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 {
type FileReferenceV1,
buildGfsFileReference,
classifyBytes,
} from '@clerum/gfs-interaction-policy'
import { resolveFileReferences } from '../fileReferenceResolver'
import type { FileReferenceGfscClient } from '../fileReferenceResolver'
import { classifyGfsReferenceSurfaces } from '../gfsReferenceSurfaces'
describe('classifyGfsReferenceSurfaces', () => {
it('keeps inline text independent from workspace delivery', () => {
expect(
classifyGfsReferenceSurfaces(
{ byteLength: 8192, reader: 'text', modelImageInput: 'unsupported' },
{ workspaceFile: false, localExecutor: false, visual: false }
)
).toEqual({
metadata: true,
inline: true,
workspace: false,
localExecutor: false,
visual: false,
})
})
it('separates workspace delivery from approved local execution for the incident size', () => {
const reference = {
byteLength: 3_836_961,
reader: 'text' as const,
modelImageInput: 'unsupported' as const,
}
expect(
classifyGfsReferenceSurfaces(reference, {
workspaceFile: true,
localExecutor: false,
visual: false,
})
).toMatchObject({ inline: false, workspace: true, localExecutor: false })
expect(
classifyGfsReferenceSurfaces(reference, {
workspaceFile: true,
localExecutor: true,
visual: false,
})
).toMatchObject({ workspace: true, localExecutor: true })
})
it('requires producer image candidacy and runtime visual support', () => {
const reference = {
byteLength: 3_836_961,
reader: 'none' as const,
modelImageInput: 'candidate' as const,
}
expect(
classifyGfsReferenceSurfaces(reference, {
workspaceFile: true,
localExecutor: true,
visual: true,
})
).toMatchObject({ visual: true })
expect(
classifyGfsReferenceSurfaces(reference, {
workspaceFile: true,
localExecutor: true,
visual: false,
})
).toMatchObject({ visual: false })
expect(
classifyGfsReferenceSurfaces(
{ ...reference, modelImageInput: 'unsupported' as const },
{ workspaceFile: true, localExecutor: true, visual: true }
)
).toMatchObject({ visual: false })
})
it('rejects workspace delivery above the generic GFS source limit', () => {
expect(
classifyGfsReferenceSurfaces(
{ byteLength: 16 * 1024 * 1024 + 1, reader: 'text', modelImageInput: 'unsupported' },
{ workspaceFile: true, localExecutor: true, visual: true }
)
).toMatchObject({ inline: false, workspace: false, localExecutor: false })
})
it('attaches fail-closed delivery surfaces to available resolver results', async () => {
const resourceId = '0123456789abcdef0123456789abcdef'
const built = buildGfsFileReference({
drive: 'main',
resourceId,
gfsUri: `gfs://main/${resourceId}`,
version: 7,
name: 'incident.csv',
declaredMediaType: 'text/csv',
byteLength: 3_836_961,
classification: classifyBytes({
bytes: new Uint8Array(),
totalByteLength: 3_836_961,
declaredMediaType: 'text/csv',
filename: 'incident.csv',
}),
})
if (!built.ok) throw new Error(built.message)
const reference = built.value as FileReferenceV1
const client: FileReferenceGfscClient = {
resolve: async () => ({
ok: true,
data: {
kind: 'file',
rid: resourceId,
drive: 'main',
resourceId,
gfsUri: `gfs://main/${resourceId}`,
name: 'incident.csv',
version: 7,
bytes: 3_836_961,
},
}),
}
const failClosed = await resolveFileReferences([reference], client)
expect(failClosed.ok).toBe(true)
if (!failClosed.ok) return
expect(failClosed.resolutions[0]!.surfaces).toEqual({
metadata: true,
inline: false,
workspace: false,
localExecutor: false,
visual: false,
})
const enabled = await resolveFileReferences([reference], client, undefined, {
workspaceFile: true,
localExecutor: true,
visual: false,
})
expect(enabled.ok).toBe(true)
if (!enabled.ok) return
expect(enabled.resolutions[0]!.surfaces).toMatchObject({
inline: false,
workspace: true,
localExecutor: true,
visual: false,
})
})
})