Self Healing Agentic GitOps Pipeline

Overview

Production-style local demo of a small GitOps pipeline where Argo CD deploys a broken Kubernetes app, Prometheus detects the failure, and a Python AI agent proposes a manifest-only fix through a GitHub pull request.

The project is intentionally small:

• 1 microservice: self-healing-api
• 1 failure mode: missing REQUIRED_GREETING env var causing CrashLoopBackOff
• 1 safe fix: add the missing env var to manifests/apps/self-healing-api/deployment.yaml

GitOps Flow

Code -> Docker image -> Git manifest -> Argo CD -> Kubernetes cluster

Failure -> Prometheus alert -> Agent observes -> Agent reads logs -> LLM diagnoses -> Agent edits Git manifest -> Agent opens PR -> Human merges -> Argo CD auto-syncs -> Pod recovers

The agent never patches the cluster. It only edits files in the manifests Git repository.