@cursor/bdk 0.2.1

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

@@ -0,0 +1,104 @@
+/**
+ * HTTP client for playground eval batches (`/v1/dev/evals/runs`).
+ *
+ * Used by `agent-sdk eval --prod` (or `--url`) so the batch runs on the serve
+ * process and is mirrored to `defineStorage` when an `evals` table is
+ * configured. Without those flags, eval keeps the in-process ephemeral path
+ * in `cli-ax.ts`.
+ */
+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" | "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;
+    readonly activeRunId?: string | undefined;
+    constructor(message: string, status?: number | undefined, code?: RemoteEvalErrorCode | undefined, activeRunId?: string | undefined);
+}
+export interface RemoteEvalListItem {
+    id: string;
+    fileId: string;
+    description?: string;
+    tags?: string[];
+}
+export interface RunRemoteEvalsOptions extends RequestAuth {
+    baseUrl: string;
+    filterIds?: string[];
+    tags?: string[];
+    /** Per-case timeout override forwarded to the serve process. */
+    timeoutMs?: number;
+    /** Ask the serve harness to emit verbose case logs (server-side). */
+    verbose?: boolean;
+    fetchImpl?: typeof fetch;
+    /** Poll interval while the batch is `running`. */
+    pollMs?: number;
+    /** Abort the wait (not the remote batch). */
+    signal?: AbortSignal;
+    /** Called each time the polled snapshot changes. */
+    onSnapshot?: (snap: EvalRunSnapshot) => void;
+}
+/**
+ * Start a batch on the target serve via `POST /v1/dev/evals/runs`, poll until
+ * terminal, and return CLI-shaped case results.
+ */
+export declare function runRemoteEvals(options: RunRemoteEvalsOptions): Promise<{
+    snapshot: EvalRunSnapshot;
+    results: EvalRunResult[];
+}>;
+/** List datapoints from `GET /v1/dev/evals` (same surface a remote run uses). */
+export declare function listRemoteEvals(args: {
+    baseUrl: string;
+    auth?: RequestAuth;
+    fetchImpl?: typeof fetch;
+    filterIds?: string[];
+    tags?: string[];
+    signal?: AbortSignal;
+}): Promise<RemoteEvalListItem[]>;
+export declare function startRemoteEvalRun(args: {
+    baseUrl: string;
+    auth?: RequestAuth;
+    fetchImpl?: typeof fetch;
+    filterIds?: string[];
+    tags?: string[];
+    timeoutMs?: number;
+    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;
+    runId: string;
+    auth?: RequestAuth;
+    fetchImpl?: typeof fetch;
+    signal?: AbortSignal;
+}): Promise<EvalRunSnapshot>;
+/** Recent batches from `GET /v1/dev/evals/runs`. */
+export declare function listRemoteEvalRuns(args: {
+    baseUrl: string;
+    auth?: RequestAuth;
+    fetchImpl?: typeof fetch;
+    signal?: AbortSignal;
+}): Promise<{
+    runs: EvalRunSnapshot[];
+    activeRunId?: string;
+}>;
+export declare function pollRemoteEvalRun(args: {
+    baseUrl: string;
+    runId: string;
+    auth?: RequestAuth;
+    fetchImpl?: typeof fetch;
+    pollMs?: number;
+    signal?: AbortSignal;
+    onSnapshot?: (snap: EvalRunSnapshot) => void;
+}): Promise<EvalRunSnapshot>;
+export declare function snapshotToEvalRunResults(snapshot: EvalRunSnapshot): EvalRunResult[];
+//# sourceMappingURL=evals-client.d.ts.map

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

@@ -0,0 +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,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,CA2DhC;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,CA2E3B;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,CAuD3B;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,CA4B3B;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,CAuC7D;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,CAAC,QAAQ,EAAE,eAAe,GAAG,aAAa,EAAE,CAsBnF"}

package/dist/internal/evals-client.js

@@ -0,0 +1,303 @@
+/**
+ * HTTP client for playground eval batches (`/v1/dev/evals/runs`).
+ *
+ * Used by `agent-sdk eval --prod` (or `--url`) so the batch runs on the serve
+ * process and is mirrored to `defineStorage` when an `evals` table is
+ * configured. Without those flags, eval keeps the in-process ephemeral path
+ * in `cli-ax.ts`.
+ */
+var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, generator) {
+    function adopt(value) { return value instanceof P ? value : new P(function (resolve) { resolve(value); }); }
+    return new (P || (P = Promise))(function (resolve, reject) {
+        function fulfilled(value) { try { step(generator.next(value)); } catch (e) { reject(e); } }
+        function rejected(value) { try { step(generator["throw"](value)); } catch (e) { reject(e); } }
+        function step(result) { result.done ? resolve(result.value) : adopt(result.value).then(fulfilled, rejected); }
+        step((generator = generator.apply(thisArg, _arguments || [])).next());
+    });
+};
+import { expandSummarySuiteSelection, isEvalSummaryCase, matchesEvalIdFilter } from "../evals.js";
+import { mergeRequestHeaders } from "./request-headers.js";
+const DEFAULT_POLL_MS = 1000;
+export class RemoteEvalError extends Error {
+    constructor(message, status, code, activeRunId) {
+        super(message);
+        this.status = status;
+        this.code = code;
+        this.activeRunId = activeRunId;
+        this.name = "RemoteEvalError";
+    }
+}
+/**
+ * Start a batch on the target serve via `POST /v1/dev/evals/runs`, poll until
+ * terminal, and return CLI-shaped case results.
+ */
+export function runRemoteEvals(options) {
+    return __awaiter(this, void 0, void 0, function* () {
+        var _a;
+        var _b, _c;
+        const baseUrl = options.baseUrl.replace(/\/$/, "");
+        const fetchImpl = (_b = options.fetchImpl) !== null && _b !== void 0 ? _b : fetch;
+        const auth = {
+            bearerToken: options.bearerToken,
+            headers: options.headers,
+        };
+        const started = yield startRemoteEvalRun({
+            baseUrl,
+            auth,
+            fetchImpl,
+            filterIds: options.filterIds,
+            tags: options.tags,
+            timeoutMs: options.timeoutMs,
+            verbose: options.verbose,
+            signal: options.signal,
+        });
+        (_a = options.onSnapshot) === null || _a === void 0 ? void 0 : _a.call(options, started);
+        const snapshot = yield pollRemoteEvalRun({
+            baseUrl,
+            runId: started.runId,
+            auth,
+            fetchImpl,
+            pollMs: (_c = options.pollMs) !== null && _c !== void 0 ? _c : DEFAULT_POLL_MS,
+            signal: options.signal,
+            onSnapshot: options.onSnapshot,
+        });
+        return { snapshot, results: snapshotToEvalRunResults(snapshot) };
+    });
+}
+/** List datapoints from `GET /v1/dev/evals` (same surface a remote run uses). */
+export function listRemoteEvals(args) {
+    return __awaiter(this, void 0, void 0, function* () {
+        var _a, _b;
+        const baseUrl = args.baseUrl.replace(/\/$/, "");
+        const fetchImpl = (_a = args.fetchImpl) !== null && _a !== void 0 ? _a : fetch;
+        let resp;
+        try {
+            resp = yield fetchImpl(`${baseUrl}/v1/dev/evals`, {
+                headers: mergeRequestHeaders(args.auth),
+                signal: args.signal,
+            });
+        }
+        catch (error) {
+            throw new RemoteEvalError(error instanceof Error ? error.message : "Could not list evals.", undefined, "start_failed");
+        }
+        if (!resp.ok) {
+            throw new RemoteEvalError(`Could not list evals (HTTP ${resp.status}).`, resp.status, "start_failed");
+        }
+        const body = (yield resp.json());
+        const listed = [];
+        for (const row of (_b = body.evals) !== null && _b !== void 0 ? _b : []) {
+            if (typeof row.id !== "string" || row.id === "") {
+                continue;
+            }
+            if (typeof row.fileId !== "string" || row.fileId === "") {
+                continue;
+            }
+            const tags = Array.isArray(row.tags)
+                ? row.tags.filter((t) => typeof t === "string")
+                : undefined;
+            listed.push(Object.assign(Object.assign({ id: row.id, fileId: row.fileId }, (typeof row.description === "string" ? { description: row.description } : {})), (tags === undefined || tags.length === 0 ? {} : { tags })));
+        }
+        let selected = listed;
+        if (args.filterIds !== undefined && args.filterIds.length > 0) {
+            selected = selected.filter(e => args.filterIds.some(f => matchesEvalIdFilter(e.id, f)));
+        }
+        if (args.tags !== undefined && args.tags.length > 0) {
+            selected = selected.filter(e => { var _a; return ((_a = e.tags) !== null && _a !== void 0 ? _a : []).some(t => args.tags.includes(t)); });
+        }
+        // Match server-side filterDiscoveredEvals: summary → whole file.
+        return expandSummarySuiteSelection(listed, selected, e => isEvalSummaryCase(e.tags));
+    });
+}
+export function startRemoteEvalRun(args) {
+    return __awaiter(this, void 0, void 0, function* () {
+        var _a;
+        const baseUrl = args.baseUrl.replace(/\/$/, "");
+        const fetchImpl = (_a = args.fetchImpl) !== null && _a !== void 0 ? _a : fetch;
+        const body = {};
+        if (args.filterIds !== undefined && args.filterIds.length > 0) {
+            body.filterIds = args.filterIds;
+        }
+        if (args.tags !== undefined && args.tags.length > 0) {
+            body.tags = args.tags;
+        }
+        if (args.timeoutMs !== undefined) {
+            body.timeoutMs = args.timeoutMs;
+        }
+        if (args.verbose === true) {
+            body.verbose = true;
+        }
+        let resp;
+        try {
+            resp = yield fetchImpl(`${baseUrl}/v1/dev/evals/runs`, {
+                method: "POST",
+                headers: Object.assign({ "content-type": "application/json" }, mergeRequestHeaders(args.auth)),
+                body: JSON.stringify(body),
+                signal: args.signal,
+            });
+        }
+        catch (error) {
+            throw new RemoteEvalError(error instanceof Error ? error.message : "Could not start eval run.", undefined, "start_failed");
+        }
+        const json = (yield resp.json().catch(() => ({})));
+        if (json.error === "no_matching_evals") {
+            throw new RemoteEvalError("No matching evals found.", 404, "no_matching_evals");
+        }
+        if (resp.status === 409 || json.error === "eval_run_in_progress") {
+            throw new RemoteEvalError(json.activeRunId !== undefined && json.activeRunId !== ""
+                ? `An eval run is already in progress (${json.activeRunId}).`
+                : "An eval run is already in progress.", 409, "eval_run_in_progress", typeof json.activeRunId === "string" ? json.activeRunId : undefined);
+        }
+        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 start eval run: ${detail}`, resp.status, "start_failed");
+        }
+        if (typeof json.runId !== "string" || json.runId === "") {
+            throw new RemoteEvalError("Eval start response missing runId.", undefined, "start_failed");
+        }
+        return json;
+    });
+}
+/** Cancel a running batch via `POST /v1/dev/evals/runs/:runId/cancel`. */
+export function cancelRemoteEvalRun(args) {
+    return __awaiter(this, void 0, void 0, function* () {
+        var _a;
+        const baseUrl = args.baseUrl.replace(/\/$/, "");
+        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: ${detail}`, resp.status, "cancel_failed");
+        }
+        if (typeof json.runId !== "string" || json.runId === "") {
+            throw new RemoteEvalError("Eval cancel response missing runId.", undefined, "cancel_failed");
+        }
+        return json;
+    });
+}
+/** One-shot `GET /v1/dev/evals/runs/:runId` (no polling). */
+export function getRemoteEvalRun(args) {
+    return __awaiter(this, void 0, void 0, function* () {
+        var _a;
+        const baseUrl = args.baseUrl.replace(/\/$/, "");
+        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)}`, {
+                headers: mergeRequestHeaders(args.auth),
+                signal: args.signal,
+            });
+        }
+        catch (error) {
+            throw new RemoteEvalError(error instanceof Error ? error.message : "Could not fetch eval run.", undefined, "poll_failed");
+        }
+        if (resp.status === 404) {
+            throw new RemoteEvalError(`Unknown eval run ${args.runId}.`, 404, "unknown_eval_run");
+        }
+        if (!resp.ok) {
+            throw new RemoteEvalError(`Could not fetch eval run (HTTP ${resp.status}).`, resp.status, "poll_failed");
+        }
+        return (yield resp.json());
+    });
+}
+/** Recent batches from `GET /v1/dev/evals/runs`. */
+export function listRemoteEvalRuns(args) {
+    return __awaiter(this, void 0, void 0, function* () {
+        var _a;
+        const baseUrl = args.baseUrl.replace(/\/$/, "");
+        const fetchImpl = (_a = args.fetchImpl) !== null && _a !== void 0 ? _a : fetch;
+        let resp;
+        try {
+            resp = yield fetchImpl(`${baseUrl}/v1/dev/evals/runs`, {
+                headers: mergeRequestHeaders(args.auth),
+                signal: args.signal,
+            });
+        }
+        catch (error) {
+            throw new RemoteEvalError(error instanceof Error ? error.message : "Could not list eval runs.", undefined, "poll_failed");
+        }
+        if (!resp.ok) {
+            throw new RemoteEvalError(`Could not list eval runs (HTTP ${resp.status}).`, resp.status, "poll_failed");
+        }
+        const body = (yield resp.json());
+        const runs = Array.isArray(body.runs)
+            ? body.runs.filter((r) => r !== null && typeof r === "object" && typeof r.runId === "string")
+            : [];
+        return Object.assign({ runs }, (typeof body.activeRunId === "string" && body.activeRunId !== ""
+            ? { activeRunId: body.activeRunId }
+            : {}));
+    });
+}
+export function pollRemoteEvalRun(args) {
+    return __awaiter(this, void 0, void 0, function* () {
+        var _a, _b;
+        var _c;
+        const pollMs = (_c = args.pollMs) !== null && _c !== void 0 ? _c : DEFAULT_POLL_MS;
+        let lastJson = "";
+        for (;;) {
+            if (((_a = args.signal) === null || _a === void 0 ? void 0 : _a.aborted) === true) {
+                throw new RemoteEvalError("Eval wait aborted.", undefined, "aborted");
+            }
+            const snap = yield getRemoteEvalRun(args);
+            const serialized = JSON.stringify(snap);
+            if (serialized !== lastJson) {
+                lastJson = serialized;
+                (_b = args.onSnapshot) === null || _b === void 0 ? void 0 : _b.call(args, snap);
+            }
+            if (snap.status !== "running") {
+                return snap;
+            }
+            yield sleep(pollMs, args.signal);
+        }
+    });
+}
+export function snapshotToEvalRunResults(snapshot) {
+    return snapshot.cases.map(c => { var _a, _b, _c, _d, _e, _f; return (Object.assign(Object.assign(Object.assign(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, 
+        // Older serve processes send no verdict; infer it from `ok` so a remote
+        // batch still prints and reports consistently.
+        verdict: (_a = c.verdict) !== null && _a !== void 0 ? _a : (c.ok === true ? "passed" : "failed") }), (c.skipReason === undefined ? {} : { skipReason: c.skipReason })), (c.metadata === undefined ? {} : { metadata: c.metadata })), { assertions: (_b = c.assertions) !== null && _b !== void 0 ? _b : [] }), (c.sessionId === undefined ? {} : { sessionId: c.sessionId })), { inputs: (_c = c.inputs) !== null && _c !== void 0 ? _c : [] }), (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: (_d = c.logs) !== null && _d !== void 0 ? _d : [], metrics: (_e = c.metrics) !== null && _e !== void 0 ? _e : {}, durationMs: (_f = c.durationMs) !== null && _f !== void 0 ? _f : 0 })); });
+}
+function sleep(ms, signal) {
+    return new Promise((resolve, reject) => {
+        if ((signal === null || signal === void 0 ? void 0 : signal.aborted) === true) {
+            reject(new RemoteEvalError("Eval wait aborted.", undefined, "aborted"));
+            return;
+        }
+        const timer = setTimeout(() => {
+            signal === null || signal === void 0 ? void 0 : signal.removeEventListener("abort", onAbort);
+            resolve();
+        }, ms);
+        const onAbort = () => {
+            clearTimeout(timer);
+            reject(new RemoteEvalError("Eval wait aborted.", undefined, "aborted"));
+        };
+        signal === null || signal === void 0 ? void 0 : signal.addEventListener("abort", onAbort, { once: true });
+    });
+}

package/dist/internal/event-log.d.ts

@@ -0,0 +1,94 @@
+/**
+ * Durable, replayable per-session event log. `events.ndjson` is the source
+ * of truth. The process keeps `nextIndex`, live subscribers, and a short
+ * unflushed tail — not a cache of the full transcript. Replay reads the
+ * file on demand; live streams subscribe first, then catch up by index.
+ *
+ * Hosted engines mint thousands of sessions (one per PR under
+ * agentic-owners). Retaining every streamed tool output in RAM for the
+ * process lifetime is what OOM'd those engines at the ~4 GB heap cap.
+ */
+import type { SessionEvent, SessionEventPayload } from "../types.js";
+export type AppendableEvent = SessionEventPayload & {
+    turnId?: string;
+};
+export declare class SessionEventLog {
+    private readonly sessionId;
+    private readonly filePath;
+    private readonly onPersistError;
+    private nextIndex;
+    private readonly listeners;
+    private persistChain;
+    /** Appended events whose `appendFile` has not finished. */
+    private readonly inflight;
+    /**
+     * In-process patches (cloud cost backfill) that are not rewritten to
+     * `events.ndjson`. Applied on read.
+     */
+    private readonly replacements;
+    private constructor();
+    static open(args: {
+        sessionId: string;
+        filePath: string;
+        onPersistError?: (error: unknown) => void;
+    }): Promise<SessionEventLog>;
+    private readNextIndex;
+    get count(): number;
+    /**
+     * In-process residency vs recorded length. `eventsTotal` is the durable
+     * count (`nextIndex`). `eventsRetained` is the unflushed tail — the
+     * transcript is not cached after `appendFile` lands.
+     */
+    memoryStats(): {
+        eventsRetained: number;
+        eventsTotal: number;
+    };
+    /**
+     * Assign the next index and fan out to live subscribers. Ordering is
+     * synchronous; file persistence is chained in the background and awaited
+     * by {@link flush}. The event is not retained after the write lands.
+     */
+    append(payload: AppendableEvent): SessionEvent;
+    private dropInflight;
+    /** Wait for pending file writes (used on close). */
+    flush(): Promise<void>;
+    /**
+     * Read recorded events from `startIndex`. Does not retain the result on
+     * the log; callers own the returned array.
+     */
+    snapshot(startIndex?: number): Promise<SessionEvent[]>;
+    /**
+     * Replace one already-recorded event in this process. Used to backfill
+     * usage/cost onto historical `turn.completed` rows without rewriting
+     * `events.ndjson`. Live subscribers are not notified — patch before
+     * opening a replay stream.
+     */
+    replace(index: number, event: SessionEvent): boolean;
+    subscribe(listener: (event: SessionEvent) => void): () => void;
+    /**
+     * NDJSON stream: replays recorded events from `startIndex`, then follows
+     * live events until the consumer cancels.
+     *
+     * Subscribe first (buffer live events), then read the file, then drain
+     * the buffer by index so an async file read cannot drop or duplicate
+     * events that arrive during catch-up.
+     */
+    stream(startIndex?: number): ReadableStream<Uint8Array>;
+}
+export type EventLogMemoryStats = {
+    residentSessions: number;
+    eventsRetained: number;
+    eventsTotal: number;
+};
+/** Lazily opens one {@link SessionEventLog} per session. */
+export declare class EventLogStore {
+    private readonly eventFilePathFor;
+    private readonly onPersistError;
+    private readonly logs;
+    private readonly opened;
+    constructor(eventFilePathFor: (sessionId: string) => string, onPersistError?: (sessionId: string, error: unknown) => void);
+    get(sessionId: string): Promise<SessionEventLog>;
+    memoryStats(): EventLogMemoryStats;
+    flushAll(): Promise<void>;
+}
+//# sourceMappingURL=event-log.d.ts.map

package/dist/internal/event-log.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"event-log.d.ts","sourceRoot":"","sources":["../../src/internal/event-log.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAKH,OAAO,KAAK,EAAE,YAAY,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AAErE,MAAM,MAAM,eAAe,GAAG,mBAAmB,GAAG;IAAE,MAAM,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC;AAqDxE,qBAAa,eAAe;IAaxB,OAAO,CAAC,QAAQ,CAAC,SAAS;IAC1B,OAAO,CAAC,QAAQ,CAAC,QAAQ;IACzB,OAAO,CAAC,QAAQ,CAAC,cAAc;IAdjC,OAAO,CAAC,SAAS,CAAK;IACtB,OAAO,CAAC,QAAQ,CAAC,SAAS,CAA4C;IACtE,OAAO,CAAC,YAAY,CAAoC;IACxD,2DAA2D;IAC3D,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAsB;IAC/C;;;OAGG;IACH,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAmC;IAEhE,OAAO,eAIH;IAEJ,OAAa,IAAI,CAAC,IAAI,EAAE;QACtB,SAAS,EAAE,MAAM,CAAC;QAClB,QAAQ,EAAE,MAAM,CAAC;QACjB,cAAc,CAAC,EAAE,CAAC,KAAK,EAAE,OAAO,KAAK,IAAI,CAAC;KAC3C,GAAG,OAAO,CAAC,eAAe,CAAC,CAQ3B;YAEa,aAAa;IAU3B,IAAI,KAAK,IAAI,MAAM,CAElB;IAED;;;;OAIG;IACH,WAAW,IAAI;QAAE,cAAc,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAK7D;IAED;;;;OAIG;IACH,MAAM,CAAC,OAAO,EAAE,eAAe,GAAG,YAAY,CA2B7C;IAED,OAAO,CAAC,YAAY;IAWpB,oDAAoD;IAC9C,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAE3B;IAED;;;OAGG;IACG,QAAQ,CAAC,UAAU,SAAI,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC,CAkBtD;IAED;;;;;OAKG;IACH,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,YAAY,GAAG,OAAO,CAMnD;IAED,SAAS,CAAC,QAAQ,EAAE,CAAC,KAAK,EAAE,YAAY,KAAK,IAAI,GAAG,MAAM,IAAI,CAK7D;IAED;;;;;;;OAOG;IACH,MAAM,CAAC,UAAU,SAAI,GAAG,cAAc,CAAC,UAAU,CAAC,CAiDjD;CACF;AAED,MAAM,MAAM,mBAAmB,GAAG;IAChC,gBAAgB,EAAE,MAAM,CAAC;IACzB,cAAc,EAAE,MAAM,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;CACrB,CAAC;AAEF,4DAA4D;AAC5D,qBAAa,aAAa;IAKtB,OAAO,CAAC,QAAQ,CAAC,gBAAgB;IACjC,OAAO,CAAC,QAAQ,CAAC,cAAc;IALjC,OAAO,CAAC,QAAQ,CAAC,IAAI,CAA+C;IACpE,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAsC;IAE7D,YACmB,gBAAgB,EAAE,CAAC,SAAS,EAAE,MAAM,KAAK,MAAM,EAC/C,cAAc,GAAE,CAAC,SAAS,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,KAAK,IAAe,EACrF;IAEJ,GAAG,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAAC,eAAe,CAAC,CAmB/C;IAED,WAAW,IAAI,mBAAmB,CAajC;IAEK,QAAQ,IAAI,OAAO,CAAC,IAAI,CAAC,CAM9B;CACF"}