fix(gateway): add chat touchpoint to zhipu recipe so GLM subagents work (#1157)

The zhipu recipe was embedding-only, so models.tier.subagent=zhipu:glm-5.1
threw "does not offer a chat touchpoint" — while the error hint falsely
listed zhipu (and dashscope/minimax, also embedding-only) among providers
with chat.

- zhipu recipe: add a chat touchpoint (glm-5.1 family, supports_tools +
  supports_subagent_loop; no Anthropic-style prompt cache on the
  OpenAI-compat path, so the loop runs with the degraded:no_caching warn).
  openai-compat tier means newer GLM ids pass without a recipe edit.
- capabilities.ts: compute the "Known providers with chat" hint from the
  recipe registry instead of a hardcoded list, so it can never drift into
  naming chat-less providers again.
- Declines the originally requested models.anthropic_compatible_prefixes
  config: v0.38's recipe-driven capability gate already replaced the
  Anthropic-only enforcement, so a recipe chat touchpoint is the whole fix.

Fixes #1157

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/commands/init.ts

@@ -161,7 +161,7 @@ interface ResolveAIOptionsArgs {
   nonInteractive: boolean;       // --non-interactive (forces D3 fail-loud, no picker)
 }
 
-export interface ResolvedAIOptions {
+interface ResolvedAIOptions {
   embedding_model?: string;
   embedding_dimensions?: number;
   expansion_model?: string;
@@ -170,41 +170,6 @@ export interface ResolvedAIOptions {
   noEmbedding?: boolean;
 }
 
-/**
- * Seed init's AI options from persisted config, falling back to the raw env
- * vars when loadConfig() returned null (#1058). On a cold install (no
- * config.json AND no DATABASE_URL) loadConfig short-circuits BEFORE its env
- * merge, so GBRAIN_EMBEDDING_MODEL / GBRAIN_EMBEDDING_DIMENSIONS /
- * GBRAIN_EXPANSION_MODEL / GBRAIN_CHAT_MODEL were silently ignored by init
- * and Tier-3 detection auto-picked by API key instead. Exported for unit
- * tests (env injectable).
- */
-export function seedAIOptionsFromConfig(
-  cfg: GBrainConfig | null,
-  env: NodeJS.ProcessEnv = process.env,
-): ResolvedAIOptions {
-  const envDims = env.GBRAIN_EMBEDDING_DIMENSIONS
-    ? parseInt(env.GBRAIN_EMBEDDING_DIMENSIONS, 10)
-    : NaN;
-  const seed = cfg ?? {
-    embedding_disabled: undefined,
-    embedding_model: env.GBRAIN_EMBEDDING_MODEL,
-    embedding_dimensions: Number.isFinite(envDims) ? envDims : undefined,
-    expansion_model: env.GBRAIN_EXPANSION_MODEL,
-    chat_model: env.GBRAIN_CHAT_MODEL,
-  };
-  const out: ResolvedAIOptions = {};
-  if (seed.embedding_disabled) {
-    out.noEmbedding = true;
-  } else if (seed.embedding_model) {
-    out.embedding_model = seed.embedding_model;
-    if (seed.embedding_dimensions) out.embedding_dimensions = seed.embedding_dimensions;
-  }
-  if (seed.expansion_model) out.expansion_model = seed.expansion_model;
-  if (seed.chat_model) out.chat_model = seed.chat_model;
-  return out;
-}
-
 /**
  * Resolve AI provider options for `gbrain init`.
  *
@@ -238,13 +203,18 @@ async function resolveAIOptions(opts: ResolveAIOptionsArgs): Promise<ResolvedAIO
   // user already opted into deferred mode.
   try {
     const { loadConfig } = await import('../core/config.ts');
-    // #1058: loadConfig() returns null on a cold install (no config.json AND
-    // no DATABASE_URL) — before it ever reaches its env merge. The seed helper
-    // falls back to the same GBRAIN_* env vars directly in that case.
-    Object.assign(out, seedAIOptionsFromConfig(loadConfig()));
+    const cfg = loadConfig();
+    if (cfg?.embedding_disabled) {
+      out.noEmbedding = true;
+    } else if (cfg?.embedding_model) {
+      out.embedding_model = cfg.embedding_model;
+      if (cfg.embedding_dimensions) out.embedding_dimensions = cfg.embedding_dimensions;
+    }
+    if (cfg?.expansion_model) out.expansion_model = cfg.expansion_model;
+    if (cfg?.chat_model) out.chat_model = cfg.chat_model;
   } catch {
-    // loadConfig threw — treat as first-time install, fall through to env
-    // detection.
+    // loadConfig throws when no brain configured — first-time install, fall
+    // through to env detection.
   }
 
   // --- Tier 1+2: explicit flags ---------------------------------------------

src/core/ai/capabilities.ts

@@ -22,6 +22,7 @@
  */
 
 import { resolveRecipe } from './model-resolver.ts';
+import { listRecipes } from './recipes/index.ts';
 import { AIConfigError } from './errors.ts';
 
 export interface ProviderCapabilities {
@@ -77,7 +78,10 @@ export function getProviderCapabilities(modelString: string): ProviderCapabiliti
   if (!chat) {
     throw new AIConfigError(
       `Provider "${recipe.id}" does not offer a chat touchpoint.`,
-      `Known providers with chat: openai, anthropic, google, openrouter, litellm-proxy, deepseek, groq, together, azure-openai, dashscope, minimax, zhipu, ollama, llama-server. Pick one for models.tier.subagent.`,
+      // Computed from the registry so the hint can't drift into listing
+      // chat-less providers (the pre-fix list falsely included embedding-only
+      // recipes, sending users in circles — #1157).
+      `Known providers with chat: ${listRecipes().filter(r => r.touchpoints.chat).map(r => r.id).join(', ')}. Pick one for models.tier.subagent.`,
     );
   }
 

src/core/ai/recipes/zhipu.ts

@@ -1,9 +1,10 @@
 import type { Recipe } from '../types.ts';
 
 /**
- * Zhipu AI (智谱AI) BigModel Open Platform. OpenAI-compatible /embeddings
- * endpoint at open.bigmodel.cn. Hosts embedding-2 (1024d) and embedding-3
- * (Matryoshka up to 2048d).
+ * Zhipu AI (智谱AI) BigModel Open Platform. OpenAI-compatible /embeddings and
+ * /chat/completions endpoints at open.bigmodel.cn. Hosts embedding-2 (1024d),
+ * embedding-3 (Matryoshka up to 2048d), and the GLM chat family (glm-5.1 etc.)
+ * with native tool calling — usable for models.tier.subagent (#1157).
  *
  * embedding-3 at 2048 dims exceeds pgvector's HNSW cap of 2000 — those
  * brains fall back to exact vector scans (see
@@ -25,6 +26,20 @@ export const zhipu: Recipe = {
     setup_url: 'https://open.bigmodel.cn/',
   },
   touchpoints: {
+    chat: {
+      // Informational list (openai-compat tier: assertTouchpoint doesn't
+      // enforce it), so newer GLM ids pass without a recipe edit.
+      models: ['glm-5.1', 'glm-4.6', 'glm-4.5'],
+      supports_tools: true,
+      // gbrain-side stable tool ids (v0.38 D11) decoupled the loop from
+      // Anthropic response formats; GLM tool calling is stable through the
+      // OpenAI-compat path, same as deepseek/groq.
+      supports_subagent_loop: true,
+      // Anthropic-style cache_control markers are not honored on the
+      // OpenAI-compat path — the loop runs hot (degraded:no_caching warn).
+      supports_prompt_cache: false,
+      max_context_tokens: 128000,
+    },
     embedding: {
       models: ['embedding-3', 'embedding-2'],
       default_dims: 1024,
@@ -36,5 +51,5 @@ export const zhipu: Recipe = {
     },
   },
   setup_hint:
-    'Get an API key at https://open.bigmodel.cn/, then `export ZHIPUAI_API_KEY=...`',
+    'Get an API key at https://open.bigmodel.cn/, then `export ZHIPUAI_API_KEY=...`. Chat/subagent: use `zhipu:glm-5.1`.',
 };

src/core/operations.ts

@@ -332,15 +332,6 @@ export interface OperationContext {
    * remote/untrusted (defense in depth in case the type is bypassed via cast).
    */
   remote: boolean;
-  /**
-   * Transport marker for auth-less remote surfaces (#1061). The stdio MCP
-   * dispatch sets 'stdio' — it is deliberately `remote: true` (agent-facing,
-   * untrusted) but has no per-token auth (local pipe), so identity ops like
-   * whoami need a way to distinguish "known auth-less transport" from "a
-   * transport bug forgot to thread ctx.auth". Trust decisions MUST NOT key
-   * off this field — only `ctx.remote === false` grants trust.
-   */
-  transport?: 'stdio';
   /**
    * Subagent runtime context (v0.16+). Set by the subagent tool dispatcher when
    * dispatching an op as a tool call from an LLM loop. Used to enforce per-op
@@ -3722,10 +3713,9 @@ const whoami: Operation = {
     'Introspect the calling identity. Returns one of three transport shapes: ' +
     '{transport: "oauth", client_id, client_name, scopes, expires_at}, ' +
     '{transport: "legacy", token_name, scopes, expires_at: null}, or ' +
-    '{transport: "local", scopes: []}, or {transport: "stdio", scopes: []} ' +
-    'for the auth-less stdio MCP pipe. Throws unknown_transport when the ' +
-    'context is ambiguous (remote=true without auth and no transport marker) ' +
-    '— fail-closed posture mirroring the v0.26.9 trust-boundary contract.',
+    '{transport: "local", scopes: []}. Throws unknown_transport when the ' +
+    'context is ambiguous (remote=true without auth) — fail-closed posture ' +
+    'mirroring the v0.26.9 trust-boundary contract.',
   params: {},
   scope: 'read',
   handler: async (ctx) => {
@@ -3737,12 +3727,6 @@ const whoami: Operation = {
     if (ctx.remote === false) {
       return { transport: 'local', scopes: [] };
     }
-    // #1061: stdio MCP is remote/untrusted by design but has no per-token
-    // auth (local pipe) — a known transport, not a bug. Report it instead of
-    // throwing. Empty scopes: nothing here may be used to gate anything.
-    if (!ctx.auth && ctx.transport === 'stdio') {
-      return { transport: 'stdio', scopes: [] };
-    }
     if (!ctx.auth) {
       throw new OperationError(
         'unknown_transport',

src/mcp/dispatch.ts

@@ -32,12 +32,6 @@ export interface DispatchOpts {
   remote?: boolean;
   /** Override the default stderr logger (e.g. CLI uses console.* directly). */
   logger?: OperationContext['logger'];
-  /**
-   * #1061: transport marker for auth-less remote surfaces. The stdio MCP
-   * server passes 'stdio' so identity ops (whoami) can report the transport
-   * instead of throwing unknown_transport. Never used for trust decisions.
-   */
-  transport?: OperationContext['transport'];
   /**
    * v0.28: per-token allow-list for the takes.holder field. Threaded by
    * the HTTP/stdio transport from `access_tokens.permissions.takes_holders`.
@@ -209,7 +203,6 @@ export function buildOperationContext(
     logger: opts.logger || stderrLogger,
     dryRun: !!params.dry_run,
     remote: opts.remote ?? true,
-    transport: opts.transport,
     takesHoldersAllowList: opts.takesHoldersAllowList,
     // v0.34 D4: sourceId is REQUIRED at the type level. Auto-fill 'default'
     // for single-source brains and any caller who didn't resolve a sourceId.

src/mcp/server.ts

@@ -42,10 +42,6 @@ export async function startMcpServer(engine: BrainEngine) {
     // `gbrain call <op>` (sets remote=false in src/cli.ts).
     return dispatchToolCall(engine, name, params, {
       remote: true,
-      // #1061: mark the transport so whoami can report {transport: 'stdio'}
-      // instead of throwing unknown_transport. Trust posture unchanged —
-      // stdio stays remote/untrusted.
-      transport: 'stdio',
       takesHoldersAllowList: ['world'],
       // v0.31: source defaults to 'default' for stdio (no per-token scope).
       // Operators who want a different source on stdio MCP should set

test/ai/recipe-zhipu.test.ts

@@ -69,6 +69,45 @@ describe('recipe: zhipu', () => {
     expect(sql.toLowerCase()).toContain('hnsw');
   });
 
+  test('chat touchpoint declares GLM models with tool + subagent-loop support (#1157)', () => {
+    const r = getRecipe('zhipu')!;
+    expect(r.touchpoints.chat).toBeDefined();
+    expect(r.touchpoints.chat!.models).toContain('glm-5.1');
+    expect(r.touchpoints.chat!.supports_tools).toBe(true);
+    expect(r.touchpoints.chat!.supports_subagent_loop).toBe(true);
+    expect(r.touchpoints.chat!.supports_prompt_cache).toBe(false);
+  });
+
+  test('zhipu:glm-5.1 passes the subagent capability gate (degraded:no_caching, not refused)', async () => {
+    // Pre-fix: getProviderCapabilities threw "does not offer a chat touchpoint"
+    // and classifyCapabilities returned 'unknown' → subagent submit refused.
+    const { getProviderCapabilities, classifyCapabilities } =
+      await import('../../src/core/ai/capabilities.ts');
+    const caps = getProviderCapabilities('zhipu:glm-5.1');
+    expect(caps.supportsToolCalling).toBe(true);
+    expect(classifyCapabilities('zhipu:glm-5.1')).toBe('degraded:no_caching');
+  });
+
+  test('no-chat-touchpoint error hint lists only providers that actually have chat', async () => {
+    // The hint is computed from the registry; every provider it names must
+    // really carry a chat touchpoint (pre-fix it hardcoded zhipu/dashscope/
+    // minimax, all embedding-only at the time).
+    const { getProviderCapabilities } = await import('../../src/core/ai/capabilities.ts');
+    const { listRecipes } = await import('../../src/core/ai/recipes/index.ts');
+    let hint = '';
+    try {
+      getProviderCapabilities('voyage:voyage-3');
+      throw new Error('expected AIConfigError for embedding-only provider');
+    } catch (e) {
+      hint = (e as { fix?: string }).fix ?? String(e);
+    }
+    const listed = hint.match(/chat: ([^.]+)\./)?.[1]?.split(', ') ?? [];
+    expect(listed.length).toBeGreaterThan(0);
+    const withChat = new Set(listRecipes().filter(r => r.touchpoints.chat).map(r => r.id));
+    for (const id of listed) expect(withChat.has(id)).toBe(true);
+    expect(listed).toContain('zhipu');
+  });
+
   test('dimsProviderOptions threads dimensions for embedding-3 (Matryoshka)', async () => {
     // Codex finding #1: Zhipu embedding-3 is Matryoshka 256-2048. Without
     // `dimensions` on the wire, user-selected non-default dims are

test/init-env-detection.test.ts

@@ -12,7 +12,7 @@
  */
 
 import { describe, test, expect } from 'bun:test';
-import { groupReadyByProvider, findEnvKeyTypos, seedAIOptionsFromConfig } from '../src/commands/init.ts';
+import { groupReadyByProvider, findEnvKeyTypos } from '../src/commands/init.ts';
 
 describe('groupReadyByProvider — embedding touchpoint', () => {
   test('OPENAI_API_KEY alone → openai is ready', async () => {
@@ -149,47 +149,3 @@ describe('findEnvKeyTypos', () => {
     expect(got.find(t => t.userSet === 'COMPLETELY_UNRELATED_KEY')).toBeUndefined();
   });
 });
-
-describe('seedAIOptionsFromConfig — #1058 cold-install env fallback', () => {
-  test('null config (no config.json, no DATABASE_URL) falls back to GBRAIN_* env vars', () => {
-    const got = seedAIOptionsFromConfig(null, {
-      GBRAIN_EMBEDDING_MODEL: 'voyage:voyage-3-large',
-      GBRAIN_EMBEDDING_DIMENSIONS: '1024',
-      GBRAIN_EXPANSION_MODEL: 'openai:gpt-5-mini',
-      GBRAIN_CHAT_MODEL: 'anthropic:claude-sonnet-4-6',
-    });
-    expect(got.embedding_model).toBe('voyage:voyage-3-large');
-    expect(got.embedding_dimensions).toBe(1024);
-    expect(got.expansion_model).toBe('openai:gpt-5-mini');
-    expect(got.chat_model).toBe('anthropic:claude-sonnet-4-6');
-  });
-
-  test('null config + no env vars → empty seed (Tier-3 detection takes over)', () => {
-    const got = seedAIOptionsFromConfig(null, {});
-    expect(got).toEqual({});
-  });
-
-  test('persisted config wins (loadConfig already merged env when non-null)', () => {
-    const got = seedAIOptionsFromConfig(
-      { engine: 'pglite', embedding_model: 'openai:text-embedding-3-small', embedding_dimensions: 1536 } as any,
-      { GBRAIN_EMBEDDING_MODEL: 'voyage:voyage-3-large' },
-    );
-    expect(got.embedding_model).toBe('openai:text-embedding-3-small');
-    expect(got.embedding_dimensions).toBe(1536);
-  });
-
-  test('embedding_disabled sentinel honored on re-init', () => {
-    const got = seedAIOptionsFromConfig({ engine: 'pglite', embedding_disabled: true } as any, {});
-    expect(got.noEmbedding).toBe(true);
-    expect(got.embedding_model).toBeUndefined();
-  });
-
-  test('non-numeric GBRAIN_EMBEDDING_DIMENSIONS ignored, model still seeds', () => {
-    const got = seedAIOptionsFromConfig(null, {
-      GBRAIN_EMBEDDING_MODEL: 'voyage:voyage-3-large',
-      GBRAIN_EMBEDDING_DIMENSIONS: 'not-a-number',
-    });
-    expect(got.embedding_model).toBe('voyage:voyage-3-large');
-    expect(got.embedding_dimensions).toBeUndefined();
-  });
-});

test/whoami.test.ts

@@ -94,35 +94,6 @@ describe('whoami op contract', () => {
     expect(result.expires_at).toBeNull();
   });
 
-  // #1061: stdio MCP is remote/untrusted by design but has no per-token auth
-  // (local pipe). The stdio dispatch marks ctx.transport='stdio'; whoami
-  // reports it instead of throwing unknown_transport.
-  test('stdio transport (remote=true, no auth, transport marker) reports stdio', async () => {
-    const result = (await whoami.handler(
-      ctxWith({ remote: true, auth: undefined, transport: 'stdio' }),
-      {},
-    )) as any;
-    expect(result.transport).toBe('stdio');
-    expect(result.scopes).toEqual([]);
-  });
-
-  test('stdio marker does not mask real auth (auth still wins)', async () => {
-    const result = (await whoami.handler(
-      ctxWith({
-        remote: true,
-        transport: 'stdio',
-        auth: {
-          token: 'gbrain_at_xxx',
-          clientId: 'gbrain_cl_abc',
-          scopes: ['read'],
-          expiresAt: 1,
-        } as AuthInfo,
-      }),
-      {},
-    )) as any;
-    expect(result.transport).toBe('oauth');
-  });
-
   // Q3: ambiguous transport — fail-closed. The footgun this guards against
   // is a future transport that lands without threading auth, where a buggy
   // caller might trust whoami's output to gate sensitive ops.