feat(config): support object-style fallback_models with per-model settings
Add support for object-style entries in fallback_models arrays, enabling per-model configuration of variant, reasoningEffort, temperature, top_p, maxTokens, and thinking settings. - Zod schema for FallbackModelObject with full validation - normalizeFallbackModels() and flattenToFallbackModelStrings() utilities - Provider-agnostic model resolution pipeline with fallback chain - Session prompt params state management - Fallback chain construction with prefix-match lookup - Integration across delegate-task, background-agent, and plugin layers
This commit is contained in:
Ravi Tharuma
@@ -114,4 +114,260 @@ describe("resolveCategoryExecution", () => {
|
||||
{ providers: ["openai"], model: "gpt-5.2", variant: "high" },
|
||||
])
|
||||
})
|
||||
|
||||
test("promotes object-style fallback model settings to categoryModel when fallback becomes initial model", async () => {
|
||||
//#given
|
||||
const cacheSpy = spyOn(connectedProvidersCache, "readProviderModelsCache").mockReturnValue({
|
||||
models: { openai: ["gpt-5.4"] },
|
||||
connected: ["openai"],
|
||||
updatedAt: "2026-03-03T00:00:00.000Z",
|
||||
})
|
||||
const agentsSpy = spyOn(connectedProvidersCache, "readConnectedProvidersCache").mockReturnValue(["openai"])
|
||||
const args = {
|
||||
category: "deep",
|
||||
prompt: "test prompt",
|
||||
description: "Test task",
|
||||
run_in_background: false,
|
||||
load_skills: [],
|
||||
blockedBy: undefined,
|
||||
enableSkillTools: false,
|
||||
}
|
||||
const executorCtx = createMockExecutorContext()
|
||||
executorCtx.userCategories = {
|
||||
deep: {
|
||||
fallback_models: [
|
||||
{
|
||||
model: "openai/gpt-5.4 high",
|
||||
variant: "low",
|
||||
reasoningEffort: "high",
|
||||
temperature: 0.4,
|
||||
top_p: 0.7,
|
||||
maxTokens: 4096,
|
||||
thinking: { type: "disabled" },
|
||||
},
|
||||
],
|
||||
},
|
||||
}
|
||||
|
||||
//#when
|
||||
const result = await resolveCategoryExecution(args, executorCtx, undefined, "anthropic/claude-sonnet-4-6")
|
||||
|
||||
//#then
|
||||
expect(result.error).toBeUndefined()
|
||||
expect(result.actualModel).toBe("openai/gpt-5.4")
|
||||
expect(result.categoryModel).toEqual({
|
||||
providerID: "openai",
|
||||
modelID: "gpt-5.4",
|
||||
variant: "low",
|
||||
reasoningEffort: "high",
|
||||
temperature: 0.4,
|
||||
top_p: 0.7,
|
||||
maxTokens: 4096,
|
||||
thinking: { type: "disabled" },
|
||||
})
|
||||
cacheSpy.mockRestore()
|
||||
agentsSpy.mockRestore()
|
||||
})
|
||||
|
||||
test("matches promoted fallback settings after fuzzy model resolution", async () => {
|
||||
//#given
|
||||
const cacheSpy = spyOn(connectedProvidersCache, "readProviderModelsCache").mockReturnValue({
|
||||
models: { openai: ["gpt-5.4-preview"] },
|
||||
connected: ["openai"],
|
||||
updatedAt: "2026-03-03T00:00:00.000Z",
|
||||
})
|
||||
const agentsSpy = spyOn(connectedProvidersCache, "readConnectedProvidersCache").mockReturnValue(["openai"])
|
||||
const args = {
|
||||
category: "deep",
|
||||
prompt: "test prompt",
|
||||
description: "Test task",
|
||||
run_in_background: false,
|
||||
load_skills: [],
|
||||
blockedBy: undefined,
|
||||
enableSkillTools: false,
|
||||
}
|
||||
const executorCtx = createMockExecutorContext()
|
||||
executorCtx.userCategories = {
|
||||
deep: {
|
||||
fallback_models: [
|
||||
{
|
||||
model: "openai/gpt-5.4",
|
||||
variant: "low",
|
||||
reasoningEffort: "high",
|
||||
temperature: 0.6,
|
||||
top_p: 0.5,
|
||||
maxTokens: 1234,
|
||||
thinking: { type: "disabled" },
|
||||
},
|
||||
],
|
||||
},
|
||||
}
|
||||
|
||||
//#when
|
||||
const result = await resolveCategoryExecution(args, executorCtx, undefined, "anthropic/claude-sonnet-4-6")
|
||||
|
||||
//#then
|
||||
expect(result.error).toBeUndefined()
|
||||
expect(result.actualModel).toBe("openai/gpt-5.4-preview")
|
||||
expect(result.categoryModel).toEqual({
|
||||
providerID: "openai",
|
||||
modelID: "gpt-5.4-preview",
|
||||
variant: "low",
|
||||
reasoningEffort: "high",
|
||||
temperature: 0.6,
|
||||
top_p: 0.5,
|
||||
maxTokens: 1234,
|
||||
thinking: { type: "disabled" },
|
||||
})
|
||||
cacheSpy.mockRestore()
|
||||
agentsSpy.mockRestore()
|
||||
})
|
||||
|
||||
test("prefers exact promoted fallback match over earlier fuzzy prefix match", async () => {
|
||||
//#given
|
||||
const cacheSpy = spyOn(connectedProvidersCache, "readProviderModelsCache").mockReturnValue({
|
||||
models: { openai: ["gpt-5.4-preview"] },
|
||||
connected: ["openai"],
|
||||
updatedAt: "2026-03-03T00:00:00.000Z",
|
||||
})
|
||||
const agentsSpy = spyOn(connectedProvidersCache, "readConnectedProvidersCache").mockReturnValue(["openai"])
|
||||
const args = {
|
||||
category: "deep",
|
||||
prompt: "test prompt",
|
||||
description: "Test task",
|
||||
run_in_background: false,
|
||||
load_skills: [],
|
||||
blockedBy: undefined,
|
||||
enableSkillTools: false,
|
||||
}
|
||||
const executorCtx = createMockExecutorContext()
|
||||
executorCtx.userCategories = {
|
||||
deep: {
|
||||
fallback_models: [
|
||||
{
|
||||
model: "openai/gpt-5.4",
|
||||
variant: "low",
|
||||
reasoningEffort: "medium",
|
||||
},
|
||||
{
|
||||
model: "openai/gpt-5.4-preview",
|
||||
variant: "max",
|
||||
reasoningEffort: "high",
|
||||
},
|
||||
],
|
||||
},
|
||||
}
|
||||
|
||||
//#when
|
||||
const result = await resolveCategoryExecution(args, executorCtx, undefined, "anthropic/claude-sonnet-4-6")
|
||||
|
||||
//#then
|
||||
expect(result.error).toBeUndefined()
|
||||
expect(result.actualModel).toBe("openai/gpt-5.4-preview")
|
||||
expect(result.categoryModel).toEqual({
|
||||
providerID: "openai",
|
||||
modelID: "gpt-5.4-preview",
|
||||
variant: "max",
|
||||
reasoningEffort: "high",
|
||||
})
|
||||
cacheSpy.mockRestore()
|
||||
agentsSpy.mockRestore()
|
||||
})
|
||||
|
||||
test("matches promoted fallback settings when fuzzy resolution extends configured model without hyphen", async () => {
|
||||
//#given
|
||||
const cacheSpy = spyOn(connectedProvidersCache, "readProviderModelsCache").mockReturnValue({
|
||||
models: { openai: ["gpt-5.4o"] },
|
||||
connected: ["openai"],
|
||||
updatedAt: "2026-03-03T00:00:00.000Z",
|
||||
})
|
||||
const agentsSpy = spyOn(connectedProvidersCache, "readConnectedProvidersCache").mockReturnValue(["openai"])
|
||||
const args = {
|
||||
category: "deep",
|
||||
prompt: "test prompt",
|
||||
description: "Test task",
|
||||
run_in_background: false,
|
||||
load_skills: [],
|
||||
blockedBy: undefined,
|
||||
enableSkillTools: false,
|
||||
}
|
||||
const executorCtx = createMockExecutorContext()
|
||||
executorCtx.userCategories = {
|
||||
deep: {
|
||||
fallback_models: [
|
||||
{
|
||||
model: "openai/gpt-5.4",
|
||||
variant: "low",
|
||||
reasoningEffort: "high",
|
||||
},
|
||||
],
|
||||
},
|
||||
}
|
||||
|
||||
//#when
|
||||
const result = await resolveCategoryExecution(args, executorCtx, undefined, "anthropic/claude-sonnet-4-6")
|
||||
|
||||
//#then
|
||||
expect(result.error).toBeUndefined()
|
||||
expect(result.actualModel).toBe("openai/gpt-5.4o")
|
||||
expect(result.categoryModel).toEqual({
|
||||
providerID: "openai",
|
||||
modelID: "gpt-5.4o",
|
||||
variant: "low",
|
||||
reasoningEffort: "high",
|
||||
})
|
||||
cacheSpy.mockRestore()
|
||||
agentsSpy.mockRestore()
|
||||
})
|
||||
|
||||
test("prefers the most specific prefix match when fallback entries share a prefix", async () => {
|
||||
//#given
|
||||
const cacheSpy = spyOn(connectedProvidersCache, "readProviderModelsCache").mockReturnValue({
|
||||
models: { openai: ["gpt-4o"] },
|
||||
connected: ["openai"],
|
||||
updatedAt: "2026-03-03T00:00:00.000Z",
|
||||
})
|
||||
const agentsSpy = spyOn(connectedProvidersCache, "readConnectedProvidersCache").mockReturnValue(["openai"])
|
||||
const args = {
|
||||
category: "deep",
|
||||
prompt: "test prompt",
|
||||
description: "Test task",
|
||||
run_in_background: false,
|
||||
load_skills: [],
|
||||
blockedBy: undefined,
|
||||
enableSkillTools: false,
|
||||
}
|
||||
const executorCtx = createMockExecutorContext()
|
||||
executorCtx.userCategories = {
|
||||
deep: {
|
||||
fallback_models: [
|
||||
{
|
||||
model: "openai/gpt-4",
|
||||
variant: "low",
|
||||
reasoningEffort: "medium",
|
||||
},
|
||||
{
|
||||
model: "openai/gpt-4o",
|
||||
variant: "max",
|
||||
reasoningEffort: "high",
|
||||
},
|
||||
],
|
||||
},
|
||||
}
|
||||
|
||||
//#when
|
||||
const result = await resolveCategoryExecution(args, executorCtx, undefined, "anthropic/claude-sonnet-4-6")
|
||||
|
||||
//#then
|
||||
expect(result.error).toBeUndefined()
|
||||
expect(result.actualModel).toBe("openai/gpt-4o")
|
||||
expect(result.categoryModel).toEqual({
|
||||
providerID: "openai",
|
||||
modelID: "gpt-4o",
|
||||
variant: "max",
|
||||
reasoningEffort: "high",
|
||||
})
|
||||
cacheSpy.mockRestore()
|
||||
agentsSpy.mockRestore()
|
||||
})
|
||||
})
|
||||
|
||||
Reference in New Issue
Block a user