@botbotgo/agent-harness 0.0.463 → 0.0.464

package/dist/package-version.d.ts

@@ -1,2 +1,2 @@
-export declare const AGENT_HARNESS_VERSION = "0.0.463";
+export declare const AGENT_HARNESS_VERSION = "0.0.464";
 export declare const AGENT_HARNESS_RELEASE_DATE = "2026-05-04";

package/dist/package-version.js

@@ -1,2 +1,2 @@
-export const AGENT_HARNESS_VERSION = "0.0.463";
+export const AGENT_HARNESS_VERSION = "0.0.464";
 export const AGENT_HARNESS_RELEASE_DATE = "2026-05-04";

package/dist/resources/prompts/runtime/write-todos-full-entry.md

@@ -1 +1 @@
-When calling write_todos, every todo item must include both content and status. Do not send status-only updates. Retry by resending the full todo entry with the original content preserved.
+When calling write_todos, every todo item must include both content and status. Use only these status values: pending, in_progress, completed. Do not send aliases such as not_started, open, active, done, blocked, failed, or cancelled. Do not send status-only updates. Retry by resending the full todo entry with the original content preserved.

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

@@ -1,5 +1,5 @@
 import { extractVisibleOutput, isToolCallRecoveryFailure, isRetrySafeInvalidToolSelectionError, appendToolRecoveryInstruction, resolveMissingPlanRecoveryInstruction, resolveExecutionWithoutToolEvidenceTextInstruction, shouldValidateExecutionWithoutToolEvidence, resolveToolCallRecoveryInstruction, sanitizeVisibleText, EXECUTION_WITH_TOOL_EVIDENCE_RETRY_INSTRUCTION, INVALID_TOOL_SELECTION_RECOVERY_INSTRUCTION, } from "../../parsing/output-parsing.js";
-import { DELEGATED_TASK_FAILURE_RECOVERY_INSTRUCTION, DELEGATION_ONLY_RECOVERY_INSTRUCTION, } from "../../prompts/runtime-prompts.js";
+import { DELEGATED_TASK_FAILURE_RECOVERY_INSTRUCTION, DELEGATION_ONLY_RECOVERY_INSTRUCTION, REQUIRED_EXECUTION_CONTINUATION_INSTRUCTION, } from "../../prompts/runtime-prompts.js";
 import { buildInvocationRequest } from "../model/invocation-request.js";
 import { PROMPTED_JSON_TOOL_POLICY_KEY, withPromptedJsonToolPolicy } from "../model/prompted-json-tool-policy.js";
 import { buildRawModelMessages } from "../model/message-assembly.js";
@@ -257,6 +257,14 @@ function buildDeterministicFinalFromStreamToolEvidence(executedToolResults) {
         evidence.length > 0 ? evidence.join("\n\n") : "(no non-planning tool evidence captured)",
     ].join("\n");
 }
+function latestStreamToolErrorRecoveryInstruction(executedToolResults) {
+    const latest = [...executedToolResults].reverse().find((item) => item.isError === true);
+    if (!latest) {
+        return null;
+    }
+    const message = typeof latest.output === "string" ? latest.output : JSON.stringify(latest.output);
+    return resolveToolCallRecoveryInstruction(new Error(message)) ?? REQUIRED_EXECUTION_CONTINUATION_INSTRUCTION;
+}
 function hasUsefulVisibleSynthesis(value) {
     const trimmed = value.trim();
     if (trimmed.length < 80) {
@@ -699,6 +707,7 @@ export async function* streamRuntimeExecution(options) {
                         const eventContainsNonRetrySafeChunk = projectedChunks.some((chunk) => chunk.kind !== "upstream-event"
                             && chunk.kind !== "content"
                             && !(chunk.kind === "tool-result" && isPlanToolName(chunk.toolName))
+                            && !(chunk.kind === "tool-result" && chunk.isError === true)
                             && !(chunk.kind === "tool-result" && chunk.isError === true && isRetrySafeInvalidToolSelectionError(chunk.output)));
                         const hadPriorPlanToolResult = completedPlanToolResultCount > 0;
                         const repeatedPlanToolResultBeforeEvidence = requiresPlanEvidence(options.binding)
@@ -796,10 +805,23 @@ export async function* streamRuntimeExecution(options) {
                                 })
                                 : null;
                             const terminalDelegationOnlyRecoveryInstruction = resolveDelegationOnlyRecoveryInstruction(options.binding, terminalExecutionEvidence);
+                            const terminalPrematurePlanCloseRecoveryInstruction = requiresPlanEvidence(options.binding)
+                                && terminalExecutionEvidence.hasPlanStateEvidence
+                                && !terminalExecutionEvidence.hasSuccessfulNonTodoToolResultEvidence
+                                ? buildRunEvidenceAfterPlanInstruction(options.primaryTools)
+                                : null;
                             if (!emittedUnsafeStreamSideEffects
-                                && (terminalMissingPlanRecoveryInstruction || terminalDelegationOnlyRecoveryInstruction)) {
+                                && (terminalMissingPlanRecoveryInstruction
+                                    || terminalPrematurePlanCloseRecoveryInstruction
+                                    || terminalDelegationOnlyRecoveryInstruction)) {
                                 earlyStreamRecoveryInstruction =
-                                    terminalMissingPlanRecoveryInstruction ?? terminalDelegationOnlyRecoveryInstruction;
+                                    terminalMissingPlanRecoveryInstruction
+                                        ?? terminalPrematurePlanCloseRecoveryInstruction
+                                        ?? terminalDelegationOnlyRecoveryInstruction;
+                                earlyStreamRecoverySuppressInitialPlan = terminalPrematurePlanCloseRecoveryInstruction !== null;
+                                if (terminalPrematurePlanCloseRecoveryInstruction) {
+                                    earlyStreamExternalPlanEvidenceTools = resolveSingleConfiguredPlanEvidenceTool(options.primaryTools);
+                                }
                                 break;
                             }
                             if (requiresPlanEvidence(options.binding) && hasCompletedPlanWithEvidence(terminalExecutionEvidence)) {
@@ -820,6 +842,7 @@ export async function* streamRuntimeExecution(options) {
                                 && !hasMissingDelegatedExecutionEvidence(terminalExecutionEvidence)
                                 && !hasMissingDelegatedFindings(terminalExecutionEvidence)
                                 && !terminalMissingPlanRecoveryInstruction
+                                && !terminalPrematurePlanCloseRecoveryInstruction
                                 && !terminalDelegationOnlyRecoveryInstruction) {
                                 if (deferredStreamContent.length > 0) {
                                     yield* flushDeferredStreamContent();
@@ -931,6 +954,9 @@ export async function* streamRuntimeExecution(options) {
             const streamedRuntimeFailureRecoveryInstruction = projectionState.emittedOutput
                 ? resolveStreamedRuntimeFailureRecoveryInstruction(projectionState.emittedOutput, streamedExecutionEvidence)
                 : null;
+            const streamedToolErrorRecoveryInstruction = !emittedUnsafeStreamSideEffects
+                ? latestStreamToolErrorRecoveryInstruction(streamedToolResults)
+                : null;
             const missingPlanRecoveryInstruction = !hasUnresolvedExecution(streamedExecutionEvidence) && !delegatedExecutionRecoveryInstruction
                 ? resolveMissingPlanRecoveryInstruction({
                     request,
@@ -943,7 +969,8 @@ export async function* streamRuntimeExecution(options) {
                 : null;
             const retryInstruction = !emittedUnsafeStreamSideEffects && sawRetrySafeInvalidToolSelectionError
                 ? INVALID_TOOL_SELECTION_RECOVERY_INSTRUCTION
-                : delegatedExecutionRecoveryInstruction
+                : streamedToolErrorRecoveryInstruction
+                    ?? delegatedExecutionRecoveryInstruction
                     ?? streamedIncompletePlanRecoveryInstruction
                     ?? streamedPrematurePlanCloseRecoveryInstruction
                     ?? streamedRuntimeFailureRecoveryInstruction

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

@@ -39,7 +39,8 @@ function hasIncompleteStateSnapshotPlan(stateSnapshot) {
 function hasStateSnapshotPlan(stateSnapshot) {
     return typeof stateSnapshot === "object"
         && stateSnapshot !== null
-        && Array.isArray(stateSnapshot.todos);
+        && Array.isArray(stateSnapshot.todos)
+        && (stateSnapshot.todos).length > 0;
 }
 function normalizePlanToolName(toolName) {
     return typeof toolName === "string" ? toolName.trim().toLowerCase().replace(/[\s-]+/gu, "_") : "";
@@ -54,7 +55,19 @@ function isPlanToolName(toolName) {
         || normalized === "call_read_todos";
 }
 function hasPlanToolEvidence(executedToolResults) {
-    return executedToolResults.some((item) => isPlanToolName(item.toolName));
+    return executedToolResults.some((item) => {
+        if (!isPlanToolName(item.toolName)) {
+            return false;
+        }
+        const output = typeof item.output === "object" && item.output !== null ? item.output : null;
+        const summaryContainer = typeof output?.summary === "object" && output.summary !== null
+            ? output.summary
+            : null;
+        const counts = typeof summaryContainer?.summary === "object" && summaryContainer.summary !== null
+            ? summaryContainer.summary
+            : null;
+        return !(typeof counts?.total === "number" && counts.total <= 0);
+    });
 }
 function hasExecutionToolEvidence(executedToolResults) {
     return executedToolResults.some((item) => item.isError !== true && !isPlanToolName(item.toolName));

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

@@ -108,7 +108,11 @@ function readPlanStateSummary(output) {
         return null;
     }
     const typedCounts = counts;
+    if (typeof typedCounts.total === "number" && typedCounts.total <= 0) {
+        return null;
+    }
     return {
+        ...(typeof typedCounts.total === "number" ? { total: typedCounts.total } : {}),
         pending: typeof typedCounts.pending === "number" ? typedCounts.pending : 0,
         inProgress: typeof typedCounts.inProgress === "number" ? typedCounts.inProgress : 0,
     };

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

@@ -5,7 +5,6 @@ export type BuiltinTodoSnapshot = {
 };
 export declare function truncateLines(lines: string[], maxChars?: number): string;
 export declare function summarizeBuiltinWriteTodosArgs(args: Record<string, unknown>): BuiltinTodoSnapshot;
-export declare function isLowSignalTodoContent(content: string): boolean;
 export declare function formatBuiltinTodoSnapshot(snapshot: BuiltinTodoSnapshot): string;
 export declare function buildRequestPlanState(input: {
     sessionId: string;

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

@@ -119,13 +119,6 @@ export function summarizeBuiltinWriteTodosArgs(args) {
         summary,
     };
 }
-export function isLowSignalTodoContent(content) {
-    const normalized = content.trim().toLowerCase();
-    if (!normalized) {
-        return true;
-    }
-    return /^\d+$/.test(normalized) || /^step\s*\d+$/.test(normalized) || /^todo\s*\d+$/.test(normalized);
-}
 export function formatBuiltinTodoSnapshot(snapshot) {
     if (snapshot.summary.total === 0) {
         return "No todos tracked.";

package/dist/runtime/adapter/stream-event-projection.js

@@ -1,6 +1,5 @@
 import { extractToolFallbackContext, extractVisibleOutput, readTextContent, sanitizeVisibleText } from "../parsing/output-parsing.js";
 import { salvageFunctionLikeToolCall } from "../parsing/output-tool-args.js";
-import { isLowSignalTodoContent, summarizeBuiltinWriteTodosArgs } from "./runtime-adapter-support.js";
 import { computeIncrementalOutput, extractInterruptPayload, extractReasoningStreamOutput, sanitizeRetainedUpstreamEvent, extractStateStreamOutput, extractTerminalStreamOutput, extractToolResult, extractVisibleStreamOutput, normalizeTerminalOutputKey, } from "../parsing/stream-event-parsing.js";
 import { resolveModelFacingToolName } from "./tool/tool-name-mapping.js";
 export function createStreamEventProjectionState() {
@@ -141,6 +140,9 @@ function readSummaryCounts(summary) {
         return null;
     }
     const typed = summary;
+    if (typeof typed.total === "number" && typed.total <= 0) {
+        return null;
+    }
     const hasAnyCountField = typeof typed.pending === "number"
         || typeof typed.inProgress === "number";
     if (!hasAnyCountField) {
@@ -155,6 +157,9 @@ function hasIncompleteTodosArray(value) {
     if (!Array.isArray(value)) {
         return null;
     }
+    if (value.length === 0) {
+        return null;
+    }
     return value.some((item) => {
         if (typeof item !== "object" || item === null) {
             return false;
@@ -394,12 +399,6 @@ function isPlanToolName(toolName) {
         || normalized === "call_write_todos"
         || normalized === "call_read_todos";
 }
-function isWriteTodosToolName(toolName) {
-    const normalized = normalizePlanToolName(toolName);
-    return normalized === "write_todos"
-        || normalized === "tool_call_write_todos"
-        || normalized === "call_write_todos";
-}
 function extractTodoToolStart(event) {
     if (typeof event !== "object" || event === null) {
         return null;
@@ -413,15 +412,6 @@ function extractTodoToolStart(event) {
         return null;
     }
     const input = unwrapPossibleToolInput(typed.data?.input);
-    if (isWriteTodosToolName(toolName) && typeof input === "object" && input !== null && !Array.isArray(input)) {
-        const summary = summarizeBuiltinWriteTodosArgs(input);
-        if (summary.summary.total === 0) {
-            throw new Error("Error invoking tool 'write_todos' with kwargs {\"todos\":[]} with error: Error: Initial write_todos call cannot use an empty todo list. Send the concrete task steps with both content and status.");
-        }
-        if (summary.items.every((item) => isLowSignalTodoContent(item.content))) {
-            throw new Error("Error invoking tool 'write_todos' with placeholder todo content with error: Error: Initial write_todos call must use descriptive task content. Do not use placeholder entries like '1', '2', or 'step 1'.");
-        }
-    }
     return { toolName, input };
 }
 export function projectRuntimeStreamEvent(params) {

package/dist/runtime/adapter/tool/builtin-middleware-tools.js

@@ -2,7 +2,7 @@ import path from "node:path";
 import { z } from "zod";
 import { isSandboxBackend } from "deepagents";
 import { isRecord } from "../../../utils/object.js";
-import { formatBuiltinTodoSnapshot, isLowSignalTodoContent, summarizeBuiltinWriteTodosArgs, truncateLines } from "../runtime-adapter-support.js";
+import { formatBuiltinTodoSnapshot, summarizeBuiltinWriteTodosArgs, truncateLines } from "../runtime-adapter-support.js";
 import { maybePersistLargeToolOutput, resolveToolRuntimeContext } from "./tool-output-artifacts.js";
 function buildTaskToolDescription(subagents) {
     const lines = [
@@ -272,14 +272,6 @@ export async function createBuiltinMiddlewareTools(backend, options) {
         invoke: async (input) => {
             const args = isRecord(input) ? input : {};
             const summary = summarizeBuiltinWriteTodosArgs(args);
-            if (summary.summary.total === 0 && todoSnapshot.summary.total === 0) {
-                throw new Error("Initial write_todos call cannot use an empty todo list. Send the concrete task steps with both content and status.");
-            }
-            if (summary.summary.total > 0 &&
-                todoSnapshot.summary.total === 0 &&
-                summary.items.every((item) => isLowSignalTodoContent(item.content))) {
-                throw new Error("Initial write_todos call must use descriptive task content. Do not use placeholder entries like '1', '2', or 'step 1'.");
-            }
             todoSnapshot = summary;
             return {
                 ok: true,

package/dist/runtime/parsing/output-recovery.js

@@ -59,7 +59,8 @@ export function isToolCallValidationFailure(error) {
     if (/Invalid input:\s*expected .* received undefined/i.test(message) && /"path"\s*:\s*\[/.test(message)) {
         return true;
     }
-    return /Received tool input did not match expected schema/i.test(message) && /(?:→\s*at|at)\s+[\w[\].]+/i.test(message);
+    return /Received tool input did not match expected schema/i.test(message)
+        && (/(?:→\s*at|at)\s+[\w[\].]+/i.test(message) || /\bDetails:/i.test(message));
 }
 function collectErrorMessages(error) {
     if (!error)

package/dist/runtime/parsing/output-tool-args.js

@@ -574,11 +574,30 @@ function normalizeWriteTodosArgs(args) {
             const normalized = {};
             if (content !== undefined)
                 normalized.content = content;
-            normalized.status = typeof record.status === "string" && record.status.trim().length > 0 ? record.status : "pending";
+            normalized.status = normalizeWriteTodoStatus(record.status);
             return Object.keys(normalized).length > 0 ? normalized : todo;
         }),
     };
 }
+function normalizeWriteTodoStatus(value) {
+    if (typeof value !== "string") {
+        return "pending";
+    }
+    const normalized = value.trim().toLowerCase().replace(/[\s-]+/gu, "_");
+    if (!normalized) {
+        return "pending";
+    }
+    if (normalized === "not_started" || normalized === "todo" || normalized === "open") {
+        return "pending";
+    }
+    if (normalized === "started" || normalized === "active" || normalized === "in_progress") {
+        return "in_progress";
+    }
+    if (normalized === "done" || normalized === "complete" || normalized === "completed") {
+        return "completed";
+    }
+    return normalized;
+}
 function normalizeTaskArgs(args) {
     const description = typeof args.description === "string" && args.description.trim().length > 0
         ? args.description

package/dist/runtime/parsing/stream-event-parsing.js

@@ -1,4 +1,3 @@
-import { isLowSignalTodoContent } from "../adapter/runtime-adapter-support.js";
 import { extractReasoningText, extractVisibleOutput, hasToolCalls, readTextContent } from "./output-parsing.js";
 const MAX_STREAM_INLINE_TEXT_CHARS = 12_000;
 const STREAM_PREVIEW_TEXT_CHARS = 2_000;
@@ -205,31 +204,6 @@ function isErrorLikeToolOutput(value) {
         || /^command failed:/i.test(firstNonEmptyLine)
         || /^stderr:/i.test(firstNonEmptyLine);
 }
-function isEmptyInitialWriteTodosResult(value) {
-    if (typeof value !== "object" || !value || Array.isArray(value)) {
-        return false;
-    }
-    const typed = value;
-    if (typeof typed.summary === "object" && typed.summary && Array.isArray(typed.summary.items)) {
-        return (typed.summary.items).length === 0;
-    }
-    if (typeof typed.update === "object" && typed.update && Array.isArray(typed.update.todos)) {
-        return (typed.update.todos).length === 0;
-    }
-    return false;
-}
-function hasLowSignalInitialWriteTodos(value) {
-    if (typeof value !== "object" || !value || Array.isArray(value)) {
-        return false;
-    }
-    const typed = value;
-    const items = typeof typed.summary === "object" && typed.summary && Array.isArray(typed.summary.items)
-        ? (typed.summary.items ?? [])
-        : typeof typed.update === "object" && typed.update && Array.isArray(typed.update.todos)
-            ? (typed.update.todos ?? [])
-            : [];
-    return items.length > 0 && items.every((item) => typeof item?.content === "string" && isLowSignalTodoContent(item.content));
-}
 function isToolMessageLike(value) {
     if (typeof value !== "object" || value === null) {
         return false;
@@ -326,12 +300,6 @@ export function extractToolResult(event) {
     const normalizedOutput = typeof rawOutput === "string"
         ? parseMaybeJson(rawOutput)
         : unwrapToolMessageOutput(rawOutput);
-    if (!isToolError && toolName === "write_todos" && isEmptyInitialWriteTodosResult(normalizedOutput)) {
-        throw new Error("Error invoking tool 'write_todos' with kwargs {\"todos\":[]} with error: Error: Initial write_todos call cannot use an empty todo list. Send the concrete task steps with both content and status.");
-    }
-    if (!isToolError && toolName === "write_todos" && hasLowSignalInitialWriteTodos(normalizedOutput)) {
-        throw new Error("Error invoking tool 'write_todos' with placeholder todo content with error: Error: Initial write_todos call must use descriptive task content. Do not use placeholder entries like '1', '2', or 'step 1'.");
-    }
     return {
         toolName,
         output: sanitizeStreamPayload(normalizedOutput),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@botbotgo/agent-harness",
-  "version": "0.0.463",
+  "version": "0.0.464",
   "description": "Workspace runtime for multi-agent applications",
   "license": "MIT",
   "type": "module",