fix: keep step tests and multi-agent chats from inventing an effort override

frontend/src/lib/components/ModuleTest.svelte

@@ -20,7 +20,7 @@
 		type LinkedAgentDraft
 	} from './flows/linkedAgentDrafts'
 	import { AGENT_LINKED_STEP_KEYS } from './flows/agentResourceUtils'
-	import { agentTestInputTransforms } from './flows/agentFormFields'
+	import { AGENT_OVERRIDE_KEYS, agentTestInputTransforms } from './flows/agentFormFields'
 	import { sendUserToast } from '$lib/toast'
 	import { quickjsReady } from '$lib/utils/quickjsEval.svelte'
 	import { useOperatingWorkspace } from '$lib/components/operatingWorkspace.svelte'
@@ -186,7 +186,13 @@
 			const inputTransforms = agentTestInputTransforms(
 				agentVal.agent ? {} : ((agentVal.input_transforms ?? {}) as Record<string, InputTransform>),
 				args,
-				draft ? AGENT_LINKED_STEP_KEYS : Object.keys(args)
+				// An override the step does not hold is left out: fed from the form it would arrive as an
+				// authored empty value, which clears the agent's saved effort.
+				(draft ? AGENT_LINKED_STEP_KEYS : Object.keys(args)).filter(
+					(key) =>
+						!(AGENT_OVERRIDE_KEYS as readonly string[]).includes(key) ||
+						val.input_transforms?.[key] != undefined
+				)
 			)
 
 			await jobLoader?.runFlowPreview(

frontend/src/lib/components/flows/conversations/agentChatInputs.test.ts

@@ -551,6 +551,12 @@ describe('a step linked to a saved agent', () => {
 		).toBeUndefined()
 	})
 
+	it('does not show the saved effort for steps that fix different ones', () => {
+		const effort = (value: string) => linked({ reasoning_effort: { type: 'static', value } })
+		const wiring = resolveAgentModelWiring([effort('low'), effort('medium')], saved)!
+		expect(wiredFieldValue(wiring, {}, 'reasoning_effort')).toBeUndefined()
+	})
+
 	it('cannot be read until the agent has loaded, or with an expression it cannot place', () => {
 		expect(resolveAgentModelWiring([wired], {})).toBeUndefined()
 		expect(

frontend/src/lib/components/flows/conversations/agentChatInputs.ts

@@ -312,26 +312,28 @@ export function resolveAgentModelWiring(
 
 	const fields: AgentModelWiring['fields'] = {}
 	const fixed: AgentModelWiring['fixed'] = {}
+	const disagreed = new Set<ProviderField>()
 	for (const field of PROVIDER_FIELDS) {
 		// Every agent has to supply the field the same way for the composer to speak for
 		// them all. One wired name among agents that otherwise fix it is not agreement:
 		// the control would move that one agent and leave the others where they are.
 		const supplies = new Set(wirings.map((w) => JSON.stringify(fieldSupply(w, field))))
+
 		// Disagreement leaves the field neither editable nor known: a control offered here
 		// would govern one agent while the rest ran on something else.
-		if (supplies.size > 1) continue
+		if (supplies.size > 1) {
+			disagreed.add(field)
+			continue
+		}
 		const supply: FieldSupply = JSON.parse([...supplies][0])
 		if (supply.kind === 'wired') fields[field] = supply.name
 		else if (supply.kind === 'fixed') fixed[field] = supply.value
 	}
-	// An empty input means one thing only where every agent falls back to the same saved pair.
+	// An empty input means one thing only where every agent falls back to the same saved pair,
+	// and a saved value must not stand in for a field the agents disagree about.
 	const saved = new Set(wirings.map((w) => JSON.stringify(w.saved)))
-	return {
-		fields,
-		fixed,
-		someAgentCannotRun,
-		saved: saved.size === 1 ? wirings[0].saved : undefined
-	}
+	const agreed = saved.size === 1 && !disagreed.has('model') && !disagreed.has('reasoning_effort')
+	return { fields, fixed, someAgentCannotRun, saved: agreed ? wirings[0].saved : undefined }
 }
 
 /**

openflow.openapi.yaml

@@ -1225,7 +1225,8 @@ components:
             Path of a reusable `ai_agent` resource (hybrid linking). When set, the agent brain
             config (provider/model/system prompt/etc.) and tool set are resolved at runtime from
             that resource; the module's input_transforms then only carry the flow-local inputs
-            (user_message, user_attachments, enabled_tools and the history inputs memory_id and previous_messages).
+            (user_message, user_attachments, enabled_tools and the history inputs memory_id and previous_messages)
+            and, optionally, the `model` and `reasoning_effort` overrides of what the agent saved.
         tool_inputs:
           type: object
           description: |