@botbotgo/agent-harness 0.0.351 → 0.0.353

package/README.md

@@ -1218,9 +1218,10 @@ ACP transport notes:
 - `serveAgUiHttp(runtime)` exposes an AG-UI-compatible HTTP SSE bridge that projects runtime lifecycle, safe progress commentary, text output, upstream thinking, step progress, and tool calls onto `RUN_*`, `STATUS_UPDATE`, `TEXT_MESSAGE_*`, `THINKING_TEXT_MESSAGE_*`, `STEP_*`, and `TOOL_CALL_*` events for UI clients. `botbotgo ag-ui start|stop` now manages that HTTP bridge in the same workspace-local service registry as ACP, A2A, and runtime MCP.
 - `createRuntimeMcpServer(runtime)`, `serveRuntimeMcpOverStdio(runtime)`, and `serveRuntimeMcpOverStreamableHttp(runtime)` expose the persisted runtime control surface itself as MCP tools, including sessions, requests, approvals, artifacts, events, and package export helpers. `botbotgo mcp serve --transport streamable-http --host 127.0.0.1 --port 8090` serves the same tool surface over Streamable HTTP, and `botbotgo mcp start|stop` manages that background endpoint for one workspace.
 - `listRequestEvents(...)`, `listRequestTraceItems(...)`, and `exportRequestPackage(...)` are the request-first inspection helpers.
+- `analyzeBoundaries(runtime)` returns a workspace boundary report covering agent, subagent, tool, and skill surfaces, including structural findings such as missing subagent references, unreferenced tools or skills, and skill allow-lists that do not match the owning agent's exposed tools.
 - `exportRequestPackage(...)` and `exportSessionPackage(...)` package stable runtime records, transcript, approvals, events, artifacts, and governance evidence for operator tooling without reaching into persistence internals.
 - `runtime/default.governance.remoteMcp` can now deny or allow specific MCP servers, raise approval requirements by transport, and stamp transport-based risk tiers into runtime governance bundles. MCP server catalogs can also declare trust tier, access mode, tenant scope, approval policy, prompt-injection risk, and OAuth scope metadata so governance bundles capture why one remote tool is treated as high-risk. Tool policy overrides can also set `decisionMode: manual | auto-approve | auto-reject | deny-and-continue` so operator evidence and execution behavior stay aligned.
 - Protocol responsibilities stay split on purpose: ACP is the primary editor/client runtime boundary, A2A is the streaming-capable agent-platform bridge with polling compatibility, AG-UI is the UI event surface, and runtime MCP is the operator-facing control plane exported as MCP tools.
 - `runtime/default.observability.tracing` can now describe exporter metadata such as OTLP endpoints and propagation mode, so frozen runtime snapshots keep trace-correlation plus operator-visible export context without exposing backend-private span internals.
-- `agent-harness runtime overview`, `agent-harness runtime health`, `agent-harness runtime approvals list|watch`, `agent-harness runtime requests list|tail`, and `agent-harness runtime export request|session` provide a thin operator CLI over persisted runtime health, queue pressure, governance risk, approval queues, active request state, and audit-ready evidence packages.
+- `agent-harness runtime overview`, `agent-harness runtime boundaries`, `agent-harness runtime health`, `agent-harness runtime approvals list|watch`, `agent-harness runtime requests list|tail`, and `agent-harness runtime export request|session` provide a thin operator CLI over workspace boundary analysis, persisted runtime health, queue pressure, governance risk, approval queues, active request state, and audit-ready evidence packages.
 - detailed A2A adapter guidance lives in [`docs/a2a-bridge.md`](docs/a2a-bridge.md)

package/README.zh.md

@@ -1175,9 +1175,10 @@ ACP transport 说明：
 - `serveAgUiHttp(runtime)` 提供 AG-UI HTTP SSE bridge，把 runtime 生命周期、安全进度播报、文本输出、upstream thinking、step 进度与 tool call 投影成 `RUN_*`、`STATUS_UPDATE`、`TEXT_MESSAGE_*`、`THINKING_TEXT_MESSAGE_*`、`STEP_*` 与 `TOOL_CALL_*` 事件，便于 UI 客户端直接接入。`botbotgo ag-ui start|stop` 现在也会把这条 HTTP bridge 托管进与 ACP、A2A、runtime MCP 相同的 workspace 本地服务注册表中。
 - `createRuntimeMcpServer(runtime)`、`serveRuntimeMcpOverStdio(runtime)` 与 `serveRuntimeMcpOverStreamableHttp(runtime)` 会把持久化 runtime 控制面本身暴露成 MCP tools，包括 sessions、requests、approvals、artifacts、events 与 package export helpers。`botbotgo mcp serve --transport streamable-http --host 127.0.0.1 --port 8090` 会把同一套控制面作为 Streamable HTTP 暴露出去，而 `botbotgo mcp start|stop` 可直接托管该后台 endpoint。
 - `listRequestEvents(...)`、`listRequestTraceItems(...)` 与 `exportRequestPackage(...)` 是 request-first 的检查 helper。
+- `analyzeBoundaries(runtime)` 会返回 workspace boundary report，覆盖 agent、subagent、tool 与 skill 的可见边界，并指出缺失 subagent 引用、未被引用的 tools/skills、以及 skill allow-list 与所属 agent 暴露工具不匹配等结构性问题。
 - `exportRequestPackage(...)` 与 `exportSessionPackage(...)` 可把稳定 runtime 记录、transcript、approvals、events、artifacts 与 governance evidence 一起打包给管理工具，而不必直接访问 persistence 内部实现。
 - `runtime/default.governance.remoteMcp` 现在可以按 MCP server 或 transport 做 allow/deny、审批升级，并把 transport 风险等级写进 runtime governance bundles。MCP server catalog 也可以声明 trust tier、access mode、tenant scope、approval policy、prompt-injection risk 与 OAuth scope 元数据，让治理快照能解释为什么某个远端工具被视为高风险。tool policy override 也可以声明 `decisionMode: manual | auto-approve | auto-reject | deny-and-continue`，让治理快照与实际执行路径保持一致。
 - 协议分工要继续保持清晰：ACP 是 editor / client 的主运行时边界，A2A 是支持 streaming 且兼容轮询的 agent-platform bridge，AG-UI 是 UI 事件面，runtime MCP 是以 MCP tools 暴露的 operator control plane。
 - `runtime/default.observability.tracing` 现在可描述 OTLP endpoint 和 propagation mode 这类 exporter 元数据，使冻结的 runtime snapshot 在保留 trace correlation 的同时，也能保留有用的导出上下文，而不暴露 backend 私有 span 细节。
-- `agent-harness runtime overview`、`agent-harness runtime health`、`agent-harness runtime approvals list|watch`、`agent-harness runtime requests list|tail` 与 `agent-harness runtime export request|session` 提供了一层轻量 CLI，可直接查看 runtime health、queue pressure、governance risk、审批队列、运行状态与可审计证据包。
+- `agent-harness runtime overview`、`agent-harness runtime boundaries`、`agent-harness runtime health`、`agent-harness runtime approvals list|watch`、`agent-harness runtime requests list|tail` 与 `agent-harness runtime export request|session` 提供了一层轻量 CLI，可直接查看 workspace boundary analysis、runtime health、queue pressure、governance risk、审批队列、运行状态与可审计证据包。
 - 更详细的 A2A 适配层开发说明见 [`docs/a2a-bridge.md`](docs/a2a-bridge.md)

package/dist/api.d.ts

@@ -1,6 +1,8 @@
 import type { ArtifactListing, CancelOptions, InvocationEnvelope, ListMemoriesInput, ListMemoriesResult, MemoryRecord, MemorizeInput, MemorizeResult, MessageContent, RequestDataEvent, RecallInput, RecallResult, RemoveMemoryInput, RequestEventSnapshot, RequestPlanState, RequestDecisionOptions, RequestRecord, RequestResult as InternalRequestResult, RequestStartOptions, RequestSummary, RuntimeArtifactWriteInput, ResumeOptions, RequestListeners, RuntimeHealthSnapshot, RuntimeGovernanceEvidence, RuntimeGovernanceDiagnostics, RuntimeOperatorOverview, RuntimeQueueDiagnostics, RuntimeAdapterOptions, RuntimeEvaluationExport, RuntimeEvaluationExportInput, RuntimeEvaluationArtifact, RuntimeEvaluationReplayInput, RuntimeEvaluationReplayResult as InternalRuntimeEvaluationReplayResult, RuntimeSessionPackage, RuntimeSessionPackageInput, SessionListSummary, SessionRecord, SessionSummary, TranscriptMessage, UpdateMemoryInput, WorkspaceLoadOptions } from "./contracts/types.js";
 import { AgentHarnessRuntime } from "./runtime/harness.js";
 import type { InventoryAgentRecord, InventorySkillRecord } from "./runtime/harness/system/inventory.js";
+import type { BoundaryAnalysisOptions, WorkspaceBoundaryAnalysis } from "./runtime/harness/system/boundary-analysis.js";
+export type { BoundaryAgentSurface, BoundaryAnalysisOptions, BoundaryAnalysisSummary, BoundaryFinding, BoundaryFindingSeverity, WorkspaceBoundaryAnalysis, } from "./runtime/harness/system/boundary-analysis.js";
 import type { RequirementAssessmentOptions } from "./runtime/harness/system/skill-requirements.js";
 import type { RuntimeMcpServerOptions, RuntimeMcpStreamableHttpServerOptions, ToolMcpServerOptions } from "./mcp.js";
 export { AgentHarnessAcpServer, createAcpServer } from "./acp.js";
@@ -202,6 +204,7 @@ export declare function describeInventory(runtime: AgentHarnessRuntime, options?
     workspaceRoot: string;
     agents: InventoryAgentRecord[];
 };
+export declare function analyzeBoundaries(runtime: AgentHarnessRuntime, options?: BoundaryAnalysisOptions): WorkspaceBoundaryAnalysis;
 export declare function resolveApproval(runtime: AgentHarnessRuntime, options: ResumeOptions): Promise<PublicRequestResult>;
 export declare function cancelRequest(runtime: AgentHarnessRuntime, options: CancelOptions): Promise<PublicRequestResult>;
 export declare function stop(runtime: AgentHarnessRuntime): Promise<void>;

package/dist/api.js

@@ -366,6 +366,9 @@ export function getAgent(runtime, agentId, options) {
 export function describeInventory(runtime, options) {
     return runtime.describeWorkspaceInventory(options);
 }
+export function analyzeBoundaries(runtime, options) {
+    return runtime.analyzeWorkspaceBoundaries(options);
+}
 export async function resolveApproval(runtime, options) {
     return toPublicRequestResult(await runtime.resume(toInternalResumeOptions(options)));
 }

package/dist/cli/options.js

@@ -16,6 +16,7 @@ export function renderUsage() {
   agent-harness ag-ui start [--workspace <path>] [--host <hostname>] [--port <port>]
   agent-harness ag-ui stop [--workspace <path>]
   agent-harness runtime overview [--workspace <path>] [--limit <n>] [--json]
+  agent-harness runtime boundaries [--workspace <path>] [--json]
   agent-harness runtime health [--workspace <path>] [--json]
   agent-harness runtime approvals list [--workspace <path>] [--status <pending|approved|edited|rejected|expired>] [--json]
   agent-harness runtime approvals watch [--workspace <path>] [--status <pending|approved|edited|rejected|expired>] [--poll-ms <ms>] [--once] [--json]

package/dist/cli/runtime-commands.js

@@ -1,4 +1,4 @@
-import { renderApprovalList, renderHealthSnapshot, renderJson, renderOperatorOverview, renderRequestList } from "./runtime-output.js";
+import { renderApprovalList, renderBoundaryAnalysis, renderHealthSnapshot, renderJson, renderOperatorOverview, renderRequestList, } from "./runtime-output.js";
 import { parseRuntimeExportOptions, parseRuntimeInspectOptions, parseScheduledRunOptions, renderUsage } from "./options.js";
 import { resolveCliWorkspaceRoot, validateCliWorkspaceRoot } from "./workspace.js";
 function resolveValidatedWorkspace(cwd, workspaceRoot, stderr) {
@@ -128,6 +128,17 @@ export async function handleRuntimeCommand(subcommandAndArgs, io, deps) {
         if (!workspacePath) {
             return 1;
         }
+        if (subcommand === "boundaries") {
+            const runtime = await createHarness(workspacePath);
+            try {
+                const analysis = runtime.analyzeWorkspaceBoundaries();
+                stdout(parsed.json ? renderJson(analysis) : renderBoundaryAnalysis(analysis, workspacePath));
+                return analysis.summary.errorCount > 0 ? 2 : 0;
+            }
+            finally {
+                await runtime.stop();
+            }
+        }
         const shouldUseClient = subcommand === "health"
             || subcommand === "overview"
             || (subcommand === "approvals" && (nestedCommand === "list" || nestedCommand === "watch"))

package/dist/cli/runtime-output.d.ts

@@ -3,3 +3,4 @@ export declare function renderHealthSnapshot(snapshot: Record<string, unknown>,
 export declare function renderApprovalList(approvals: Array<Record<string, unknown>>): string;
 export declare function renderRequestList(requests: Array<Record<string, unknown>>): string;
 export declare function renderOperatorOverview(overview: Record<string, unknown>, workspacePath: string): string;
+export declare function renderBoundaryAnalysis(analysis: Record<string, unknown>, workspacePath: string): string;

package/dist/cli/runtime-output.js

@@ -122,3 +122,34 @@ export function renderOperatorOverview(overview, workspacePath) {
     }
     return `${lines.join("\n")}\n`;
 }
+export function renderBoundaryAnalysis(analysis, workspacePath) {
+    const lines = [];
+    const summary = isObject(analysis.summary) ? analysis.summary : {};
+    const findings = Array.isArray(analysis.findings) ? analysis.findings.filter(isObject) : [];
+    const errors = typeof summary.errorCount === "number" ? summary.errorCount : 0;
+    const warnings = typeof summary.warningCount === "number" ? summary.warningCount : 0;
+    const info = typeof summary.infoCount === "number" ? summary.infoCount : 0;
+    lines.push(`Runtime boundary analysis ${workspacePath}`);
+    lines.push(`Findings: errors=${errors} warnings=${warnings} info=${info}`);
+    const agents = typeof summary.agentCount === "number" ? summary.agentCount : undefined;
+    const tools = typeof summary.toolCount === "number" ? summary.toolCount : undefined;
+    const skills = typeof summary.skillCount === "number" ? summary.skillCount : undefined;
+    if (agents !== undefined || tools !== undefined || skills !== undefined) {
+        lines.push(`Inventory: agents=${agents ?? "unknown"} tools=${tools ?? "unknown"} skills=${skills ?? "unknown"}`);
+    }
+    if (findings.length === 0) {
+        lines.push("No boundary findings.");
+        return `${lines.join("\n")}\n`;
+    }
+    lines.push("Boundary findings:");
+    for (const finding of findings) {
+        const severity = typeof finding.severity === "string" ? finding.severity : "unknown";
+        const code = typeof finding.code === "string" ? finding.code : "unknown";
+        const message = typeof finding.message === "string" ? finding.message : "";
+        const agent = typeof finding.agentId === "string" ? ` agent=${finding.agentId}` : "";
+        const tool = typeof finding.toolName === "string" ? ` tool=${finding.toolName}` : "";
+        const skill = typeof finding.skillName === "string" ? ` skill=${finding.skillName}` : "";
+        lines.push(`  - ${severity} ${code}:${agent}${tool}${skill}${message ? ` ${message}` : ""}`);
+    }
+    return `${lines.join("\n")}\n`;
+}

package/dist/index.d.ts

@@ -1,9 +1,9 @@
-export { AgentHarnessAcpServer, AgentHarnessRuntime, cancelRequest, createAgentHarness, createAcpServer, createAcpStdioClient, createRuntimeMcpServer, createUpstreamTimelineReducer, createToolMcpServer, deleteSession, describeInventory, exportEvaluationBundle, exportFlow, exportSequence, exportRequestPackage, exportSessionPackage, replayEvaluationBundle, getArtifact, getAgent, getApproval, getOperatorOverview, getRequestPlanState, getRequest, getHealth, listMemories, listRequestTraceItems, getSession, listAgentSkills, listRequestArtifacts, listApprovals, listRequests, listRequestEvents, listSessionSummaries, listSessions, memorize, normalizeUserChatInput, recordArtifact, request, recall, removeMemory, resolveApproval, serveA2aHttp, serveAcpHttp, serveAcpStdio, serveAgUiHttp, serveRuntimeMcpOverStdio, serveRuntimeMcpOverStreamableHttp, serveToolsOverStdio, subscribe, stop, updateMemory, } from "./api.js";
+export { AgentHarnessAcpServer, AgentHarnessRuntime, cancelRequest, createAgentHarness, createAcpServer, createAcpStdioClient, createRuntimeMcpServer, createUpstreamTimelineReducer, createToolMcpServer, analyzeBoundaries, deleteSession, describeInventory, exportEvaluationBundle, exportFlow, exportSequence, exportRequestPackage, exportSessionPackage, replayEvaluationBundle, getArtifact, getAgent, getApproval, getOperatorOverview, getRequestPlanState, getRequest, getHealth, listMemories, listRequestTraceItems, getSession, listAgentSkills, listRequestArtifacts, listApprovals, listRequests, listRequestEvents, listSessionSummaries, listSessions, memorize, normalizeUserChatInput, recordArtifact, request, recall, removeMemory, resolveApproval, serveA2aHttp, serveAcpHttp, serveAcpStdio, serveAgUiHttp, serveRuntimeMcpOverStdio, serveRuntimeMcpOverStreamableHttp, serveToolsOverStdio, subscribe, stop, updateMemory, } from "./api.js";
 export { AcpHarnessClient, InProcessHarnessClient, createAcpHarnessClient, createAcpHttpHarnessClient, createAcpStdioHarnessClient, createAgentHarnessClient, createInProcessHarnessClient, } from "./client.js";
 export { createKnowledgeModule, readKnowledgeRuntimeConfig } from "./knowledge/index.js";
 export { readProceduralMemoryRuntimeConfig } from "./knowledge/procedural/index.js";
 export type { AcpApproval, AcpHttpClient, AcpHttpClientOptions, AcpArtifact, AcpEventNotification, AcpNotification, AcpJsonRpcError, AcpJsonRpcRequest, AcpJsonRpcResponse, AcpJsonRpcSuccess, AcpRequestRecord, AcpRequestParams, AcpServerCapabilities, AcpSessionRecord, AcpStreamNotification, AcpStdioClient, AcpStdioClientOptions, } from "./acp.js";
-export type { Approval, CreateAgentHarnessOptions, ListMemoriesInput, ListMemoriesResult, MemoryDecision, MemoryKind, MemoryRecord, MemoryScope, MemorizeInput, MemorizeResult, NormalizeUserChatInputOptions, OperatorOverview, RecordArtifactInput, PublicRequestListeners, RequestData, RequestArtifactListing, RequestEvent, RequestEventType, RequestSnapshot, RequestPlanState, RequestPackage, RequestPackageInput, RequestFlowGraphInput, RequestResult, RequestTraceItem, RecallInput, RecallResult, RemoveMemoryInput, RuntimeEvaluationExport, RuntimeEvaluationExportInput, RuntimeEvaluationReplayInput, RuntimeEvaluationReplayResult, SessionListSummary, RuntimeSessionPackage, RuntimeSessionPackageInput, UpdateMemoryInput, UserChatInput, UserChatMessage, } from "./api.js";
+export type { Approval, BoundaryAgentSurface, BoundaryAnalysisOptions, BoundaryAnalysisSummary, BoundaryFinding, BoundaryFindingSeverity, CreateAgentHarnessOptions, ListMemoriesInput, ListMemoriesResult, MemoryDecision, MemoryKind, MemoryRecord, MemoryScope, MemorizeInput, MemorizeResult, NormalizeUserChatInputOptions, OperatorOverview, RecordArtifactInput, PublicRequestListeners, RequestData, RequestArtifactListing, RequestEvent, RequestEventType, RequestSnapshot, RequestPlanState, RequestPackage, RequestPackageInput, RequestFlowGraphInput, RequestResult, RequestTraceItem, RecallInput, RecallResult, RemoveMemoryInput, RuntimeEvaluationExport, RuntimeEvaluationExportInput, RuntimeEvaluationReplayInput, RuntimeEvaluationReplayResult, SessionListSummary, RuntimeSessionPackage, RuntimeSessionPackageInput, WorkspaceBoundaryAnalysis, UpdateMemoryInput, UserChatInput, UserChatMessage, } from "./api.js";
 export type { AcpHarnessTransport, HarnessClient, HarnessClientApprovalFilter, HarnessClientRequestFilter, HarnessClientRequestOptions, HarnessClientRequestResult, HarnessClientRequestStartOptions, } from "./client.js";
 export type { KnowledgeListInput, KnowledgeMemorizeInput, KnowledgeModule, KnowledgeModuleDependencies, KnowledgeRecallInput, KnowledgeRuntimeConfig, KnowledgeRuntimeContext, } from "./knowledge/index.js";
 export type { ProceduralMemoryBackgroundConfig, ProceduralMemoryFormationConfig, ProceduralMemoryMaintenanceConfig, ProceduralMemoryMaintenanceIdleConfig, ProceduralMemoryMaintenanceScheduleConfig, ProceduralMemoryProviderConfig, ProceduralMemoryRetrievalConfig, ProceduralMemoryRuntimeConfig, } from "./knowledge/procedural/index.js";

package/dist/index.js

@@ -1,4 +1,4 @@
-export { AgentHarnessAcpServer, AgentHarnessRuntime, cancelRequest, createAgentHarness, createAcpServer, createAcpStdioClient, createRuntimeMcpServer, createUpstreamTimelineReducer, createToolMcpServer, deleteSession, describeInventory, exportEvaluationBundle, exportFlow, exportSequence, exportRequestPackage, exportSessionPackage, replayEvaluationBundle, getArtifact, getAgent, getApproval, getOperatorOverview, getRequestPlanState, getRequest, getHealth, listMemories, listRequestTraceItems, getSession, listAgentSkills, listRequestArtifacts, listApprovals, listRequests, listRequestEvents, listSessionSummaries, listSessions, memorize, normalizeUserChatInput, recordArtifact, request, recall, removeMemory, resolveApproval, serveA2aHttp, serveAcpHttp, serveAcpStdio, serveAgUiHttp, serveRuntimeMcpOverStdio, serveRuntimeMcpOverStreamableHttp, serveToolsOverStdio, subscribe, stop, updateMemory, } from "./api.js";
+export { AgentHarnessAcpServer, AgentHarnessRuntime, cancelRequest, createAgentHarness, createAcpServer, createAcpStdioClient, createRuntimeMcpServer, createUpstreamTimelineReducer, createToolMcpServer, analyzeBoundaries, deleteSession, describeInventory, exportEvaluationBundle, exportFlow, exportSequence, exportRequestPackage, exportSessionPackage, replayEvaluationBundle, getArtifact, getAgent, getApproval, getOperatorOverview, getRequestPlanState, getRequest, getHealth, listMemories, listRequestTraceItems, getSession, listAgentSkills, listRequestArtifacts, listApprovals, listRequests, listRequestEvents, listSessionSummaries, listSessions, memorize, normalizeUserChatInput, recordArtifact, request, recall, removeMemory, resolveApproval, serveA2aHttp, serveAcpHttp, serveAcpStdio, serveAgUiHttp, serveRuntimeMcpOverStdio, serveRuntimeMcpOverStreamableHttp, serveToolsOverStdio, subscribe, stop, updateMemory, } from "./api.js";
 export { AcpHarnessClient, InProcessHarnessClient, createAcpHarnessClient, createAcpHttpHarnessClient, createAcpStdioHarnessClient, createAgentHarnessClient, createInProcessHarnessClient, } from "./client.js";
 export { createKnowledgeModule, readKnowledgeRuntimeConfig } from "./knowledge/index.js";
 export { readProceduralMemoryRuntimeConfig } from "./knowledge/procedural/index.js";

package/dist/package-version.d.ts

@@ -1,2 +1,2 @@
-export declare const AGENT_HARNESS_VERSION = "0.0.351";
+export declare const AGENT_HARNESS_VERSION = "0.0.353";
 export declare const AGENT_HARNESS_RELEASE_DATE = "2026-04-24";

package/dist/package-version.js

@@ -1,2 +1,2 @@
-export const AGENT_HARNESS_VERSION = "0.0.351";
+export const AGENT_HARNESS_VERSION = "0.0.353";
 export const AGENT_HARNESS_RELEASE_DATE = "2026-04-24";

package/dist/resources/prompts/runtime/execution-with-tool-evidence-retry.md

@@ -1 +1,5 @@
-Your previous response was rejected because it claimed execution without any tool calls. Your next response must contain real tool calls only. If this is a multi-step task and the runtime todo board does not exist yet, call write_todos first. If the todo board already exists or prior tool evidence is already available, do not restart planning or ask the user for more details; continue with the next concrete tool call instead. Do not describe completed work until the tool calls have actually run.
+Your previous response was rejected because it did not produce concrete execution evidence from the tools configured for this agent. Your next response must contain real execution tool calls only.
+
+Do not call planning-only tools such as write_todos or read_todos in this retry. Do not restart the plan, ask for more details, or describe completed work. Select one configured non-planning tool that can advance the original request, call it with concrete arguments, and then continue from that tool result.
+
+If this agent has no configured non-planning tool that can advance the original request, return a blocker instead of pretending the work is complete.

package/dist/runtime/adapter/flow/invocation-flow.js

@@ -5,6 +5,8 @@ import { invokeRuntimeWithLocalTools } from "./invoke-runtime.js";
 import { buildInvocationRequest } from "../model/invocation-request.js";
 import { UPSTREAM_REQUEST_CONFIG_KEY, UPSTREAM_SESSION_CONFIG_KEY } from "../upstream-configurable-keys.js";
 import { extractVisibleOutput, tryParseJson } from "../../parsing/output-parsing.js";
+import { salvageJsonToolCalls } from "../../parsing/output-tool-args.js";
+import { isEmptyFinalAiMessageError } from "../resilience.js";
 function readBindingExecutionParams(binding) {
     const params = binding.execution?.params ?? binding.deepAgentParams ?? binding.langchainAgentParams;
     return {
@@ -48,6 +50,34 @@ function hasNativeTaskDelegationIntent(value) {
     return hasNativeTaskDelegationIntent(typed.tool_calls) || hasNativeTaskDelegationIntent(typed.messages);
 }
 function readStructuredToolCall(value) {
+    const salvaged = salvageJsonToolCalls(value)[0];
+    if (salvaged) {
+        return salvaged;
+    }
+    if (Array.isArray(value)) {
+        for (const item of value) {
+            const nested = readStructuredToolCall(item);
+            if (nested) {
+                return nested;
+            }
+        }
+        return null;
+    }
+    if (typeof value === "object" && value !== null) {
+        const typed = value;
+        const fromOutput = typed.output !== undefined ? readStructuredToolCall(typed.output) : null;
+        if (fromOutput) {
+            return fromOutput;
+        }
+        const fromContent = typed.content !== undefined ? readStructuredToolCall(typed.content) : null;
+        if (fromContent) {
+            return fromContent;
+        }
+        const fromMessages = typed.messages !== undefined ? readStructuredToolCall(typed.messages) : null;
+        if (fromMessages) {
+            return fromMessages;
+        }
+    }
     const text = typeof value === "string" ? value.trim() : "";
     const parsed = text ? (tryParseJson(text) ?? extractFirstJsonObject(text)) : value;
     if (typeof parsed !== "object" || parsed === null || Array.isArray(parsed)) {
@@ -116,7 +146,7 @@ async function replayStructuredTaskToolCall(input) {
         return input.invocation;
     }
     const visibleOutput = extractVisibleOutput(input.invocation.result);
-    const toolCall = readStructuredToolCall(visibleOutput);
+    const toolCall = readStructuredToolCall(visibleOutput) ?? readStructuredToolCall(input.invocation.result);
     if (toolCall?.name !== "task") {
         return input.invocation;
     }
@@ -179,6 +209,20 @@ function buildDelegationOnlyRecoveryInstruction(binding, input) {
         JSON.stringify(input),
     ].join("\n");
 }
+function buildEmptyAssistantRecoveryInstruction() {
+    return [
+        "/no_think",
+        "Runtime correction: your previous assistant response was empty.",
+        "Continue the original request now.",
+        "If an available tool is needed, call exactly one valid tool with complete arguments.",
+        "If no tool is needed, return a non-empty final answer.",
+        "Do not return empty content.",
+    ].join("\n");
+}
+function isEmptyAssistantOutputError(error) {
+    const message = error instanceof Error ? error.message : String(error);
+    return isEmptyFinalAiMessageError(error) || message === "empty_final_output";
+}
 function appendUserRecoveryInstruction(input, instruction) {
     const content = [
         "Runtime correction:",
@@ -246,12 +290,39 @@ export async function executeRequestInvocation(options) {
     if (!result) {
         throw new Error("Agent invocation returned no result");
     }
-    return finalizeRequestResult({
-        bindingAgentId: options.binding.agent.id,
-        sessionId: options.sessionId,
-        requestId: options.requestId,
-        result,
-        executedToolResults,
-    });
+    try {
+        return finalizeRequestResult({
+            bindingAgentId: options.binding.agent.id,
+            sessionId: options.sessionId,
+            requestId: options.requestId,
+            binding: options.binding,
+            result,
+            executedToolResults,
+        });
+    }
+    catch (error) {
+        if (options.resumePayload !== undefined || !isEmptyAssistantOutputError(error)) {
+            throw error;
+        }
+        const shouldUseDelegationRecovery = isDelegationOnlyBinding(options.binding)
+            && !hasTaskDelegationEvidence(executedToolResults)
+            && !hasNativeTaskDelegationIntent(result);
+        const recoveredRequest = appendUserRecoveryInstruction(request, shouldUseDelegationRecovery
+            ? buildDelegationOnlyRecoveryInstruction(options.binding, options.input)
+            : buildEmptyAssistantRecoveryInstruction());
+        const recoveredInvocation = await replayStructuredTaskToolCall({
+            invocation: await invokeOnce(recoveredRequest),
+            builtinExecutableTools: builtinExecutableTools,
+            toolRuntimeContext: invokeOptions.toolRuntimeContext,
+        });
+        return finalizeRequestResult({
+            bindingAgentId: options.binding.agent.id,
+            sessionId: options.sessionId,
+            requestId: options.requestId,
+            binding: options.binding,
+            result: recoveredInvocation.result,
+            executedToolResults: recoveredInvocation.executedToolResults,
+        });
+    }
 }
 export const executeRuntimeInvocation = executeRequestInvocation;

package/dist/runtime/adapter/invocation-result.d.ts

@@ -16,6 +16,13 @@ export declare function finalizeRequestResult(params: {
     bindingAgentId: string;
     sessionId: string;
     requestId: string;
+    binding?: {
+        harnessRuntime?: {
+            executionContract?: {
+                requiresPlan?: boolean;
+            };
+        };
+    };
     result: Record<string, unknown>;
     executedToolResults: ExecutedToolResult[];
 }): RequestResult;

package/dist/runtime/adapter/invocation-result.js

@@ -1,5 +1,5 @@
 import { containsLikelySkillDocument, extractContentBlocks, extractEmptyAssistantMessageFailure, extractOutputContent, extractToolFallbackContext, extractVisibleOutput, isLikelyToolArgsObject, sanitizeVisibleText, tryParseJson, } from "../parsing/output-parsing.js";
-import { salvageFunctionLikeToolCall } from "../parsing/output-tool-args.js";
+import { salvageFunctionLikeToolCall, salvageJsonToolCalls } from "../parsing/output-tool-args.js";
 import { buildStateSnapshot } from "./model/message-assembly.js";
 import { asRecord } from "./tool/resolved-tool.js";
 import { renderToolFailure } from "../support/harness-support.js";
@@ -12,9 +12,30 @@ function looksLikeLeakedToolCallText(value) {
     if (salvageFunctionLikeToolCall(normalized)) {
         return true;
     }
+    if (salvageJsonToolCalls(normalized).length > 0) {
+        return true;
+    }
     const prefixedToolCallMatch = /^(?:\s*(?:Ready|Understood|Okay|Ok|Got it|Sure|All set|What is your request|Please provide a task for me to orchestrate)[.:?!]?\s*)+([A-Za-z_][A-Za-z0-9_]*\([\s\S]*\))\s*$/u.exec(normalized);
     return !!(prefixedToolCallMatch && salvageFunctionLikeToolCall(prefixedToolCallMatch[1]));
 }
+function hasIncompleteStateSnapshotPlan(stateSnapshot) {
+    if (typeof stateSnapshot !== "object" || stateSnapshot === null) {
+        return false;
+    }
+    const todos = stateSnapshot.todos;
+    if (!Array.isArray(todos)) {
+        return false;
+    }
+    return todos.some((todo) => {
+        if (typeof todo !== "object" || todo === null) {
+            return false;
+        }
+        const status = typeof todo.status === "string"
+            ? todo.status.trim().toLowerCase()
+            : "";
+        return status === "pending" || status === "in_progress";
+    });
+}
 function isPlaceholderTaskCompletion(value) {
     const normalized = sanitizeVisibleText(value).trim();
     return normalized === "Task completed";
@@ -113,6 +134,62 @@ function extractLatestSuccessfulNonTodoToolResultText(executedToolResults) {
         ?? candidates.at(-1)
         ?? "";
 }
+function readSerializedMessageType(value) {
+    if (typeof value !== "object" || value === null) {
+        return "";
+    }
+    const id = value.id;
+    if (!Array.isArray(id)) {
+        return "";
+    }
+    return id.map((item) => typeof item === "string" ? item : "").filter(Boolean).join(".");
+}
+function readToolMessageRecord(value) {
+    if (typeof value !== "object" || value === null) {
+        return null;
+    }
+    const typed = value;
+    const messageType = readSerializedMessageType(value);
+    const kwargs = typeof typed.kwargs === "object" && typed.kwargs !== null ? typed.kwargs : undefined;
+    const isToolMessage = typed.role === "tool"
+        || typed.type === "tool"
+        || messageType.endsWith("ToolMessage")
+        || kwargs?.name !== undefined && kwargs?.content !== undefined;
+    if (!isToolMessage) {
+        return null;
+    }
+    const toolNameCandidate = kwargs?.name ?? typed.name;
+    const toolName = typeof toolNameCandidate === "string" ? toolNameCandidate.trim() : "";
+    if (!toolName) {
+        return null;
+    }
+    const output = kwargs?.content ?? typed.content ?? "";
+    const status = kwargs?.status ?? typed.status;
+    const outputText = typeof output === "string" ? output : extractVisibleOutput(output);
+    return {
+        toolName,
+        output,
+        ...(status === "error" || looksLikeToolBlocker(outputText) ? { isError: true } : {}),
+    };
+}
+function extractUpstreamToolResults(value, seen = new Set()) {
+    if (typeof value !== "object" || value === null || seen.has(value)) {
+        return [];
+    }
+    seen.add(value);
+    if (Array.isArray(value)) {
+        return value.flatMap((item) => extractUpstreamToolResults(item, seen));
+    }
+    const direct = readToolMessageRecord(value);
+    if (direct) {
+        return [direct];
+    }
+    const typed = value;
+    return [
+        ...extractUpstreamToolResults(typed.messages, seen),
+        ...extractUpstreamToolResults(typed.output, seen),
+    ];
+}
 function hasDelegationBlocker(executedToolResults) {
     return executedToolResults.some((toolResult) => {
         if (toolResult.toolName !== "task") {
@@ -236,7 +313,11 @@ export function extractToolResultFindingsText(executedToolResults) {
     return extractLatestSuccessfulNonTodoToolResultText(executedToolResults);
 }
 export function finalizeRequestResult(params) {
-    const { bindingAgentId, sessionId, requestId, result, executedToolResults } = params;
+    const { bindingAgentId, sessionId, requestId, binding, result, executedToolResults } = params;
+    const allExecutedToolResults = [
+        ...executedToolResults,
+        ...extractUpstreamToolResults(result),
+    ];
     const interruptContent = Array.isArray(result.__interrupt__) && result.__interrupt__.length > 0 ? JSON.stringify(result.__interrupt__) : undefined;
     const extractedOutput = extractVisibleOutput(result);
     const visibleOutput = extractedOutput && !isLikelyToolArgsObject(tryParseJson(extractedOutput)) ? extractedOutput : "";
@@ -257,7 +338,7 @@ export function finalizeRequestResult(params) {
         && contentBlocks.length === 0
         && structuredResponse === undefined
         && !files
-        && executedToolResults.length === 0
+        && allExecutedToolResults.length === 0
         && hasEmptyFinalMessage(result)
         && !hasFinalMessageToolCalls(result)) {
         throw new Error("empty_final_output");
@@ -266,20 +347,27 @@ export function finalizeRequestResult(params) {
     const output = resolveDeterministicFinalOutput({
         visibleOutput,
         toolFallback,
-        executedToolResults,
+        executedToolResults: allExecutedToolResults,
     })
         || (containsLikelySkillDocument(result) ? "" : serializedResult);
     const finalMessageText = sanitizeVisibleText(output);
     const terminalStatus = structuredTerminalStatus ?? readTerminalExecutionStatus(finalMessageText);
     const stateSnapshot = buildStateSnapshot(result);
-    const memoryCandidates = executedToolResults.flatMap((toolResult) => toolResult.memoryCandidates ?? []);
+    const hasIncompleteRequiredPlan = binding?.harnessRuntime?.executionContract?.requiresPlan === true
+        && hasIncompleteStateSnapshotPlan(stateSnapshot);
+    const hasTerminalToolBlocker = looksLikeToolBlocker(finalMessageText);
+    const memoryCandidates = allExecutedToolResults.flatMap((toolResult) => toolResult.memoryCandidates ?? []);
     return {
         sessionId,
         requestId,
         agentId: bindingAgentId,
         state: Array.isArray(result.__interrupt__) && result.__interrupt__.length > 0
             ? "waiting_for_approval"
-            : mapTerminalStatusToRequestState(terminalStatus),
+            : hasIncompleteRequiredPlan
+                ? "failed"
+                : hasTerminalToolBlocker
+                    ? "failed"
+                    : mapTerminalStatusToRequestState(terminalStatus),
         interruptContent,
         output: finalMessageText,
         finalMessageText,
@@ -287,7 +375,7 @@ export function finalizeRequestResult(params) {
         ...(contentBlocks.length > 0 ? { contentBlocks } : {}),
         ...(structuredResponse !== undefined ? { structuredResponse } : {}),
         metadata: {
-            ...(executedToolResults.length > 0 ? { executedToolResults } : {}),
+            ...(allExecutedToolResults.length > 0 ? { executedToolResults: allExecutedToolResults } : {}),
             ...(memoryCandidates.length > 0 ? { memoryCandidates } : {}),
             ...(structuredResponse !== undefined ? { structuredResponse } : {}),
             ...(terminalStatus ? { terminalStatus } : {}),

package/dist/runtime/adapter/local-tool-invocation.js

@@ -4,7 +4,8 @@ import { canReplayToolCallsLocally } from "./tool/tool-replay.js";
 import { extractToolCallsFromResult, normalizeToolArgsForSchema, stringifyToolOutput } from "./tool/tool-arguments.js";
 import { extractMemoryCandidatesFromToolOutput } from "../harness/system/runtime-memory-candidates.js";
 import { maybePersistLargeToolOutput } from "./tool/tool-output-artifacts.js";
-import { appendToolRecoveryInstruction, extractVisibleOutput, resolveMissingPlanRecoveryInstruction, resolveExecutionWithoutToolEvidenceTextInstruction, sanitizeVisibleText, } from "../parsing/output-parsing.js";
+import { appendToolRecoveryInstruction, extractVisibleOutput, resolveMissingPlanRecoveryInstruction, resolveExecutionWithoutToolEvidenceTextInstruction, resolveToolCallRecoveryInstruction, sanitizeVisibleText, STRICT_TOOL_JSON_INSTRUCTION, } from "../parsing/output-parsing.js";
+import { salvageJsonToolCalls } from "../parsing/output-tool-args.js";
 import { AUTONOMOUS_INVESTIGATION_RECOVERY_INSTRUCTION } from "../prompts/runtime-prompts.js";
 const TOOL_FOLLOW_UP_INSTRUCTION = "One or more tool results are already available in this conversation. Answer the user's current request directly from the existing context and tool results. Do not ask the user to repeat inputs that are already present above.";
 function readPlanStateSummary(output) {
@@ -43,6 +44,17 @@ function hasNonTodoToolEvidence(executedToolResults) {
 function hasPlanStateEvidence(executedToolResults) {
     return executedToolResults.some((item) => item.toolName === "write_todos" || item.toolName === "read_todos" || readPlanStateSummary(item.output) !== null);
 }
+function latestToolErrorRecoveryInstruction(executedToolResults) {
+    const latest = executedToolResults.at(-1);
+    if (!latest || latest.isError !== true) {
+        return null;
+    }
+    const message = typeof latest.output === "string" ? latest.output : JSON.stringify(latest.output);
+    return resolveToolCallRecoveryInstruction(new Error(message)) ?? AUTONOMOUS_INVESTIGATION_RECOVERY_INSTRUCTION;
+}
+function terminalToolErrorRecoveryInstruction(terminalText) {
+    return resolveToolCallRecoveryInstruction(new Error(terminalText));
+}
 function requiresPlanEvidence(binding) {
     return binding.harnessRuntime.executionContract?.requiresPlan === true;
 }
@@ -83,18 +95,24 @@ export async function runLocalToolInvocationLoop({ binding, request, primaryTool
         if (toolCalls.length === 0) {
             const terminalText = sanitizeVisibleText(extractVisibleOutput(result) || "");
             const hasIncompletePlanState = hasIncompleteExecutedPlan(executedToolResults);
+            const shouldEnforceIncompletePlan = requiresPlanEvidence(binding) && hasIncompletePlanState;
             const hasExecutionBeyondTodoPlanning = hasNonTodoToolEvidence(executedToolResults);
-            const recoveryInstruction = terminalText
+            const toolErrorRecoveryInstruction = latestToolErrorRecoveryInstruction(executedToolResults)
+                ?? terminalToolErrorRecoveryInstruction(terminalText);
+            const leakedJsonToolCallRecoveryInstruction = terminalText && salvageJsonToolCalls(terminalText).length > 0
+                ? STRICT_TOOL_JSON_INSTRUCTION
+                : null;
+            const recoveryInstruction = toolErrorRecoveryInstruction ?? leakedJsonToolCallRecoveryInstruction ?? (terminalText
                 ? resolveExecutionWithoutToolEvidenceTextInstruction(activeRequest, terminalText, false, {
                     hasWriteTodosEvidence: executedToolResults.some((item) => item.toolName === "write_todos"),
                     hasToolResultEvidence: hasExecutionBeyondTodoPlanning,
                     hasPlanStateEvidence: hasPlanStateEvidence(executedToolResults),
-                    hasIncompletePlanState: hasExecutionBeyondTodoPlanning && hasIncompletePlanState,
+                    hasIncompletePlanState: shouldEnforceIncompletePlan,
                     requiresPlan: requiresPlanEvidence(binding),
                 })
-                : hasIncompletePlanState && hasExecutionBeyondTodoPlanning
+                : shouldEnforceIncompletePlan
                     ? AUTONOMOUS_INVESTIGATION_RECOVERY_INSTRUCTION
-                    : null;
+                    : null);
             if (recoveryInstruction) {
                 if (iteration + 1 === maxToolIterations) {
                     throw new Error(`Tool-calling loop exceeded the maximum of ${maxToolIterations} iterations`);

package/dist/runtime/adapter/middleware-assembly.js

@@ -12,6 +12,32 @@ import { materializeDeepAgentSkillSourcePaths } from "./compat/deepagent-compat.
 import { DEFAULT_SUBAGENT_PROMPT, EXECUTION_WITH_TOOL_EVIDENCE_RETRY_INSTRUCTION, } from "../prompts/runtime-prompts.js";
 import { createContextHygieneMiddleware } from "./middleware/context-hygiene.js";
 const INVALID_TOOL_MESSAGE_BLOCK_TYPES = new Set(["tool_use", "thinking", "redacted_thinking"]);
+const DEFAULT_BUILTIN_TASK_TIMEOUT_MS = 180_000;
+function resolveBuiltinTaskTimeoutMs(model) {
+    const timeout = model?.init?.timeout;
+    return typeof timeout === "number" && Number.isFinite(timeout) && timeout > 0
+        ? Math.floor(timeout)
+        : DEFAULT_BUILTIN_TASK_TIMEOUT_MS;
+}
+async function withBuiltinTaskTimeout(producer, timeoutMs, subagentName) {
+    let timeoutHandle;
+    try {
+        return await Promise.race([
+            producer(),
+            new Promise((_, reject) => {
+                timeoutHandle = setTimeout(() => {
+                    reject(new Error(`Delegated agent ${subagentName} timed out after ${timeoutMs}ms.`));
+                }, timeoutMs);
+                timeoutHandle.unref?.();
+            }),
+        ]);
+    }
+    finally {
+        if (timeoutHandle) {
+            clearTimeout(timeoutHandle);
+        }
+    }
+}
 function extractDeepAgentTaskContent(result) {
     if (typeof result !== "object" || result === null) {
         return undefined;
@@ -369,9 +395,10 @@ export async function invokeBuiltinTaskTool(input) {
         configurable: { [UPSTREAM_SESSION_CONFIG_KEY]: `${input.binding.agent.id}:builtin-task` },
         ...(input.options?.context ? { context: input.options.context } : {}),
     };
-    const invokeSubagent = (content) => runnable.invoke({
+    const taskTimeoutMs = resolveBuiltinTaskTimeoutMs(selectedCompiledSubagent?.model ?? primaryModel);
+    const invokeSubagent = (content) => withBuiltinTaskTimeout(() => runnable.invoke({
         messages: [new HumanMessage({ content })],
-    }, invokeConfig);
+    }, invokeConfig), taskTimeoutMs, selectedSubagent.name);
     let result = await invokeSubagent(description);
     if (!hasSubagentExecutionToolEvidence(result, resolvedSubagentTools, selectedCompiledSubagent?.tools)) {
         result = await invokeSubagent([description, EXECUTION_WITH_TOOL_EVIDENCE_RETRY_INSTRUCTION].filter(Boolean).join("\n\n"));

package/dist/runtime/adapter/resilience.d.ts

@@ -8,5 +8,6 @@ export type ProviderRetryPolicy = {
     backoffMs: number;
     retryableMessages: string[];
 };
+export declare function isEmptyFinalAiMessageError(error: unknown): boolean;
 export declare function resolveProviderRetryPolicy(binding: CompiledAgentBinding): ProviderRetryPolicy;
 export declare function isRetryableProviderError(binding: CompiledAgentBinding, error: unknown): boolean;

package/dist/runtime/adapter/resilience.js

@@ -31,13 +31,14 @@ export function resolveStreamIdleTimeout(binding) {
     return 60_000;
 }
 const BUILTIN_RETRYABLE_PROVIDER_MESSAGES = [
+    "eof",
     "unexpected eof",
     "other side closed",
     "socket hang up",
     "connection reset",
     "econnreset",
 ];
-function isEmptyFinalAiMessageError(error) {
+export function isEmptyFinalAiMessageError(error) {
     const message = error instanceof Error ? error.message : String(error);
     return message.toLowerCase().startsWith("empty_final_ai_message:");
 }

package/dist/runtime/adapter/terminal-status.js

@@ -11,10 +11,10 @@ function normalizeTerminalStatus(value) {
 function readStatusLine(value) {
     for (const line of value.split("\n")) {
         const [key, ...rest] = line.split(":");
-        if (key?.trim().toLowerCase() !== "status") {
+        if (key?.trim().replaceAll("*", "").toLowerCase() !== "status") {
             continue;
         }
-        const statusValue = rest.join(":").trim().split(/\s+/)[0];
+        const statusValue = rest.join(":").trim().replaceAll("*", "").split(/\s+/)[0];
         const status = normalizeTerminalStatus(statusValue);
         if (status) {
             return status;