman/zypper repos; the CI package-name check is now blocking.
• New scripts/test-distros.sh (make test-distros) smoke-tests install across all four package managers via rootless podman.

Other

• scripts/agent-guard.sh enforces MCP-first target work; scripts/check_doc_links.py reports dead doc links.
• Pinned pip-audit and markdownlint; refreshed AGENTS/CLAUDE guidance; trimmed skill descriptions under the spec cap.

Full Changelog: v1.0.0...v1.1.0

v1.0.0

First stable release. Modular installer for 580+ security tools on Linux and Termux, plus an authorization-gated MCP server and 872 agent skills.