Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
import { afterEach, describe, expect, it, vi } from 'vitest'
const originalEnv = process.env
async function loadConfig(raw: string | undefined) {
vi.resetModules()
process.env = { ...originalEnv }
if (raw === undefined) {
delete process.env.CONTEXT_MAPPER_RUNTIME_NAMESPACES
} else {
process.env.CONTEXT_MAPPER_RUNTIME_NAMESPACES = raw
}
return import('./config')
}
afterEach(() => {
process.env = originalEnv
vi.resetModules()
})
describe('CONTEXT_MAPPER_RUNTIME_NAMESPACES', () => {
it('defaults to the four platform runtime namespaces', async () => {
const { config } = await loadConfig(undefined)
expect(config.runtimeNamespaces).toEqual([
'mcp-server',
'mcp-host',
'sandbox-recipes',
'rpc-proxy',
])
})
// N5 regression guard: listing llm-hooks here makes ensureInfrastructurePolicies
// emit `allow-dns-egress-llm-hooks` with podSelector: {} (namespace-wide DNS)
// and `allow-hcc-api-egress-llm-hooks` selecting clerum.io/managed-by, which
// hook pod templates carry. Both reach a pure `/v1` responder that declared no
// egressBindings, contradicting "no implicit DNS". The llm-hooks baseline is
// static (deploy/base/llm-hooks/networkpolicies.yaml) precisely so HCC's
// namespace-wide infra pass never selects those pods.
it('omits llm-hooks so hook pods get no implicit DNS or HCC-gateway egress (N5)', async () => {
const { config } = await loadConfig(undefined)
expect(config.runtimeNamespaces).not.toContain('llm-hooks')
})
it('honours an explicit override', async () => {
const { config } = await loadConfig('mcp-server,sandbox-recipes,rpc-proxy,sandbox-ui')
expect(config.runtimeNamespaces).toEqual([
'mcp-server',
'sandbox-recipes',
'rpc-proxy',
'sandbox-ui',
])
})
})