Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
import { describe, expect, it } from 'vitest'
import { readFileSync } from 'node:fs'
import path from 'node:path'
describe('host-context-controller RBAC manifest', () => {
it('can patch McpServer metadata annotations in the mcp-server namespace', () => {
const manifest = readFileSync(
path.join(__dirname, '../../deploy/base/mcp-server/rbac.yaml'),
'utf8'
)
const role = manifest
.split('name: host-context-controller\n namespace: mcp-server\nrules:')[1]
?.split('---')[0]
expect(role).toBeTruthy()
const mcpServerRule = role?.match(/resources: \["mcpservers"\]\n\s+verbs: \[([^\]]+)\]/)
expect(mcpServerRule?.[1]).toContain('"patch"')
})
it('can reconcile LlmHook workloads and read pods in the llm-hooks namespace', () => {
const manifest = readFileSync(
path.join(__dirname, '../../deploy/base/llm-hooks/rbac.yaml'),
'utf8'
)
const role = manifest
.split('name: host-context-controller\n namespace: llm-hooks\nrules:')[1]
?.split('---')[0]
expect(role).toBeTruthy()
// llmhooks: watch + patch (status write-back path)
const llmhooksRule = role?.match(/resources: \["llmhooks"\]\n\s+verbs: \[([^\]]+)\]/)
expect(llmhooksRule?.[1]).toContain('"watch"')
expect(llmhooksRule?.[1]).toContain('"patch"')
// llmhooks/status: patch
const statusRule = role?.match(/resources: \["llmhooks\/status"\]\n\s+verbs: \[([^\]]+)\]/)
expect(statusRule?.[1]).toContain('"patch"')
// pods get/list — required for status.observedDigest (live pod image)
const podsRule = role?.match(/resources: \["pods"\]\n\s+verbs: \[([^\]]+)\]/)
expect(podsRule?.[1]).toContain('"get"')
expect(podsRule?.[1]).toContain('"list"')
// networkpolicies + deployments + services managed by HCC
expect(role).toContain('resources: ["networkpolicies"]')
expect(role).toContain('resources: ["deployments"]')
expect(role).toContain('resources: ["services"]')
// Bound to the HCC ServiceAccount in control-plane
expect(manifest).toContain('kind: ServiceAccount')
expect(manifest).toMatch(/name: host-context-controller\n\s+namespace: control-plane/)
})
it('can reconcile GFS writer PodDisruptionBudgets in the gfs namespace', () => {
const manifest = readFileSync(path.join(__dirname, '../../deploy/base/gfs/rbac.yaml'), 'utf8')
const role = manifest
.split('name: host-context-controller-gfs-runtime\n namespace: gfs')[1]
?.split('---')[0]
expect(role).toBeTruthy()
expect(role).toContain("apiGroups: ['policy']")
expect(role).toContain("resources: ['poddisruptionbudgets']")
expect(role).toContain("verbs: ['get', 'create', 'update', 'delete']")
})
it('can read the shared LLM allowlist ConfigMap in the mcp-host namespace', () => {
const manifest = readFileSync(
path.join(__dirname, '../../deploy/base/mcp-host/rbac.yaml'),
'utf8'
)
const role = manifest
.split('name: host-context-controller-host-runtime\n namespace: mcp-host')[1]
?.split('---')[0]
expect(role).toBeTruthy()
expect(role).toContain("resources: ['configmaps']")
expect(role).toMatch(/verbs: \['get', 'watch', 'list'\]/)
expect(role).not.toContain('llm:codex:execute')
})
})