Learn

AI for DevSecOps and CI/CD pipelines

Pipeline files, Dockerfiles and policy files are repository files. Topsoil can plan a change, edit them in a worktree and open it for a person to review. Your CI system still runs the jobs.

Pipeline files are project files

Topsoil is a private AI platform for engineering. On a pipeline, that means the same loop you use for application code: read the repository, write a plan, wait for approval, edit in an isolated git worktree, and open a pull request or a merge request. A person merges. Topsoil does not approve the change and does not merge it.

Topsoil recognises the CI configuration a repository already carries:

File in the repositoryWhat Topsoil calls it
.github/workflowsGitHub Actions
.gitlab-ci.ymlGitLab CI
bitbucket-pipelines.ymlBitbucket Pipelines
azure-pipelines.ymlAzure Pipelines
JenkinsfileJenkins
.circleci/config.ymlCircleCI

Dockerfiles and policy files that live in the same repository are edited the same way. Plan mode stays read-only until you approve, which is the pause you want before a workflow definition changes. A command such as a test runner, terraform plan or cfn-lint runs only when it matches your allowlist.

The change opens on GitHub, self-managed GitLab, or Bitbucket Data Center. Bitbucket Cloud is not supported. Topsoil recognises an Azure DevOps remote and does not open the pull request there. How the gates work is in how engineering changes get approved.

What a review can read

In the Terminal, /review 42 reads pull request or merge request 42: the title, the description, the comments, the CI status and the diff. /review with no number lists the open ones and reviews the working tree. /pr-comments 42 reads the comments. The review is read-only. It does not approve, decline or merge.

ForgeCI status Topsoil reads
GitHubChecks on the pull request.
Self-managed GitLabThe head pipeline and its jobs. The first 200 jobs are the ones reported.
Bitbucket Data CenterA build status already posted on the pull request.

GitHub checks, self-managed GitLab pipelines and a Bitbucket build status come from source control. Bamboo and Jenkins are separate servers. In the desktop app, Settings for CI offers Source control, Bamboo or Jenkins. Bamboo asks for the base URL and the plan key. Jenkins asks for the base URL, the username the API token belongs to, and the job path. The token goes in the operating system keychain.

In the Terminal, /ci lists the same three choices. /ci bamboo and /ci jenkins save the server in your config. The token is not an argument on that command.

A CI job that posts a review

Your pipeline can run one Topsoil step that reviews the pull request or merge request and posts the review as a comment:

topsoil --ci-policy ci/review-only.policy.json --post-review "/review 42"

That comments. It does not approve. The policy file is an allowlist you commit in the repository. A denial exits 2, and the line names the rule: commands, paths, hosts, writes, ask, or execution-config. The same step targets GitHub, self-managed GitLab, or Bitbucket Data Center, using the forge settings on the machine that runs the job.

An unattended run needs a budget ceiling or it does not start. A budget stop exits 3, distinct from a failure (1) or a run that did not start (2), so the pipeline can tell a spend limit from a failed test. Budgeting on Amazon Bedrock names each code. There is no scheduler inside Topsoil. A job that has to survive a reboot is cron or launchd running topsoil --yes.

The same step is what a GitLab CI job, a Jenkinsfile or a Bamboo script task runs. On GitLab, the job token cannot post the note, so the job uses a token allowed to write the merge-request note. Mask that variable in CI.

What stays with the tools you already run

Your CI system runs the jobs. Topsoil edits the definitions in git, can read the status the forge exposes, and can post a review comment. SAST, SCA and the other scanners you already run keep doing that work. Applying a cloud change stays on the review and deploy path you already use.

Model calls for that work go to Amazon Bedrock in your own AWS account, commercial or AWS GovCloud (US). How the call is made, including tool use, is on Amazon Bedrock. Which engineering files are in scope beyond pipelines is in software, infrastructure and pipeline work.

Questions

Frequently asked questions

Does Topsoil run our CI pipelines?

Your CI system runs the jobs. Topsoil can edit pipeline files in git, read CI status on a pull request or merge request where the forge supports it, and post a review as a comment.

Can a pipeline job ask Topsoil to review a change?

Yes. A step runs topsoil with a CI policy and --post-review. It posts a comment. It does not approve or merge. A policy denial exits 2. A budget stop exits 3.

Which pipeline files can it edit?

The CI configuration in the repository, including GitHub Actions, GitLab CI, Bitbucket Pipelines, Azure Pipelines, Jenkinsfile and CircleCI, plus Dockerfiles and policy files that live beside them. The change opens on GitHub, self-managed GitLab, or Bitbucket Data Center.

Does this replace our scanners?

No. Topsoil edits repository files and can post a review. SAST, SCA and the scanners you already run keep their jobs.

Try it on your own codebase

A pilot runs in your own AWS account, with your models and your material, and ends with findings you can act on.