Classify the OpenResponses surface and the failing test provider

34cc32225094 · AtlantisPleb · 4d ago · parent 2c26298299a9

Classify the OpenResponses surface and the failing test provider

Two enumeration proofs went red on main. PERSONA-001's adapter and
request-builder populations are exact sets, and the OpenResponses stub and
the fallback bug's failing test provider both arrived without being named
in them.

FailingTestProvider is in-process like the other test providers: it never
leaves the VM, so it has no model to instruct. The OpenResponses surface
relays a caller's own instructions, the same classification the inference
proxy carries, so it joins that list rather than earning a new one — the
property PERSONA-001 needs from it is identical, that it reach no persona
module, and the test proves that for every relay rather than for one named
controller.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KnhfrafYx5ZGaMbzZEJQ2d

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KnhfrafYx5ZGaMbzZEJQ2d

Deploy story

What this commit did to the running system — joined from the forge receipt chain, the part a commit page elsewhere cannot show.

pushed: by user · WAL seq 402 · 2026-08-25T15:35:11.612629Z

Changed files

modified test/openagents/providers/persona_boundary_test.exs

Diff

1 file changed, +17 -3

    OpenAgents.Providers.OpenAI => :outbound_http,
    OpenAgents.Providers.OpenRouter => :outbound_http,
    OpenAgents.Providers.VercelGateway => :outbound_http,
    OpenAgents.Providers.FailingTestProvider => :in_process,
    OpenAgents.Providers.FallbackTestProvider => :in_process,
    OpenAgents.Providers.RecordingTestProvider => :in_process,
    OpenAgents.Providers.Test => :in_process,

    OpenAgents.CompensationFixtures => :composes_installed_persona,
    OpenAgents.Persona.Evaluation.Runner => :composes_persona_candidate,
    OpenAgentsWeb.InferenceProxyController => :relays_caller_instructions,
    OpenAgentsWeb.ResponsesController => :relays_caller_instructions,
    OpenAgents.Conversations => :pins_a_composed_request,
    OpenAgents.Providers.OpenAI => :adapter,
    OpenAgents.Providers.OpenRouter => :adapter,
    OpenAgents.Providers.VercelGateway => :adapter,
    OpenAgents.Providers.FailingTestProvider => :adapter,
    OpenAgents.Providers.FallbackTestProvider => :adapter,
    OpenAgents.Providers.RecordingTestProvider => :adapter,
    OpenAgents.Providers.Test => :adapter,

      end
    end

    test "the inference proxy composes no OpenAgents persona" do
      relays = for {module, :relays_caller_instructions} <- @request_sites, do: module
      assert relays == [OpenAgentsWeb.InferenceProxyController]
    test "a surface that relays a caller's instructions composes no OpenAgents persona" do
      relays =
        for({module, :relays_caller_instructions} <- @request_sites, do: module)
        |> Enum.sort()

      # Both surfaces take a caller's own instructions and answer them. The
      # OpenResponses surface joined the inference proxy here rather than
      # earning a classification of its own, because the property PERSONA-001
      # needs from it is identical: it must reach no persona module.
      assert relays ==
               Enum.sort([
                 OpenAgentsWeb.InferenceProxyController,
                 OpenAgentsWeb.ResponsesController
               ])

      for module <- relays do
        reached =

66	66		`OpenAgents.Providers.OpenAI => :outbound_http,`
67	67		`OpenAgents.Providers.OpenRouter => :outbound_http,`
68	68		`OpenAgents.Providers.VercelGateway => :outbound_http,`
	69	+	`OpenAgents.Providers.FailingTestProvider => :in_process,`
69	70		`OpenAgents.Providers.FallbackTestProvider => :in_process,`
70	71		`OpenAgents.Providers.RecordingTestProvider => :in_process,`
71	72		`OpenAgents.Providers.Test => :in_process,`

98	99		`OpenAgents.CompensationFixtures => :composes_installed_persona,`
99	100		`OpenAgents.Persona.Evaluation.Runner => :composes_persona_candidate,`
100	101		`OpenAgentsWeb.InferenceProxyController => :relays_caller_instructions,`
	102	+	`OpenAgentsWeb.ResponsesController => :relays_caller_instructions,`
101	103		`OpenAgents.Conversations => :pins_a_composed_request,`
102	104		`OpenAgents.Providers.OpenAI => :adapter,`
103	105		`OpenAgents.Providers.OpenRouter => :adapter,`
104	106		`OpenAgents.Providers.VercelGateway => :adapter,`
	107	+	`OpenAgents.Providers.FailingTestProvider => :adapter,`
105	108		`OpenAgents.Providers.FallbackTestProvider => :adapter,`
106	109		`OpenAgents.Providers.RecordingTestProvider => :adapter,`
107	110		`OpenAgents.Providers.Test => :adapter,`

317	320		`end`
318	321		`end`
319	322
320		-	`test "the inference proxy composes no OpenAgents persona" do`
321		-	`relays = for {module, :relays_caller_instructions} <- @request_sites, do: module`
322		-	`assert relays == [OpenAgentsWeb.InferenceProxyController]`
	323	+	`test "a surface that relays a caller's instructions composes no OpenAgents persona" do`
	324	+	`relays =`
	325	+	`for({module, :relays_caller_instructions} <- @request_sites, do: module)`
	326	+	`\|> Enum.sort()`
	327	+
	328	+	`# Both surfaces take a caller's own instructions and answer them. The`
	329	+	`# OpenResponses surface joined the inference proxy here rather than`
	330	+	`# earning a classification of its own, because the property PERSONA-001`
	331	+	`# needs from it is identical: it must reach no persona module.`
	332	+	`assert relays ==`
	333	+	`Enum.sort([`
	334	+	`OpenAgentsWeb.InferenceProxyController,`
	335	+	`OpenAgentsWeb.ResponsesController`
	336	+	`])`
323	337
324	338		`for module <- relays do`
325	339		`reached =`