@xopcai/xopc 0.0.385 → 0.0.387

package/dist/src/agent/tools/delegate-tool.js

@@ -1,3 +1,6 @@
+import { init_research_checkpoint, researchResumeContext } from "../orchestration/research-checkpoint.js";
+import { claimResearchCheckpoint, createResearchCheckpoint, finishResearchCheckpoint, getResearchCheckpoint, init_research_checkpoint_repository, recordResearchEvidence, recordResearchSearch, saveResearchCheckpoint } from "../../storage/sqlite/research-checkpoint-repository.js";
+import { resolveAgentAbortReason } from "../orchestration/run-termination.js";
 import { readWorkspaceRevision } from "../coding/workspace-revision.js";
 import { EXTERNAL_TOOL_NAMES } from "../external-tools/gateway-tools.js";
 import "../external-tools/index.js";
@@ -8,8 +11,11 @@ import { runGit } from "../../execution-environments/git.js";
 import { LocalWorktreeManager } from "../../execution-environments/local-worktree-manager.js";
 import { SessionEnvironmentService } from "../../execution-environments/session-environment-service.js";
 import { DELEGATION_CAPABILITIES, DELEGATION_TOOL_CAPABILITIES, protectDelegatedTool, resolveDelegationTools } from "./delegation-policy.js";
+import { resolve } from "node:path";
 import { Type } from "@sinclair/typebox";
 //#region src/agent/tools/delegate-tool.ts
+init_research_checkpoint();
+init_research_checkpoint_repository();
 const DEFAULT_DELEGATE_TOOLS = resolveDelegationTools(Object.keys(DELEGATION_TOOL_CAPABILITIES), "research").granted;
 /** Tools never passed to a delegated sub-agent (even if requested). */
 const DELEGATE_BLOCKED_TOOLS = new Set([
@@ -25,10 +31,23 @@ const DELEGATE_BLOCKED_TOOLS = new Set([
 	"skill_install",
 	EXTERNAL_TOOL_NAMES.requireConnection,
 	EXTERNAL_TOOL_NAMES.updateConnectionObjective,
-	"managed_job"
+	"managed_job",
+	"research_checkpoint"
 ]);
 const DelegateTaskSchema = Type.Object({
-	goal: Type.String({ minLength: 1 }),
+	goal: Type.String({
+		minLength: 1,
+		maxLength: 8e3
+	}),
+	resumeId: Type.Optional(Type.String({
+		format: "uuid",
+		description: "Resume a saved research checkpoint owned by this conversation and workspace; preserve the original goal."
+	})),
+	maxAttempts: Type.Optional(Type.Integer({
+		minimum: 1,
+		maximum: 3,
+		description: "Research attempts in this call, including bounded automatic continuation after budget exhaustion. Default 2; cancellation, permissions and provider failures are not automatically retried."
+	})),
 	context: Type.Optional(Type.String()),
 	mode: Type.Optional(Type.Union([
 		"inspect",
@@ -54,16 +73,48 @@ function createDelegateTool(deps) {
 		name: "delegate_task",
 		label: "Delegate task",
 		parameters: DelegateTaskSchema,
-		description: "Delegate a bounded task with a fresh context; supply relevant facts and constraints in context. Default research inherits the parent's local, web, knowledge, skill and approved external read tools. read/inspect/review are local read-only. toolset and capabilities only narrow parent access; custom requires explicit capabilities. Browser interaction is opt-in. implement requires a project and clean Git workspace and returns changes in a separate worktree for review. No nested delegation, external writes, installation or background jobs. budgetTier selects light, standard (default), or deep; token budgets sum all model requests, not context size. Reports stop reason, budget usage, partial handoff and rejected tools.",
+		description: "Delegate a bounded task with a fresh context; supply relevant facts and constraints in context. Default research inherits the parent's local, web, knowledge, skill and approved external read tools. read/inspect/review are local read-only. toolset and capabilities only narrow parent access; custom requires explicit capabilities. Browser interaction is opt-in. implement requires a project and clean Git workspace and returns changes in a separate worktree for review. No nested delegation, external writes, installation or background jobs. budgetTier selects light, standard (default), or deep; token budgets sum all model requests, not context size. Research saves structured findings and source evidence durably for the current conversation. Use resumeId to continue unfinished work; maxAttempts bounds automatic budget continuation (default 2). Reports stop reason, budget usage, partial handoff and rejected tools.",
 		supportsParallel: false,
 		idempotent: false,
+		timeoutMs: DELEGATION_BUDGETS.deep.timeoutMs * 3 + 3e4,
 		async execute(_id, input, signal) {
 			signal?.throwIfAborted();
 			const mode = input.mode ?? "research";
+			if (input.resumeId && mode !== "research") throw new Error("resumeId is only supported for research");
 			const scope = getDelegationScope();
 			const parentTools = scope?.tools ?? deps.getParentTools();
 			const access = resolveDelegationTools(parentTools.map((tool) => tool.name), mode, input.toolset, input.capabilities);
 			const tools = access.granted;
+			const parentConversationId = deps.getCurrentContext?.()?.conversationId;
+			let workspace = deps.workspace, environmentId;
+			if (mode === "implement") {
+				if (!parentConversationId) throw new Error("Implementation delegation requires a project session");
+				const parent = new SessionEnvironmentService().get(parentConversationId);
+				if (!parent?.projectId) throw new Error("Attach this session to a project before implementation delegation");
+				const environment = await new LocalWorktreeManager().provisionManagedWorktree({
+					projectId: parent.projectId,
+					repositoryPath: workspace
+				});
+				workspace = environment.rootPath;
+				environmentId = environment.id;
+			}
+			const revision = mode === "review" ? await readWorkspaceRevision(workspace) : void 0;
+			const authorizeToolCall = scope?.authorizeToolCall ?? deps.createParentPolicy?.().beforeToolCall;
+			let research = mode === "research" && parentConversationId && (tools.length > 0 || input.resumeId) ? input.resumeId ? getResearchCheckpoint(input.resumeId, parentConversationId, resolve(workspace)) : createResearchCheckpoint(parentConversationId, resolve(workspace), input.goal) : void 0;
+			if (input.resumeId && !research) throw new Error("Research checkpoint not found in this conversation and workspace");
+			if (research && research.goal !== input.goal) throw new Error("Resuming research must preserve the original goal");
+			if (research?.status === "completed") return {
+				content: [{
+					type: "text",
+					text: research.summary + "\n" + JSON.stringify(research.checkpoint)
+				}],
+				details: {
+					status: "success",
+					stopReason: "completed",
+					resumeId: research.id,
+					checkpoint: research.checkpoint
+				}
+			};
 			if (!tools.length) return {
 				content: [{
 					type: "text",
@@ -77,130 +128,174 @@ function createDelegateTool(deps) {
 					stopReason: "parent_policy",
 					mode,
 					grantedTools: tools,
-					rejectedTools: access.rejected
+					rejectedTools: access.rejected,
+					...research ? {
+						resumeId: research.id,
+						checkpoint: research.checkpoint
+					} : {}
 				}
 			};
-			const parentConversationId = deps.getCurrentContext?.()?.conversationId;
 			const model = deps.getSubagentModel();
-			let workspace = deps.workspace, environmentId;
-			if (mode === "implement") {
-				if (!parentConversationId) throw new Error("Implementation delegation requires a project session");
-				const parent = new SessionEnvironmentService().get(parentConversationId);
-				if (!parent?.projectId) throw new Error("Attach this session to a project before implementation delegation");
-				const environment = await new LocalWorktreeManager().provisionManagedWorktree({
-					projectId: parent.projectId,
-					repositoryPath: workspace
-				});
-				workspace = environment.rootPath;
-				environmentId = environment.id;
-			}
-			const revision = mode === "review" ? await readWorkspaceRevision(workspace) : void 0;
-			const authorizeToolCall = scope?.authorizeToolCall ?? deps.createParentPolicy?.().beforeToolCall;
-			const child = createDelegateChildHandle({
-				workspace,
-				goal: mode === "review" ? `Independently review the actual changes. Inspect review_workspace and relevant source/tests. Report actionable defects with file and line evidence, severity and a concrete failure example. If there are no findings, state the checks and remaining uncertainty. Do not treat the parent's completion claim as evidence.\n\n${input.goal}` : input.goal,
-				context: input.context,
-				requesterConversationId: parentConversationId,
-				allowedToolNames: tools,
-				budgetTier: input.budgetTier,
-				maxIterations: input.maxIterations ?? DELEGATION_BUDGETS[input.budgetTier ?? "standard"].maxToolCalls,
-				model,
-				bus: deps.bus,
-				getConfig: deps.getConfig,
-				toolExecutorConfig: deps.toolExecutorConfig,
-				authorizeToolCall,
-				buildChildTools: (opts) => {
-					const local = mode === "implement" ? new Map(deps.buildChildTools(opts).map((tool) => [tool.name, tool])) : void 0;
-					const rebound = new Set([
-						"read_file",
-						"grep",
-						"find",
-						"list_dir",
-						"review_workspace",
-						"write_file",
-						"apply_patch",
-						"exec_command",
-						"language_diagnostics"
-					]);
-					return parentTools.flatMap((tool) => {
-						const candidate = local && rebound.has(tool.name) ? local.get(tool.name) : tool;
-						return candidate ? [protectDelegatedTool(candidate)] : [];
+			const attemptLimit = research ? Math.min(3, Math.max(1, input.maxAttempts ?? 2)) : 1;
+			let attempt = 0;
+			const attemptBudgets = [];
+			let result;
+			do {
+				signal?.throwIfAborted();
+				attempt++;
+				if (research) research = claimResearchCheckpoint(research, DELEGATION_BUDGETS[input.budgetTier ?? "standard"].timeoutMs + 1e4);
+				const claimedResearch = research;
+				const checkpointSink = claimedResearch ? {
+					read: () => getResearchCheckpoint(claimedResearch.id, claimedResearch.conversationId, claimedResearch.workspace).checkpoint,
+					save: (update) => saveResearchCheckpoint(claimedResearch, update),
+					recordEvidence: (evidence) => recordResearchEvidence(claimedResearch, evidence),
+					recordSearch: (entry) => recordResearchSearch(claimedResearch, entry)
+				} : void 0;
+				let child;
+				const abort = () => child?.abort(resolveAgentAbortReason(signal?.reason));
+				try {
+					child = createDelegateChildHandle({
+						workspace,
+						goal: mode === "review" ? `Independently review the actual changes. Inspect review_workspace and relevant source/tests. Report actionable defects with file and line evidence, severity and a concrete failure example. If there are no findings, state the checks and remaining uncertainty. Do not treat the parent's completion claim as evidence.\n\n${input.goal}` : input.goal,
+						context: [input.context, research ? researchResumeContext(research.checkpoint) : void 0].filter(Boolean).join("\n\n") || void 0,
+						researchCheckpoint: checkpointSink,
+						requesterConversationId: parentConversationId,
+						allowedToolNames: tools,
+						budgetTier: input.budgetTier,
+						maxIterations: input.maxIterations ?? DELEGATION_BUDGETS[input.budgetTier ?? "standard"].maxToolCalls,
+						model,
+						bus: deps.bus,
+						getConfig: deps.getConfig,
+						toolExecutorConfig: deps.toolExecutorConfig,
+						authorizeToolCall,
+						buildChildTools: (opts) => {
+							const local = mode === "implement" ? new Map(deps.buildChildTools(opts).map((tool) => [tool.name, tool])) : void 0;
+							const rebound = new Set([
+								"data_batch",
+								"read_file",
+								"grep",
+								"find",
+								"list_dir",
+								"review_workspace",
+								"write_file",
+								"apply_patch",
+								"exec_command",
+								"language_diagnostics"
+							]);
+							return parentTools.flatMap((tool) => {
+								const candidate = local && rebound.has(tool.name) ? local.get(tool.name) : tool;
+								return candidate ? [protectDelegatedTool(candidate)] : [];
+							});
+						},
+						verifyChanges: mode === "implement"
 					});
-				},
-				verifyChanges: mode === "implement"
-			});
-			const abort = () => child.abort();
-			signal?.addEventListener("abort", abort, { once: true });
-			if (signal?.aborted) abort();
-			try {
-				const result = await child.run();
-				const currentRevision = mode === "review" ? await readWorkspaceRevision(workspace) : void 0;
-				const stale = mode === "review" && (!revision || revision !== currentRevision);
-				const patch = mode === "implement" ? await runGit(workspace, [
-					"diff",
-					"--no-ext-diff",
-					"--no-textconv",
-					"--binary",
-					"HEAD"
-				]).catch(() => "") : void 0;
-				return {
-					content: [{
-						type: "text",
-						text: [
-							result.summary,
-							`Budget (${result.budget.tier}): ${JSON.stringify({
-								stopReason: result.stopReason,
-								usage: result.budget.usage,
-								limits: result.budget.limits,
-								handoffRequested: result.handoff.requested
-							})}`,
-							stale ? "Workspace changed or could not be fingerprinted; review findings require revalidation." : "",
-							access.rejected.length ? `Rejected tools: ${JSON.stringify(access.rejected)}` : "",
-							environmentId ? `Changes retained in ${workspace} (environment ${environmentId}). Review and integrate from this workspace, then verify the parent workspace again.` : "",
-							patch ? `Patch preview:\n${patch.slice(0, 3e4)}` : ""
-						].filter(Boolean).join("\n\n")
-					}],
-					details: {
-						status: stale && result.status === "success" ? "partial" : result.status,
-						mode,
-						summary: result.summary,
-						grantedTools: tools,
-						rejectedTools: access.rejected,
-						iterations: result.toolIterations,
-						revision: currentRevision,
-						reviewedRevision: revision,
-						stopReason: result.stopReason,
-						budget: result.budget,
-						handoff: result.handoff,
-						...result.verification ? { childVerification: result.verification } : {},
-						...environmentId ? {
-							environmentId,
-							workspace,
-							patchTruncated: (patch?.length ?? 0) > 3e4
-						} : {}
+					signal?.addEventListener("abort", abort, { once: true });
+					if (signal?.aborted) abort();
+					result = await child.run();
+					attemptBudgets.push(result.budget);
+					if (research && checkpointSink) {
+						const checkpoint = checkpointSink.read();
+						if (result.status === "success" && (!checkpoint.findings.length || checkpoint.nextSteps.length || checkpoint.findings.some((finding) => finding.status === "unverified"))) result = {
+							...result,
+							status: "partial",
+							stopReason: "verification",
+							summary: result.summary + "\nResearch checkpoint still has unfinished work; do not claim completion."
+						};
+						finishResearchCheckpoint(research, result.status === "success" ? "completed" : result.status, result.stopReason, result.summary);
+						research = getResearchCheckpoint(research.id, research.conversationId, research.workspace);
 					}
-				};
-			} catch (error) {
-				return {
-					content: [{
-						type: "text",
-						text: `Delegation failed: ${String(error)}${environmentId ? `; work retained in ${workspace} (${environmentId})` : ""}`
-					}],
-					details: {
-						status: signal?.aborted ? "cancelled" : "failed",
-						stopReason: signal?.aborted ? "cancelled" : "runtime_error",
-						mode,
-						grantedTools: tools,
-						rejectedTools: access.rejected,
-						...environmentId ? {
-							environmentId,
-							workspace
-						} : {}
+				} catch (error) {
+					const stopReason = signal?.aborted ? resolveAgentAbortReason(signal.reason) : "runtime_error";
+					const status = stopReason === "deadline_exceeded" ? "partial" : signal?.aborted ? "cancelled" : "failed";
+					if (research) {
+						if (getResearchCheckpoint(research.id, research.conversationId, research.workspace)?.leaseToken === research.leaseToken) finishResearchCheckpoint(research, status, stopReason, String(error));
+						research = getResearchCheckpoint(research.id, research.conversationId, research.workspace) ?? research;
 					}
-				};
-			} finally {
-				signal?.removeEventListener("abort", abort);
-			}
+					return {
+						content: [{
+							type: "text",
+							text: `Delegation failed: ${String(error)}${environmentId ? `; work retained in ${workspace} (${environmentId})` : ""}`
+						}],
+						details: {
+							status,
+							stopReason,
+							mode,
+							grantedTools: tools,
+							rejectedTools: access.rejected,
+							...research ? {
+								resumeId: research.id,
+								checkpoint: research.checkpoint
+							} : {},
+							...environmentId ? {
+								environmentId,
+								workspace
+							} : {}
+						}
+					};
+				} finally {
+					signal?.removeEventListener("abort", abort);
+				}
+			} while (research && attempt < attemptLimit && !signal?.aborted && result.status === "partial" && [
+				"tool_calls",
+				"total_tokens",
+				"model_turns",
+				"timeout",
+				"deadline_exceeded"
+			].includes(result.stopReason) && research.checkpoint.nextSteps.length > 0);
+			const currentRevision = mode === "review" ? await readWorkspaceRevision(workspace) : void 0;
+			const stale = mode === "review" && (!revision || revision !== currentRevision);
+			const patch = mode === "implement" ? await runGit(workspace, [
+				"diff",
+				"--no-ext-diff",
+				"--no-textconv",
+				"--binary",
+				"HEAD"
+			]).catch(() => "") : void 0;
+			return {
+				content: [{
+					type: "text",
+					text: [
+						result.summary,
+						research ? `Research checkpoint ${research.id}: ${JSON.stringify(research.checkpoint)}. Resume unfinished work with resumeId and the same goal.` : "",
+						`Budget (${result.budget.tier}): ${JSON.stringify({
+							stopReason: result.stopReason,
+							usage: result.budget.usage,
+							limits: result.budget.limits,
+							handoffRequested: result.handoff.requested
+						})}`,
+						stale ? "Workspace changed or could not be fingerprinted; review findings require revalidation." : "",
+						access.rejected.length ? `Rejected tools: ${JSON.stringify(access.rejected)}` : "",
+						environmentId ? `Changes retained in ${workspace} (environment ${environmentId}). Review and integrate from this workspace, then verify the parent workspace again.` : "",
+						patch ? `Patch preview:\n${patch.slice(0, 3e4)}` : ""
+					].filter(Boolean).join("\n\n")
+				}],
+				details: {
+					status: stale && result.status === "success" ? "partial" : result.status,
+					mode,
+					summary: result.summary,
+					grantedTools: tools,
+					rejectedTools: access.rejected,
+					iterations: attemptBudgets.reduce((total, budget) => total + (budget.usage.toolCalls ?? 0), 0),
+					revision: currentRevision,
+					reviewedRevision: revision,
+					stopReason: result.stopReason,
+					budget: result.budget,
+					handoff: result.handoff,
+					...research ? {
+						resumeId: research.id,
+						checkpoint: research.checkpoint,
+						attempts: attempt,
+						totalAttempts: research.attempts,
+						attemptBudgets
+					} : {},
+					...result.verification ? { childVerification: result.verification } : {},
+					...environmentId ? {
+						environmentId,
+						workspace,
+						patchTruncated: (patch?.length ?? 0) > 3e4
+					} : {}
+				}
+			};
 		}
 	};
 }

package/dist/src/agent/tools/delegation-policy.js

@@ -11,6 +11,7 @@ const DELEGATION_CAPABILITIES = [
 ];
 /** Host-owned classifications. New tools fail closed until classified. */
 const DELEGATION_TOOL_CAPABILITIES = {
+	data_batch: "local_read",
 	read_file: "local_read",
 	grep: "local_read",
 	find: "local_read",
@@ -43,6 +44,16 @@ const DELEGATION_TOOL_CAPABILITIES = {
 	language_diagnostics: "command",
 	browser_use: "browser"
 };
+const BATCH_SOURCES = new Set([
+	"read_file",
+	"grep",
+	"exec_command",
+	"knowledge_search",
+	"knowledge_get",
+	"web_search",
+	"web_fetch",
+	"xopc_tool_execute"
+]);
 const RESEARCH = [
 	"local_read",
 	"web_read",
@@ -67,7 +78,8 @@ function resolveDelegationTools(parentNames, mode, requested, capabilities) {
 	const rejected = [];
 	for (const tool of new Set((requested ?? parentNames).map((name) => name.trim()).filter(Boolean))) {
 		const capability = DELEGATION_TOOL_CAPABILITIES[tool];
-		const reason = !parent.has(tool) ? "Unavailable to the parent agent" : !capability ? "Tool has no delegable capability" : !allowed.has(capability) ? `Capability ${capability} is outside this delegation` : void 0;
+		const capabilityAllowed = tool === "data_batch" ? parentNames.some((name) => BATCH_SOURCES.has(name) && allowed.has(DELEGATION_TOOL_CAPABILITIES[name])) : capability !== void 0 && allowed.has(capability);
+		const reason = !parent.has(tool) ? "Unavailable to the parent agent" : !capability ? "Tool has no delegable capability" : !capabilityAllowed ? `Capability ${capability} is outside this delegation` : void 0;
 		if (reason) {
 			if (requested) rejected.push({
 				tool,

package/dist/src/agent/tools/exec-command.js

@@ -1,3 +1,4 @@
+import { runProcess } from "../../process/run-process.js";
 import { commandRegistry, commandTimeout } from "../commands/command-registry.js";
 import { evaluateExecPolicy } from "../sandbox/exec-policy.js";
 import { formatSize, truncateTail } from "./truncate.js";
@@ -58,6 +59,7 @@ function createExecCommandTool(workspaceCwd, options) {
 			"Unix and Docker use /bin/sh, not your interactive shell. Do not use Bash process substitution <(...) or other Bash-only syntax unless invoking an installed bash explicitly. Prefer separate portable commands or a script file. Windows hosts use cmd.exe.",
 			"Use apply_patch for file edits; do not edit files by shell redirection unless explicitly necessary.",
 			"Relative cwd values resolve under the current agent workspace.",
+			"Use a fresh output directory with mkdir -p. Avoid destructive cleanup such as rm -rf on variable paths; blocked cleanup should be removed, never disguised or moved to bypass policy. A syntax preflight on Unix host commands prevents partial writes from malformed shell syntax.",
 			"When the command creates final user-facing files, declare all final paths in outputs so they are persisted with this command."
 		].join(" "),
 		parameters: ExecCommandSchema,
@@ -99,7 +101,7 @@ function createExecCommandTool(workspaceCwd, options) {
 				allowedEnvVars: options?.getSkillPassthroughEnvVarNames?.() ?? []
 			});
 			if (!policy.allowed) {
-				const text = `Sandbox blocked command: ${policy.reason}`;
+				const text = `Sandbox blocked command: ${policy.reason}\nNothing was executed. Remove destructive cleanup and use a fresh output directory with mkdir -p, or narrow the operation. Do not repeat unchanged or rephrase the blocked operation to bypass policy.`;
 				return {
 					content: [{
 						type: "text",
@@ -125,12 +127,55 @@ function createExecCommandTool(workspaceCwd, options) {
 			}
 			const timeoutMs = commandTimeout(params.timeoutMs, policy.timeoutMs);
 			const maxOutputChars = clampMaxOutputChars(params.maxOutputChars);
+			const isolation = options?.getCommandIsolation?.();
+			if (process.platform !== "win32" && isolation?.mode !== "docker") {
+				const started = performance.now();
+				const syntax = await runProcess({
+					program: "/bin/sh",
+					args: [
+						"-n",
+						"-c",
+						command
+					],
+					cwd: policy.effectiveCwd,
+					env: policy.sanitizedEnv,
+					timeoutMs: 5e3,
+					maxOutputBytes: 8e3,
+					signal,
+					terminationPolicy: "tree"
+				});
+				if (syntax.exitCode !== 0) {
+					const details = {
+						command,
+						cwd: policy.effectiveCwd,
+						status: syntax.timedOut ? "timed_out" : "failed",
+						exitCode: syntax.exitCode,
+						durationMs: performance.now() - started,
+						timedOut: syntax.timedOut,
+						truncated: false,
+						outputBytes: Buffer.byteLength(syntax.stderr),
+						totalOutputBytes: Buffer.byteLength(syntax.stderr),
+						captureTruncated: syntax.outputTruncated,
+						stdout: "",
+						stderr: syntax.stderr,
+						aggregatedOutput: syntax.stderr,
+						failureHint: "Shell syntax preflight failed; nothing was executed. Use portable POSIX sh syntax (for example diff file1 file2), or explicitly invoke an installed bash for Bash syntax. Fix the command; no side effects need replaying."
+					};
+					return {
+						content: [{
+							type: "text",
+							text: formatOutputForModel(details, maxOutputChars).text
+						}],
+						details
+					};
+				}
+			}
 			const runtimeEnv = options?.prepareEnv ? await options.prepareEnv(policy.sanitizedEnv, policy.effectiveCwd) : policy.sanitizedEnv;
 			const owner = options?.getConversationId?.() ?? workspaceCwd;
 			const registry = commandRegistry();
 			let commandResult = await registry.start({
 				workspace: workspaceCwd,
-				isolation: options?.getCommandIsolation?.(),
+				isolation,
 				owner,
 				command,
 				cwd: policy.effectiveCwd,

package/dist/src/agent/tools/factory.js

@@ -8,9 +8,10 @@ import { getSessionMetadata } from "../../storage/sqlite/session-read-repository
 import { patchSessionMetadata } from "../../storage/sqlite/session-repository.js";
 import { getSessionTaskPlan, setSessionTaskPlan } from "../../storage/sqlite/session-task-plan-repository.js";
 import { init_sqlite } from "../../storage/sqlite/index.js";
-import { init_agent_scope, resolveDefaultAgentId } from "../agent-scope.js";
 import { resolveUserContextSessionAccess } from "../../user-context/access-policy.js";
+import { init_agent_scope, resolveDefaultAgentId } from "../agent-scope.js";
 import { resolveEffectiveAgentConfigForAgent, resolveEffectiveAgentConfigForSession } from "../../config/agent-profile.js";
+import { TaskConversationRepository } from "../../tasks/task-conversation-repository.js";
 import { LOCAL_READ_MAX_CHARS } from "./local-read-budget.js";
 import { createPersonalAttentionTool } from "./personal-attention-tool.js";
 import { markCodemodeCoreRead } from "./codemode-permissions.js";
@@ -40,6 +41,7 @@ import { createWebFetchTool, createWebSearchTool } from "./web.js";
 import { createWebExtractTool } from "./web-extract.js";
 import { createMessageTool } from "./communication.js";
 import { createSendMediaTool } from "./send-media.js";
+import { createTaskResultTool } from "./task-result-tool.js";
 import { createReadMediaTool } from "./media-read-tool.js";
 import { createCreateShareTool, isShareToolAvailable } from "./create-share-tool.js";
 import { createTodoTool } from "./todo-tool.js";
@@ -71,7 +73,10 @@ import { createImageGenerateTool } from "./image-generate-tool.js";
 import { createPersonalReadTool } from "./personal-read-tool.js";
 import { createTaskCollaborationTool } from "./task-collaboration-tool.js";
 import "./index.js";
+import { createPersonalResultTool } from "./personal-result-tool.js";
 import { createPersonalTaskTool } from "./personal-task-tool.js";
+import { withTaskEvidenceIdentity } from "./task-evidence.js";
+import { withPersonalPerformance } from "./personal-performance.js";
 import { createPersonalRequestTools } from "./personal-request-tool.js";
 import { createPersonalRequestConnectionTool, createPersonalRequestResultTool } from "./personal-request-result-tool.js";
 import { createPersonalPreferenceTool } from "./personal-preference-tool.js";
@@ -427,6 +432,12 @@ var AgentToolsFactory = class AgentToolsFactory {
 			})] : [],
 			createSendMediaTool(workspace, bus, deliveryContext),
 			createPublishArtifactsTool(workspace),
+			createTaskResultTool(currentConversationId),
+			createTaskCollaborationTool({
+				...productToolDeps,
+				getWorkspace: () => workspace,
+				getCurrentConversationId: currentConversationId
+			}),
 			createReadMediaTool(personalToolsConfigured ? {
 				textOnly: true,
 				maxChars: LOCAL_READ_MAX_CHARS,
@@ -530,6 +541,7 @@ var AgentToolsFactory = class AgentToolsFactory {
 			...this.deps.getAutomationService || this.deps.getSceneAccess || this.deps.getProjectService || this.deps.getNotesService || this.deps.getLocalAppService || this.deps.getChatPreviewService || this.deps.dispatchTaskEvents || this.deps.dispatchTaskRuns ? [createXopcUseTool(productToolDeps), ...personalToolsConfigured ? [
 				createPersonalTaskTool(productToolDeps),
 				createPersonalReadTool(productToolDeps),
+				createPersonalResultTool(currentConversationId),
 				...createPersonalRequestTools(productToolDeps),
 				createPersonalAttentionTool({
 					getConversationId: () => this.deps.getCurrentContext()?.conversationId,
@@ -614,14 +626,7 @@ var AgentToolsFactory = class AgentToolsFactory {
 				}
 			})] : [],
 			...optionalTools,
-			...personalExecution ? [
-				createPersonalRequestResultTool(currentConversationId),
-				createPersonalRequestConnectionTool(currentConversationId),
-				createTaskCollaborationTool({
-					...productToolDeps,
-					getCurrentConversationId: currentConversationId
-				})
-			] : []
+			...personalExecution ? [createPersonalRequestResultTool(currentConversationId), createPersonalRequestConnectionTool(currentConversationId)] : []
 		];
 		if ([
 			"read_file",
@@ -666,7 +671,11 @@ var AgentToolsFactory = class AgentToolsFactory {
 		return ["create_desktop_pet"];
 	}
 	createAllTools(coreOptions) {
-		return sortToolsForPromptCache(wrapToolsWithProtection(this.createCoreTools(coreOptions), this.deps.toolExecutorConfig));
+		const wrapped = wrapToolsWithProtection(this.createCoreTools(coreOptions), this.deps.toolExecutorConfig);
+		const executionId = coreOptions?.conversationId ?? getEmbeddedExecutionSession() ?? this.deps.getCurrentContext()?.conversationId;
+		const taskExecution = executionId && isXopcDatabaseOpen() && new TaskConversationRepository().resolveActiveExecutionSession(executionId);
+		const observed = wrapped.map((tool) => taskExecution || tool.name.startsWith("personal_") ? withPersonalPerformance(tool, executionId, taskExecution ? taskExecution.taskId : void 0, taskExecution ? taskExecution.runId : void 0, taskExecution ? taskExecution.assignmentEpoch : void 0) : tool);
+		return sortToolsForPromptCache(taskExecution ? observed.map((tool) => withTaskEvidenceIdentity(tool)) : observed);
 	}
 };
 function filterToolsByDisabledSet(tools, disabled) {

package/dist/src/agent/tools/index.d.ts

@@ -17,6 +17,7 @@ export { createWebExtractTool, stripHtmlBoilerplate, DEFAULT_WEB_EXTRACT_MAX_LEN
 export { createMessageTool } from './communication.js';
 export { createSendMediaTool } from './send-media.js';
 export { createPublishArtifactsTool } from './publish-artifacts.js';
+export { createTaskResultTool } from './task-result-tool.js';
 export { createReadMediaTool } from './media-read-tool.js';
 export { createCreateShareTool, isShareToolAvailable, type CreateShareToolDeps, } from './create-share-tool.js';
 export { createTodoTool, TodoStore, type TodoItem, type TodoStatus } from './todo-tool.js';

package/dist/src/agent/tools/index.js

@@ -19,6 +19,7 @@ import { createWebFetchTool, createWebSearchTool } from "./web.js";
 import { DEFAULT_WEB_EXTRACT_MAX_LENGTH, MAX_RAW_HTML_CHARS_FOR_WEB_EXTRACT, createWebExtractTool, stripHtmlBoilerplate } from "./web-extract.js";
 import { createMessageTool } from "./communication.js";
 import { createSendMediaTool } from "./send-media.js";
+import { createTaskResultTool } from "./task-result-tool.js";
 import { createReadMediaTool } from "./media-read-tool.js";
 import { createCreateShareTool, isShareToolAvailable } from "./create-share-tool.js";
 import { TodoStore, createTodoTool } from "./todo-tool.js";
@@ -47,4 +48,4 @@ import { createImageTool } from "./image-tool.js";
 import { createImageGenerateTool, resolveImageGenerationModelConfigForTool } from "./image-generate-tool.js";
 import { createPersonalReadTool } from "./personal-read-tool.js";
 import { createTaskCollaborationTool } from "./task-collaboration-tool.js";
-export { BrowserRuntime, DEFAULT_DELEGATE_TOOLS, DEFAULT_MAX_BYTES, DEFAULT_MAX_LINES, DEFAULT_TOOL_METADATA, DEFAULT_WEB_EXTRACT_MAX_LENGTH, DELEGATE_BLOCKED_TOOLS, GREP_MAX_LINE_LENGTH, MAX_RAW_HTML_CHARS_FOR_WEB_EXTRACT, TodoStore, ToolConcurrencyController, assertBrowserUrlAllowed, createApplyPatchTool, createAutomationTool, createBrowserAutomationTool, createClarifyTool, createCreateShareTool, createDataBatchTool, createDelegateTool, createDesktopPetTool, createExecCommandTool, createFindTool, createGrepTool, createImageGenerateTool, createImageTool, createKnowledgeGetTool, createKnowledgeReadTool, createKnowledgeSearchTool, createKnowledgeWriteTool, createListDirTool, createManagedJobTool, createMemoryMaintenanceTool, createMessageTool, createPersonalReadTool, createPublishArtifactsTool, createReadFileTool, createReadMediaTool, createSendMediaTool, createSessionRecallTool, createSessionSearchTool, createSessionStatusTool, createSkillInstallTool, createSkillManageTool, createSkillViewTool, createSkillsListTool, createSkillsMarketplaceSearchTool, createTaskCollaborationTool, createTodoTool, createToolManualTool, createUpdatePlanTool, createUserContextGetTool, createUserContextReadTool, createUserContextSearchTool, createUserContextUpdateTool, createWebExtractTool, createWebFetchTool, createWebSearchTool, createWorkflowManageTool, createWorkflowTool, createWriteFileTool, createXopcUseTool, findTool, formatSize, fuzzyFindText, generateDiffString, getToolMetadata, grepTool, isShareToolAvailable, listDirTool, normalizeForFuzzyMatch, normalizeToLF, resolveImageGenerationModelConfigForTool, resolveImageModelConfigForTool, resolveToolLockMode, restoreLineEndings, stripBom, stripHtmlBoilerplate, truncateHead, truncateLine, truncateTail, writeFileTool };
+export { BrowserRuntime, DEFAULT_DELEGATE_TOOLS, DEFAULT_MAX_BYTES, DEFAULT_MAX_LINES, DEFAULT_TOOL_METADATA, DEFAULT_WEB_EXTRACT_MAX_LENGTH, DELEGATE_BLOCKED_TOOLS, GREP_MAX_LINE_LENGTH, MAX_RAW_HTML_CHARS_FOR_WEB_EXTRACT, TodoStore, ToolConcurrencyController, assertBrowserUrlAllowed, createApplyPatchTool, createAutomationTool, createBrowserAutomationTool, createClarifyTool, createCreateShareTool, createDataBatchTool, createDelegateTool, createDesktopPetTool, createExecCommandTool, createFindTool, createGrepTool, createImageGenerateTool, createImageTool, createKnowledgeGetTool, createKnowledgeReadTool, createKnowledgeSearchTool, createKnowledgeWriteTool, createListDirTool, createManagedJobTool, createMemoryMaintenanceTool, createMessageTool, createPersonalReadTool, createPublishArtifactsTool, createReadFileTool, createReadMediaTool, createSendMediaTool, createSessionRecallTool, createSessionSearchTool, createSessionStatusTool, createSkillInstallTool, createSkillManageTool, createSkillViewTool, createSkillsListTool, createSkillsMarketplaceSearchTool, createTaskCollaborationTool, createTaskResultTool, createTodoTool, createToolManualTool, createUpdatePlanTool, createUserContextGetTool, createUserContextReadTool, createUserContextSearchTool, createUserContextUpdateTool, createWebExtractTool, createWebFetchTool, createWebSearchTool, createWorkflowManageTool, createWorkflowTool, createWriteFileTool, createXopcUseTool, findTool, formatSize, fuzzyFindText, generateDiffString, getToolMetadata, grepTool, isShareToolAvailable, listDirTool, normalizeForFuzzyMatch, normalizeToLF, resolveImageGenerationModelConfigForTool, resolveImageModelConfigForTool, resolveToolLockMode, restoreLineEndings, stripBom, stripHtmlBoilerplate, truncateHead, truncateLine, truncateTail, writeFileTool };

package/dist/src/agent/tools/language-diagnostics.js

@@ -1,7 +1,7 @@
 import { createExecCommandTool } from "./exec-command.js";
 import { relative, resolve, sep } from "node:path";
-import { access } from "node:fs/promises";
 import { Type } from "@sinclair/typebox";
+import { access } from "node:fs/promises";
 //#region src/agent/tools/language-diagnostics.ts
 function parseTypeScriptDiagnostics(output) {
 	return [...output.matchAll(/^(.+)\((\d+),(\d+)\): error (TS\d+): (.+)$/gm)].slice(0, 200).map((match) => ({

package/dist/src/agent/tools/personal-performance.d.ts

@@ -0,0 +1,2 @@
+import type { AgentTool } from '@earendil-works/pi-agent-core';
+export declare function withPersonalPerformance<T extends AgentTool>(tool: T, conversationId?: string, taskId?: string, runId?: string, assignmentEpoch?: number): T;