Putting Cordon to work: choose your path
The tutorials from here on are ready-made implementations: each one is something you copy into a machine, a repository or a pipeline and it works. They start with one developer's laptop and end with an organisation-wide rollout, and each level adds to the one before rather than replacing it.
┌──────────────────────────────────────────────────────────────────────────┐ │ LEVEL 1 one machine │ │ 33 On your laptop ........ vet a package before you install it, │ │ scan a project, git hooks that fail closed│ │ 34 Guard your installs ... pnpm, npm and pip: checked before fetch, │ │ read before anything runs │ ├──────────────────────────────────────────────────────────────────────────┤ │ LEVEL 2 every repository's pipeline │ │ 35 GitHub Actions ........ the action, SARIF, dependency review │ │ 36 GitLab CI ............. the template, code quality, merge gate │ │ 37 Other CI systems ...... Jenkins, Azure, Bitbucket, CircleCI, │ │ Buildkite, and any shell │ │ 38 Containers and images . an image before it is pushed or deployed │ ├──────────────────────────────────────────────────────────────────────────┤ │ LEVEL 3 the organisation │ │ 39 Rolling out ........... one policy ceiling, baselines, expiring │ │ suppressions, notifications │ │ 40 Air-gapped ............ mirrored intel, offline scans, a verified │ │ install │ │ 41 AI agents ............. the MCP server for each coding agent │ │ 42 From your own code .... the Python API and the JSON result │ └──────────────────────────────────────────────────────────────────────────┘
Do you run code from a package registry on this machine?
│
├── yes ──► 33, then 34: nothing new lands unread
│
Does a pipeline build or deploy what you write?
│
├── yes ──► 35, 36 or 37, whichever runs your pipeline; 38 if it ships images
│
Are there many repositories, or rules you must prove you follow?
│
└── yes ──► 39 for the policy; 40 if the machines are offline; 41 if
coding agents install things for you; 42 to wire results
into your own systemsCordon READS. It never runs the code it is scanning, and nothing about your code leaves the machine: default asks OSV for advisories published since this release: the same requests for everyone, naming nothing you scan --offline no request at all (an air-gap; tutorial 40) --online also fetches the packages being checked, from their own registries, to read them Every level uses the same exit codes, so a script, a hook and a pipeline all decide the same way: 0 clean 1 findings at or above --fail-on 2 scanner error 3 configuration error 4 incomplete
An incomplete scan is never a pass: a file that could not be read, or intel older than its limit, is said, and with --fail-on-incomplete it fails the step. Tutorial 29 explains each coverage note and what to do about it.
Next: 33 · On your laptop.