feat: per-agent model routing — route different agents to different providers (#238)

* feat: add agentModels and agentRouting to SettingsSchema Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * feat: add agentRouting module for per-agent provider resolution Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * feat: thread providerOverride through OpenAI shim for per-agent routing Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * feat: getAnthropicClient accepts providerOverride for agent routing Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * feat: thread providerOverride through Options and queryModel calls Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * feat: thread providerOverride through query loop and ToolUseContext Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * feat: resolve agent routing in runAgent and inject providerOverride Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * docs: add Agent Routing configuration guide to README Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: add unit tests for resolveAgentProvider + plaintext api_key note - 15 tests covering priority chain (name > subagentType > default > null) - normalize() case-insensitive and hyphen/underscore equivalence - Edge cases: null settings, missing config sections, non-existent model - README note about api_key stored in plaintext Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * security: address code review — SSRF, credential leak, key collision - base_url schema now uses z.string().url() for SSRF mitigation - Strip auth headers (Authorization, x-api-key, api-key) from defaultHeaders when providerOverride is active, preventing Anthropic credentials from leaking to third-party endpoints - Warn on duplicate normalized routing keys to prevent silent shadowing - providerOverride.apiKey is never logged (verified via grep) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> --------- Co-authored-by: 冯俊辉 <fengjunhui@shiyanjia.com> Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-03 21:47:26 +08:00
parent 59ab2701f7
commit fb32e3f829
11 changed files with 319 additions and 9 deletions
--- a/src/services/api/agentRouting.test.ts
+++ b/src/services/api/agentRouting.test.ts
@@ -0,0 +1,125 @@
+import { describe, expect, test } from 'bun:test'
+import { resolveAgentProvider } from './agentRouting.js'
+import type { SettingsJson } from '../../utils/settings/types.js'
+
+const baseSettings = {
+  agentModels: {
+    'deepseek-chat': { base_url: 'https://api.deepseek.com/v1', api_key: 'sk-ds' },
+    'gpt-4o': { base_url: 'https://api.openai.com/v1', api_key: 'sk-oai' },
+  },
+  agentRouting: {
+    Explore: 'deepseek-chat',
+    'general-purpose': 'gpt-4o',
+    'frontend-dev': 'deepseek-chat',
+    default: 'gpt-4o',
+  },
+} as unknown as SettingsJson
+
+describe('resolveAgentProvider', () => {
+  // ── Priority chain ──────────────────────────────────────────
+
+  test('name takes priority over subagentType', () => {
+    const result = resolveAgentProvider('frontend-dev', 'Explore', baseSettings)
+    expect(result).toEqual({
+      model: 'deepseek-chat',
+      baseURL: 'https://api.deepseek.com/v1',
+      apiKey: 'sk-ds',
+    })
+  })
+
+  test('subagentType used when name has no match', () => {
+    const result = resolveAgentProvider('unknown-name', 'Explore', baseSettings)
+    expect(result).toEqual({
+      model: 'deepseek-chat',
+      baseURL: 'https://api.deepseek.com/v1',
+      apiKey: 'sk-ds',
+    })
+  })
+
+  test('falls back to "default" when neither name nor subagentType match', () => {
+    const result = resolveAgentProvider('nobody', 'unknown-type', baseSettings)
+    expect(result).toEqual({
+      model: 'gpt-4o',
+      baseURL: 'https://api.openai.com/v1',
+      apiKey: 'sk-oai',
+    })
+  })
+
+  test('returns null when no routing match and no default', () => {
+    const settings = {
+      agentModels: baseSettings.agentModels,
+      agentRouting: { Explore: 'deepseek-chat' },
+    } as unknown as SettingsJson
+    const result = resolveAgentProvider('nobody', 'unknown-type', settings)
+    expect(result).toBeNull()
+  })
+
+  test('returns null when name and subagentType are both undefined', () => {
+    const settings = {
+      agentModels: baseSettings.agentModels,
+      agentRouting: { Explore: 'deepseek-chat' },
+    } as unknown as SettingsJson
+    const result = resolveAgentProvider(undefined, undefined, settings)
+    expect(result).toBeNull()
+  })
+
+  // ── normalize() matching ────────────────────────────────────
+
+  test('matching is case-insensitive', () => {
+    const result = resolveAgentProvider(undefined, 'explore', baseSettings)
+    expect(result?.model).toBe('deepseek-chat')
+  })
+
+  test('matching is case-insensitive (UPPER)', () => {
+    const result = resolveAgentProvider(undefined, 'EXPLORE', baseSettings)
+    expect(result?.model).toBe('deepseek-chat')
+  })
+
+  test('hyphen and underscore are equivalent', () => {
+    const result = resolveAgentProvider(undefined, 'general_purpose', baseSettings)
+    expect(result?.model).toBe('gpt-4o')
+  })
+
+  test('underscore in config matches hyphen in input', () => {
+    const settings = {
+      agentModels: baseSettings.agentModels,
+      agentRouting: { general_purpose: 'deepseek-chat' },
+    } as unknown as SettingsJson
+    const result = resolveAgentProvider(undefined, 'general-purpose', settings)
+    expect(result?.model).toBe('deepseek-chat')
+  })
+
+  // ── Edge cases ──────────────────────────────────────────────
+
+  test('returns null when settings is null', () => {
+    expect(resolveAgentProvider('Explore', 'Explore', null)).toBeNull()
+  })
+
+  test('returns null when agentRouting is missing', () => {
+    const settings = { agentModels: baseSettings.agentModels } as unknown as SettingsJson
+    expect(resolveAgentProvider(undefined, 'Explore', settings)).toBeNull()
+  })
+
+  test('returns null when agentModels is missing', () => {
+    const settings = { agentRouting: baseSettings.agentRouting } as unknown as SettingsJson
+    expect(resolveAgentProvider(undefined, 'Explore', settings)).toBeNull()
+  })
+
+  test('returns null when routing references non-existent model', () => {
+    const settings = {
+      agentModels: {},
+      agentRouting: { Explore: 'non-existent-model' },
+    } as unknown as SettingsJson
+    expect(resolveAgentProvider(undefined, 'Explore', settings)).toBeNull()
+  })
+
+  test('subagentType only (no name)', () => {
+    const result = resolveAgentProvider(undefined, 'Explore', baseSettings)
+    expect(result?.model).toBe('deepseek-chat')
+  })
+
+  test('name only (no subagentType)', () => {
+    const result = resolveAgentProvider('frontend-dev', undefined, baseSettings)
+    expect(result?.model).toBe('deepseek-chat')
+  })
+})
--- a/src/services/api/agentRouting.ts
+++ b/src/services/api/agentRouting.ts
@@ -0,0 +1,75 @@
+import type { SettingsJson } from '../../utils/settings/types.js'
+
+/**
+ * Provider override resolved from agent routing config.
+ * When present, the API client should use these instead of global env vars.
+ */
+export interface ProviderOverride {
+  /** Model name to send to the API (e.g. "deepseek-chat", "gpt-4o") */
+  model: string
+  /** OpenAI-compatible base URL */
+  baseURL: string
+  /** API key for this provider */
+  apiKey: string
+}
+
+/**
+ * Normalize an agent identifier for case-insensitive, hyphen/underscore-agnostic matching.
+ */
+function normalize(key: string): string {
+  return key.toLowerCase().replace(/[-_]/g, '')
+}
+
+/**
+ * Look up agent.routing by name or subagent_type, then resolve via agent.models.
+ *
+ * Priority: name > subagentType > "default" > null (use global provider)
+ */
+export function resolveAgentProvider(
+  name: string | undefined,
+  subagentType: string | undefined,
+  settings: SettingsJson | null,
+): ProviderOverride | null {
+  if (!settings) return null
+
+  const routing = settings.agentRouting
+  const models = settings.agentModels
+  if (!routing || !models) return null
+
+  // Build normalized lookup from routing config.
+  // Warn on duplicate normalized keys (e.g. "explore-agent" and "explore_agent"
+  // both normalize to "exploreagent") to prevent silent shadowing.
+  const normalizedRouting = new Map<string, string>()
+  for (const [key, value] of Object.entries(routing)) {
+    const nk = normalize(key)
+    if (normalizedRouting.has(nk)) {
+      console.error(`[agentRouting] Warning: routing key "${key}" collides with an existing key after normalization (both map to "${nk}"). First entry wins.`)
+    }
+    if (!normalizedRouting.has(nk)) {
+      normalizedRouting.set(nk, value)
+    }
+  }
+
+  // Try name first, then subagentType, then "default"
+  const candidates = [name, subagentType, 'default'].filter(Boolean) as string[]
+  let modelName: string | undefined
+
+  for (const candidate of candidates) {
+    const match = normalizedRouting.get(normalize(candidate))
+    if (match) {
+      modelName = match
+      break
+    }
+  }
+
+  if (!modelName) return null
+
+  const modelConfig = models[modelName]
+  if (!modelConfig) return null
+
+  return {
+    model: modelName,
+    baseURL: modelConfig.base_url,
+    apiKey: modelConfig.api_key,
+  }
+}
--- a/src/services/api/claude.ts
+++ b/src/services/api/claude.ts
@@ -704,6 +704,7 @@ export type Options = {
  // so the model can pace itself. `remaining` is computed by the caller
  // (query.ts decrements across the agentic loop).
  taskBudget?: { total: number; remaining?: number }
+  providerOverride?: { model: string; baseURL: string; apiKey: string }
 }

 export async function queryModelWithoutStreaming({
@@ -820,6 +821,7 @@ export async function* executeNonStreamingRequest(
    model: string
    fetchOverride?: Options['fetchOverride']
    source: string
+    providerOverride?: Options['providerOverride']
  },
  retryOptions: {
    model: string
@@ -847,6 +849,7 @@ export async function* executeNonStreamingRequest(
        model: clientOptions.model,
        fetchOverride: clientOptions.fetchOverride,
        source: clientOptions.source,
+        providerOverride: clientOptions.providerOverride,
      }),
    async (anthropic, attempt, context) => {
      const start = Date.now()
@@ -1782,6 +1785,7 @@ async function* queryModel(
          model: options.model,
          fetchOverride: options.fetchOverride,
          source: options.querySource,
+          providerOverride: options.providerOverride,
        }),
      async (anthropic, attempt, context) => {
        attemptNumber = attempt
@@ -2549,7 +2553,7 @@ async function* queryModel(
          : 'other') as AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS,
      })
      const result = yield* executeNonStreamingRequest(
-        { model: options.model, source: options.querySource },
+        { model: options.model, source: options.querySource, providerOverride: options.providerOverride },
        {
          model: options.model,
          fallbackModel: options.fallbackModel,
--- a/src/services/api/client.ts
+++ b/src/services/api/client.ts
@@ -95,12 +95,14 @@ export async function getAnthropicClient({
  model,
  fetchOverride,
  source,
+  providerOverride,
 }: {
  apiKey?: string
  maxRetries: number
  model?: string
  fetchOverride?: ClientOptions['fetch']
  source?: string
+  providerOverride?: { model: string; baseURL: string; apiKey: string }
 }): Promise<Anthropic> {
  const containerId = process.env.CLAUDE_CODE_CONTAINER_ID
  const remoteSessionId = process.env.CLAUDE_CODE_REMOTE_SESSION_ID
@@ -154,6 +156,24 @@ export async function getAnthropicClient({
      fetch: resolvedFetch,
    }),
  }
+  // Agent routing override: use per-agent provider when configured.
+  // Strip auth-related headers to prevent leaking Anthropic credentials
+  // to third-party endpoints (SSRF / credential forwarding mitigation).
+  if (providerOverride) {
+    const { createOpenAIShimClient } = await import('./openaiShim.js')
+    const safeHeaders: Record<string, string> = {}
+    for (const [k, v] of Object.entries(defaultHeaders)) {
+      const lower = k.toLowerCase()
+      if (lower === 'authorization' || lower === 'x-api-key' || lower === 'api-key') continue
+      safeHeaders[k] = v
+    }
+    return createOpenAIShimClient({
+      defaultHeaders: safeHeaders,
+      maxRetries,
+      timeout: parseInt(process.env.API_TIMEOUT_MS || String(600 * 1000), 10),
+      providerOverride,
+    }) as unknown as Anthropic
+  }
  if (
    isEnvTruthy(process.env.CLAUDE_CODE_USE_OPENAI) ||
    isEnvTruthy(process.env.CLAUDE_CODE_USE_GITHUB) ||
--- a/src/services/api/openaiShim.ts
+++ b/src/services/api/openaiShim.ts
@@ -683,10 +683,12 @@ class OpenAIShimStream {
 class OpenAIShimMessages {
  private defaultHeaders: Record<string, string>
  private reasoningEffort?: 'low' | 'medium' | 'high' | 'xhigh'
+  private providerOverride?: { model: string; baseURL: string; apiKey: string }

-  constructor(defaultHeaders: Record<string, string>, reasoningEffort?: 'low' | 'medium' | 'high' | 'xhigh') {
+  constructor(defaultHeaders: Record<string, string>, reasoningEffort?: 'low' | 'medium' | 'high' | 'xhigh', providerOverride?: { model: string; baseURL: string; apiKey: string }) {
    this.defaultHeaders = defaultHeaders
    this.reasoningEffort = reasoningEffort
+    this.providerOverride = providerOverride
  }

  create(
@@ -698,7 +700,7 @@ class OpenAIShimMessages {
    let httpResponse: Response | undefined

    const promise = (async () => {
-      const request = resolveProviderRequest({ model: params.model, reasoningEffortOverride: self.reasoningEffort })
+      const request = resolveProviderRequest({ model: self.providerOverride?.model ?? params.model, baseUrl: self.providerOverride?.baseURL, reasoningEffortOverride: self.reasoningEffort })
      const response = await self._doRequest(request, params, options)
      httpResponse = response

@@ -857,7 +859,7 @@ class OpenAIShimMessages {
      ...(options?.headers ?? {}),
    }

-    const apiKey = process.env.OPENAI_API_KEY ?? ''
+    const apiKey = this.providerOverride?.apiKey ?? process.env.OPENAI_API_KEY ?? ''
    // Detect Azure endpoints by hostname (not raw URL) to prevent bypass via
    // path segments like https://evil.com/cognitiveservices.azure.com/
    let isAzure = false
@@ -1056,8 +1058,8 @@ class OpenAIShimBeta {
  messages: OpenAIShimMessages
  reasoningEffort?: 'low' | 'medium' | 'high' | 'xhigh'

-  constructor(defaultHeaders: Record<string, string>, reasoningEffort?: 'low' | 'medium' | 'high' | 'xhigh') {
-    this.messages = new OpenAIShimMessages(defaultHeaders, reasoningEffort)
+  constructor(defaultHeaders: Record<string, string>, reasoningEffort?: 'low' | 'medium' | 'high' | 'xhigh', providerOverride?: { model: string; baseURL: string; apiKey: string }) {
+    this.messages = new OpenAIShimMessages(defaultHeaders, reasoningEffort, providerOverride)
    this.reasoningEffort = reasoningEffort
  }
 }
@@ -1067,6 +1069,7 @@ export function createOpenAIShimClient(options: {
  maxRetries?: number
  timeout?: number
  reasoningEffort?: 'low' | 'medium' | 'high' | 'xhigh'
+  providerOverride?: { model: string; baseURL: string; apiKey: string }
 }): unknown {
  hydrateGithubModelsTokenFromSecureStorage()

@@ -1089,7 +1092,7 @@ export function createOpenAIShimClient(options: {

  const beta = new OpenAIShimBeta({
    ...(options.defaultHeaders ?? {}),
-  }, options.reasoningEffort)
+  }, options.reasoningEffort, options.providerOverride)

  return {
    beta,