@botbotgo/agent-harness 0.0.41 → 0.0.42

package/README.md

@@ -2,36 +2,26 @@
 
 ## Product Overview
 
-`@botbotgo/agent-harness` solves the gap between an agent execution engine and a real multi-agent application runtime.
+`@botbotgo/agent-harness` is a workspace-shaped application runtime for real agent products.
 
-The problem:
+It is not a new agent framework. It is the layer that sits around LangChain v1 and DeepAgents so an application can be assembled, configured, operated, and recovered as one runtime.
 
-- agent logic, tools, MCP, routing, memory, and approvals usually end up scattered across code and scripts
-- execution frameworks can run agents, but they do not give you a workspace-shaped product runtime by themselves
-- applications need persisted threads, resumability, approvals, and operational control, not just model calls
+The package is built for the gap between:
 
-What this package provides:
+- agent execution semantics owned by LangChain v1 and DeepAgents
+- application runtime concerns such as persisted threads, approvals, recovery, routing, maintenance, and local resource loading
 
-- a workspace runtime for multi-agent applications
-- YAML-defined agents, models, routing, and maintenance policy
-- local tools and SKILL packages loaded from `resources/`
-- DeepAgents-first execution with LangChain v1 compatibility
-- feature-level runtime APIs for runs, threads, approvals, and events
-
-Core API:
+What it provides:
 
-- `createAgentHarness(...)`
-- `run(...)`
-- `subscribe(...)`
-- `listThreads(...)`
-- `getThread(...)`
-- `listApprovals(...)`
-- `getApproval(...)`
-- `stop(...)`
+- a small runtime API centered on `createAgentHarness(...)`, `run(...)`, `subscribe(...)`, inspection methods, and `stop(...)`
+- YAML-defined runtime assembly for hosts, models, routing, recovery, concurrency, MCP, and maintenance policy
+- DeepAgents-first execution with LangChain v1 compatibility
+- local `resources/tools/` and `resources/skills/` loading
+- persisted runs, threads, approvals, events, and resumable checkpoints
 
 ## Quick Start
 
-Install the package:
+Install:
 
 ```bash
 npm install @botbotgo/agent-harness
@@ -54,19 +44,17 @@ your-workspace/
     skills/
 ```
 
-Use the standard layout only. Agent entry files must live under `config/agents/`.
-
 Minimal usage:
 
 ```ts
 import { createAgentHarness, run, stop } from "@botbotgo/agent-harness";
 
-const runtime = await createAgentHarness("/absolute/path/to/your-workspace");
+const runtime = await createAgentHarness("/absolute/path/to/workspace");
 
 try {
   const result = await run(runtime, {
     agentId: "auto",
-    input: "Summarize what this workspace is for.",
+    input: "Explain what this workspace is for.",
   });
 
   console.log(result.output);
@@ -77,13 +65,16 @@ try {
 
 ## Feature List
 
-- Workspace runtime with DeepAgents-first execution and LangChain v1 compatibility
-- YAML-defined host routing through `config/workspace.yaml`
-- Auto-discovered local tools and SKILL packages from `resources/tools/` and `resources/skills/`
-- MCP bridge support for agent-declared servers
-- MCP server support for exposing an agent toolset to other clients
-- Persisted threads, approvals, run lifecycle, and resumable checkpoints
+- Workspace runtime for multi-agent applications
+- DeepAgents-first execution with LangChain v1 compatibility
+- YAML-defined host routing
+- Auto-discovered local tools and SKILL packages
+- MCP bridge support for agent-declared MCP servers
+- MCP server support for exposing harness tools outward
+- Persisted threads, runs, approvals, and lifecycle events
+- Recovery policy and resumable checkpoints
 - Background checkpoint maintenance
+- Runtime-level concurrency control
 
 ## How To Use
 
@@ -95,13 +86,9 @@ import { AgentHarnessRuntime, createAgentHarness } from "@botbotgo/agent-harness
 const runtime: AgentHarnessRuntime = await createAgentHarness("/absolute/path/to/workspace");
 ```
 
-Or:
+You can also create a runtime from a precompiled `WorkspaceBundle`.
 
-```ts
-const runtime = await createAgentHarness(workspaceBundle);
-```
-
-`createAgentHarness(workspaceRoot)` loads one `WorkspaceBundle`, resolves `resources/`, initializes persistence under `runRoot`, and starts runtime maintenance.
+`createAgentHarness(...)` loads one workspace, resolves `resources/`, initializes persistence under `runRoot`, and starts runtime maintenance.
 
 ### Run A Request
 
@@ -110,7 +97,7 @@ import { run } from "@botbotgo/agent-harness";
 
 const result = await run(runtime, {
   agentId: "orchestra",
-  input: "Explain the available agents in this workspace.",
+  input: "Summarize the runtime design.",
   context: {
     requestId: "req-123",
   },
@@ -120,24 +107,24 @@ const result = await run(runtime, {
 });
 ```
 
-Each run creates or continues a persisted thread. The result includes `threadId`, `runId`, `state`, and `output`.
+Each run creates or continues a persisted thread and returns `threadId`, `runId`, `state`, and `output`.
 
-When the underlying LangChain or DeepAgents graph expects extra invocation data, `run(...)` can pass it through without inventing a parallel harness abstraction:
+When the upstream LangChain v1 or DeepAgents graph expects extra invocation data, the runtime passes it through directly instead of inventing a parallel harness abstraction:
 
-- `context`: forwarded as LangChain v1 runtime context
-- `state`: merged into the agent invocation input for stateful graphs
-- `files`: merged into the agent invocation input so DeepAgents state-backed skills or memories can be provided at call time
+- `context` is forwarded as runtime context
+- `state` is merged into the invocation input
+- `files` is merged into the invocation input for state-backed DeepAgents files, memories, and skills
 
 ### Let The Runtime Route
 
 ```ts
 const result = await run(runtime, {
   agentId: "auto",
-  input: "Inspect this repository and explain the release flow.",
+  input: "Inspect the repository and explain the release flow.",
 });
 ```
 
-`agentId: "auto"` evaluates ordered `routing.rules` first, then falls back to `routing.defaultAgentId`, and only uses model routing when `routing.modelRouting: true`.
+`agentId: "auto"` evaluates ordered `routing.rules`, then `routing.defaultAgentId`, and only falls back to model routing when `routing.modelRouting: true`.
 
 ### Stream Output And Events
 
@@ -156,9 +143,9 @@ const result = await run(runtime, {
 });
 ```
 
-The runtime emits typed lifecycle events for output, state changes, approvals, and delegations. `subscribe(...)` is a read-only observer surface.
+`subscribe(...)` is a read-only observer surface over stored lifecycle events.
 
-### Query Threads And Approvals
+### Inspect Threads And Approvals
 
 ```ts
 import {
@@ -174,21 +161,7 @@ const approvals = await listApprovals(runtime, { status: "pending" });
 const approval = approvals[0] ? await getApproval(runtime, approvals[0].approvalId) : null;
 ```
 
-These methods return stored runtime data, not raw persistence, workspace, or backend instances.
-
-### Use Skills And Local Tools
-
-`agent-harness` treats `resources/skills/` as SKILL packages and `resources/tools/` as executable local tools.
-
-```yaml
-spec:
-  skills:
-    - path: resources/skills/reviewer
-  tools:
-    - ref: tool/local-toolset
-```
-
-Tool modules are discovered from `resources/tools/*.js`, `resources/tools/*.mjs`, and `resources/tools/*.cjs`. The preferred format is exporting `tool({...})`.
+These methods return runtime-facing records, not raw persistence or backend objects.
 
 ### Bridge MCP Servers Into Agents
 
@@ -201,10 +174,9 @@ spec:
     - name: docs
       transport: http
       url: https://example.com/mcp
-      token: ${DOCS_MCP_TOKEN}
 ```
 
-The runtime discovers MCP tools, filters them through agent config, and exposes them like any other tool.
+The runtime discovers MCP tools, filters them through agent configuration, and exposes them like other tools.
 
 ### Expose Runtime Tools As An MCP Server
 
@@ -227,33 +199,41 @@ await stop(runtime);
 
 Core workspace files:
 
-- `config/workspace.yaml`: runtime defaults such as `runRoot`, routing, and maintenance
-- `config/agent-context.md`: shared bootstrap context for agents
-- `config/models.yaml`: named model presets
-- `config/agents/direct.yaml`: optional lightweight side-path host
-- `config/agents/orchestra.yaml`: default DeepAgent execution host
-- `resources/package.json`: resource package boundary
-- `resources/tools/`: local tool modules
-- `resources/skills/`: local skills
+- `config/workspace.yaml`
+- `config/agent-context.md`
+- `config/models.yaml`
+- `config/agents/direct.yaml`
+- `config/agents/orchestra.yaml`
+- `resources/package.json`
+- `resources/tools/`
+- `resources/skills/`
 
 ### `config/workspace.yaml`
 
-Use this file for:
+Use this file for runtime-level policy:
 
 - `runRoot`
-- ordered `routing.rules`
+- `routing.rules`
+- `routing.defaultAgentId`
 - `routing.systemPrompt`
+- `routing.modelRouting`
+- `concurrency.maxConcurrentRuns`
+- `recovery.enabled`
+- `recovery.resumeOnStartup`
+- `recovery.maxRecoveryAttempts`
 - `maintenance.checkpoints.*`
 
 If `runRoot` is omitted, the runtime defaults to `<workspace-root>/run-data`.
 
 ### `config/agent-context.md`
 
-Use this file for shared bootstrap context that agents read at construction time. Put stable project context here, not long-term mutable memory.
+Use this file for shared startup context loaded into agents at construction time.
 
-### Agent YAML
+Put stable project context here. Do not use it as mutable long-term memory.
 
-Use `config/agents/*.yaml` to configure agents. Common fields include:
+### `config/agents/*.yaml`
+
+Use agent YAML for upstream-facing agent assembly. Common fields include:
 
 - `modelRef`
 - `systemPrompt`
@@ -263,19 +243,39 @@ Use `config/agents/*.yaml` to configure agents. Common fields include:
 - `checkpointer`
 - `store`
 - `backend`
+- `middleware`
 - `subagents`
+- `mcpServers`
 
-### Resource Package
+### `resources/`
 
-Use `resources/` for executable extensions:
+Use `resources/` for executable local extensions:
 
-- `resources/tools/` for local tool modules
+- `resources/tools/` for tool modules
 - `resources/skills/` for SKILL packages
 
-Keep runtime extension source under `resources/`. Keep tests outside the published source tree, for example under the repository `test/` folder.
+Tool modules are discovered from `resources/tools/*.js`, `resources/tools/*.mjs`, and `resources/tools/*.cjs`.
+
+The preferred tool module format is exporting `tool({...})`.
+
+Keep runtime extension source under `resources/`. Keep tests outside the published source tree, for example under repository `test/`.
+
+## Design Notes
 
-### Skills And MCP
+- `agent-harness` should express LangChain v1 and DeepAgents concepts as directly as possible
+- agent-level execution behavior stays upstream
+- application-level orchestration and lifecycle management stays in the harness
+- checkpoint resume is treated as a system-managed runtime behavior, not a primary public abstraction
 
-- Use `resources/skills/` for reusable skill packages
-- Use `mcpServers:` in `config/agents/*.yaml` to bridge external MCP tools into an agent
-- Use `createToolMcpServer(...)` or `serveToolsOverStdio(...)` when you want the runtime to act as an MCP server for another client
+## API Summary
+
+- `createAgentHarness(...)`
+- `run(...)`
+- `subscribe(...)`
+- `listThreads(...)`
+- `getThread(...)`
+- `listApprovals(...)`
+- `getApproval(...)`
+- `createToolMcpServer(...)`
+- `serveToolsOverStdio(...)`
+- `stop(...)`

package/dist/contracts/types.d.ts

@@ -182,7 +182,7 @@ export type CompiledAgentBinding = {
         runRoot: string;
         workspaceRoot?: string;
         hostFacing: boolean;
-        checkpointer?: Record<string, unknown>;
+        checkpointer?: Record<string, unknown> | boolean;
         store?: Record<string, unknown>;
     };
 };
@@ -405,6 +405,12 @@ export type RuntimeModelResolver = (modelId: string) => unknown;
 export type RuntimeEmbeddingModelResolver = (embeddingModelId: string) => unknown;
 export type RuntimeVectorStoreResolver = (vectorStoreId: string) => unknown;
 export type RuntimeMiddlewareResolver = (binding: CompiledAgentBinding) => unknown[];
+export type RuntimeDeclaredMiddlewareResolver = (input: {
+    kind: string;
+    config: Record<string, unknown>;
+    binding?: CompiledAgentBinding;
+    resolveModel: (model: CompiledModel) => Promise<unknown>;
+}) => unknown | unknown[] | null | undefined | Promise<unknown | unknown[] | null | undefined>;
 export type RuntimeCheckpointerResolver = (binding: CompiledAgentBinding) => unknown;
 export type RuntimeStoreResolver = (binding: CompiledAgentBinding) => unknown;
 export type RuntimeBackendResolver = (binding: CompiledAgentBinding) => unknown;
@@ -414,6 +420,7 @@ export type RuntimeAdapterOptions = {
     embeddingModelResolver?: RuntimeEmbeddingModelResolver;
     vectorStoreResolver?: RuntimeVectorStoreResolver;
     middlewareResolver?: RuntimeMiddlewareResolver;
+    declaredMiddlewareResolver?: RuntimeDeclaredMiddlewareResolver;
     checkpointerResolver?: RuntimeCheckpointerResolver;
     storeResolver?: RuntimeStoreResolver;
     backendResolver?: RuntimeBackendResolver;

package/dist/package-version.d.ts

@@ -1 +1 @@
-export declare const AGENT_HARNESS_VERSION = "0.0.40";
+export declare const AGENT_HARNESS_VERSION = "0.0.41";

package/dist/package-version.js

@@ -1 +1 @@
-export const AGENT_HARNESS_VERSION = "0.0.40";
+export const AGENT_HARNESS_VERSION = "0.0.41";

package/dist/runtime/agent-runtime-adapter.d.ts

@@ -15,7 +15,6 @@ declare class RuntimeOperationTimeoutError extends Error {
 export declare class AgentRuntimeAdapter {
     private readonly options;
     constructor(options?: RuntimeAdapterOptions);
-    private canUseSimpleDeepAgentFastPath;
     private resolveBindingTimeout;
     private resolveStreamIdleTimeout;
     private withTimeout;

package/dist/runtime/agent-runtime-adapter.js

@@ -12,6 +12,9 @@ import { computeIncrementalOutput, extractAgentStep, extractInterruptPayload, ex
 import { wrapToolForExecution } from "./tool-hitl.js";
 import { resolveDeclaredMiddleware } from "./declared-middleware.js";
 import { extractMessageText, normalizeMessageContent } from "../utils/message-content.js";
+function countConfiguredTools(binding) {
+    return binding.langchainAgentParams?.tools.length ?? binding.deepAgentParams?.tools.length ?? 0;
+}
 const AGENT_INTERRUPT_SENTINEL_PREFIX = "__agent_harness_interrupt__:";
 const MODEL_SAFE_TOOL_NAME_PATTERN = /^[a-zA-Z0-9_-]+$/;
 class RuntimeOperationTimeoutError extends Error {
@@ -163,33 +166,11 @@ function asStructuredExecutableTool(resolvedTool, modelFacingName, description)
         schema: normalizeResolvedToolSchema(resolvedTool),
     });
 }
-function countConfiguredTools(binding) {
-    return binding.langchainAgentParams?.tools.length ?? binding.deepAgentParams?.tools.length ?? 0;
-}
 export class AgentRuntimeAdapter {
     options;
     constructor(options = {}) {
         this.options = options;
     }
-    canUseSimpleDeepAgentFastPath(binding) {
-        const params = binding.deepAgentParams;
-        if (!params) {
-            return false;
-        }
-        if ((params.subagents?.length ?? 0) > 0) {
-            return false;
-        }
-        if ((params.memory?.length ?? 0) > 0) {
-            return false;
-        }
-        if ((params.skills?.length ?? 0) > 0) {
-            return false;
-        }
-        if (params.backend || params.store) {
-            return false;
-        }
-        return true;
-    }
     resolveBindingTimeout(binding) {
         return resolveTimeoutMs(binding.langchainAgentParams?.model.init.timeout ?? binding.deepAgentParams?.model.init.timeout);
     }
@@ -452,6 +433,8 @@ export class AgentRuntimeAdapter {
         const declarativeMiddleware = await resolveDeclaredMiddleware(binding.langchainAgentParams?.middleware ??
             binding.deepAgentParams?.middleware, {
             resolveModel: (model) => this.resolveModel(model),
+            resolveCustom: this.options.declaredMiddlewareResolver,
+            binding,
         });
         const middleware = [
             ...declarativeMiddleware,
@@ -485,7 +468,7 @@ export class AgentRuntimeAdapter {
             .replaceAll("{{secondaryAgentId}}", secondaryBinding.agent.id)
             .replaceAll("{{secondaryDescription}}", secondaryBinding.agent.description);
     }
-    async resolveSubagents(subagents) {
+    async resolveSubagents(subagents, binding) {
         return Promise.all(subagents.map(async (subagent) => ({
             ...subagent,
             model: subagent.model ? (await this.resolveModel(subagent.model)) : undefined,
@@ -495,6 +478,8 @@ export class AgentRuntimeAdapter {
             contextSchema: subagent.contextSchema,
             middleware: (await resolveDeclaredMiddleware(subagent.middleware, {
                 resolveModel: (model) => this.resolveModel(model),
+                resolveCustom: this.options.declaredMiddlewareResolver,
+                binding,
             })),
         })));
     }
@@ -526,23 +511,6 @@ export class AgentRuntimeAdapter {
         if (!params) {
             throw new Error(`Agent ${binding.agent.id} has no runnable params`);
         }
-        if (this.canUseSimpleDeepAgentFastPath(binding)) {
-            const interruptOn = this.resolveInterruptOn(binding);
-            const model = (await this.resolveModel(params.model));
-            const tools = this.resolveTools(params.tools, binding);
-            if (tools.length > 0 && typeof model.bindTools !== "function") {
-                throw new Error(`Agent ${binding.agent.id} configures ${tools.length} tool(s), but resolved model ${params.model.id} does not support tool binding.`);
-            }
-            return createAgent({
-                model: model,
-                tools: tools,
-                systemPrompt: params.systemPrompt,
-                responseFormat: params.responseFormat,
-                contextSchema: params.contextSchema,
-                middleware: (await this.resolveMiddleware(binding, interruptOn)),
-                checkpointer: this.resolveCheckpointer(binding),
-            });
-        }
         const deepAgentConfig = {
             model: (await this.resolveModel(params.model)),
             tools: this.resolveTools(params.tools, binding),
@@ -550,7 +518,7 @@ export class AgentRuntimeAdapter {
             responseFormat: params.responseFormat,
             contextSchema: params.contextSchema,
             middleware: (await this.resolveMiddleware(binding)),
-            subagents: (await this.resolveSubagents(params.subagents)),
+            subagents: (await this.resolveSubagents(params.subagents, binding)),
             checkpointer: this.resolveCheckpointer(binding),
             store: this.options.storeResolver?.(binding),
             backend: this.options.backendResolver?.(binding),

package/dist/runtime/checkpoint-maintenance.js

@@ -46,7 +46,7 @@ export function readCheckpointMaintenanceConfig(workspace) {
 }
 function resolveSqliteCheckpointPath(binding) {
     const config = binding.harnessRuntime.checkpointer;
-    if (!config) {
+    if (!config || typeof config === "boolean") {
         return null;
     }
     const kind = typeof config.kind === "string" ? config.kind : "FileCheckpointer";

package/dist/runtime/declared-middleware.d.ts

@@ -1,4 +1,11 @@
-import type { CompiledModel } from "../contracts/types.js";
+import type { CompiledAgentBinding, CompiledModel } from "../contracts/types.js";
 export declare function resolveDeclaredMiddleware(middlewareConfigs: unknown[] | undefined, options: {
     resolveModel: (model: CompiledModel) => Promise<unknown>;
+    resolveCustom?: (input: {
+        kind: string;
+        config: Record<string, unknown>;
+        binding?: CompiledAgentBinding;
+        resolveModel: (model: CompiledModel) => Promise<unknown>;
+    }) => unknown | unknown[] | null | undefined | Promise<unknown | unknown[] | null | undefined>;
+    binding?: CompiledAgentBinding;
 }): Promise<unknown[]>;

package/dist/runtime/declared-middleware.js

@@ -83,8 +83,25 @@ export async function resolveDeclaredMiddleware(middlewareConfigs, options) {
             case "anthropicPromptCaching":
                 resolved.push(anthropicPromptCachingMiddleware(runtimeConfig));
                 break;
-            default:
+            default: {
+                const customResolved = options.resolveCustom
+                    ? await options.resolveCustom({
+                        kind,
+                        config: runtimeConfig,
+                        binding: options.binding,
+                        resolveModel: options.resolveModel,
+                    })
+                    : undefined;
+                if (Array.isArray(customResolved)) {
+                    resolved.push(...customResolved);
+                    break;
+                }
+                if (customResolved) {
+                    resolved.push(customResolved);
+                    break;
+                }
                 throw new Error(`Unsupported declarative middleware kind ${kind}`);
+            }
         }
     }
     return resolved;

package/dist/runtime/harness.d.ts

@@ -21,6 +21,9 @@ export declare class AgentHarnessRuntime {
     private readonly resolvedRuntimeAdapterOptions;
     private readonly checkpointMaintenance;
     private readonly recoveryConfig;
+    private readonly concurrencyConfig;
+    private activeRunSlots;
+    private readonly pendingRunSlots;
     private listHostBindings;
     private defaultRunRoot;
     private heuristicRoute;
@@ -68,6 +71,7 @@ export declare class AgentHarnessRuntime {
     private resolveApprovalRecord;
     private isDecisionRun;
     private notifyListener;
+    private acquireRunSlot;
     private dispatchRunListeners;
     run(options: RunOptions): Promise<RunResult>;
     streamEvents(options: RunStartOptions): AsyncGenerator<HarnessStreamItem>;