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 { CODEX_UNASSIGNED_CONNECTION_KEY } from '@clerum/codex-catalog-projection'
import { computeCodexPolicyHash } from '@clerum/llm-provider-attempt-contract'
import {
isPluginWorkloadSdkCodexBindingProof,
readVerifiedSdkOnlyCodexBinding,
} from './sdkOnlyCodexBinding'
describe('mcp-host sdk-only Codex binding proof', () => {
const model = 'gpt-5.6-luna'
const proof = {
connectionKey: 'team-plus',
catalogRevision: 4,
credentialRevision: 1,
model,
bindingHash: computeCodexPolicyHash({
model,
catalogRevision: 4,
credentialRevision: 1,
connectionKey: 'team-plus',
}),
}
it('rejects the unassigned connection sentinel', () => {
expect(
isPluginWorkloadSdkCodexBindingProof({
...proof,
connectionKey: CODEX_UNASSIGNED_CONNECTION_KEY,
})
).toBe(false)
expect(
readVerifiedSdkOnlyCodexBinding(
{
...proof,
connectionKey: CODEX_UNASSIGNED_CONNECTION_KEY,
bindingHash: computeCodexPolicyHash({
model,
catalogRevision: 4,
credentialRevision: 1,
connectionKey: CODEX_UNASSIGNED_CONNECTION_KEY,
}),
},
model
)
).toBeNull()
})
it('strips extra keys after the hash verifies', () => {
expect(readVerifiedSdkOnlyCodexBinding({ ...proof, extra: 'drop-me' }, model)).toEqual(proof)
})
})