chpasswd with ubuntu:changeme and ssh_pwauth: true gives every new server a login password that other users and processes on it can often read from the metadata service, and opens SSH to password guessing. And "# cloud-config" with one extra space makes cloud-init skip the whole file without an error, so none of your users, keys or security settings are applied. This skill finds those and the other cloud-init mistakes below before you launch, and tells you how to fix each one.

Your agent runs a scanner on #cloud-config files, cloud.cfg.d files, Ubuntu autoinstall files, files named like user-data, and cloud-config embedded in Terraform heredocs, CloudFormation UserData, KubeVirt, LXD profiles and Vagrantfiles. It knows cloud-init's defaults (passwords locked, root login off, chpasswd type hash), the hash formats the legacy chpasswd list accepts, the legacy hyphenated keys, and Terraform cloudinit_config parts that need no header; it decodes base64 and gzip write_files content and treats template placeholders as set but unknown. It never runs cloud-init or the commands. It returns BLOCK, REVIEW, or NOT FLAGGED, with the file, line, reason, and a safer approach. Secrets are never printed.

What it checks

BLOCK (fix before shipping)

• C01: passwords in clear text (plain_text_passwd, chpasswd text or legacy list, password, a plain value where a hash belongs, the autoinstall identity password, and commands such as echo or printf into chpasswd or passwd, chpasswd <<< and openssl passwd with a literal)
• C02: SSH password login turned on (ssh_pwauth, PasswordAuthentication yes) while an account gets a usable password (a root-only password counts only with PermitRootLogin yes, since OpenSSH defaults to prohibit-password)
• C03: code run as root that is fetched over plain HTTP or with certificate checks off: curl or wget piped into a shell (also through tee) or saved and then run, made executable or installed with dpkg, rpm or apt, rpm or dnf installs from http URLs, apt trusted=yes or yum/zypper gpgcheck off over http, executable write_files sources, Ansible, Puppet and Chef installer URLs
• C04: private keys and credentials in user-data: PEM keys (also in base64 or gzip content and ssh_keys), known token formats, passwords in URLs, Authorization headers, WireGuard keys, secret-named settings, and secrets on command lines (docker or az login -p, --token, htpasswd -b, rabbitmqctl add_user and others)

REVIEW (check before release)

• C11: scripts or packages downloaded over https and run or installed with no checksum or signature check
• C12: root login allowed (disable_root: false, root users entry or password, PermitRootLogin yes)
• C13: NOPASSWD sudo for every user, a group, a system account or a user who also has a password, in users entries or sudoers files written by write_files or echo/tee (key-only admins are not reported)
• C14: lock_passwd: false for a user who gets a password
• C15: TLS or signature checks off over https (curl -k, apt trusted=yes, gpgcheck off, --allow-unauthenticated, snap --dangerous, pip --trusted-host and others)
• C16: write_files modes that everyone can write, or an unquoted mode like 644 that YAML reads as decimal
• C17: files cloud-init will not read as cloud-config (misspelled header, also inside a heredoc, a comment or --- before it, no header at all, a byte-order mark or UTF-16)

What's included

• The scanner (Python 3.9+, standard library only)
• A rules guide with the reason, safer approach and official documentation link for each rule
• A risky and a safe example project (cloud-config, Terraform and a NoCloud seed), and the real report the scanner produces on the risky one

Limits

It reads each file as written. Values a template fills in (Terraform, Jinja, Helm) are unknown, MIME multi-part user-data, #include URLs, compressed user-data, user-data shell scripts and vendor-data are not read, and cloud-config stored as a JSON string is not found. Shell commands are matched by pattern, not parsed. Ubuntu Pro tokens and Red Hat activation keys are not reported. Some official cloud-init examples are reported on purpose because they show a plain password, an account password, a Basic auth header, a demonstration private key or a host private key in user-data. NOT FLAGGED is not a guarantee.

Contents

One ZIP file with 16 files: SKILL.md, the scanner in scripts/ (7 Python files), references/rules.md (every rule with its reason and safer approach), and examples/ (6 example files plus report.md, the real scanner output on the risky example).

How to use: This is an agent skill: a folder with a SKILL.md file and Python scripts that an AI coding agent such as Claude Code, Cursor, or Codex CLI can run. Unzip it into your agent's skills folder (for Claude Code: ~/.claude/skills/), then ask your agent to check your project. The scripts also run on their own with Python 3.9 or newer, with nothing to install. They read your files only: no network access, and nothing is sent anywhere.