@cursor/july 0.1.12 → 0.1.15

package/dist/internal/eval-runner.js

@@ -139,27 +139,79 @@ export function runDiscoveredEvals(options) {
         }
         const maxConcurrency = resolveEvalMaxConcurrency(config);
         const defaultTimeoutMs = (_c = (_b = options.timeoutMs) !== null && _b !== void 0 ? _b : config.timeoutMs) !== null && _c !== void 0 ? _c : 180000;
-        return mapPool(selected, maxConcurrency, (discovered) => __awaiter(this, void 0, void 0, function* () {
-            var _a, _b;
-            if (options.verbose) {
-                process.stderr.write(`  eval ${discovered.id}…\n`);
+        // Iterations of the same authored datapoint share one concurrency slot and
+        // run sequentially. maxConcurrency is therefore the max number of distinct
+        // datapoints in flight, not datapoints × iterations.
+        const groups = groupEvalsForIterationSchedule(selected);
+        const nested = yield mapPool(groups, maxConcurrency, (group) => __awaiter(this, void 0, void 0, function* () {
+            var _a, _b, _c;
+            const results = [];
+            for (const discovered of group) {
+                if ((_a = options.signal) === null || _a === void 0 ? void 0 : _a.aborted) {
+                    break;
+                }
+                if (options.verbose) {
+                    process.stderr.write(`  eval ${discovered.id}…\n`);
+                }
+                (_b = options.onCaseStart) === null || _b === void 0 ? void 0 : _b.call(options, discovered.id);
+                const result = yield runOneEval(discovered, {
+                    baseUrl: options.baseUrl,
+                    defaultTimeoutMs,
+                    bearerToken: options.bearerToken,
+                    headers: options.headers,
+                    verbose: Boolean(options.verbose),
+                    signal: options.signal,
+                    onEvent: (event) => {
+                        var _a, _b;
+                        (_a = options.onCaseEvent) === null || _a === void 0 ? void 0 : _a.call(options, discovered.id, event);
+                        (_b = options.onEvent) === null || _b === void 0 ? void 0 : _b.call(options, event);
+                    },
+                });
+                (_c = options.onCaseDone) === null || _c === void 0 ? void 0 : _c.call(options, result);
+                results.push(result);
             }
-            (_a = options.onCaseStart) === null || _a === void 0 ? void 0 : _a.call(options, discovered.id);
-            const result = yield runOneEval(discovered, {
-                baseUrl: options.baseUrl,
-                defaultTimeoutMs,
-                bearerToken: options.bearerToken,
-                headers: options.headers,
-                verbose: Boolean(options.verbose),
-                onEvent: (event) => {
-                    var _a, _b;
-                    (_a = options.onCaseEvent) === null || _a === void 0 ? void 0 : _a.call(options, discovered.id, event);
-                    (_b = options.onEvent) === null || _b === void 0 ? void 0 : _b.call(options, event);
-                },
-            });
-            (_b = options.onCaseDone) === null || _b === void 0 ? void 0 : _b.call(options, result);
-            return result;
-        }));
+            return results;
+        }), { signal: options.signal });
+        return nested.flat();
+    });
+}
+/**
+ * Schedule key for {@link groupEvalsForIterationSchedule}: iteration siblings
+ * (`weather/nyc/1`…`/3`) share a key; single-iteration ids keep their full id.
+ */
+export function iterationScheduleKey(discovered) {
+    var _a, _b;
+    const iterations = (_a = discovered.definition.iterations) !== null && _a !== void 0 ? _a : 1;
+    const iteration = (_b = discovered.definition.iteration) !== null && _b !== void 0 ? _b : 1;
+    if (iterations <= 1) {
+        return discovered.id;
+    }
+    const suffix = `/${iteration}`;
+    if (discovered.id.endsWith(suffix)) {
+        return discovered.id.slice(0, -suffix.length);
+    }
+    return discovered.id;
+}
+/**
+ * Group expanded iteration cases so each authored datapoint's iterations run
+ * back-to-back under one {@link mapPool} slot (sorted by iteration index).
+ */
+export function groupEvalsForIterationSchedule(evals) {
+    const groups = new Map();
+    const order = [];
+    for (const discovered of evals) {
+        const key = iterationScheduleKey(discovered);
+        let group = groups.get(key);
+        if (group === undefined) {
+            group = [];
+            groups.set(key, group);
+            order.push(key);
+        }
+        group.push(discovered);
+    }
+    return order.map((key) => {
+        const group = groups.get(key);
+        return group.sort((a, b) => { var _a, _b; return ((_a = a.definition.iteration) !== null && _a !== void 0 ? _a : 1) - ((_b = b.definition.iteration) !== null && _b !== void 0 ? _b : 1); });
     });
 }
 function resolveEvalMaxConcurrency(config) {
@@ -173,16 +225,25 @@ function resolveEvalMaxConcurrency(config) {
 function runOneEval(discovered, opts) {
     return __awaiter(this, void 0, void 0, function* () {
         var _a;
+        var _b;
         const logs = [];
+        const metrics = {};
         const assertions = [];
         const def = discovered.definition;
-        const timeoutMs = (_a = def.timeoutMs) !== null && _a !== void 0 ? _a : opts.defaultTimeoutMs;
+        const timeoutMs = (_b = def.timeoutMs) !== null && _b !== void 0 ? _b : opts.defaultTimeoutMs;
         const startedAt = Date.now();
         let sessionId;
         let continuationToken;
         const allEvents = [];
         const inputs = [];
         let reply = null;
+        if ((_a = opts.signal) === null || _a === void 0 ? void 0 : _a.aborted) {
+            return Object.assign(Object.assign({ id: discovered.id, path: discovered.path }, (def.description === undefined
+                ? {}
+                : { description: def.description })), { ok: false, assertions, error: "cancelled", logs,
+                metrics,
+                inputs, toolCalls: [], tools: [], durationMs: Date.now() - startedAt });
+        }
         const record = (name, passed, detail) => {
             assertions.push(Object.assign({ name,
                 passed }, (detail === undefined ? {} : { detail })));
@@ -199,6 +260,14 @@ function runOneEval(discovered, opts) {
             get events() {
                 return allEvents;
             },
+            get iteration() {
+                var _a;
+                return (_a = def.iteration) !== null && _a !== void 0 ? _a : 1;
+            },
+            get iterations() {
+                var _a;
+                return (_a = def.iterations) !== null && _a !== void 0 ? _a : 1;
+            },
             send(message_1) {
                 return __awaiter(this, arguments, void 0, function* (message, options = {}) {
                     inputs.push(message);
@@ -213,7 +282,9 @@ function runOneEval(discovered, opts) {
                         headers: opts.headers,
                         workspaceDir: options.workspaceDir,
                         workspaceFiles: options.workspaceFiles,
+                        cloud: options.cloud,
                         timeoutMs,
+                        signal: opts.signal,
                         onEvent: opts.onEvent,
                     });
                     sessionId = result.sessionId;
@@ -258,6 +329,16 @@ function runOneEval(discovered, opts) {
                     process.stderr.write(`    ${message}\n`);
                 }
             },
+            metric(name, value) {
+                const key = name.trim();
+                if (key === "") {
+                    return;
+                }
+                metrics[key] = value;
+                if (opts.verbose) {
+                    process.stderr.write(`    metric ${key}=${String(value)}\n`);
+                }
+            },
         };
         try {
             yield def.test(t);
@@ -270,6 +351,7 @@ function runOneEval(discovered, opts) {
             return Object.assign(Object.assign(Object.assign(Object.assign(Object.assign({ id: discovered.id, path: discovered.path }, (def.description === undefined
                 ? {}
                 : { description: def.description })), { ok: false, assertions, error: message, logs,
+                metrics,
                 inputs, toolCalls: toolCallIo() }), (sessionId === undefined ? {} : { sessionId })), (reply === null ? {} : { finalText: reply })), { tools: toolsUsed(), durationMs: Date.now() - startedAt });
         }
         const traj = buildTrajectory(allEvents);
@@ -279,6 +361,7 @@ function runOneEval(discovered, opts) {
         return Object.assign(Object.assign(Object.assign(Object.assign({ id: discovered.id, path: discovered.path }, (def.description === undefined ? {} : { description: def.description })), { ok,
             assertions,
             logs,
+            metrics,
             inputs, toolCalls: toolCallIo() }), (sessionId === undefined ? {} : { sessionId })), { finalText: reply !== null && reply !== void 0 ? reply : traj.finalText, tools: toolsUsed(), durationMs: Date.now() - startedAt });
     });
 }
@@ -297,7 +380,13 @@ function matchExpectation(value, assertion) {
         }
         case "equals": {
             const passed = deepEqual(value, assertion.value);
-            return Object.assign({ passed }, (passed ? {} : { detail: `got ${JSON.stringify(value)}` }));
+            // Always attach a short detail so values like recall `40.0% (4/10)`
+            // show up on PASS lines / playground, not only on failure.
+            const rendered = typeof value === "string" ? value : JSON.stringify(value);
+            return {
+                passed,
+                detail: passed ? rendered : `got ${rendered}`,
+            };
         }
         case "satisfies": {
             const passed = assertion.predicate(value);

package/dist/internal/evals-client.d.ts

@@ -9,7 +9,7 @@
 import type { EvalRunSnapshot } from "../evals.js";
 import type { EvalRunResult } from "./eval-runner.js";
 import { type RequestAuth } from "./request-headers.js";
-export type RemoteEvalErrorCode = "no_matching_evals" | "eval_run_in_progress" | "unknown_eval_run" | "start_failed" | "poll_failed" | "aborted";
+export type RemoteEvalErrorCode = "no_matching_evals" | "eval_run_in_progress" | "eval_run_not_running" | "unknown_eval_run" | "start_failed" | "poll_failed" | "cancel_failed" | "aborted";
 export declare class RemoteEvalError extends Error {
     readonly status?: number | undefined;
     readonly code?: RemoteEvalErrorCode | undefined;
@@ -65,6 +65,14 @@ export declare function startRemoteEvalRun(args: {
     verbose?: boolean;
     signal?: AbortSignal;
 }): Promise<EvalRunSnapshot>;
+/** Cancel a running batch via `POST /v1/dev/evals/runs/:runId/cancel`. */
+export declare function cancelRemoteEvalRun(args: {
+    baseUrl: string;
+    runId: string;
+    auth?: RequestAuth;
+    fetchImpl?: typeof fetch;
+    signal?: AbortSignal;
+}): Promise<EvalRunSnapshot>;
 /** One-shot `GET /v1/dev/evals/runs/:runId` (no polling). */
 export declare function getRemoteEvalRun(args: {
     baseUrl: string;

package/dist/internal/evals-client.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"evals-client.d.ts","sourceRoot":"","sources":["../../src/internal/evals-client.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAEnD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AACtD,OAAO,EAAuB,KAAK,WAAW,EAAE,MAAM,sBAAsB,CAAC;AAI7E,MAAM,MAAM,mBAAmB,GAC3B,mBAAmB,GACnB,sBAAsB,GACtB,kBAAkB,GAClB,cAAc,GACd,aAAa,GACb,SAAS,CAAC;AAEd,qBAAa,eAAgB,SAAQ,KAAK;IAGtC,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM;IACxB,QAAQ,CAAC,IAAI,CAAC,EAAE,mBAAmB;IACnC,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM;IAJ/B,YACE,OAAO,EAAE,MAAM,EACN,MAAM,CAAC,EAAE,MAAM,YAAA,EACf,IAAI,CAAC,EAAE,mBAAmB,YAAA,EAC1B,WAAW,CAAC,EAAE,MAAM,YAAA,EAI9B;CACF;AAED,MAAM,WAAW,kBAAkB;IACjC,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;CACjB;AAED,MAAM,WAAW,qBAAsB,SAAQ,WAAW;IACxD,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,gEAAgE;IAChE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,qEAAqE;IACrE,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,kDAAkD;IAClD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,6CAA6C;IAC7C,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,oDAAoD;IACpD,UAAU,CAAC,EAAE,CAAC,IAAI,EAAE,eAAe,KAAK,IAAI,CAAC;CAC9C;AAED;;;GAGG;AACH,wBAAsB,cAAc,CAClC,OAAO,EAAE,qBAAqB,GAC7B,OAAO,CAAC;IAAE,QAAQ,EAAE,eAAe,CAAC;IAAC,OAAO,EAAE,aAAa,EAAE,CAAA;CAAE,CAAC,CA+BlE;AAED,iFAAiF;AACjF,wBAAsB,eAAe,CAAC,IAAI,EAAE;IAC1C,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,GAAG,OAAO,CAAC,kBAAkB,EAAE,CAAC,CAgEhC;AAED,wBAAsB,kBAAkB,CAAC,IAAI,EAAE;IAC7C,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,GAAG,OAAO,CAAC,eAAe,CAAC,CAuF3B;AAED,6DAA6D;AAC7D,wBAAsB,gBAAgB,CAAC,IAAI,EAAE;IAC3C,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,GAAG,OAAO,CAAC,eAAe,CAAC,CAmC3B;AAED,oDAAoD;AACpD,wBAAsB,kBAAkB,CAAC,IAAI,EAAE;IAC7C,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,GAAG,OAAO,CAAC;IAAE,IAAI,EAAE,eAAe,EAAE,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC,CAyC7D;AAED,wBAAsB,iBAAiB,CAAC,IAAI,EAAE;IAC5C,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,UAAU,CAAC,EAAE,CAAC,IAAI,EAAE,eAAe,KAAK,IAAI,CAAC;CAC9C,GAAG,OAAO,CAAC,eAAe,CAAC,CAoB3B;AAED,wBAAgB,wBAAwB,CACtC,QAAQ,EAAE,eAAe,GACxB,aAAa,EAAE,CAgBjB"}
+{"version":3,"file":"evals-client.d.ts","sourceRoot":"","sources":["../../src/internal/evals-client.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAEnD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AACtD,OAAO,EAAuB,KAAK,WAAW,EAAE,MAAM,sBAAsB,CAAC;AAI7E,MAAM,MAAM,mBAAmB,GAC3B,mBAAmB,GACnB,sBAAsB,GACtB,sBAAsB,GACtB,kBAAkB,GAClB,cAAc,GACd,aAAa,GACb,eAAe,GACf,SAAS,CAAC;AAEd,qBAAa,eAAgB,SAAQ,KAAK;IAGtC,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM;IACxB,QAAQ,CAAC,IAAI,CAAC,EAAE,mBAAmB;IACnC,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM;IAJ/B,YACE,OAAO,EAAE,MAAM,EACN,MAAM,CAAC,EAAE,MAAM,YAAA,EACf,IAAI,CAAC,EAAE,mBAAmB,YAAA,EAC1B,WAAW,CAAC,EAAE,MAAM,YAAA,EAI9B;CACF;AAED,MAAM,WAAW,kBAAkB;IACjC,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;CACjB;AAED,MAAM,WAAW,qBAAsB,SAAQ,WAAW;IACxD,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,gEAAgE;IAChE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,qEAAqE;IACrE,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,kDAAkD;IAClD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,6CAA6C;IAC7C,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,oDAAoD;IACpD,UAAU,CAAC,EAAE,CAAC,IAAI,EAAE,eAAe,KAAK,IAAI,CAAC;CAC9C;AAED;;;GAGG;AACH,wBAAsB,cAAc,CAClC,OAAO,EAAE,qBAAqB,GAC7B,OAAO,CAAC;IAAE,QAAQ,EAAE,eAAe,CAAC;IAAC,OAAO,EAAE,aAAa,EAAE,CAAA;CAAE,CAAC,CA+BlE;AAED,iFAAiF;AACjF,wBAAsB,eAAe,CAAC,IAAI,EAAE;IAC1C,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,GAAG,OAAO,CAAC,kBAAkB,EAAE,CAAC,CAgEhC;AAED,wBAAsB,kBAAkB,CAAC,IAAI,EAAE;IAC7C,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,GAAG,OAAO,CAAC,eAAe,CAAC,CAuF3B;AAED,0EAA0E;AAC1E,wBAAsB,mBAAmB,CAAC,IAAI,EAAE;IAC9C,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,GAAG,OAAO,CAAC,eAAe,CAAC,CAsE3B;AAED,6DAA6D;AAC7D,wBAAsB,gBAAgB,CAAC,IAAI,EAAE;IAC3C,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,GAAG,OAAO,CAAC,eAAe,CAAC,CAmC3B;AAED,oDAAoD;AACpD,wBAAsB,kBAAkB,CAAC,IAAI,EAAE;IAC7C,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,GAAG,OAAO,CAAC;IAAE,IAAI,EAAE,eAAe,EAAE,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC,CAyC7D;AAED,wBAAsB,iBAAiB,CAAC,IAAI,EAAE;IAC5C,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,KAAK,CAAC;IACzB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,UAAU,CAAC,EAAE,CAAC,IAAI,EAAE,eAAe,KAAK,IAAI,CAAC;CAC9C,GAAG,OAAO,CAAC,eAAe,CAAC,CAoB3B;AAED,wBAAgB,wBAAwB,CACtC,QAAQ,EAAE,eAAe,GACxB,aAAa,EAAE,CAiBjB"}

package/dist/internal/evals-client.js

@@ -162,6 +162,49 @@ export function startRemoteEvalRun(args) {
         return json;
     });
 }
+/** Cancel a running batch via `POST /v1/dev/evals/runs/:runId/cancel`. */
+export function cancelRemoteEvalRun(args) {
+    return __awaiter(this, void 0, void 0, function* () {
+        var _a;
+        const baseUrl = args.baseUrl.replace(/\/$/, "");
+        const fetchImpl = (_a = args.fetchImpl) !== null && _a !== void 0 ? _a : fetch;
+        let resp;
+        try {
+            resp = yield fetchImpl(`${baseUrl}/v1/dev/evals/runs/${encodeURIComponent(args.runId)}/cancel`, {
+                method: "POST",
+                headers: mergeRequestHeaders(args.auth),
+                signal: args.signal,
+            });
+        }
+        catch (error) {
+            throw new RemoteEvalError(error instanceof Error ? error.message : "Could not cancel eval run.", undefined, "cancel_failed");
+        }
+        const json = (yield resp.json().catch(() => ({})));
+        if (resp.status === 404 || json.error === "unknown_eval_run") {
+            throw new RemoteEvalError(`Unknown eval run ${args.runId}.`, 404, "unknown_eval_run");
+        }
+        if (resp.status === 409 || json.error === "eval_run_not_running") {
+            const status = typeof json.status === "string" && json.status !== ""
+                ? json.status
+                : "unknown";
+            throw new RemoteEvalError(typeof json.message === "string" && json.message !== ""
+                ? json.message
+                : `Eval run ${args.runId} is not running (status: ${status}).`, 409, "eval_run_not_running");
+        }
+        if (!resp.ok) {
+            const detail = typeof json.message === "string" && json.message !== ""
+                ? json.message
+                : typeof json.error === "string" && json.error !== ""
+                    ? json.error
+                    : `HTTP ${resp.status}`;
+            throw new RemoteEvalError(`Could not cancel eval run: ${detail}`, resp.status, "cancel_failed");
+        }
+        if (typeof json.runId !== "string" || json.runId === "") {
+            throw new RemoteEvalError("Eval cancel response missing runId.", undefined, "cancel_failed");
+        }
+        return json;
+    });
+}
 /** One-shot `GET /v1/dev/evals/runs/:runId` (no polling). */
 export function getRemoteEvalRun(args) {
     return __awaiter(this, void 0, void 0, function* () {
@@ -241,7 +284,7 @@ export function pollRemoteEvalRun(args) {
     });
 }
 export function snapshotToEvalRunResults(snapshot) {
-    return snapshot.cases.map((c) => { var _a, _b, _c, _d; return (Object.assign(Object.assign(Object.assign(Object.assign(Object.assign(Object.assign(Object.assign(Object.assign(Object.assign({ id: c.id, path: c.fileId }, (c.description === undefined ? {} : { description: c.description })), { ok: c.ok === true, assertions: (_a = c.assertions) !== null && _a !== void 0 ? _a : [] }), (c.sessionId === undefined ? {} : { sessionId: c.sessionId })), { inputs: (_b = c.inputs) !== null && _b !== void 0 ? _b : [] }), (c.finalText === undefined ? {} : { finalText: c.finalText })), (c.tools === undefined ? {} : { tools: c.tools })), (c.toolCalls === undefined ? {} : { toolCalls: c.toolCalls })), (c.error === undefined ? {} : { error: c.error })), { logs: (_c = c.logs) !== null && _c !== void 0 ? _c : [], durationMs: (_d = c.durationMs) !== null && _d !== void 0 ? _d : 0 })); });
+    return snapshot.cases.map((c) => { var _a, _b, _c, _d, _e; return (Object.assign(Object.assign(Object.assign(Object.assign(Object.assign(Object.assign(Object.assign(Object.assign(Object.assign({ id: c.id, path: c.fileId }, (c.description === undefined ? {} : { description: c.description })), { ok: c.ok === true, assertions: (_a = c.assertions) !== null && _a !== void 0 ? _a : [] }), (c.sessionId === undefined ? {} : { sessionId: c.sessionId })), { inputs: (_b = c.inputs) !== null && _b !== void 0 ? _b : [] }), (c.finalText === undefined ? {} : { finalText: c.finalText })), (c.tools === undefined ? {} : { tools: c.tools })), (c.toolCalls === undefined ? {} : { toolCalls: c.toolCalls })), (c.error === undefined ? {} : { error: c.error })), { logs: (_c = c.logs) !== null && _c !== void 0 ? _c : [], metrics: (_d = c.metrics) !== null && _d !== void 0 ? _d : {}, durationMs: (_e = c.durationMs) !== null && _e !== void 0 ? _e : 0 })); });
 }
 function sleep(ms, signal) {
     return new Promise((resolve, reject) => {

package/dist/internal/http-channel.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"http-channel.d.ts","sourceRoot":"","sources":["../../src/internal/http-channel.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,KAAK,EACV,WAAW,EACX,UAAU,EACV,qBAAqB,EAErB,aAAa,EACd,MAAM,aAAa,CAAC;AAQrB,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAC/C,OAAO,EAML,KAAK,aAAa,EACnB,MAAM,qBAAqB,CAAC;AAQ7B,wBAAgB,sBAAsB,CAAC,IAAI,EAAE;IAC3C,MAAM,EAAE,aAAa,CAAC;IACtB,MAAM,EAAE,qBAAqB,CAAC;IAC9B,IAAI,EAAE,UAAU,EAAE,CAAC;IACnB,4DAA4D;IAC5D,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB;;;;;;OAMG;IACH,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAC/B;;;OAGG;IACH,kCAAkC,CAAC,EAAE,OAAO,CAAC;IAC7C;;;;;OAKG;IACH,wBAAwB,CAAC,EAAE,OAAO,CAAC;CACpC,GAAG,WAAW,EAAE,CAkYhB;AAED,wBAAgB,mBAAmB,CACjC,MAAM,EAAE,aAAa,EACrB,MAAM,EAAE,WAAW,GAAG,IAAI,EAC1B,OAAO,CAAC,EAAE;IACR,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CAC3B,GACA,QAAQ,GAAG,SAAS,CAiBtB;AAED,wBAAgB,YAAY,CAC1B,IAAI,EAAE,OAAO,EACb,IAAI,GAAE;IAAE,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CAAO,GAC/D,QAAQ,CAQV;AAED,wBAAgB,aAAa,CAC3B,MAAM,EAAE,MAAM,EACd,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,MAAM,GACd,QAAQ,CAEV;AA0BD,wBAAsB,cAAc,CAClC,OAAO,EAAE,OAAO,GACf,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,SAAS,CAAC,CAc9C"}
+{"version":3,"file":"http-channel.d.ts","sourceRoot":"","sources":["../../src/internal/http-channel.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,KAAK,EAEV,WAAW,EACX,UAAU,EACV,qBAAqB,EAErB,aAAa,EACd,MAAM,aAAa,CAAC;AAQrB,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAC/C,OAAO,EAML,KAAK,aAAa,EACnB,MAAM,qBAAqB,CAAC;AAQ7B,wBAAgB,sBAAsB,CAAC,IAAI,EAAE;IAC3C,MAAM,EAAE,aAAa,CAAC;IACtB,MAAM,EAAE,qBAAqB,CAAC;IAC9B,IAAI,EAAE,UAAU,EAAE,CAAC;IACnB,4DAA4D;IAC5D,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB;;;;;;OAMG;IACH,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAC/B;;;OAGG;IACH,kCAAkC,CAAC,EAAE,OAAO,CAAC;IAC7C;;;;;OAKG;IACH,wBAAwB,CAAC,EAAE,OAAO,CAAC;CACpC,GAAG,WAAW,EAAE,CAwYhB;AAED,wBAAgB,mBAAmB,CACjC,MAAM,EAAE,aAAa,EACrB,MAAM,EAAE,WAAW,GAAG,IAAI,EAC1B,OAAO,CAAC,EAAE;IACR,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CAC3B,GACA,QAAQ,GAAG,SAAS,CAiBtB;AAED,wBAAgB,YAAY,CAC1B,IAAI,EAAE,OAAO,EACb,IAAI,GAAE;IAAE,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CAAO,GAC/D,QAAQ,CAQV;AAED,wBAAgB,aAAa,CAC3B,MAAM,EAAE,MAAM,EACd,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,MAAM,GACd,QAAQ,CAEV;AAoGD,wBAAsB,cAAc,CAClC,OAAO,EAAE,OAAO,GACf,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,SAAS,CAAC,CAc9C"}

package/dist/internal/http-channel.js

@@ -50,7 +50,10 @@ export function buildHttpChannelRoutes(args) {
             }
             const message = yield transformMessage(body.message, ctx.auth);
             const workspaceFiles = readWorkspaceFiles(body.workspaceFiles);
-            const session = yield engine.send(HTTP_CHANNEL_ID, message, Object.assign(Object.assign(Object.assign(Object.assign({ auth: ctx.auth }, (body.state === undefined ? {} : { state: body.state })), (typeof body.title === "string" ? { title: body.title } : {})), (body.purpose === "eval" ? { purpose: "eval" } : {})), (workspaceFiles === undefined ? {} : { workspaceFiles })));
+            // Per-session cloud attach (create only). Eval harness uses this to pin
+            // fixture repos without putting them on the agent's default cloud.repos.
+            const cloud = readCloudOptions(body.cloud);
+            const session = yield engine.send(HTTP_CHANNEL_ID, message, Object.assign(Object.assign(Object.assign(Object.assign(Object.assign({ auth: ctx.auth }, (body.state === undefined ? {} : { state: body.state })), (typeof body.title === "string" ? { title: body.title } : {})), (body.purpose === "eval" ? { purpose: "eval" } : {})), (workspaceFiles === undefined ? {} : { workspaceFiles })), (cloud === undefined ? {} : { cloud })));
             return jsonResponse({
                 ok: true,
                 sessionId: session.id,
@@ -67,7 +70,7 @@ export function buildHttpChannelRoutes(args) {
         handler: (request, ctx) => __awaiter(this, void 0, void 0, function* () {
             var _a;
             const sessionId = (_a = ctx.params.sessionId) !== null && _a !== void 0 ? _a : "";
-            const record = yield engine.sessions.get(sessionId);
+            const record = yield engine.getOrRestoreSession(sessionId);
             if (record === undefined) {
                 return errorResponse(404, "session_not_found", `Unknown session ${sessionId}.`);
             }
@@ -138,7 +141,9 @@ export function buildHttpChannelRoutes(args) {
             // custom channels or schedules are observable here too when the caller
             // matches the session's stored principal (or `--dev` relax). Follow-ups
             // use the same ownership gate plus the session's continuation token.
-            const record = yield engine.sessions.get(sessionId);
+            // Lazy-restore by id so playground Open-trace works after redeploy
+            // when the session is durable but not in the local state root yet.
+            const record = yield engine.getOrRestoreSession(sessionId);
             if (record === undefined) {
                 return errorResponse(404, "session_not_found", `Unknown session ${sessionId}.`);
             }
@@ -175,7 +180,7 @@ export function buildHttpChannelRoutes(args) {
         handler: (request, ctx) => __awaiter(this, void 0, void 0, function* () {
             var _a;
             const sessionId = (_a = ctx.params.sessionId) !== null && _a !== void 0 ? _a : "";
-            const record = yield engine.sessions.get(sessionId);
+            const record = yield engine.getOrRestoreSession(sessionId);
             if (record === undefined) {
                 return errorResponse(404, "session_not_found", `Unknown session ${sessionId}.`);
             }
@@ -216,7 +221,7 @@ export function buildHttpChannelRoutes(args) {
         handler: (_request, ctx) => __awaiter(this, void 0, void 0, function* () {
             var _a;
             const sessionId = (_a = ctx.params.sessionId) !== null && _a !== void 0 ? _a : "";
-            const record = yield engine.sessions.get(sessionId);
+            const record = yield engine.getOrRestoreSession(sessionId);
             if (record === undefined) {
                 return errorResponse(404, "session_not_found", `Unknown session ${sessionId}.`);
             }
@@ -241,7 +246,7 @@ export function buildHttpChannelRoutes(args) {
             var _b, _c;
             const sessionId = (_b = ctx.params.sessionId) !== null && _b !== void 0 ? _b : "";
             const callId = (_c = ctx.params.callId) !== null && _c !== void 0 ? _c : "";
-            const record = yield engine.sessions.get(sessionId);
+            const record = yield engine.getOrRestoreSession(sessionId);
             if (record === undefined) {
                 return errorResponse(404, "session_not_found", `Unknown session ${sessionId}.`);
             }
@@ -296,7 +301,7 @@ export function buildHttpChannelRoutes(args) {
         handler: (_request, ctx) => __awaiter(this, void 0, void 0, function* () {
             var _a;
             const sessionId = (_a = ctx.params.sessionId) !== null && _a !== void 0 ? _a : "";
-            const record = yield engine.sessions.get(sessionId);
+            const record = yield engine.getOrRestoreSession(sessionId);
             if (record === undefined) {
                 return errorResponse(404, "session_not_found", `Unknown session ${sessionId}.`);
             }
@@ -364,6 +369,68 @@ function readWorkspaceFiles(value) {
     }
     return Object.keys(files).length === 0 ? undefined : files;
 }
+function isRecord(value) {
+    return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+/** Best-effort parse of {@link AgentCloudOptions} from a JSON body field. */
+function readCloudOptions(value) {
+    if (!isRecord(value)) {
+        return undefined;
+    }
+    const cloud = {};
+    if (isRecord(value.env) && typeof value.env.type === "string") {
+        const name = typeof value.env.name === "string" ? value.env.name : undefined;
+        if (value.env.type === "cloud") {
+            cloud.env =
+                name === undefined ? { type: "cloud" } : { type: "cloud", name };
+        }
+        else if (value.env.type === "pool") {
+            cloud.env =
+                name === undefined ? { type: "pool" } : { type: "pool", name };
+        }
+        else if (value.env.type === "machine") {
+            cloud.env =
+                name === undefined ? { type: "machine" } : { type: "machine", name };
+        }
+    }
+    if (Array.isArray(value.repos)) {
+        const repos = [];
+        for (const entry of value.repos) {
+            if (!isRecord(entry) ||
+                typeof entry.url !== "string" ||
+                entry.url === "") {
+                continue;
+            }
+            repos.push(Object.assign(Object.assign({ url: entry.url }, (typeof entry.startingRef === "string"
+                ? { startingRef: entry.startingRef }
+                : {})), (typeof entry.prUrl === "string" ? { prUrl: entry.prUrl } : {})));
+        }
+        if (repos.length > 0) {
+            cloud.repos = repos;
+        }
+    }
+    if (typeof value.workOnCurrentBranch === "boolean") {
+        cloud.workOnCurrentBranch = value.workOnCurrentBranch;
+    }
+    if (typeof value.autoCreatePR === "boolean") {
+        cloud.autoCreatePR = value.autoCreatePR;
+    }
+    if (typeof value.skipReviewerRequest === "boolean") {
+        cloud.skipReviewerRequest = value.skipReviewerRequest;
+    }
+    if (isRecord(value.envVars)) {
+        const envVars = {};
+        for (const [key, envValue] of Object.entries(value.envVars)) {
+            if (typeof key === "string" && typeof envValue === "string") {
+                envVars[key] = envValue;
+            }
+        }
+        if (Object.keys(envVars).length > 0) {
+            cloud.envVars = envVars;
+        }
+    }
+    return Object.keys(cloud).length === 0 ? undefined : cloud;
+}
 export function readJsonObject(request) {
     return __awaiter(this, void 0, void 0, function* () {
         try {

package/dist/internal/map-pool.d.ts

@@ -1,6 +1,11 @@
 /**
  * Run `fn` over `items` with at most `concurrency` in flight; preserve order
  * of results (and of side-effect completion relative to index assignment).
+ *
+ * When `signal` aborts, workers stop claiming new items; in-flight `fn` calls
+ * are responsible for honoring the same signal.
  */
-export declare function mapPool<T, R>(items: readonly T[], concurrency: number, fn: (item: T, index: number) => Promise<R>): Promise<R[]>;
+export declare function mapPool<T, R>(items: readonly T[], concurrency: number, fn: (item: T, index: number) => Promise<R>, options?: {
+    signal?: AbortSignal;
+}): Promise<R[]>;
 //# sourceMappingURL=map-pool.d.ts.map

package/dist/internal/map-pool.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"map-pool.d.ts","sourceRoot":"","sources":["../../src/internal/map-pool.ts"],"names":[],"mappings":"AAAA;;;GAGG;AACH,wBAAsB,OAAO,CAAC,CAAC,EAAE,CAAC,EAChC,KAAK,EAAE,SAAS,CAAC,EAAE,EACnB,WAAW,EAAE,MAAM,EACnB,EAAE,EAAE,CAAC,IAAI,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,GACzC,OAAO,CAAC,CAAC,EAAE,CAAC,CAuBd"}
+{"version":3,"file":"map-pool.d.ts","sourceRoot":"","sources":["../../src/internal/map-pool.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,wBAAsB,OAAO,CAAC,CAAC,EAAE,CAAC,EAChC,KAAK,EAAE,SAAS,CAAC,EAAE,EACnB,WAAW,EAAE,MAAM,EACnB,EAAE,EAAE,CAAC,IAAI,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,EAC1C,OAAO,CAAC,EAAE;IAAE,MAAM,CAAC,EAAE,WAAW,CAAA;CAAE,GACjC,OAAO,CAAC,CAAC,EAAE,CAAC,CA0Bd"}

package/dist/internal/map-pool.js

@@ -10,14 +10,21 @@ var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, ge
 /**
  * Run `fn` over `items` with at most `concurrency` in flight; preserve order
  * of results (and of side-effect completion relative to index assignment).
+ *
+ * When `signal` aborts, workers stop claiming new items; in-flight `fn` calls
+ * are responsible for honoring the same signal.
  */
-export function mapPool(items, concurrency, fn) {
+export function mapPool(items, concurrency, fn, options) {
     return __awaiter(this, void 0, void 0, function* () {
         const results = new Array(items.length);
         let next = 0;
         function worker() {
             return __awaiter(this, void 0, void 0, function* () {
+                var _a;
                 for (;;) {
+                    if ((_a = options === null || options === void 0 ? void 0 : options.signal) === null || _a === void 0 ? void 0 : _a.aborted) {
+                        return;
+                    }
                     const index = next;
                     next += 1;
                     if (index >= items.length) {

package/dist/internal/mcp-host.d.ts

@@ -22,7 +22,20 @@ export declare class McpHost implements HostMcpRegistry {
     listTools(connectionName: string): Promise<HostMcpToolInfo[]>;
     callTool(connectionName: string, toolName: string, args?: Record<string, unknown>): Promise<HostMcpCallResult>;
     close(): Promise<void>;
+    /**
+     * Run against a cached host MCP client, reconnecting once when the remote
+     * Streamable HTTP session is gone (`session_terminated` / HTTP 404).
+     *
+     * Anytool (and the MCP spec) return 404 for unknown/terminated
+     * `Mcp-Session-Id`s so clients can re-initialize. The SDK transport keeps
+     * the dead session id, so without this invalidation every later host.mcp
+     * call fails until the serve process restarts.
+     */
+    private withClient;
     private getClient;
+    private invalidateClient;
     private connect;
 }
+/** Exported for unit tests. */
+export declare function isStaleMcpSessionError(error: unknown): boolean;
 //# sourceMappingURL=mcp-host.d.ts.map

package/dist/internal/mcp-host.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"mcp-host.d.ts","sourceRoot":"","sources":["../../src/internal/mcp-host.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAOH,OAAO,KAAK,EACV,oBAAoB,EACpB,iBAAiB,EACjB,eAAe,EACf,eAAe,EAEhB,MAAM,aAAa,CAAC;AASrB,qBAAa,OAAQ,YAAW,eAAe;IAC7C,OAAO,CAAC,QAAQ,CAAC,MAAM,CAA2C;IAClE,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAgD;IAExE,YAAY,WAAW,EAAE,oBAAoB,EAAE,EAI9C;IAED;;;OAGG;IACH,aAAa,CAAC,UAAU,EAAE,oBAAoB,GAAG,IAAI,CAEpD;IAED,KAAK,IAAI,MAAM,EAAE,CAEhB;IAEK,SAAS,CAAC,cAAc,EAAE,MAAM,GAAG,OAAO,CAAC,eAAe,EAAE,CAAC,CAYlE;IAEK,QAAQ,CACZ,cAAc,EAAE,MAAM,EACtB,QAAQ,EAAE,MAAM,EAChB,IAAI,GAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAM,GACjC,OAAO,CAAC,iBAAiB,CAAC,CAO5B;IAEK,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAU3B;IAED,OAAO,CAAC,SAAS;YAYH,OAAO;CAUtB"}
+{"version":3,"file":"mcp-host.d.ts","sourceRoot":"","sources":["../../src/internal/mcp-host.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAOH,OAAO,KAAK,EACV,oBAAoB,EACpB,iBAAiB,EACjB,eAAe,EACf,eAAe,EAEhB,MAAM,aAAa,CAAC;AASrB,qBAAa,OAAQ,YAAW,eAAe;IAC7C,OAAO,CAAC,QAAQ,CAAC,MAAM,CAA2C;IAClE,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAgD;IAExE,YAAY,WAAW,EAAE,oBAAoB,EAAE,EAI9C;IAED;;;OAGG;IACH,aAAa,CAAC,UAAU,EAAE,oBAAoB,GAAG,IAAI,CAEpD;IAED,KAAK,IAAI,MAAM,EAAE,CAEhB;IAEK,SAAS,CAAC,cAAc,EAAE,MAAM,GAAG,OAAO,CAAC,eAAe,EAAE,CAAC,CAalE;IAEK,QAAQ,CACZ,cAAc,EAAE,MAAM,EACtB,QAAQ,EAAE,MAAM,EAChB,IAAI,GAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAM,GACjC,OAAO,CAAC,iBAAiB,CAAC,CAQ5B;IAEK,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAU3B;IAED;;;;;;;;OAQG;YACW,UAAU;IAiBxB,OAAO,CAAC,SAAS;YAYH,gBAAgB;YAiBhB,OAAO;CAUtB;AAED,+BAA+B;AAC/B,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,CAU9D"}

package/dist/internal/mcp-host.js

@@ -44,23 +44,25 @@ export class McpHost {
     }
     listTools(connectionName) {
         return __awaiter(this, void 0, void 0, function* () {
-            const { client } = yield this.getClient(connectionName);
-            const listed = yield client.listTools();
-            return listed.tools.map((tool) => (Object.assign(Object.assign({ name: tool.name }, (tool.description === undefined
-                ? {}
-                : { description: tool.description })), (tool.inputSchema === undefined
-                ? {}
-                : { inputSchema: tool.inputSchema }))));
+            return this.withClient(connectionName, (client) => __awaiter(this, void 0, void 0, function* () {
+                const listed = yield client.listTools();
+                return listed.tools.map((tool) => (Object.assign(Object.assign({ name: tool.name }, (tool.description === undefined
+                    ? {}
+                    : { description: tool.description })), (tool.inputSchema === undefined
+                    ? {}
+                    : { inputSchema: tool.inputSchema }))));
+            }));
         });
     }
     callTool(connectionName_1, toolName_1) {
         return __awaiter(this, arguments, void 0, function* (connectionName, toolName, args = {}) {
-            const { client } = yield this.getClient(connectionName);
-            const result = yield client.callTool({
-                name: toolName,
-                arguments: args,
-            });
-            return normalizeCallResult(result);
+            return this.withClient(connectionName, (client) => __awaiter(this, void 0, void 0, function* () {
+                const result = yield client.callTool({
+                    name: toolName,
+                    arguments: args,
+                });
+                return normalizeCallResult(result);
+            }));
         });
     }
     close() {
@@ -74,6 +76,31 @@ export class McpHost {
             })));
         });
     }
+    /**
+     * Run against a cached host MCP client, reconnecting once when the remote
+     * Streamable HTTP session is gone (`session_terminated` / HTTP 404).
+     *
+     * Anytool (and the MCP spec) return 404 for unknown/terminated
+     * `Mcp-Session-Id`s so clients can re-initialize. The SDK transport keeps
+     * the dead session id, so without this invalidation every later host.mcp
+     * call fails until the serve process restarts.
+     */
+    withClient(name, run) {
+        return __awaiter(this, void 0, void 0, function* () {
+            const opened = yield this.getClient(name);
+            try {
+                return yield run(opened.client);
+            }
+            catch (error) {
+                if (!isStaleMcpSessionError(error)) {
+                    throw error;
+                }
+                yield this.invalidateClient(name, opened);
+                const reopened = yield this.getClient(name);
+                return yield run(reopened.client);
+            }
+        });
+    }
     getClient(name) {
         let pending = this.clients.get(name);
         if (pending === undefined) {
@@ -85,6 +112,22 @@ export class McpHost {
         }
         return pending;
     }
+    invalidateClient(name, opened) {
+        return __awaiter(this, void 0, void 0, function* () {
+            const pending = this.clients.get(name);
+            if (pending !== undefined) {
+                try {
+                    if ((yield pending) === opened) {
+                        this.clients.delete(name);
+                    }
+                }
+                catch (_a) {
+                    this.clients.delete(name);
+                }
+            }
+            yield opened.close();
+        });
+    }
     connect(name) {
         return __awaiter(this, void 0, void 0, function* () {
             const connection = this.byName.get(name);
@@ -96,6 +139,18 @@ export class McpHost {
         });
     }
 }
+/** Exported for unit tests. */
+export function isStaleMcpSessionError(error) {
+    if (!(error instanceof Error)) {
+        return false;
+    }
+    if (error.message.includes("session_terminated")) {
+        return true;
+    }
+    // StreamableHTTPError from the MCP SDK carries the HTTP status as `.code`.
+    const code = error.code;
+    return code === 404;
+}
 function openConnection(connection) {
     return __awaiter(this, void 0, void 0, function* () {
         const client = new Client({