feature: helm chart ingestor

[george-lhj]

overall, the entire helm goal is a bit ambiguous and difficult, so I'm trying to create small PR's and agents and get immediate feedback.

this first agent takes in a github helm url, and then turns the helm chart into structured, json structured metadata.

Example:

input: https://github.com/Qiskit/qiskit-serverless/tree/main/charts/qiskit-serverless
output:

{'meta': {'repo': 'Qiskit/qiskit-serverless', 'ref': 'main', 'chartRoot': 'charts/qiskit-serverless', 'chartName': 'qiskit-serverless', 'version': '0.25.4'}, 'defaults': {'global': {'version': '0.25.4'}, 'platform': 'default', 'nginxIngressControllerEnable': False, 'nginx-ingress-controller': {'nameOverride': 'nginx-ingress-controller', 'fullnameOverride': 'nginx-ingress-controller'}, 'ingress': {'className': 'nginx', 'annotations': {'nginx.ingress.kubernetes.io/proxy-buffers-number': '4', 'nginx.ingress.kubernetes.io/proxy-buffer-size': '512k', 'nginx.ingress.kubernetes.io/proxy-body-size': '50m'}, 'tls': [], 'hosts': []}, 'gatewayEnable': True, 'gateway': {'nameOverride': 'gateway', 'fullnameOverride': 'gateway', 'useCertManager': False, 'image': {'repository': 'icr.io/quantum-public/qiskit-serverless/gateway', 'pullPolicy': 'IfNotPresent', 'tag': '0.25.4'}, 'application': {'siteHost': 'http://gateway:8000', 'rayHost': 'http://kuberay-head-svc:8265', 'auth': {'mechanism': 'mock_token'}, 'superuser': {'enable': True}, 'service': {'type': 'ClusterIP', 'port': 8000}, 'ray': {'nodeImage': 'icr.io/quantum-public/qiskit-serverless/ray-node:0.25.4', 'opensslImage': 'registry.access.redhat.com/ubi8/openssl:8.9-13', 'kubectlImage': 'alpine/k8s:1.33.4', 'gpuJobsConfig': '/tmp/gpujobs/gpu-jobs.json'}, 'limits': {'maxJobsPerUser': 2, 'maxComputeResources': 4}, 'nodeSelector': {'cpu': 'has-cpu: cpu', 'gpu': 'has-gpu: gpu'}, 'dependencies': {'dynamicDependencies': 'requirements-dynamic-dependencies.txt'}}, 'cos': {'claimName': 'gateway-claim'}, 'secrets': {'secretKey': {'create': True, 'name': 'gateway-secret-key', 'key': 'django-secret-key', 'value': 'CHANGE_ME'}, 'servicePsql': {'create': True, 'name': 'service-psql-binding', 'key': {'adminPassword': 'admin-password', 'databaseName': 'database-name', 'databasePassword': 'database-password', 'host': 'database-host', 'port': 'database-port', 'userName': 'user-name'}, 'value': {'databaseName': 'serverlessdb', 'databasePassword': 'serverlesspassword', 'host': 'postgresql', 'port': 5432, 'userName': 'serverlessuser'}}, 'superuser': {'create': True, 'name': 'gateway-superuser', 'key': {'name': 'name', 'password': 'password', 'email': 'email'}, 'value': {'name': 'admin', 'password': 'passw0rd', 'email': 'admin@examplemail.io'}}}}, 'kuberayOperatorEnable': True, 'kuberay-operator': {'image': {'repository': 'quay.io/kuberay/operator', 'tag': 'v1.3.2', 'pullPolicy': 'IfNotPresent'}}, 'postgresqlEnable': True, 'postgresql': {'nameOverride': 'postgresql', 'fullnameOverride': 'postgresql', 'auth': {'enablePostgresUser': False, 'database': 'serverlessdb', 'username': 'serverlessuser', 'password': 'serverlesspassword'}}, 'createPvc': True, 'storageClassName': 'manual', 'claimName': 'gateway-claim', 'cos': {'bucket': 'BUCKETNAME-CHANGEME', 'endpoint': 'ENDPOINT-CHANGEME', 'secretName': 'SECRETNAME-CHANGEME', 'storageSize': '10Gi'}}, 'schema': None, 'discoveredKeys': ['claimName', 'cos.bucket', 'cos.endpoint', 'cos.secretName', 'cos.storageSize', 'createPvc', 'gateway.application.ray.nodeImage', 'ingress.annotations', 'ingress.hosts', 'ingress.tls', 'platform', 'storageClassName'], 'requiredHints': [], 'docs': None}

[george-lhj]

would appreciate any feedback regarding the output strucutre, if im missing anything regarding helm chart logic itself.

Current Gap - The raw output is comprehensive but not actionable. We have all the chart data, but it's not organized in a way that downstream automation can easily use.

my next step if this is approved:

HelmSchemaResolver Agent
Goal: Transform the raw chart data into a clean, typed field catalog
Input: ChartBundle JSON (what we have now)
Output: Normalized field definitions with types, defaults, and requirements

[psschwei]

I think this is fine. At some point though you probably want to see if some of this can be moved down to the LLM (or maybe as a tool for an LLM)