A proof-of-concept that showcases the best of agentic DevOps on GitHub: a real application that ships to Azure through a secure, multi-environment pipeline, with GitHub Continuous AI agents automating triage, documentation, diagnostics, and test improvement.
Continuous AI is to collaborative software what CI/CD is to builds and deployments: routine, automated, AI-assisted work that runs continuously in the background of your repository. This repo demonstrates that vision end-to-end.
| Capability | How it's shown here |
|---|---|
| Ships to Azure | .NET 10 minimal API → container → Azure Container Apps via Terraform |
| Secure by default | GitHub OIDC (no cloud secrets), managed identity, ACR without admin creds |
| Quality gates | dotnet format, xUnit tests, 60% coverage gate, tflint, actionlint, markdownlint |
| DevSecOps scanning | Trivy (filesystem, IaC config, container image) + CodeQL + Dependabot |
| Multi-environment | Dev → Test → Prod promotion with manual approval before production |
| Continuous AI | Four agentic workflows (gh-aw) running on GitHub Copilot |
These live as Markdown in .github/workflows/ and compile to
locked GitHub Actions (*.lock.yml) via gh aw compile. They use safe-outputs
— the agent never gets write tokens directly; its proposed actions (labels,
comments, PRs, issues) are applied by a separate, minimally-scoped job.
| Agent | File | Trigger | Continuous AI pillar |
|---|---|---|---|
| Issue Triage | issue-triage.md | Issue opened/reopened | Continuous Triage |
| Doc Updater | doc-updater.md | Push to main |
Continuous Documentation |
| CI Doctor | ci-doctor.md | CI/CD run fails | Continuous Repair |
| Test Improver | test-improver.md | Weekly + manual | Continuous Quality |
The engine is GitHub Copilot, so no third-party model API keys are required.
flowchart TB
dev[Developer] -->|push / PR| repo[(GitHub Repository)]
subgraph GH[GitHub]
repo --> ci[CI: format, test, coverage, tflint, Trivy, CodeQL]
repo --> cd[CD: build, scan, promote]
repo --> agents[Continuous AI agents<br/>triage · docs · ci-doctor · tests]
end
cd -->|OIDC, no secrets| azlogin{{Azure OIDC login}}
subgraph AZ[Azure]
azlogin --> tf[Terraform: ACR · Log Analytics · Container Apps]
tf --> aca[Azure Container App<br/>.NET 10 API]
acr[(Azure Container Registry)] --> aca
mi[User-assigned Managed Identity<br/>AcrPull] --> aca
end
agents -.safe-outputs.-> repo
flowchart LR
build[Build image<br/>+ Trivy scan gate] --> dev[Deploy Dev<br/>auto]
dev --> test[Deploy Test<br/>auto]
test -->|manual approval| prod[Deploy Prod<br/>protected]
Each environment is fully isolated: its own resource group, ACR, Container Apps environment, managed identity, and Terraform state file.
.
├── src/Api/ # .NET 10 minimal API (Task CRUD) + Dockerfile
├── tests/Api.Tests/ # xUnit unit + integration tests
├── infra/ # Terraform for Azure Container Apps (multi-env)
│ ├── envs/ # dev/test/prod tfvars + backend configs
│ └── bootstrap/ # remote state + GitHub OIDC identities (run once)
├── .github/
│ ├── workflows/ # CI, CD, CodeQL + agentic (*.md → *.lock.yml)
│ └── dependabot.yml
├── AgenticDevOps.sln
└── README.md
A small Task API (in-memory store) that is realistic enough to exercise the whole pipeline:
| Method | Route | Description |
|---|---|---|
GET |
/healthz |
Liveness/readiness probe |
GET |
/tasks |
List tasks |
GET |
/tasks/{id} |
Get a task |
POST |
/tasks |
Create a task |
PUT |
/tasks/{id} |
Update a task |
DELETE |
/tasks/{id} |
Delete a task |
OpenAPI is exposed at /openapi/v1.json in Development.
Prerequisites: .NET 10 SDK (see global.json).
dotnet restore AgenticDevOps.sln
dotnet test AgenticDevOps.sln -c Release --settings coverlet.runsettings
dotnet run --project src/ApiOn corporate machines with a restricted global NuGet feed, add
--configfile nuget.configto the restore command to force nuget.org.
Deployment is OIDC-based and runs from GitHub Actions — there are no cloud secrets stored in the repo. See docs/SETUP.md for the full one-time setup (bootstrap state + OIDC identities, GitHub Environments, and variables). In short:
terraform -chdir=infra/bootstrap apply— creates the Terraform state account and one GitHub-federated managed identity per environment.- Configure GitHub Environments (
dev,test,prod) and set theAZURE_CLIENT_ID/AZURE_TENANT_ID/AZURE_SUBSCRIPTION_ID/STATE_STORAGE_ACCOUNTvariables. Add required reviewers toprod. - Push to
main— CD builds, scans, and promotes Dev → Test → Prod.
- No long-lived cloud credentials — GitHub OIDC federation only.
- No ACR admin user — image pulls use a user-assigned managed identity with
AcrPull. - Defense-in-depth scanning — Trivy (deps, IaC, image) + CodeQL, all reported as SARIF to GitHub code scanning; CRITICAL/HIGH findings fail the build.
- Least-privilege agents — Continuous AI workflows read-only by default; mutations go through scoped safe-outputs jobs.
See SECURITY.md for details.