@botbotgo/agent-harness 0.0.365 → 0.0.367

package/dist/package-version.d.ts

@@ -1,2 +1,2 @@
-export declare const AGENT_HARNESS_VERSION = "0.0.365";
-export declare const AGENT_HARNESS_RELEASE_DATE = "2026-04-28";
+export declare const AGENT_HARNESS_VERSION = "0.0.367";
+export declare const AGENT_HARNESS_RELEASE_DATE = "2026-04-30";

package/dist/package-version.js

@@ -1,2 +1,2 @@
-export const AGENT_HARNESS_VERSION = "0.0.365";
-export const AGENT_HARNESS_RELEASE_DATE = "2026-04-28";
+export const AGENT_HARNESS_VERSION = "0.0.367";
+export const AGENT_HARNESS_RELEASE_DATE = "2026-04-30";

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

@@ -8,6 +8,19 @@ import { appendToolRecoveryInstruction, extractVisibleOutput, tryParseJson } fro
 import { salvageJsonToolCalls } from "../../parsing/output-tool-args.js";
 import { isEmptyFinalAiMessageError } from "../resilience.js";
 import { AUTONOMOUS_INVESTIGATION_RECOVERY_INSTRUCTION, EXECUTION_WITH_TOOL_EVIDENCE_RETRY_INSTRUCTION, } from "../../prompts/runtime-prompts.js";
+const CLOSE_REQUIRED_PLAN_RECOVERY_INSTRUCTION = [
+    "The current required todo board still has unfinished work.",
+    "Do not broaden the investigation, restart planning, or ask the user what to do next.",
+    "Use the existing tool evidence already available in this run.",
+    "Your next action must be write_todos: update every remaining pending or in_progress item to completed if evidence was gathered, or failed if it cannot be completed with the available tools.",
+    "After that write_todos call, provide the final answer required by the agent response format.",
+].join("\n");
+const INITIAL_REQUIRED_PLAN_INSTRUCTION = [
+    "This agent has a required visible planning contract.",
+    "Your first action for this request must be write_todos with concrete task steps and statuses.",
+    "Do not call any domain/evidence tool and do not provide a final answer before the initial write_todos call succeeds.",
+    "After each evidence step, update the todo board. Before the final answer, close every todo as completed or failed.",
+].join("\n");
 function readBindingExecutionParams(binding) {
     const params = binding.execution?.params ?? binding.deepAgentParams ?? binding.langchainAgentParams;
     return {
@@ -280,9 +293,13 @@ function appendUserRecoveryInstruction(input, instruction) {
 export async function executeRequestInvocation(options) {
     const history = options.history ?? [];
     const invokeOptions = options.invokeOptions ?? {};
-    const request = options.resumePayload === undefined
+    let request = options.resumePayload === undefined
         ? buildInvocationRequest(options.binding, history, options.input, invokeOptions)
         : new Command({ resume: options.resumePayload });
+    if (options.resumePayload === undefined
+        && options.binding.harnessRuntime.executionContract?.requiresPlan === true) {
+        request = appendToolRecoveryInstruction(request, INITIAL_REQUIRED_PLAN_INSTRUCTION);
+    }
     const { primaryTools, toolNameMapping, executableTools, defersToUpstreamHitlExecution, } = buildBindingToolExecutionContext({
         binding: options.binding,
         resolveTools: options.resolveTools,
@@ -335,7 +352,7 @@ export async function executeRequestInvocation(options) {
             ? result.messages
             : undefined;
         const recoveryBase = messages ? { messages } : request;
-        const recoveredRequest = appendToolRecoveryInstruction(recoveryBase, AUTONOMOUS_INVESTIGATION_RECOVERY_INSTRUCTION);
+        const recoveredRequest = appendToolRecoveryInstruction(recoveryBase, CLOSE_REQUIRED_PLAN_RECOVERY_INSTRUCTION);
         const recoveredInvocation = await invokeOnce(recoveredRequest);
         localOrUpstreamInvocation = recoveredInvocation;
         result = recoveredInvocation.result;
@@ -343,8 +360,7 @@ export async function executeRequestInvocation(options) {
     }
     if (options.resumePayload === undefined
         && options.binding.harnessRuntime.executionContract?.requiresPlan === true
-        && hasIncompleteUpstreamPlan(result)
-        && !extractVisibleOutput(result).trim()) {
+        && hasIncompleteUpstreamPlan(result)) {
         const messages = Array.isArray(result.messages)
             ? result.messages
             : undefined;

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

@@ -164,6 +164,11 @@ function hasDelegatedPlanEvidence(result) {
     return Array.isArray(toolResults)
         && toolResults.some((item) => item.toolName === "write_todos" || item.toolName === "read_todos");
 }
+const DELEGATED_PLAN_EVIDENCE_RETRY_INSTRUCTION = [
+    "The delegated task requires visible TODO planning evidence.",
+    "Before any other tool call or final answer, call write_todos with concrete task steps and statuses.",
+    "Then continue the task to completion, update TODO statuses after evidence steps, and close every TODO as completed or failed before the final answer.",
+].join("\n");
 function resolveDelegatedResultOutput(result) {
     const executedToolResults = Array.isArray(result.metadata?.executedToolResults)
         ? result.metadata.executedToolResults
@@ -936,21 +941,62 @@ export class AgentRuntimeAdapter {
             delegatedResult = await runDelegatedRequest(requestText);
         }
         catch (error) {
-            const output = error instanceof Error ? error.message : String(error);
-            return {
-                toolOutput: output,
-                delegatedSubagentType: subagentType,
-                delegatedResult: {
-                    sessionId,
-                    requestId,
-                    agentId: selectedBinding.agent.id,
-                    state: "failed",
-                    output,
-                    finalMessageText: output,
-                },
-            };
+            if (selectedBinding.harnessRuntime.executionContract?.requiresPlan === true) {
+                try {
+                    delegatedResult = await runDelegatedRequest([requestText, DELEGATED_PLAN_EVIDENCE_RETRY_INSTRUCTION].filter(Boolean).join("\n\n"), ":plan-evidence-retry");
+                }
+                catch (recoveryError) {
+                    const output = recoveryError instanceof Error ? recoveryError.message : String(recoveryError);
+                    return {
+                        toolOutput: output,
+                        delegatedSubagentType: subagentType,
+                        delegatedResult: {
+                            sessionId,
+                            requestId,
+                            agentId: selectedBinding.agent.id,
+                            state: "failed",
+                            output,
+                            finalMessageText: output,
+                        },
+                    };
+                }
+            }
+            else {
+                const output = error instanceof Error ? error.message : String(error);
+                return {
+                    toolOutput: output,
+                    delegatedSubagentType: subagentType,
+                    delegatedResult: {
+                        sessionId,
+                        requestId,
+                        agentId: selectedBinding.agent.id,
+                        state: "failed",
+                        output,
+                        finalMessageText: output,
+                    },
+                };
+            }
         }
         const targetRequiresExecutionToolEvidence = getBindingPrimaryTools(selectedBinding).length > 0;
+        if (selectedBinding.harnessRuntime.executionContract?.requiresPlan === true
+            && !hasDelegatedPlanEvidence(delegatedResult)) {
+            try {
+                delegatedResult = await runDelegatedRequest([requestText, DELEGATED_PLAN_EVIDENCE_RETRY_INSTRUCTION].filter(Boolean).join("\n\n"), ":plan-evidence-retry");
+            }
+            catch (error) {
+                const output = error instanceof Error ? error.message : String(error);
+                return {
+                    toolOutput: output,
+                    delegatedSubagentType: subagentType,
+                    delegatedResult: {
+                        ...delegatedResult,
+                        state: "failed",
+                        output,
+                        finalMessageText: output,
+                    },
+                };
+            }
+        }
         if (targetRequiresExecutionToolEvidence && !hasDelegatedExecutionToolEvidence(delegatedResult)) {
             try {
                 delegatedResult = await runDelegatedRequest([requestText, EXECUTION_WITH_TOOL_EVIDENCE_RETRY_INSTRUCTION].filter(Boolean).join("\n\n"), ":tool-evidence-retry");
@@ -969,8 +1015,9 @@ export class AgentRuntimeAdapter {
                 };
             }
         }
-        if (targetRequiresExecutionToolEvidence && !hasDelegatedExecutionToolEvidence(delegatedResult)) {
-            const output = `runtime_error=Delegated agent ${selectedBinding.agent.id} completed without tool execution evidence.`;
+        if (selectedBinding.harnessRuntime.executionContract?.requiresPlan === true
+            && !hasDelegatedPlanEvidence(delegatedResult)) {
+            const output = "runtime_error=Delegated agent ended before producing required plan evidence.";
             return {
                 toolOutput: output,
                 delegatedSubagentType: subagentType,
@@ -982,10 +1029,8 @@ export class AgentRuntimeAdapter {
                 },
             };
         }
-        if (selectedBinding.harnessRuntime.executionContract?.requiresPlan === true
-            && !hasDelegatedPlanEvidence(delegatedResult)
-            && !hasDelegatedExecutionToolEvidence(delegatedResult)) {
-            const output = "runtime_error=Delegated agent ended before producing required plan evidence.";
+        if (targetRequiresExecutionToolEvidence && !hasDelegatedExecutionToolEvidence(delegatedResult)) {
+            const output = `runtime_error=Delegated agent ${selectedBinding.agent.id} completed without tool execution evidence.`;
             return {
                 toolOutput: output,
                 delegatedSubagentType: subagentType,
@@ -1008,12 +1053,66 @@ export class AgentRuntimeAdapter {
         const delegatedOutput = typeof compactDelegation.toolOutput === "string"
             ? [compactDelegation.toolOutput]
             : [];
+        const delegatedPayload = (() => {
+            if (delegatedOutput.length === 0) {
+                return undefined;
+            }
+            try {
+                const parsed = JSON.parse(delegatedOutput[0]);
+                return parsed && typeof parsed === "object" && !Array.isArray(parsed)
+                    ? parsed
+                    : undefined;
+            }
+            catch {
+                return undefined;
+            }
+        })();
+        const payloadStringArray = (key) => {
+            const value = delegatedPayload?.[key];
+            if (!Array.isArray(value)) {
+                return undefined;
+            }
+            const items = value.filter((item) => typeof item === "string" && item.trim().length > 0);
+            return items.length > 0 ? items : undefined;
+        };
+        const payloadReport = typeof delegatedPayload?.report === "string" && delegatedPayload.report.trim().length > 0
+            ? delegatedPayload.report
+            : undefined;
         const delegatedToolNames = Array.isArray(compactDelegation.delegatedResult?.metadata?.executedToolResults)
             ? compactDelegation.delegatedResult.metadata.executedToolResults
                 .filter((toolResult) => toolResult?.toolName)
                 .map((toolResult) => toolResult.toolName)
             : [];
         const state = compactDelegation.delegatedResult?.state === "failed" ? "failed" : "completed";
+        const uniqueToolNames = [...new Set(delegatedToolNames)];
+        const toolEvidence = uniqueToolNames.length > 0 ? uniqueToolNames.join(", ") : "none";
+        const fallbackTodoTrace = [
+            `1) TODO observed: delegated to ${delegatedSubagentType}.`,
+            uniqueToolNames.includes("write_todos")
+                ? "2) TODO evidence: delegated specialist invoked write_todos."
+                : "2) TODO evidence missing: delegated specialist did not expose write_todos in returned metadata.",
+            state === "failed"
+                ? "3) TODO closed: delegated execution failed; blocker reported."
+                : "3) TODO closed: delegated execution completed; synthesis returned.",
+        ];
+        const fallbackStepResults = [
+            `1) Delegation step: task invoked ${delegatedSubagentType}.`,
+            `2) Evidence step: delegated tool evidence = ${toolEvidence}.`,
+            state === "failed"
+                ? "3) Synthesis step: returned blocker report because delegated execution failed."
+                : "3) Synthesis step: compact delegation report assembled from delegated output.",
+        ];
+        const fallbackSummary = [
+            state === "failed"
+                ? `子代理 ${delegatedSubagentType} 委托执行失败。`
+                : `已完成子代理 ${delegatedSubagentType} 委托执行。`,
+        ];
+        const fallbackFindings = payloadReport
+            ? ["子代理返回了结构化报告，详见 report。"]
+            : delegatedOutput.length > 0
+                ? delegatedOutput.slice(0, 3)
+                : ["none"];
+        const report = payloadReport ?? (delegatedOutput.join("\n") || "委托已完成，未返回附加报告。");
         return {
             status: state,
             routing: [
@@ -1026,18 +1125,18 @@ export class AgentRuntimeAdapter {
             ],
             execution: [
                 `1) 调用 task 工具，目标子代理：${delegatedSubagentType}`,
-                delegatedToolNames.length > 0
-                    ? `2) 子代理返回工具证据：${[...new Set(delegatedToolNames)].join(", ")}`
+                uniqueToolNames.length > 0
+                    ? `2) 子代理返回工具证据：${toolEvidence}`
                     : "2) 子代理返回文本结果。",
                 "3) 产出主编排汇总并返回结构化结果。",
             ],
-            summary: [
-                `已完成子代理 ${delegatedSubagentType} 委托执行。`,
-            ],
-            findings: delegatedOutput.length > 0 ? delegatedOutput.slice(0, 3) : ["none"],
+            todoTrace: payloadStringArray("todoTrace") ?? fallbackTodoTrace,
+            stepResults: payloadStringArray("stepResults") ?? fallbackStepResults,
+            summary: payloadStringArray("summary") ?? fallbackSummary,
+            findings: payloadStringArray("findings") ?? fallbackFindings,
             blockers: state === "failed" ? ["子代理执行未能完成。"] : ["none"],
             nextActions: ["如需更深入，可继续追问该次委托的细节。"],
-            report: delegatedOutput.join("\n") || "委托已完成，未返回附加报告。",
+            report,
         };
     }
     async *stream(binding, input, sessionId, history = [], options = {}) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@botbotgo/agent-harness",
-  "version": "0.0.365",
+  "version": "0.0.367",
   "description": "Workspace runtime for multi-agent applications",
   "license": "MIT",
   "type": "module",