feat: global extract provider config

- workflow.yaml supports config section (maxDepth, extract provider)
- ExtractProviderConfig with env: prefix for apiKey resolution
- getExtractProvider(storageRoot) returns LlmProvider from config
- workflowAsAgent uses config maxDepth (fallback 3)
- Registry read/write preserves config
- 158 tests passing

Fixes #43

packages/workflow/src/extract-provider.ts

@@ -0,0 +1,35 @@
+import { readWorkflowRegistry } from "./registry.js";
+import type { WorkflowConfig } from "./registry-types.js";
+import { err, ok, type Result } from "./result.js";
+import { getDefaultWorkflowStorageRoot } from "./storage-root.js";
+import type { LlmProvider } from "./types.js";
+
+const DEFAULT_WORKFLOW_AS_AGENT_MAX_DEPTH = 3;
+
+export function getWorkflowAsAgentMaxDepth(config: WorkflowConfig | null): number {
+  if (config === null) {
+    return DEFAULT_WORKFLOW_AS_AGENT_MAX_DEPTH;
+  }
+  return config.maxDepth;
+}
+
+/** Loads `config.extract` from workflow.yaml (apiKey already resolved at registry parse time). */
+export async function getExtractProvider(
+  storageRoot: string | undefined,
+): Promise<Result<LlmProvider, string>> {
+  const root = storageRoot ?? getDefaultWorkflowStorageRoot();
+  const regResult = await readWorkflowRegistry(root);
+  if (!regResult.ok) {
+    return err(regResult.error.message);
+  }
+  const cfg = regResult.value.config;
+  if (cfg === null) {
+    return err("workflow registry has no global config section");
+  }
+  const ex = cfg.extract;
+  return ok({
+    baseUrl: ex.baseUrl,
+    apiKey: ex.apiKey,
+    model: ex.model,
+  });
+}

packages/workflow/src/index.ts

@@ -17,6 +17,7 @@ export {
 } from "./engine.js";
 export { type ExtractedBundleExports, extractBundleExports } from "./extract-bundle-exports.js";
 export { createExtract, type ExtractFn } from "./extract-fn.js";
+export { getExtractProvider } from "./extract-provider.js";
 export {
   buildForkPlan,
   type ForkHistoricalStep,
@@ -54,6 +55,7 @@ export {
   type ThreadMerklePayload,
 } from "./merkle.js";
 export {
+  type ExtractProviderConfig,
   getRegisteredWorkflow,
   listRegisteredWorkflowNames,
   parseWorkflowRegistryYaml,
@@ -62,6 +64,7 @@ export {
   rollbackWorkflowToHistoryHash,
   stringifyWorkflowRegistryYaml,
   unregisterWorkflow,
+  type WorkflowConfig,
   type WorkflowHistoryEntry,
   type WorkflowRegistryEntry,
   type WorkflowRegistryFile,

packages/workflow/src/registry-normalize.ts

@@ -1,10 +1,68 @@
 import type {
+  ExtractProviderConfig,
+  WorkflowConfig,
   WorkflowHistoryEntry,
   WorkflowRegistryEntry,
   WorkflowRegistryFile,
 } from "./registry-types.js";
 import { err, ok, type Result } from "./result.js";
 
+function resolveRegistryApiKey(raw: string): Result<string, Error> {
+  if (raw.startsWith("env:")) {
+    const name = raw.slice("env:".length);
+    if (name === "") {
+      return err(new Error('config.extract.apiKey "env:" reference must name a variable'));
+    }
+    const value = process.env[name];
+    if (value === undefined) {
+      return err(new Error(`config.extract.apiKey: environment variable "${name}" is not set`));
+    }
+    return ok(value);
+  }
+  return ok(raw);
+}
+
+function normalizeExtractProviderConfig(raw: unknown): Result<ExtractProviderConfig, Error> {
+  if (raw === null || typeof raw !== "object") {
+    return err(new Error('registry config must contain an "extract" mapping'));
+  }
+  const e = raw as Record<string, unknown>;
+  const baseUrl = e.baseUrl;
+  const model = e.model;
+  const apiKeyRaw = e.apiKey;
+  if (typeof baseUrl !== "string" || baseUrl === "") {
+    return err(new Error("config.extract.baseUrl must be a non-empty string"));
+  }
+  if (typeof model !== "string" || model === "") {
+    return err(new Error("config.extract.model must be a non-empty string"));
+  }
+  if (typeof apiKeyRaw !== "string" || apiKeyRaw === "") {
+    return err(new Error("config.extract.apiKey must be a non-empty string"));
+  }
+  const apiKeyResult = resolveRegistryApiKey(apiKeyRaw);
+  if (!apiKeyResult.ok) {
+    return apiKeyResult;
+  }
+  return ok({ baseUrl, model, apiKey: apiKeyResult.value });
+}
+
+function normalizeWorkflowConfig(raw: unknown): Result<WorkflowConfig, Error> {
+  if (raw === null || typeof raw !== "object") {
+    return err(new Error('registry "config" must be a mapping'));
+  }
+  const c = raw as Record<string, unknown>;
+  const maxDepth = c.maxDepth;
+  const extractRaw = c.extract;
+  if (typeof maxDepth !== "number" || !Number.isInteger(maxDepth) || maxDepth < 0) {
+    return err(new Error("config.maxDepth must be a non-negative integer"));
+  }
+  const extractResult = normalizeExtractProviderConfig(extractRaw);
+  if (!extractResult.ok) {
+    return extractResult;
+  }
+  return ok({ maxDepth, extract: extractResult.value });
+}
+
 export function normalizeWorkflowHistoryEntry(
   workflowName: string,
   index: number,
@@ -61,6 +119,15 @@ export function normalizeWorkflowRegistryRoot(raw: unknown): Result<WorkflowRegi
     return err(new Error("registry root must be a mapping"));
   }
   const root = raw as Record<string, unknown>;
+  const configRaw = root.config;
+  let config: WorkflowConfig | null = null;
+  if (configRaw !== undefined && configRaw !== null) {
+    const configResult = normalizeWorkflowConfig(configRaw);
+    if (!configResult.ok) {
+      return configResult;
+    }
+    config = configResult.value;
+  }
   const workflowsRaw = root.workflows;
   if (workflowsRaw === null || workflowsRaw === undefined || typeof workflowsRaw !== "object") {
     return err(new Error('registry must contain a "workflows" mapping'));
@@ -73,5 +140,5 @@ export function normalizeWorkflowRegistryRoot(raw: unknown): Result<WorkflowRegi
     }
     workflows[name] = entryResult.value;
   }
-  return ok({ workflows });
+  return ok({ config, workflows });
 }

packages/workflow/src/registry-types.ts

@@ -9,6 +9,19 @@ export type WorkflowRegistryEntry = {
   history: WorkflowHistoryEntry[];
 };
 
+/** LLM provider settings under `config.extract` in workflow.yaml (apiKey resolved after parse). */
+export type ExtractProviderConfig = {
+  baseUrl: string;
+  model: string;
+  apiKey: string;
+};
+
+export type WorkflowConfig = {
+  maxDepth: number;
+  extract: ExtractProviderConfig;
+};
+
 export type WorkflowRegistryFile = {
+  config: WorkflowConfig | null;
   workflows: Record<string, WorkflowRegistryEntry>;
 };

packages/workflow/src/registry.ts

@@ -12,6 +12,8 @@ import type {
 import { err, ok, type Result } from "./result.js";
 
 export type {
+  ExtractProviderConfig,
+  WorkflowConfig,
   WorkflowHistoryEntry,
   WorkflowRegistryEntry,
   WorkflowRegistryFile,
@@ -22,7 +24,7 @@ export function workflowRegistryPath(storageRoot: string): string {
 }
 
 function emptyRegistry(): WorkflowRegistryFile {
-  return { workflows: {} };
+  return { config: null, workflows: {} };
 }
 
 export function parseWorkflowRegistryYaml(text: string): Result<WorkflowRegistryFile, Error> {
@@ -103,6 +105,7 @@ export function registerWorkflowVersion(
       : [{ hash: prev.hash, timestamp: prev.timestamp }, ...baseHistory];
   const next: WorkflowRegistryEntry = { hash, timestamp, history };
   return {
+    config: registry.config,
     workflows: { ...registry.workflows, [name]: next },
   };
 }
@@ -150,5 +153,5 @@ export function unregisterWorkflow(
     return err(new Error(`workflow not registered: ${name}`));
   }
   const { [name]: _removed, ...rest } = registry.workflows;
-  return ok({ workflows: rest });
+  return ok({ config: registry.config, workflows: rest });
 }

packages/workflow/src/workflow-as-agent.ts

@@ -3,15 +3,13 @@ import { join } from "node:path";
 import { createCasStore } from "./cas.js";
 import { type ExecuteThreadIo, executeThread } from "./engine.js";
 import { extractBundleExports } from "./extract-bundle-exports.js";
+import { getWorkflowAsAgentMaxDepth } from "./extract-provider.js";
 import { createLogger } from "./logger.js";
 import { getRegisteredWorkflow, readWorkflowRegistry } from "./registry.js";
 import { getDefaultWorkflowStorageRoot, getGlobalCasDir } from "./storage-root.js";
 import type { AgentContext, AgentFn, ThreadInput } from "./types.js";
 import { generateUlid } from "./ulid.js";
 
-/** Maximum `WorkflowFnOptions.depth` allowed for a child spawned via `workflowAsAgent`. */
-const WORKFLOW_AS_AGENT_MAX_DEPTH = 3;
-
 export type WorkflowAsAgentOptions = {
   /** When `null`, uses `getDefaultWorkflowStorageRoot()`. */
   storageRoot: string | null;
@@ -34,9 +32,6 @@ export function workflowAsAgent(
 ): AgentFn {
   return async (ctx: AgentContext): Promise<string> => {
     const nextDepth = ctx.depth + 1;
-    if (nextDepth > WORKFLOW_AS_AGENT_MAX_DEPTH) {
-      return `ERROR: workflow-as-agent depth limit exceeded (max ${WORKFLOW_AS_AGENT_MAX_DEPTH})`;
-    }
 
     const storageRoot = resolveWorkflowAsAgentStorageRoot(options);
 
@@ -45,6 +40,11 @@ export function workflowAsAgent(
       return `ERROR: failed to read workflow registry: ${registryResult.error.message}`;
     }
 
+    const maxDepth = getWorkflowAsAgentMaxDepth(registryResult.value.config);
+    if (nextDepth > maxDepth) {
+      return `ERROR: workflow-as-agent depth limit exceeded (max ${maxDepth})`;
+    }
+
     const entry = getRegisteredWorkflow(registryResult.value, workflowName);
     if (entry === null) {
       return `ERROR: workflow "${workflowName}" not found in registry`;