dazio --version prints the version you have. Install covers upgrading. Each version’s downloads are on the Dazio releases page.

v0.1.32 - 2026-11-06

New

• The inventory now lists the remote connectors you have used in Claude Desktop, with the host each one connects to. Nothing from your conversations is kept.
• MCP servers declared inside Kiro agent files, in your home folder or in a project, now show up in the inventory, along with the agent file that declares them.
• MCP servers that come with Claude Code, Codex and Claude Desktop plugins are now inventoried on every scan, including those offered by a plugin marketplace you have added but not installed.
• The inventory now lists the MCP servers and skills that your installed Kiro powers bring.

Improved

• MCP servers and agent skills in projects kept outside your home folder, such as under /srv, are now inventoried when Claude Code, Codex or a VS Code-family editor has opened the project.
• Each agent skill in the inventory now carries a fingerprint of its files, so a skill whose contents change under the same name is no longer indistinguishable from the original.
• The inventory now records each remote MCP server’s host, whether it is switched off or still awaiting your approval, and whether it is set up for you, for one project, or by a plugin.
• Extensions for VS Code, Cursor and other VS Code-family editors that add MCP servers from their own code are now marked in the inventory, with the name of each MCP server source they declare.

Fixed

• Agent skills in a .agent/skills folder are now attributed to Antigravity, which reads them, rather than only to Factory Droid.
• Skills in OpenClaw’s former home folder (~/.clawdbot) are now credited to OpenClaw, and skills in a project’s .amp/skills folder to Amp, instead of to an unknown agent.
• MCP servers set up in a project’s .vscode/mcp.json, .cursor/mcp.json or .gemini/settings.json, and in Claude Desktop on Linux, now show up in the inventory.
• MCP servers set up in Cursor’s user settings folder (Cursor/User/mcp.json) now show up in the inventory, attributed to Cursor.
• MCP servers set up in a project’s .mcp.json or Kiro settings, or brought by a Claude Code plugin, no longer drop out of the inventory between full scans. A change to one already found shows up within minutes; a newly added one shows up at the next full scan, or right away with scan --scope mcp.
• MCP servers set up for opencode, in its global or project opencode.json or opencode.jsonc, or in a config named by OPENCODE_CONFIG or OPENCODE_CONFIG_DIR, now show up in the inventory.
• MCP servers set up for Amazon Q in ~/.aws/amazonq/mcp.json, agents/mcp.json or the default agent config now show up in the inventory as Amazon Q’s.
• MCP servers and skills an administrator installs machine-wide for Claude Code, Codex and opencode are now inventoried, under the managed scope so you can tell them from your own.
• MCP servers set up for VS Code and VS Code Insiders in ~/.vscode/mcp.json, under the “mcp” section of the user settings, or in a named profile now show up in the inventory.
• MCP servers set up for Codex in a profile config (name.config.toml) or in a project’s .codex/config.toml now show up in the inventory. An agent home or config moved by an environment variable such as CODEX_HOME, CLAUDE_CONFIG_DIR or QWEN_HOME is now found when the variable is set in the environment the scan runs in, not only in a shell startup file, for every agent that has one.
• MCP servers set up for GitHub Copilot CLI, in ~/.copilot/mcp-config.json (or the folder COPILOT_HOME points to) or in a project’s .github/mcp.json, now show up in the inventory.
• MCP servers set up for a Windsurf project in its .windsurf/mcp.json now show up in the inventory.
• MCP servers configured in a project’s .gemini/mcp_config.json are now inventoried and credited to Antigravity.

v0.1.31 - 2026-11-04

Fixed

• With analytics off, background scans no longer send package names to the malware lookup; they check only against the malware feed copy on your machine.

v0.1.30 - 2026-10-30

Fixed

• When a malware check fails for a reason other than the known-malware list itself, dazio scan no longer tells you to run dazio feed refresh, and the error is logged.
• dazio status no longer reports a scan as free of malware when that scan could not check for malware; the scan row says it went unchecked, and why.
• dazio scan no longer says the known-malware list could not be read while the background service is running. When Dazio’s online threat lookup cannot be reached for some packages, the malware result now says those packages were not checked instead of reporting them clean.

v0.1.29 - 2026-10-28

Improved

• On macOS, brew upgrade dazio now restarts the background service onto the new version, so you no longer need to run dazio service restart afterwards.

v0.1.28 - 2026-10-28

Improved

• Malware checks now also consult Dazio’s online threat lookup for packages the local feed doesn’t cover, at install time and for recently installed packages.

v0.1.27 - 2026-10-22

New

• Dazio now tells you whether your machine is affected by the vulnerabilities fixed in the October 2026 Node.js security releases, by PHP2Shell, the remote code execution flaws in PHP (CVE-2026-109618 and CVE-2026-109619), and by the weak eMudhra CA keys. It finds installed runtimes and developer tools, including Node.js, PHP, Python, Java, Go, Bun, Deno, pip, uv, rustup, pixi and desktop apps built on Electron, plus the certifi, pip, uv, webpki, composer/ca-bundle and electron packages your projects install or pin, and names the release that fixes each one affected. A runtime installed by your Linux distribution is reported as possibly affected, with a pointer to your distro’s advisory. dazio result vulnerabilities lists every finding, and dazio result inventory --type runtime or --type dev_tool lists what is installed.
• Scans now check whether your machine still trusts the weak eMudhra CA keys, which anyone can recover to impersonate any website: macOS and Windows themselves, the system and Homebrew certificate bundles, the bundles that Python, Ruby, Java, conda, curl, Git for Windows and the Google Cloud and Azure CLIs ship, and any bundle you point to with SSL_CERT_FILE, REQUESTS_CA_BUNDLE, CURL_CA_BUNDLE, NODE_EXTRA_CA_CERTS or git’s http.sslCAInfo. Each finding says what to update, and on Windows it also tells you when automatic root certificate updates are turned off. dazio result anchors lists every one.

Improved

• Continuous protection now finds tools installed outside the system PATH, in Homebrew, ~/.local/bin or ~/.cargo/bin, and uses your shell’s PATH, CA bundle and version-manager settings, which dazio service install records when it sets up the background service. Package installs it watches through cargo now honour CARGO_HOME.

Fixed

• Dazio no longer trusts the eMudhra intermediate certificates issued with weak private keys, including their cross-signed copies, even where the system trust store still accepts them.

v0.1.26 - 2026-10-21

Improved

• dazio result inventory now separates agent skills by scope: skills your agents load in every session (user) apart from skills inside a project (project), with skills no known agent reads listed last. Filter with --scope user or --scope project; --json output carries each skill’s scope.
• result inventory lists project skills under the repo that holds them, with the repo path shown in the heading, and tags each skill with the agents that read it.
• dazio result inventory now tells plugin skills apart from the rest: skills an installed plugin ships show as plugin, and skills in a plugin marketplace’s local copy of everything it offers, installed or not, show as catalog, one line per marketplace with a count. List them with --scope catalog; a malware match there is as severe as anywhere else.
• The scan summary counts the agent skills each agent actually loads, with skills sitting in plugin marketplace catalogs shown as a separate total. A catalog left behind by an agent you no longer have no longer raises a “not installed” warning, and the suggested dazio result inventory command lists exactly the skills it counts.

Fixed

• On Windows, npm and pip packages installed globally on the host are no longer also listed under each WSL distro.
• The built-in trusted root certificates no longer include emSign ECC Root CA - G3, which Mozilla removed after certificates issued under it were compromised.

v0.1.25 - 2026-10-15

Improved

• Installed packages are rescanned every 15 minutes instead of every hour.
• A package installed globally with npm, pip, cargo or NuGet is checked against the malware feed within seconds, instead of waiting for the next package rescan.
• A package installed through safe-pkg that the malware feed flags shortly afterwards is reported as soon as the feed updates, instead of at the next package rescan.

Fixed

• safe-pkg no longer breaks uv older than 0.9.23: while such a uv is installed, it stays out of uv.toml and removes what it had written there.

v0.1.24 - 2026-10-15

New

• IDE extensions are now checked in every VS Code-family editor: VS Code Insiders, VS Code Server, VSCodium, Cursor and Windsurf remote servers, Trae, Kiro, Positron and Antigravity. Each extension is reported under the editor that holds it.
• Skills installed with the ClawHub CLI or OpenClaw are checked against the malware feed by publisher, name and installed version.
• Skills bundled in Claude Code plugins are checked against the malware feed by the publisher of the repository the plugin came from.
• Skills installed with npx skills are checked against the malware feed by the repository they were installed from.
• Many more AI coding agents are detected, with their versions: Junie, Hermes, Kiro CLI, OpenClaw, Pi, Mistral Vibe, Letta Code, Cline CLI, Continue (cn), OpenHands, Qwen Code, goose, CodeBuddy Code, Kilo, Crush, Amp, Kimi Code, Freebuff, Droid, Devin CLI and Grok CLI. Their skills and MCP servers are checked too.
• Scans now list AI desktop apps and IDEs: Claude Desktop, ChatGPT, Codex, Antigravity, Warp, Goose, OpenClaw, Zed, VS Code, VS Code Insiders, Cursor, Windsurf, Kiro, Trae and Tabnine.
• An agent home moved by an environment variable in your shell config (zsh, bash, fish or PowerShell) is scanned alongside the default one. Dazio reads those files; it never runs your shell.
• Agents installed with bun, pnpm, yarn, Nix, dpkg or pacman are found, with their versions.
• Skill folders listed in Hermes, OpenClaw and Mistral Vibe configs are checked.
• LM Studio MCP servers are read, and the newer Antigravity and Cline MCP config locations are followed.
• On Windows, Dazio now scans and protects your WSL distros from Windows, with nothing to install inside them. WSL2 distros need mirrored networking for install checks; dazio status names any distro left out.

Improved

• A skill is checked against the malware feed by its publisher and install folder, not only by the name it declares. Skills under Gemini, Kiro, Copilot, Windsurf and opencode are now labelled with their agent.
• Agents are found and versioned from what is installed on disk, without running them.
• A skill in a folder several agents read is listed under each of those agents.
• A malicious skill is reported once per folder, naming every agent that reads it.

Fixed

• IDE extensions that were uninstalled but not yet deleted from disk are no longer reported.
• A full scan reports each pip configuration finding once instead of twice.

v0.1.22

This release and earlier ones shipped without notes.