@sema-agent/core 3.0.0 → 5.0.0

package/dist/core/runner/prepare-task.js

@@ -29,9 +29,9 @@ import { reconcileInterruptedSession } from "../session-reconcile.js";
 import { CacheBreakDetector, toolsToFingerprintInputs } from "../cache-break-detector.js";
 import { reservedCollisions, reservedFor } from "../../brain/request-params.js";
 import { defineTool, isDefineToolProduct } from "../tools.js";
-import { canonicalToolName } from "../tool-name-aliases.js";
+import { RETIRED_TOOL_NAMES } from "../tool-name-aliases.js";
 import { pathToUri } from "../lsp-protocol.js";
-import { DEFAULT_TOOL_RESULT_THRESHOLD_CHARS, buildToolResultRef, firstPartyOffloadPolicy, InMemoryToolResultStore, RunnerSharedToolResultStore, ScopedToolResultStore, isVolatileOffloadStore, OFFLOAD_TOOL_NAME, createReadToolResultTool, withToolResultOffload, } from "../tool-result-store.js";
+import { DEFAULT_TOOL_RESULT_THRESHOLD_CHARS, createOffloadPersist, firstPartyOffloadPolicy, InMemoryToolResultStore, RunnerSharedToolResultStore, ScopedToolResultStore, isVolatileOffloadStore, OFFLOAD_TOOL_NAME, createReadToolResultTool, withToolResultOffload, } from "../tool-result-store.js";
 import { OUTPUT_TOOL_NAME, REPORT_FINDINGS_TOOL_NAME, SKILL_CONTENT_MAX_CHARS, SKILL_TOOL_NAME, createOutputTool, createReportBlockedTool, createReportFindingsTool, createSkillTool, normalizeSkills } from "./synthetic-tools.js";
 import { compileOutputSchema } from "./strict-output-schema.js";
 import { TOOL_SEARCH_NAME, buildDeferredRegistry, classifyDeferred, extractDiscoveredToolNames, createPlaceholderTool, createToolSearchTool, } from "./tool-disclosure.js";
@@ -51,6 +51,7 @@ import { dropOrphanToolResults, guardBudget, insertTrimNotice, trimToBudget } fr
 import { StubExecutionEnv } from "../stub-env.js";
 import { hasDestroy, isIsolated, isRemoteExecutionEnv, isRetryableRemoteErrorCode, isSuspendable, missingRestoreSurface, RETRYABLE_REMOTE_ERROR_CODES } from "../remote-env.js";
 import { withRetry } from "../with-retry.js";
+import { settleTeardownLeg } from "./teardown-bounded.js";
 import { hasBackgroundShell, sweepBackgroundShells } from "../background-shell.js";
 import { createTaskOutputTool, createTaskStopTool, defaultTaskRegistry } from "../task-registry.js";
 import { createMonitorTool } from "../../tools/monitor.js";
@@ -62,7 +63,7 @@ import { createSchedulerTools } from "../../tools/scheduler-tools.js";
 import { createPresentPlanTool, createEnterPlanModeTool, PRESENT_PLAN_TOOL_NAME, ENTER_PLAN_MODE_TOOL_NAME } from "../present-plan-tool.js";
 import { isSelfOrchestrationActive, selfOrchestrationFailClosedReason } from "../../orchestration/workflow-script-runner.js";
 import { createRunWorkflowTool, RUN_WORKFLOW_TOOL_NAME } from "../../orchestration/run-workflow-tool.js";
-import { createLspTool, gitCheckIgnoreFilter } from "../lsp.js";
+import { createLspTool, gitCheckIgnoreFilter, resolveLspPath } from "../lsp.js";
 import { resolveKey } from "../../tools/fs/safety.js";
 import { BINDING_CHECKPOINT_VERSION, mintCheckpointToken, RESOURCE_CHECKPOINT_VERSION, WALLTIME_CHECKPOINT_VERSION, buildRiskDescriptor, debitLedger, remainingBudgetMicroUsd, } from "../checkpoint-store.js";
 import { boundInputHashOf } from "../canonical-json.js";
@@ -442,18 +443,21 @@ export async function prepareTask(spec, deps, sessions, resume, internals, runne
     if (extraBodyCollisions.length > 0) {
         deps.onError?.(new Error(`model.extraBody contains reserved key(s) [${extraBodyCollisions.join(", ")}] that core owns and sets itself — they are ignored. Remove them from extraBody.`), { phase: "config", sessionId });
     }
-    const nestedStats = { tokens: 0, turns: 0, tasks: 0, costMicroUsd: 0 };
+    const nestedStats = { tokens: 0, turns: 0, tasks: 0, costMicroUsd: 0, anyUnpriced: false };
     if (resume) {
         nestedStats.tokens = resume.seed.nestedStats.tokens;
         nestedStats.turns = resume.seed.nestedStats.turns;
         nestedStats.tasks = resume.seed.nestedStats.tasks;
         nestedStats.costMicroUsd = resume.seed.nestedStats.costMicroUsd;
+        nestedStats.anyUnpriced = resume.seed.nestedStats.anyUnpriced ?? true;
     }
     const reportUsage = (u) => {
         nestedStats.tokens += u.tokens;
         nestedStats.turns += u.turns;
         nestedStats.tasks += u.tasks;
         nestedStats.costMicroUsd += u.costMicroUsd ?? 0;
+        if (u.costMicroUsd === undefined)
+            nestedStats.anyUnpriced = true;
     };
     const explicitGlobalThreshold = spec.toolResultThresholdChars ?? deps.toolResultThresholdChars;
     const offloadThreshold = explicitGlobalThreshold ?? DEFAULT_TOOL_RESULT_THRESHOLD_CHARS;
@@ -482,13 +486,13 @@ export async function prepareTask(spec, deps, sessions, resume, internals, runne
     if (offloadStore && (spec.tools ?? []).some((t) => t.name === OFFLOAD_TOOL_NAME)) {
         const e = new Error(`Tool name "${OFFLOAD_TOOL_NAME}" is reserved when large-result offload is enabled.`);
         e.code = "config.reserved_tool_name";
-        await forgetOnThrow();
+        await settleTeardownLeg(() => forgetOnThrow(), "forgetOnThrow (reserved-name leg)", (err) => deps.onError?.(err, { phase: "config", sessionId }));
         throw e;
     }
     let ownedEnv;
     let mcp;
     if (internals?.requestedCwd !== undefined && deps.executionEnvFactory === undefined) {
-        await forgetOnThrow();
+        await settleTeardownLeg(() => forgetOnThrow(), "forgetOnThrow (cwd-unsupported leg)", (err) => deps.onError?.(err, { phase: "config", sessionId }));
         const e = new Error(`Agent cwd "${internals.requestedCwd}" cannot take effect: this deployment has no executionEnvFactory (a static execution environment cannot be re-rooted per agent). Drop the cwd parameter or deploy a factory.`);
         e.code = "config.cwd_unsupported";
         throw e;
@@ -504,7 +508,7 @@ export async function prepareTask(spec, deps, sessions, resume, internals, runne
             : undefined;
     }
     catch (factoryErr) {
-        await forgetOnThrow();
+        await settleTeardownLeg(() => forgetOnThrow(), "forgetOnThrow (env-factory-throw leg)", (err) => deps.onError?.(err, { phase: "config", sessionId }));
         throw factoryErr;
     }
     const executionEnv = ownedEnv ?? deps.executionEnv ?? new StubExecutionEnv();
@@ -515,13 +519,8 @@ export async function prepareTask(spec, deps, sessions, resume, internals, runne
         };
         const want = await canon(internals.requestedCwd);
         const rejectCwd = async (message, code) => {
-            await forgetOnThrow();
-            try {
-                if (ownedEnv && hasDestroy(ownedEnv))
-                    await ownedEnv.destroy();
-            }
-            catch {
-            }
+            await settleTeardownLeg(() => forgetOnThrow(), "forgetOnThrow (cwd-reject leg)", (err) => deps.onError?.(err, { phase: "config", sessionId }));
+            await settleTeardownLeg(() => (ownedEnv && hasDestroy(ownedEnv) ? ownedEnv.destroy() : undefined), "ownedEnv.destroy (cwd-reject leg)", (err) => deps.onError?.(err, { phase: "config", sessionId }));
             const e = new Error(message);
             e.code = code;
             throw e;
@@ -1403,6 +1402,7 @@ export async function prepareTask(spec, deps, sessions, resume, internals, runne
                 ...(internals?.onTaskNotification !== undefined ? { onTaskNotification: internals.onTaskNotification } : {}),
                 ...(handsCwdRef !== undefined ? { cwdRef: handsCwdRef } : {}),
                 ...(offloadStore !== undefined ? { toolResultStore: offloadStore } : {}),
+                ...(spec.retainBackgroundProcesses === true ? { retainBackgroundProcesses: true } : {}),
             })));
         }
         if (handsCwdRef !== undefined) {
@@ -1466,30 +1466,7 @@ export async function prepareTask(spec, deps, sessions, resume, internals, runne
         const nudgeLspOnEdit = lspDiagnostics
             ? (rawPath) => {
                 const baseDir = handsCwdRef?.current ?? taskRootPath;
-                const isAbsForm = rawPath.startsWith("/") || /^[A-Za-z]:[\\/]/.test(rawPath);
-                const joinSep = baseDir.includes("\\") ? "\\" : "/";
-                const normalizeSegments = (pth) => {
-                    const isUnc = pth.startsWith("\\\\");
-                    const drive = /^[A-Za-z]:/.exec(pth)?.[0] ?? "";
-                    const rest = pth.slice(drive.length + (isUnc ? 2 : 0));
-                    const sep = pth.includes("\\") ? "\\" : "/";
-                    const abs = rest.startsWith("/") || rest.startsWith("\\");
-                    const outSegs = [];
-                    for (const seg of rest.split(/[\\/]+/)) {
-                        if (!seg || seg === ".")
-                            continue;
-                        if (seg === "..") {
-                            if (outSegs.length > 0 && outSegs[outSegs.length - 1] !== "..")
-                                outSegs.pop();
-                            else if (!abs)
-                                outSegs.push("..");
-                            continue;
-                        }
-                        outSegs.push(seg);
-                    }
-                    return (isUnc ? sep + sep : "") + drive + (abs ? sep : "") + outSegs.join(sep);
-                };
-                const filePath = normalizeSegments(isAbsForm ? rawPath : baseDir.replace(/[\\/]+$/, "") + joinSep + rawPath);
+                const filePath = resolveLspPath(rawPath, baseDir);
                 lspDiagnostics.fileEdited(pathToUri(filePath));
                 void lspManager
                     .sessionFor(filePath, undefined, executionEnv)
@@ -2041,9 +2018,9 @@ export async function prepareTask(spec, deps, sessions, resume, internals, runne
             };
         }
         if (spec.agents !== undefined && spec.agents.length > 0) {
-            const known = new Set(tools.flatMap((t) => [canonicalToolName(t.name), ...(t.aliases ?? []).map((a) => canonicalToolName(a))]));
+            const known = new Set(tools.flatMap((t) => [t.name, ...(t.aliases ?? [])]));
             for (const def of spec.agents) {
-                const unknownAllow = (def.allowTools ?? []).filter((n) => n !== "*" && !known.has(canonicalToolName(n)));
+                const unknownAllow = (def.allowTools ?? []).filter((n) => n !== "*" && !known.has(n));
                 if (unknownAllow.length > 0) {
                     try {
                         deps.onError?.(new Error(`TaskSpec.agents: agent "${def.name}" allows tool(s) ${unknownAllow.join(", ")} not present in this task's assembled roster — likely a typo (the entry would be item-filtered at spawn; the agent stays usable). Advisory only: the delegation pool can differ from this roster, so a tool mounted only on the delegation tool or produced by per-spawn extraTools makes this spurious.`), { phase: "config", sessionId });
@@ -2054,7 +2031,7 @@ export async function prepareTask(spec, deps, sessions, resume, internals, runne
                 for (const n of def.denyTools ?? []) {
                     if (n === "*")
                         continue;
-                    if (!known.has(canonicalToolName(n))) {
+                    if (!known.has(n)) {
                         const e = new Error(`TaskSpec.agents: agent "${def.name}" declares tool "${n}" in its denied tools, but no such tool exists in this deployment — fix the agent's tools list or mount the tool.`);
                         e.code = "config.agent.unknown_tool";
                         throw e;
@@ -2216,15 +2193,22 @@ export async function prepareTask(spec, deps, sessions, resume, internals, runne
         {
             const nameGroups = toolPolicyNameSets(policy);
             if (nameGroups.length > 0) {
-                const known = new Set(tools.flatMap((t) => [canonicalToolName(t.name), ...(t.aliases ?? []).map((a) => canonicalToolName(a))]));
+                const known = new Set(tools.flatMap((t) => [t.name, ...(t.aliases ?? [])]));
                 for (const t of harnessTools)
-                    known.add(canonicalToolName(t.name));
+                    known.add(t.name);
                 const unmatched = new Set();
                 for (const g of nameGroups) {
                     for (const [kind, list] of [["deny", g.deny], ["ask", g.ask], ["allow", g.allow]]) {
                         for (const n of list ?? []) {
-                            if (!known.has(canonicalToolName(n)))
-                                unmatched.add(`"${n}" (${kind})`);
+                            if (known.has(n))
+                                continue;
+                            const retired = RETIRED_TOOL_NAMES.get(n);
+                            if (retired !== undefined) {
+                                const err = new Error(`tool policy ${kind}-list entry "${n}" is a RETIRED tool name (${retired}) and matches nothing in this run's roster — legacy-name normalization was removed (RB-476-A), so this entry would silently guard nothing. Update the deployment's rule to the current name.`);
+                                err.code = "config.legacy_tool_name";
+                                throw err;
+                            }
+                            unmatched.add(`"${n}" (${kind})`);
                         }
                     }
                 }
@@ -3183,6 +3167,7 @@ export async function prepareTask(spec, deps, sessions, resume, internals, runne
                     let result;
                     try {
                         result = await runToolGate({
+                            onNotifyError: (f) => emitTrace(deps.tracer, () => ({ kind: "observer.notify_failed", version: 1, taskId: spec.taskId ?? sessionId, site: f.site, message: f.error.message, ts: Date.now() })),
                             event: e,
                             preToolUse: hooks?.preToolUse,
                             adjudicate,
@@ -3296,11 +3281,7 @@ export async function prepareTask(spec, deps, sessions, resume, internals, runne
                 ...(offloadStore
                     ? {
                         offload: {
-                            persist: (toolCallId, fullText) => {
-                                const ref = buildToolResultRef(sessionId, toolCallId);
-                                void offloadStore.put(ref, fullText);
-                                return ref;
-                            },
+                            persist: createOffloadPersist(offloadStore, sessionId),
                         },
                     }
                     : {}),
@@ -3501,20 +3482,14 @@ export async function prepareTask(spec, deps, sessions, resume, internals, runne
     }
     catch (prepareErr) {
         if (mcp) {
-            try {
-                await mcp.dispose();
-            }
-            catch {
-            }
+            const mcpHandle = mcp;
+            await settleTeardownLeg(() => mcpHandle.dispose(), "mcp.dispose (prepare-throw leg)", (e) => deps.onError?.(e, { phase: "config", sessionId }));
         }
         if (ownedEnv && hasDestroy(ownedEnv)) {
-            try {
-                await ownedEnv.destroy();
-            }
-            catch {
-            }
+            const env = ownedEnv;
+            await settleTeardownLeg(() => env.destroy(), "ownedEnv.destroy (prepare-throw leg)", (e) => deps.onError?.(e, { phase: "config", sessionId }));
         }
-        await forgetOnThrow();
+        await settleTeardownLeg(() => forgetOnThrow(), "forgetOnThrow (prepare-throw leg)", (e) => deps.onError?.(e, { phase: "config", sessionId }));
         throw prepareErr;
     }
 }

package/dist/core/runner/runtask.d.ts

@@ -19,6 +19,7 @@ interface ResumeRun {
         trusted: boolean;
     };
     onEnvRestoreFailed?: (reason: ReopenReason) => Promise<void>;
+    decisionDelivered?: boolean;
 }
 export declare class Runner {
     private deps;